Demographic Methods Across The Tree of Life

Demographic Methods across the Tree of Life
Demographic
Methods across
the Tree of Life
EDITED BY
Roberto Salguero-Gómez
Associate Professor in Ecology, University of Oxford, UK
Tutorial Fellow in Ecology, Pembroke College, UK
Guest Researcher, Max Planck Institute for Demographic Research, Germany
Honorary Researcher, University of Queensland, Australia
Marlène Gamelon
Chargée de Recherche, Centre National de la Recherche Scientifique (CNRS), University
of Lyon, France
Researcher, Norwegian University of Science and Technology, Norway
3
3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Oxford University Press 2021
The moral rights of the authors have been asserted
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2021936265
ISBN 978-0-19-883860-9 (hbk.)
ISBN 978-0-19-883861-6 (pbk.)
DOI: 10.1093/oso/9780198838609.001.0001
Printed: Printed and bound by
CPI Group (UK) Ltd, Croydon, CR0 4YY
Cover image: d3sign/Getty Images (front main) and Cecilie Sonsteby/Getty Images (back)
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
Foreword
Demography is at the heart of biology. Populations These challenges should not deter you from study-
change in size over time due to temporal variation in ing demography, because the rewards from mastering
demographic rates, and spatial variation in population the subject are myriad. Demography has become pop-
dynamics is due to different demographic rates being ular in biology in recent years. The first reason for
observed across populations of the same species at dif- this popularity is in part because it is now widely
ferent locations. Community dynamics are determined accepted as a powerful way to study eco-evolution,
by the demographic rates of constituent species. Evo- evolution in structured populations, life history evo-
lution occurs when different genotypes have different lution, population dynamics, species interactions, and
demographies. A good biologist consequently needs to epidemiology. The reward from mastering the maths
have a firm grasp of demography. is worth it. The second reason is that the develop-
Despite the central role of demography in biology, ment of integral projection models (IPMs) that allow
it is not an easy subject to master for two reasons. the study of continuous phenotypic traits has solved
First, demographic data are neither cheap nor easy to a challenge that faced demography for many years—
collect. The most detailed data come from repeated how to study quantitative characters without arbitrari-
captures and observations of individuals that can be ly splitting them into wide, discrete bins. IPMs allow
individually identified. Not all species are amenable researchers to go seamlessly from the identification
to such study—try collecting such data on Caenorhabdi- of statistical patterns in data to demographic mod-
tis elegans or Anopheles gambiae in the wild—and those els. The third reason for the rise of demography is
species that are suitable for such study are often long- the technological advances that now allow increased
lived, surviving for many years. Studying evolution ease of data collection from wild populations. These
in such long-lived species is unfeasible, as it can take methods include miniaturised tags, ruggedised lower-
numerous consecutive research grants, or even careers, power wireless sensor networks, cheaper and more
to gain robust insight. Fortunately, there are approach- accurate satellite tags for animals, and even automated
es being developed to collect demographic data with- traps that target which individuals to live trap.
out marking individuals that can be used to construct I imagine that when I am asked to write the foreword
powerful demographic models. for the second edition of this book, I have no doubt that
The second reason that demography is a difficult the fourth reason I will list will be this tome. It is a very
subject is because it is mathematical. All populations well-structured book that covers not only data collec-
are structured, be it by genotypes, phenotypic traits, tion but also model construction and analysis, and the
strategies, age, sex, life history stage, or some oth- contributors are the very people who have advanced
er quantity. Modelling structured population usually the field of demography so substantially in the last
requires matrices, and the analysis of matrices requires, decade. The book succeeds, where so many fail, in not
at the bare minimum, an understanding of linear alge- only providing an accessible introduction to the field
bra. When models become frequency-dependent, and for those who have not been immersed in it for years,
include trait-mediated interactions, two sexes, or mul- but also containing some state-of-the-art methods and
tiple interacting species, analytical methods become applications. The editors, who are leaders of their gen-
even more challenging. eration in the field of demography, have put together a
v
vi FOREWORD
very thoughtful document: the book has something for good biologists will have this book, well-thumbed,
everyone, from the novice to the demographic wizard. centre-stage on their book case.
Part of demography is about prediction. As a demog-
Tim Coulson
rapher, it would consequently be remiss of me to fail
Professor of Zoology
to make a prediction. So here it is: within a year all
University of Oxford, UK
Acknowledgements
The community of individuals to whom we are grate- We warmly thank our past and present academ-
ful for their impact on this book is as heterogeneous as ic institutions: the University of Queensland (Aus-
the diversity of questions, methods, and applications tralia) and the Universities of Sheffield and Oxford
of demography itself. (UK) for Roberto, and the Centre d’Ecologie Fonc-
First and foremost, this attempt to bring together tionnelle et Evolutive in Montpellier, the Laboratoire
demographic methods across the Tree of Life would de Biométrie et Biologie Evolutive in Lyon (France),
have been impossible without the excellent contribu- and the CBD in Trondheim (Norway) for Marlène.
tions of all 77 authors. They brought a depth, breadth, All these academic institutions, although in different
and clarity in explanations that we feel confident will countries, with different cultures and languages (and
be extremely useful to the demographic communi- versions of English!), share common denominators:
ty and beyond. The diligence and resilience of these stimulating scientific environments, and open discus-
authors—even through a pandemic—kept us focused sions in ecology, evolution, and conservation biology,
on the delivery of their individual contributions and from which we greatly benefitted while editing and
the book. Likewise, we are indebted to the 44 anony- writing this book. In this regard, we are extremely
mous reviewers who provided excellent, timely, and thankful for the intellectual, logistical, and emotion-
on-time suggestions to improve the content and stylis- al support provided by Aziz Aboobaker, Mike Bon-
tic presentation of each chapter. sall, Stephanie Borrelle, Natalie Briscoe, Pol Capdev-
The inception of this book started almost as a mat- ila, Dylan Childs, Tim Coulson, Shaun Coutts, Katrina
ter of unconscious intellectual development during our Davis, John Dwyer, Jane Elith, Rob Freckleton, Ivar
respective postdoctoral times at the Max Planck Insti- Herfindal, Matt Holden, Irja Ida Ratikainen, Aline M.
tute for Demographic Research (Germany) for Rober- Lee, E.J. Milner-Gulland, Maria Paniw, Henrik Pärn,
to Salguero-Gómez and at the Centre for Biodiver- Christophe Pélabon, Hugh Possingham, Mark Rees,
sity Dynamics (CBD), Norwegian University of Sci- Ben C. Sheldon, Jarle Tufto, and Pete Vesk.
ence and Technology (NTNU) (Norway) for Marlène The maturation of the corpus of this book was car-
Gamelon. The exposure to demographers working on ried over numerous internet conference calls in dif-
a wide range of systems and with an even broader ferent time zones. However, the whole thing orig-
repertoire of methodologies and questions encouraged inated from discussions between us and other col-
us to test whether those methods would be appli- leagues at multiple international meetings. It is worth
cable within and beyond our specific study systems. here acknowledging the most influential venue: the
We owe a great deal of inspiration to Ruth Archer, Evolutionary Demography Society (EvoDemoS) annu-
Oskar Burger, Brenda Casper, Hal Caswell, Fernando al meetings. The intellectual exchange supported by
Colchero, Dalia Conde, David N. Koons, Steve Ellner, those meetings allowed us to fine-tune the content
Steinar Engen, Jean-Michel Gaillard, Olivier Gimenez, of the book as well as to recruit invaluable contribu-
Vidar Grøtan, Owen Jones, Eelke Jongejans, Hans de tions by excellent junior and established researchers in
Kroon, Jean-Dominique Lebreton, Cory Merow, Sean the field of demography. Interactions with EvoDemoS
McMahon, Jessica Metcalf, Jim Oeppen, Peter Petraitis, members, as well as members from the British Ecolog-
Brett K. Sandercock, Ralph Schaible, Alex Scheuer- ical Society and the Ecological Society of America have
lein, Iain Stott, Bernt-Erik Sæther, James Vaupel, and further allowed us to tailor this book for our target
Cyrille Violle for inspiring us to think beyond our audience.
particular one-species, one-population, one-method, We also acknowledge the financial support
one-question origins. provided by various funding agencies and schemes.
vii
viii ACKNOWLEDGEMENTS
Roberto Salguero-Gómez was supported by an Aus- This work has greatly benefitted from the time-
tralian Research Council DECRA fellowship while at ly, supportive, and encouraging advice and guidance
the University of Queensland, and a NERC Indepen- from the OUP team. Special thanks go to Ian Sherman,
dent Research Fellowship while at the University of Charles Bath, and Bethany Kershaw.
Sheffield and University of Oxford. Further grants Finally, we would like to thank the loved ones who
to him and his research group by the ARC, NERC, have provided their input to this book while being
BBSRC, Royal Society, Oxford–Africa Funds, Oxford very patient for the time that this book took away
University Press John Fell Funds, and the Spanish from them. Roberto Salguero-Gómez thanks Katrina
Ministerio de Economía y Competitividad allowed Davis and Manon Salguero-Davis for their uncondi-
for interactions with key collaborators, as well as tional support, love, critical input, and patience. Mar-
developing and applying demographic methods to a lène Gamelon dedicates this book to her family, who
diversity of species. Marlène Gamelon was supported have always been supportive along the academic sci-
by a grant of the French National Agency for Wildlife entific path.
(formerly ONCFS - now OFB), the Research Council
Roberto Salguero-Gómez
of Norway through its Centres of Excellence fund-
Marlène Gamelon
ing scheme to CBD, and the Centre National de la
Recherche Scientifique (CNRS).
Contents
Foreword v
Acknowledgements vii
List of Contributors xix
Introduction: From lions, to lion’s manes, to dandelions: how different

types of demographic data and methods are selected and why 1
Roberto Salguero-Gómez and Marlène Gamelon
1 On the origins of this book 1
2 Origins of demographic analyses: from humans to nonhuman
animals, plants, fungi, and microbes 2
3 Why use demographic methods across the Tree of Life? 2
4 From data acquisition to demographic analyses 3
5 Organisation of the book 6
6 Target audience 11
7 What next? A plea for a stronger, richer, broader, and more
innovative demography 12
References 12
Part 1 Demographic Data Collection: From Genes to Environment 15
1 Genetic data collection, pedigrees, and phylogenies 17

Emily G. Simmonds, Alina K. Niskanen, Henrik Jensen, and Steven Smith
1.1 Introduction 17
1.2 Genetic pedigrees 19
1.3 Genetic phylogenies 19
1.4 Collection of genetic samples 21
1.5 Processing of samples (DNA extraction, amplification, and
sequencing) 22
1.6 Preparation of data for analyses 23
1.7 Using the data to construct pedigrees and phylogenies 24
1.7.1 Pedigrees 24
1.7.2 Phylogenies 25
1.8 Conclusions 29
References 29
ix
x CONTENTS
2 Biochemical and physiological data collection 35

Oldřich Tomášek, Alan A. Cohen, Erola Fenollosa, Maurizio Mencuccini, Sergi
Munné-Bosch, and Fanie Pelletier
2.1 Introduction 35
2.2 General criteria for selecting biomarkers in demographic studies 36
2.3 Designing the data collection 37
2.4 Physiological traits 37
2.4.1 Animal metabolism 37
2.4.2 Plant metabolism 39
2.4.3 Semen physiology 39
2.5 Biochemical traits 40
2.5.1 General biochemistry 40
2.5.2 Hormones 41
2.5.3 Oxidative stress 42
2.6 Telomeres 43
2.7 Immune markers 44
2.8 Epigenetic and omics data 45
2.9 Biologging 45
2.10 Combining biomarkers 46
2.11 Conclusions and perspectives 46
Acknowledgements 46
References 46
3 Social data collection and analyses 53

Marie J. E. Charpentier, Marie Pelé, Julien P. Renoult, and Cédric Sueur
3.1 Introduction 53
3.2 Classic field-based methods for sampling social behaviours 54
3.2.1 A brief history of social studies, Jeanne Altmann, and the
sampling of social behaviour 54
3.2.2 The interplay between field and lab studies 55
3.3 New technologies to sample social behaviour in the wild 55
3.3.1 A brief overview of animal’s biologging 55
3.3.2 Biologging to determine individual social interactions and
spatial associations 57
3.4 Analysing social behaviour 64
3.4.1 From data collection to social network analyses 64
3.4.2 Individual level: the roles of nodes 66
3.4.3 Intermediary levels 67
3.4.4 Group level and group comparison 67
3.4.5 Network dynamics 69
3.5 Conclusions 70
References 70
4 Growth rings across the Tree of Life: demographic insights from biogenic
time series data 77
Margaret E. K. Evans, Bryan A. Black, Donald A. Falk, Courtney L. Giebink, and Emily L.
Schultz
4.1 Introduction 77
4.2 Tree and otolith growth rings—the basics 78
CONTENTS xi
4.3 Growth rings in other organisms 81

4.4 Beyond growth rings and ring widths 82
4.5 Insights into demography 82
4.6 Ecosystem and macrosystem demography 86
4.7 A research agenda 87
4.8 Conclusions 89
References 90
5 Longitudinal demographic data collection 97

Marlène Gamelon, Josh A. Firth, Mathilde Le Moullec, William K. Petry,
and Roberto Salguero-Gómez
5.1 Introduction: long-term field studies 97
5.2 Collection of longitudinal demographic data on mobile species 98
5.2.1 Procedures and type of data collected at the population level 98
5.2.2 Individual-level long-term field studies: concepts, methods,
and data 101
5.3 Collection of longitudinal demographic data on sessile species 103
5.3.1 Procedures and type of data collected at the population level 104
5.3.2 Procedures and type of data collected at the individual level 104
5.4 The future of long-term studies: new aspects, current biases, and
arising challenges 107
Acknowledgements 108
References 108
6 Drivers of demography: past challenges and a promise for a changed

future 115
Pedro F. Quintana-Ascencio, Eric S. Menges, Geoffrey S. Cook, Johan Ehrlén, and Michelle
E. Afkhami
6.1 Population dynamics and biotic and abiotic drivers 115
6.2 Challenges with incorporating drivers into structured population
models 115
6.3 Quantifying drivers by observation and by experiments 117
6.4 Continuous drivers (‘stress’) 119
6.5 Episodic drivers (‘disturbance’) 120
6.6 Population dynamics in the landscape: influence of spatial habitat
structure and heterogeneity 122
6.7 New and promising approaches 124
References 125
Part 2 Data- and Research Question-Driven Methods 131
7 Abundance-based approaches 133

Jonas Knape and Andreas Lindén
7.1 Aims of abundance analyses 133
7.2 Measures of abundance 134
7.2.1 Sampling abundance 134
xii CONTENTS
7.3 Analytic approaches 135

7.3.1 Regression-type models 135
7.3.2 Dynamic abundance models 136
7.3.3 State-space models 143
7.4 Conclusions 145
References 147
8 Life tables: construction and interpretation 151

Owen R. Jones
8.1 Introduction 151
8.2 Cohort and period life tables 152
8.2.1 Cohort life tables 153
8.2.2 Period life tables 154
8.2.3 Fertility 156
8.3 The richness of life table information 157
8.4 Complications, correlations, and limitations 157
8.5 Conclusions 158
Data and code accessibility 158
References 159
9 Introduction to matrix population models 163

Yngvild Vindenes, Christie Le Coeur, and Hal Caswell
9.2 An age-structured MPM 164
9.2.1 Life table for age structured population 164
9.2.2 Projecting growth of age structured population 165
9.2.3 Defining Leslie matrix elements for pre- and
post-reproductive censuses 165
9.3 Stage-structured MPMs 166
9.4 Basic analytical results from MPMs 167
9.4.1 Asymptotic properties from the projection matrix 167
9.4.2 Sensitivity analysis of population growth rate 168
9.4.3 Life history characteristics 169
9.5 Case studies with different demographic structures 169
9.5.1 Case study 1: Age structure in Asian elephants 169
9.5.2 Case study 2: Life cycle stages in an isopod 171
9.5.3 Case study 3: Age and habitat structure in black-headed
gulls 173
9.6 Conclusions 175
References 177
10 Integral projection models 181

Edgar J. González, Dylan Z. Childs, Pedro F. Quintana-Ascencio,
10.2 Constructing a basic IPM 181
10.3 Modelling vital rates in terms of the state variable 185
10.3.1 Linear models 185
10.3.2 Generalised linear models 185
CONTENTS xiii
10.4 Combining the vital functions in the IPM 186

10.5 Perturbation analysis: sensitivity and elasticity 187
10.5.1 Perturbation on the kernel 187
10.5.2 Perturbation on the pre-existing and newborn sections of
the kernel 187
10.5.3 Sensitivities on the vital rates 188
10.5.4 Sensitivities to vital rate parameters 188
10.6 Further considerations in the construction of vital-rate models 189
10.6.1 Extensions of the linear model 189
10.6.2 Model selection 189
10.7 Case studies 191
10.7.1 Case study 1: Soay sheep population structured by
weight 191
10.7.2 Case study 2: Perennial herb population structured by
height, elevation, and time since fire 191
References 193
11 Transient analyses of population dynamics using matrix projection

models 197
David N. Koons, David T. Iles, and Iain Stott
11.2 Examples 198
11.2.1 Example life cycle models 198
11.2.2 Transient analysis in constant environments 200
11.2.3 Transient perturbation analysis in constant environments 203
11.2.4 Population dynamics in time-varying environments 204
11.2.5 Transient perturbation analysis in time-varying
environments 207
11.3 Conclusions 209
References 210
12 Individual-based models 213

Viktoriia Radchuk, Stephanie Kramer-Schadt, Uta Berger, Cédric Scherer, Pia Backmann,
and Volker Grimm
12.2 What are individual-based models and how do we document
them? 213
12.3 Why individual-based models? 214
12.3.1 Emergent behaviour vs. imposed demographic rates 214
12.3.2 Local interactions and nonlinear feedbacks 215
12.4 The pros and cons of IBMs 216
12.5 Types of data used to build an IBM 217
12.6.1 Vole–mustelid model 218
12.6.2 Wild boar–classical swine fever (CSF) virus model 220
12.6.3 Wild tobacco–insect model (TIMELY) 221
12.8 Note on Supplementary Comparison Supplementary list 225
References 226
xiv CONTENTS
13 Survival analyses 229

Sarah Cubaynes, Simon Galas, Myriam Richaud, Ana Sanz Aguilar, Roger Pradel,
Giacomo Tavecchia, Fernando Colchero, Sebastien Roques, Richard Shefferson, and
Carlo Giovanni Camarda
13.2 What is survival analysis and why do we need it? 229
13.2.1 Time matters 229
13.2.2 Censoring and missing data 230
13.2.3 Known fate data versus imperfect detection 230
13.3 Survival analyses in the lab: when it’s (almost) all under
control 230
13.3.1 Monitoring survival in the lab 230
13.3.2 Kaplan–Meier estimator and log-rank tests 230
13.3.3 Cox’s proportional hazards to adjust for covariates 231
13.3.4 Example of a stress assay in an unconventional resistant
organism: the tardigrade 231
13.4 Survival analyses in the wild: dealing with uncertainty and
variable environments 232
13.4.1 Overview of CMR models 232
13.4.2 Environmental variability and individual heterogeneity in
CMR models 234
13.4.3 Inference framework 234
13.4.5 Case studies in animal demography 235
13.4.6 Case studies in plant demography 236
13.5 Mortality analysis in human populations 237
13.5.1 Human data and assumptions 237
13.5.2 Parametric models over age 238
13.5.3 Overparametrised models: the example of the Lee–Carter
model 239
13.5.4 Semiparametric models: the example of P-splines 239
13.5.5 Implementation example 239
References 240
14 Efficient use of demographic data: integrated population

models 245
Marlène Gamelon, Stefan J. G. Vriend, Marcel E. Visser, Caspar A. Hallmann,
Suzanne T. E. Lommen, and Eelke Jongejans
14.1 Introduction: why use integrated population models? 245
14.2 How to build an IPM? 246
14.2.1 Population model 246
14.2.2 Individual likelihoods 246
14.2.3 Joint likelihood 246
14.3.1 Blue tit 247
14.3.2 Common ragweed 251
References 254
CONTENTS xv
Part 3 Applications 257
15 Spatial demography 259

Guillaume Péron
15.2 Spatial variation in demographic rates 259
15.2.1 Statistical tools for nonparametric inference about spatial
variation 259
15.2.2 Recent implementations in spatial demography 262
15.2.3 Discussion 263
15.3 Estimating dispersal fluxes 263
15.3.1 Integrated population models 264
15.3.2 Multisite capture–recapture models and integrated
metapopulation models 264
15.3.3 Dispersal as a diffusion process along resistance surfaces 265
15.4 The feedback between space use and fitness 266
15.4.1 Principles of integral projection models as pertaining to
spatial demography 266
15.4.2 Space use or resource selection? 266
15.4.3 Formal analysis of IPMs 267
15.4.4 Individual-based simulation approach 268
References 269
16 Evolutionary Demography 273

Shripad Tuljapurkar and Wenyun Zuo
16.2 The early theory 273
16.3 Randomly varying environments 276
16.4 Post-reproductive life 279
16.5 Nonlinear models for mutation–selection 280
16.6 Density-dependent models 280
References 282
17 Reproductive value and analyses of population dynamics of

age-structured populations 285
Bernt-Erik Sæther and Steinar Engen
17.2 Stochastic theory 285
17.3 The concept of individual reproductive value 287
17.4 Temporal autocovariance in the environmental noise 289
17.5.1 Moose on the island of Vega 289
17.5.2 A practical application: estimating risk of extinction in
harvested Scandinavian lynx populations 290
xvi CONTENTS

References 294
18 Applying comparative methods to different databases: lessons from

demographic analyses across mammal species 299
Jean-Michel Gaillard, Victor Ronget, Jean-François Lemaître, Christophe Bonenfant,
Guillaume Péron, Pol Capdevila, Marlène Gamelon, and Roberto Salguero-Gómez
18.2 Methods 301
18.2.1 Building databases and calculating demographic metrics 301
18.2.2 Statistical analyses accounting for phylogenetic inertia 302
18.3 Results 303
18.3.1 Allometric analyses 303
18.3.2 Assessing axes of variation among dimensionless
demographic metrics 305
18.5 Prespectives 310
References 310
19 Adaptive management: making recurrent decisions in the face of

uncertainty 313
James D. Nichols
19.2 Structured decision-making 314
19.3 Adaptive management (AM) 314
19.3.1 Deliberative phase 315
19.3.2 Iterative phase 316
19.3.3 Double-loop learning phase 318
19.4 Demographic methods 318
19.4.1 Population size/abundance 318
19.4.2 Occupancy 319
19.4.3 Species richness 319
19.4.4 Landcover vegetative states 319
19.5.1 Mallard (Anas platyrhynchos) duck harvest in North America 320
19.5.2 Management of vegetation states for Florida scrub jays
(Aphelocoma coerulescens) 322
19.5.3 Disease control via AM 324
References 326
20 Heritability, polymorphism, and population dynamics: individual-based

eco-evolutionary simulations 329
Anna Kuparinen
20.2 Heritability and selection in the context of population dynamics 329
CONTENTS xvii
20.3 Generic ways to model the evolution of demographic rates 330

20.3.1 Parental trait values coupled with known heritability 330
20.3.2 Dummy genes coding the traits in a classic Mendelian
manner 330
20.4.1 Boreal forest adaptation to warming climate 332
20.4.2 Co-evolution of correlated traits in Atlantic cod 333
20.4.3 Evolution of a sexually polymorphic age at maturity in
Atlantic salmon 336
20.5 Conclusions and future directions 337
References 339
21 Demographic processes in socially structured populations 341

Maria Paniw, Gabriele Cozzi, Stefan Sommer, and Arpat Ozgul
21.2 A simulated, socially structured population 343
21.3 A socially structured matrix population model 344
21.4 A socially structured integral projection model 345
21.5 Explicitly accounting for group sizes in a population model 345
21.6 An individual-based model as a mechanistic approach to
modelling sociality 345
21.7 Does increased model complexity improve inferences of
population dynamics in a socially structured population? 346
21.8 Implications for life history models of socially structured
populations 346
References 349
22 Demographic methods in epidemiology 351

Petra Klepac and C. Jessica E. Metcalf
22.1 Why demography is important to epidemic dynamics 351
22.2 Modelling the intersection between epidemiology and
demography 352
22.3 Demographic underpinnings of a core parameter, R0 355
22.4 A discrete-time multistate matrix approach 355
22.5 A specific example 356
22.6 Case study 358
References 360
Index 363
List of Contributors
Michelle E. Afkhami Assistant Professor, University Erola Fenollosa Post-Doctoral Researcher, Uni-
of Miami, USA versity of Barcelona; Post-Doctoral Researcher,
Pia Backmann Post-Doctoral Researcher, Technische IRBio (Institut de Recerca de la Biodiversitat),
Universität Dresden, Germany Spain
Uta Berger Professor, Technische Universität Josh A. Firth Researcher, University of Oxford, UK
Dresden, Germany Jean-Michel Gaillard Directeur de Recherche, CNRS,
Bryan A. Black Associate Professor, University of University of Lyon, France
Arizona, USA Simon Galas Professor, University of Montpellier,
Christophe Bonenfant Chargé de Recherche, CNRS, France
University of Lyon, France Marlène Gamelon Chargée de Recherche, CNRS, Uni-
Carlo Giovanni Camarda Researcher, Institut National versity of Lyon, France; Researcher, Norwegian
d′Études Démographiques, France University of Science and Technology, Norway
Pol Capdevila Research Associate, University of Courtney L. Giebink Graduate Fellow, University of
Bristol, UK Arizona, USA
Hal Caswell Professor, University of Amsterdam, The Edgar J. González Associate Professor, Universidad
Netherlands Nacional Autónoma de México, Mexico.
Marie J. E. Charpentier Directrice de Recherche, Volker Grimm Professor, University of Potsdam;
CNRS, University of Montpellier, France Senior Researcher, Helmholtz Centre for Environ-
Dylan Z. Childs Senior Lecturer (Associate Professor), mental Research–UFZ, Germany
University of Sheffield, UK Caspar A. Hallmann Post-Doctoral Researcher,
Alan A. Cohen Associate Professor, University of Radboud University Nijmegen, The Netherlands
Sherbrooke, Canada David T. Iles Quantitative Wildlife Biologist, Cana-
Fernando Colchero Associate Professor, University of dian Wildlife Service, Environment and Climate
Southern Denmark, Denmark Change Canada, Canada
Geoffrey S. Cook Assistant Professor, University of Henrik Jensen Professor, Norwegian University of
Central Florida, USA Science and Technology, Norway
Gabriele Cozzi Post-Doctoral Researcher in Move- Owen R. Jones Associate Professor, University of
ment Ecology, University of Zurich, Switzerland Southern Denmark, Denmark
Sarah Cubaynes Maître de Conférences, University of Eelke Jongejans Assistant Professor, Radboud
Montpellier, France University Nijmegen, The Netherlands
Johan Ehrlén Professor, Stockholm University, Petra Klepac Assistant Professor, London School of
Sweden Hygiene and Tropical Medicine, UK
Steinar Engen Professor, Norwegian University of Jonas Knape Associate Professor, Swedish University
Science and Technology, Norway of Agricultural Sciences, Sweden
Margaret E. K. Evans Assistant Professor, University Stephanie Kramer-Schadt Professor, Technische Uni-
of Arizona, USA versität Berlin; Head of Department of Ecological
Dynamics, Leibniz Institute for Zoo and Wildlife
Donald A. Falk Professor, University of Arizona, USA
Research, Germany
xix
xx L I S T O F C O N T R I B U TO R S
David N. Koons Professor, James C. Kennedy Roger Pradel Directeur de Recherche, CNRS,
Endowed Chair of Wetland and Waterfowl University of Montpellier, France
Conservation, Colorado State University, Pedro F. Quintana-Ascencio Professor, Department of
USA Biology, University of Central Florida, USA
Anna Kuparinen Professor, University of Jyväskylä, Viktoriia Radchuk Senior Researcher, Leibniz
Finland Institute for Zoo and Wildlife Research, Austria
Christie Le Coeur Post-Doctoral Researcher, Julien P. Renoult Chargé de Recherche, CNRS,
University of Oslo, Norway University of Montpellier, France
Jean-François Lemaître Chargé de Recherche, CNRS, Myriam Richaud PhD, University of Montpellier,
University of Lyon, France France
Mathilde Le Moullec Post-Doctoral Researcher, Victor Ronget Post-Doctoral Researcher, University
Norwegian University of Science and Technology, of Paris, France
Norway Sébastien Roques PhD, University of Montpellier,
Andreas Lindén Senior Scientist, Wildlife Ecolo- France
gy, Natural Resources Institute Finland (Luke), Roberto Salguero-Gómez Associate Professor,
Finland University of Oxford, UK; Tutorial Fellow, Pem-
Suzanne T. E. Lommen Post-Doctoral Researcher, broke College, UK; Guest Researcher, Max Planck
Leiden University and University of Fribourg, Institute for Demographic Research, Germany;
Germany Honorary Researcher, University of Queensland,
Maurizio Mencuccini Professor, ICREA (Catalan Australia
Institute for Research and Advances Stud- Bernt-Erik Sæther Director/Professor, Norwegian
ies), Barcelona; Professor, CREAF (Research University of Science and Technology, Norway
Center in Ecology and Forestry Applications), Ana Sanz Aguilar Researcher, Ramon y Cajal
Spain Researcher, IMEDEA, Spain
Eric S Menges Program Director and Senior Cédric Scherer Post-Doctoral Researcher, Leibniz
Researcher, Archbold Biological Station, USA Institute for Zoo and Wildlife Research, Austria
C. Jessica E. Metcalf Associate Professor, Evolu- Emily L. Schultz Post-Doctoral Researcher, University
tionary Biology and Public Affairs, Princeton of Arizona, USA
University, USA
Richard Shefferson Associate Professor, University of
Sergi Munné-Bosch Professor, University of Tokyo, Japan
Barcelona, Spain
Emily G. Simmonds Post-Doctoral Researcher, Nor-
James D. Nichols Adjunct Professor, University of wegian University of Science and Technology,
Florida, USA Norway
Alina K. Niskanen Post-Doctoral Researcher, Univer- Steven Smith Head of Ecological Genomics, Univer-
sity of Oulu, Finland sity of Veterinary, Austria
Arpat Ozgul Associate Professor, University of Stefan Sommer Research Assistant, University of
Zurich, Switzerland Zurich, Switzerland
Maria Paniw Post-Doctoral Researcher, University of Iain Stott Senior Lecturer, University of Lincoln, UK
Zurich, Switzerland
Cédric Sueur Maître de Conférences, University of
Marie Pelé Chargée de Recherche, Lille Catholic Strasbourg, and Institut Universitaire de France,
University, France France
Fanie Pelletier Professor, University of Sherbrooke, Giacomo Tavecchia Researcher, IMEDEA, Spain
Canada
Oldřich Tomášek Researcher, Institute of Vertebrate
Guillaume Péron Chargé de Recherche, CNRS, Biology of the Czech Academy of Sciences,
University of Lyon, France Czechia; Researcher, Masaryk University, Czechia
William K. Petry Assistant Professor, North Carolina Shripad Tuljapurkar Professor, Stanford University,
State University, USA USA
L I S T O F C O N T R I B U TO R S xxi
Yngvild Vindenes Associate Professor, University of Stefan J. G. Vriend PhD, Norwegian University of
Oslo, Norway Science and Technology, Norway
Marcel E. Visser Professor, Groningen University and Wenyun Zuo Researcher, Stanford University,
Wageningen University, The Netherlands USA
INTRODUCTION
From lions, to lion’s manes,

to dandelions: how different types
of demographic data and methods
are selected and why
Roberto Salguero-Gómez and Marlène Gamelon
1 On the origins of this book (Latin ‘vita’), or vital rates of survival, development,
reproduction, and recruitment, are common across
The three pillars upon which this book has been
all species, from bacteria to humans. Thus, simi-
engineered are the diversity of demographic schedules
lar demographic methods can accommodate all the
found in nature, coupled with the common currencies
species across the Tree of Life to tackle important
of demography and the myriad of existing quantita-
questions regarding the universality of rules in ecol-
tive methods in demography. These pillars are present
ogy and evolution (Sutherland et al. 2013).
along the book and its 22 chapters in the following
3. High diversity in quantitative methods: demog-
aspects:
raphy, through its very origin (see Part II), is well
equipped to address all of the aforementioned ques-
1. High diversity in living forms: the vast amount tions and more (Metcalf and Pavard 2007), as it is
of diversity across the Tree of Life is expressed at strongly cemented in statistics and actuarial science.
all levels of biological organisation, from genes to Demographic methods—as the reader will have an
ecosystems. Individuals within a population, popu- opportunity to learn from this book—allow one to
lations within species, and species within a commu- accommodate a high variation in data quality and
nity differ a great deal in terms of variation in their sources, to address fundamental questions in ecol-
demographic schedules (i.e. rates of survival and ogy, evolution, and conservation/management,
reproduction through time). For instance, longevi- for all species across the Tree of Life. From that
ty varies by orders of magnitude, from days as angle, we introduce here a book that goes beyond
in the case of some mayfly species (e.g. Ephemera taxonomic boundaries. The data and applications of
simulans, Carey 2002) to millennia as in bristlecone most chapters include trans-kingdom comparisons
pines (Pinus longevata, Lanner and Connor 2001). that highlight the universality of said common
But why is there variation in demographic schedules currencies.
across individuals, populations, and species? What
are the implications of this demographic variation in We realised that there was an empty niche in try-
ecology, evolution, and conservation/management ing to bring together Demographic Methods across the
contexts? Tree of Life. This was the initial aim of this book! Later,
2. The common currencies of demography: Despite the beginnings of this book were developed via frequent
variation in demographic schedules, all individu- conversations between us and with all of the contribut-
als of all species are born, then develop, survive, ing authors, as well as rigorous peer-reviewing, to
reproduce, and, sooner or later, die. These rates of life
Roberto Salguero-Gómez and Marlène Gamelon, From lions, to lion’s manes, to dandelions: how different types of demographic data and methods
are selected and why. In: Demographic Methods Across the Tree of Life. Edited by Roberto Salguero-Gómez and Marlène Gamelon,
Oxford University Press. © Oxford University Press (2021). DOI: 10.1093/oso/9780198838609.002.0001
2 DEMOGRAPHIC METHODS ACROSS THE TREE OF LIFE
make sure that our primary motivation was fulfilled: to However, in most wild populations, knowing the fate
provide a cohesive book that brings together field ecol- of an individual remains challenging. An individual
ogists, ecological modellers, and practitioners work- that is alive is not necessarily detected by the observ-
ing with animal and plant kingdoms. To reach this er (Barthold et al. 2016; Eikelboom et al. 2019). This
goal, the book is divided into three parts: demographic component adds an additional layer of complexity, as
data collection (Part I), demographic analyses (Part II), identifying the cause of death is difficult because death
and demographic applications (Part III). This phase of is rarely observed. This example demonstrates why we
growth/development took some years and luckily end- need to develop new techniques for data acquisition, as
ed with the birth of this book. We will do our best well as new demographic methods for data analyses,
to slow down the inevitable decline of performance with across the Tree of Life to address key and timely ques-
increasing ages (but see Jones et al. 2014) by maintain- tions in ecology, evolution, and conservation biology.
ing a repository containing the materials linked to this
book to prevent it from becoming outdated. We now
3 Why use demographic methods across
hope that this book’s reproductive output will be high
with the recruitment of the next generation of demogra- the Tree of Life?
phers, where questions, methods, and systems will not A myriad of questions can be addressed from demo-
be siloed taxonomically or biogeographically. graphic data collected in natura or in the lab when anal-
ysed with the appropriate demographic methods. This
2 Origins of demographic analyses: book provides a nonexhaustive review of the ques-
from humans to nonhuman animals, tions that can be tackled, ranging from the study of
eco-evolutionary feedbacks (Chapter 20), to the assess-
plants, fungi, and microbes ment of the extinction risk in harvested populations
Classical demography is broadly concerned with the and the best way to manage them (Chapters 17 and
size, distribution, structure, and change of popula- 19), through the evaluation of the effects of environ-
tions through time. Demography (Gr. demo-: people, mental disturbances and perturbations on short- and
-graphy: writing, description, or measurement) literally long-term population dynamics (Chapters 10 and 11),
means ‘description of people’. Not surprisingly, this as well as the importance of demography in epidemic
science has long been centred on the study of human dynamics (Chapter 22). All these questions are particu-
populations (Preston et al. 2000). It is worth noting larly timely, given the ongoing context of global change
that such descriptions of people and their popula- where the Earth has entered into the sixth major extinc-
tions have been greatly enhanced through the part- tion event of its history (Lewis and Maslin 2015) and
nering of human demographers (self-named ‘formal where we are all experiencing a pandemic!
demographers’, Hinde 1998) with statisticians and The development and use of demographic tools for
actuarial scientists (i.e. researchers examining ageing the analysis of demographic data requires strong quan-
patterns). Although demographic tools were initially titative inclinations. However, the reliance of demogra-
developed to examine human populations, these tools phy on quantitative methods does not mean that if the
have been progressively applied to nonhuman species. current holder of this book is not good at math, matrix
For instance, competing risk statistical models (Crow- calculus, statistics, or the like they should immediate-
der 2001; Heisey and Patterson 2006), used to estimate ly close it. What it does mean, though, is that investing
age- and cause-specific mortality probability when the adequate time to become quantitatively trained (or col-
fate of an individual (i.e. alive or dead) and the cause laborate with somebody who is) will undeniably result
of death (e.g. cancer, car accident) are precisely known, in important benefits in the understanding of, say, (1)
have been used in medicine, human demography, and through which evolutionary mechanisms populations
more broadly on captive animals (Southey et al. 2004) of dandelions (Taraxacum officinale) became established
and on plant populations (Shefferson and Roach 2013). and invasive in New Zealand (Vitousek et al. 1997),
Though these kinds of demographic analyses are (2) the current ecological influences and socioecologi-
possible in other taxonomic corners of the Tree of Life— cal implications of hunting by male lions (Panthera leo)
perhaps because the data are scarcer—the applicabil- in South Africa (Funston et al. 1998), or (3) the projected
ity of the tool is very much there. To further build viability of the edible yet endangered lion’s mane fun-
on our example on mortality probability, advances in gus (Hericium erinaceus) in the United Kingdom (Boddy
radiotelemetry and GPS technologies now allow for et al. 2011).
the fate of wild animals to be known and the caus- The choice of examples described above regarding
es of death to be identified (Tomkiewicz et al. 2010). how demographic tools help biologists and managers
F R O M L I O N S, TO L I O N ’ S M A N E S, TO D A N D E L I O N S 3
understand the biology of three different species of 4 From data acquisition to

‘lion’, on three different topics, three different corners demographic analyses
of the Tree of Life, three different locations around
the world, and at three different temporal scales (past, Techniques for data acquisition have been developed
present, and future) is not coincidental. Rather, these across the entire Tree of Life, from microbes, to fungi,
are part of a conscious effort—also represented in plants, and animals. Several textbooks (Harper 1977;
every chapter in this book—to highlight the fact that Elzinga et al. 2001) have done a wonderful job at detail-
demographic tools need not be limited to biogeograph- ing specific data collection methods. However, these
ic, temporal, and/or taxonomic silos. Tearing down volumes have focused on specific taxonomic groups
those silos and improving cross-fertilisation among (e.g. plants, Gibson 2002; nonhuman animals, Williams
them has been the primary motivation for this book. et al. 2002; or humans, Hinde 1998). The historical tax-
Indeed, equipping oneself with robust quantitative onomic focus of demographic textbooks is based on
methods can help the researcher describe a vast range the rationale that certain protocols that are appropri-
of questions in ecology, evolution, and conservation ate for animals (e.g. GPS tracking) are a priori not
biology. Their application, we argue, pertains to a applicable for plants, and vice versa. In recent decades,
broad range of creatures and ecosystems because of the though, the number of techniques for data acquisition
universal currencies of demography: the vital rates of has drastically increased thanks, in part, to the devel-
survival, development, reproduction, and recruitment. opment of technologies such as micro radio-tracking
These universal currencies are shared across dande- unmanned aerial vehicles (i.e. drones, Hodgson et al.
lions, lions, and lion’s manes, as much as they are 2018) and satellites (Zhao et al. 2020; Chapter 5). These
between bacteria, you, and us. technologies hold great promise to unlock the potential
We believe that a significant number of questions, of classical human demographic tools, or of large mam-
protocols, and methods can be applied across the mal demographic techniques to the whole of the Tree
entire spectrum of life, and that important research of Life (e.g. Pons and Pausas 2007). As a consequence
advances will emerge via the unification and stan- of the development of techniques for data acquisition,
dardisation of these questions, protocols, and analyt- demographic data have been collected for many taxa
ical tools across taxa. This book, thus, aims to fulfil a across the Tree of Life. This key advancement has given
need to unify both field and analytical demographic rise to a large number of open-access databases gather-
methods across all major life forms. From this unifi- ing information on vital rates and/or life history traits
cation, three historically separated disciplines (formal (Table 1). However, we note that more demograph-
[human] demography, [nonhuman] animal demogra- ic works would be particularly welcome in microbes
phy, and rest of the Tree of Life [including plants (Levin 1990), certain taxonomic groups in the plant
and microbes] demography) will benefit in at least kingdom (e.g. algae, Salguero-Gómez et al. 2014), and
three foreseeable ways. First, by sharing questions and some groups with important ecological value in the
field/analytical methods, broader analyses (both glob- animal kingdom (e.g. insects, Carey 2001), as well as
ally and taxonomically) will emerge that will allow even among some charismatic tetrapods (Conde et al.
researchers to address general rules in ecology and 2019).
evolution, and ultimately develop more cost-effective The last decades have also witnessed an explosion
management plans for the eradication of invasive of demographic analyses for a large range of taxa.
species and the preservation of endangered ones. Sec- For instance, key textbooks have elaborated on demo-
ond, thanks to new techniques shared across taxa, graphic tools (e.g. matrix population models, Caswell
a series of eco-evolutionary questions will be made 2001; integrated population models, Kéry and Schaub
available to a broader range of taxa. For instance, 2011; Schaub and Kéry 2021; integral projection mod-
the improvement of pedigree analyses in areas that els, Easterling et al. 2000; Ellner et al. 2016; stochas-
have historically not implemented them (e.g. plant and tic population dynamics, Lande et al. 2003; and a
microbial biology) will allow novel eco-evolutionary large diversity of analytical and modelling approach-
questions to be tackled, questions addressed predomi- es, Williams et al. 2002; Murray and Sandercock 2020).
nantly in animal populations so far (Wilson et al. 2011; However, these textbooks assume that the data have
Bonnet et al. 2019). Third, this book strengthens the already been collected—or provide somewhat limited
link between data collectors and demographic mod- information about how to obtain them. In addition,
ellers, thus bringing a tighter biological understanding some of them are now circa 20 years old and do not
to the statistical analyses and vice versa (Gimenez et al. reflect how methods for data acquisition and analyses
2013). have improved rapidly in the last decades. As such,
Table 1 Some of the key data sources that contain open-access demographic data of species across the Tree of Life.
Data source Description Source
Amniote Life history trait database of bird, mammal, and reptile species http://onlinelibrary.wiley.com/doi/10.1890/15-0846R.1/abstract
AnAge Animal ageing and longevity database https://onlinelibrary.wiley.com/doi/full/10.1111/j.1420-
9101.2009.01783.x
BIDDABA Bird demographic database https://link.springer.com/article/10.1007/s10336-010-0582-0
BTO survey data British Trust for Ornithology http://www.bto.org/research-data-services/data-services
DATLife Life tables of animals, as well as records of longevity https://datlife.org/
demography Forecasting mortality, fertility, migration, and population data https://cran.r-project.org/web/packages/demography/index.html
COMADRE Matrix population models for over 500 animal species worldwide https://compadre-db.org/Data/Comadre
COMPADRE Matrix population models for ca. 1000 plant species worldwide https://compadre-db.org/Data/Compadre
Coral Trait Database Database containing, among other traits, life history trait information for over 1500 https://coraltraits.org
coral species
EURING databank Ringing recovery data of bird species across Europe https://euring.org/data-and-codes/euring-databank
europop Historical populations of European cities, 1500–1800 https://cran.r-project.org/web/packages/europop/index.html
fishdata A small collection of fish population data sets https://cran.r-project.org/web/packages/fishdata/index.html
FishTraits Ecological and life history traits of fishes of the United States https://pubs.er.usgs.gov/publication/70156095
Global Assessment of Reptile Trait and geographic data on lizards https://onlinelibrary.wiley.com/doi/full/10.1111/geb.12773
Distributions
Human Cause-of-Death Database Continuous human data series of causes of mortality for 16 countries https://www.causesofdeath.org
Human Fertility Database Period and cohort fertility data for human populations from 32 countries http://www.humanfertility.org
Human Life-Table Database Collection of human population life tables for multiple years across 141 countries https://www.lifetable.de/cgi-bin/index.php
Human Mortality Database Human population size and mortality data for 41 countries http://www.mortality.org
International Database of Information about supercentenarians (humans of ages 110 and above) from 13 https://www.supercentenarians.org
Longevity countries
International Data Base Demographic measures of human populations across 288 countries worldwide https://www.census.gov/programs-surveys/international-
programs/about/idb.html
LEDA A database of life history traits of the Northwest European flora https://besjournals.onlinelibrary.wiley.com/doi/10.1111/j.1365-
2745.2008.01430.x
Life history trait database of A database of traits (e.g. activity, energy, movement) of European reptile species https://datadryad.org/stash/dataset/doi:10.5061/dryad.hb4ht
European reptile species
Living Planet Index Trends of 27,232 natural populations from 4,784 species through time around the WWF International. 2012. Living Planet Report 2012. Pp. 160.
world WWF International, Gland, Switzerland
Longevity Records Life spans for mammals, birds, amphibians, reptiles, and fish https://www.demogr.mpg.de/longevityrecords/
MALDABBA Age-specific vital rates and life history traits for 200 mammalian species https://www.pnas.org/content/suppl/2020/03/18/1911999117.
DCSupplemental
PADRINO Database of IPMs for hundreds of animals and plants around the globe https://github.com/levisc8/Padrino.github.io
Pantheria Mammal life history database https://ecologicaldata.org/wiki/pantheria
LTER US Long Term Ecological Research Network, including long-term records of https://lternet.edu
population size and individual records in some cases
SCALES Securing the conservation of biodiversity across administrative levels and spatial, http://scales.ckff.si/scaletool/?menu=6
temporal, and ecological scales—a database of species traits
Serengeti bird species occurrence, Georeferenced occurrences for 568 species from 1929 to 2017. Records contain https://esajournals.onlinelibrary.wiley.com/doi/full/10.1002/ecy.
abundance and habitat feeding location, food source, distribution status, observation locality 2919
Serengeti: survey of age structure Sample counts from 1926 to 2018 of 13 ungulate species and 1 ostrich species https://www.nature.com/articles/s41597-020-00701-0
in ungulates and ostrich
SPI-Birds Database of bird breeding and mark/capture information https://nioo.knaw.nl/en/spi-birds
Traits Freshwater biological traits database https://www.epa.gov/risk/freshwater-biological-traits-database-
traits
the present book should serve as a useful crossroads of that researchers and practitioners must fill in, linking
data collection, analyses, and interpretation to a broad the acquisition of data, their analyses, and applications
range of disciplines. to their specific questions. An additional motivation
A key feature of any research programme is of this book is to provide a broad springboard from
that its analyses and results should be reproducible which any biologist may be able to walk away with
(Goodman et al. 2016). The field of demography, due a good understanding of what demographic data to
to its strong quantitative inclinations, has benefit- collect (Part I), how to analyse those data through a
ted tremendously from the move towards textbooks multitude of possible analytical approaches (Part II),
with scripts and open-access packages. Caswell, for and what kinds of applications in ecology, evolution,
instance, in his (2001) seminal overview of matrix pop- and conservation biology are pertinent (Part III).
ulation models where a detailed account of the per-
tinent theory, construction, and interpretation of this
key demographic tool is provided, also provided Mat-
Lab scripts that greatly helped readers apply this tool. 5 Organisation of the book
Likewise, conservation biologists had a unique oppor- This book will guide the readers, step by step, along
tunity to apply demographic tools in the context of 22 chapters organised into three main parts: data col-
population viability analyses in the textbook by Mor- lection, analyses, and applications. Because we have
ris and Doak (2002), where MatLab scripts were also aimed to find a common ground in demographic data
made available. Even earlier than these two references, collection, modelling, and questions that go beyond
Ebert (1999) already provided detailed comments and taxonomic boundaries, all the chapters do not spe-
code in BASIC for the analysis of structured pop- cialise according to taxonomic affiliations and rather
ulation data in both animals and plants. Following include diverse taxa.
that spirit of research reproducibility, and with the Part I of this book naturally brings together six
specific aim to provide our readers with key mate- chapters that present how data can be collected at
rials to adapt to their data sets and questions, the different levels of organisation, from genes to the
present textbook contains carefully annotated scripts environment, going through traits, individuals, and
in R (Team 2010) (a nonproprietary software plat- populations. Chapter 1 introduces the reader to
form that is widely used among ecologists, evolution- the collection and consideration of genetic data in
ary biologists, and conservation biologists), as well demographic analyses. Specifically, it provides a
as full sets of demographic data. These are available gentle introduction to pedigrees and phylogenies,
at www.oup.com/companion/SalgueroGamelonDM, now fundamental tools of demographic analyses.
and it is our intention to revisit them, update them, and Chapter 2 details methods to collect information at
add new applications, in an effort to keep up with the the biochemical and physiological levels. Examples
high speed of development of demography. include the quantification of biomarkers, hormones,
The last decades have witnessed a Malthusian and proxies of oxidative stress. Chapter 3 introduces
increase in R libraries specifically targeted at helping the reader to the collection of social data. It provides
ecologists, evolutionary biologists, and conservation sampling procedures to describe fine-grained patterns
biologists avail from demographic models. This emer- of social interactions and initiate the reader to social
gence has been further supported by the recent appear- network analyses. Chapters 4 and 5 divide the heavy
ance of journals (e.g. Methods in Ecology and Evolution, load of demographic data collection into two cate-
founded in 2010) that house a lot of software for demo- gories. In Chapter 4, methods are introduced to collect
graphic analyses (e.g. Colchero et al. 2012; Metcalf et al. demographic data from a single visit (or few visits) to
2013; Kettle et al. 2018; Shefferson et al. 2021). With the a field site using tools that allow researchers to obtain
purpose of making the reader aware of the diversity a historical perspective, such as dendrochronological
of demographic tools at their disposal via R, we have and sclerochronological methods. Chapter 5, on the
compiled a not at all comprehensive and yet rather long other hand, overviews methods to track dynamics of
list of R libraries for demographic analysis (Table 2). individuals in populations, and whole populations,
The main merit of these publications and their soft- in mobile and sessile species, including the usage of
ware is that they focus on keeping the mathematics new technologies such as small tracking devices and
and statistics behind these developments to a manage- drones. Chapter 6 then serves as an excellent comple-
able (but important!) size, while often providing useful ment to contextualise the demography of the species
vignettes with examples regarding their execution with of interest, as populations do not occur in isolation.
real-world data. However, there is still a significant gap This chapter overviews some of the key biotic and
Table 2 A nonexhaustive review of R packages developed in the last decades that have significantly improved the understanding and applicability of demographic tools by ecologists, evolutionary
biologists, and conservation biologists.
Brief description Source

Name of package
albopictus Age-structured population dynamics models https://cran.r-project.org/web/packages/albopictus/index.html

BaSTA Age-specific survival analysis from incomplete capture-recapture/recovery data https://cran.r-project.org/web/packages/BaSTA/index.html
BayeSPsurv Bayesian spatial split population survival model https://cran.r-project.org/web/packages/BayesSPsurv/index.html
befproj Makes a local-population projection https://cran.r-project.org/web/packages/befproj/index.html
bootstrapFP Bootstrap algorithms for finite population inference https://cran.r-project.org/web/packages/bootstrapFP/index.html
bssm Bayesian inference of non-Gaussian state space models https://cran.r-project.org/web/packages/bssm/index.html
BTSPAS Bayesian time-stratified population analysis https://cran.r-project.org/web/packages/BTSPAS/index.html
capm Companion animal population management https://cran.r-project.org/web/packages/capwire/index.html
capwire Estimates population size from noninvasive sampling https://cran.r-project.org/web/packages/capwire/index.html
CARE1 Statistical package for population-size estimation in capture-recapture models https://cran.r-project.org/web/packages/CARE1/index.html
colorednoise Simulate temporally autocorrelated populations https://cran.r-project.org/web/packages/colorednoise/index.html
conStruct Models spatially continuous and discrete population genetic structure https://cran.r-project.org/web/packages/conStruct/index.html
Cyclops Cyclic coordinate descent for logistic, Poisson, and survival analysis https://cran.r-project.org/web/packages/Cyclops/index.html
DPWeibull Dirichlet process Weibull mixture model for survival data https://cran.r-project.org/web/packages/DPWeibull/index.html
demogR Analysis of age-structured demographic models https://cran.r-project.org/web/packages/demogR/index.html
dendRoAnalyst A tool for processing and analysing dendrometer data https://cran.r-project.org/web/packages/dendRoAnalyst/index.html
dendrometeR Analysing dendrometer data https://cran.r-project.org/web/packages/dendrometeR/index.html
dendroTools Linear and nonlinear methods for analysing daily and monthly dendroclimatological data https://cran.r-project.org/web/packages/dendroTools/index.html
dynamichazard Dynamic hazard models using state space models https://cran.r-project.org/web/packages/dynamichazard/index.html
dynsurv Dynamic models for survival data https://cran.r-project.org/web/packages/dynsurv/index.html
FinePop Fine-scale population analysis https://cran.r-project.org/web/packages/FinePop/index.html
FREQ Estimate population size from capture frequencies https://cran.r-project.org/web/packages/FREQ/index.html
gauseR Lotka-Volterra models for Gause’s ‘struggle for existence’ https://cran.r-project.org/web/packages/gauseR/index.html
IBMPopSim Individual-based model population simulation https://cran.r-project.org/web/packages/IBMPopSim/index.html
IPMpack Analysis and interpretation of simple IPMs https://cran.r-project.org/src/contrib/Archive/IPMpack/
continued
Table 2 Continued

Name of package
IPMR Analysis and interpretation of simple and complex IPMs https://github.com/levisc8/ipmr

jacpop Jaccard index for population structure identification https://cran.r-project.org/web/packages/jacpop/index.html
lefko3 Historical and ahistorical population projection matrix analysis https://cran.r-project.org/web/packages/lefko3/index.html
lhmixr Fit sex-specific life history models with missing classifications https://cran.r-project.org/web/packages/lhmixr/index.html
LifeHist Life history models of individuals https://cran.r-project.org/web/packages/LifeHist/index.html
lmf Functions for estimation and inference of selection in age-structured populations https://cran.r-project.org/web/packages/lmf/index.html
LTRCforests Ensemble methods for survival data with time-varying covariates https://cran.r-project.org/web/packages/LTRCforests/index.html
metafolio Metapopulation simulations for conserving salmon through portfolio optimisation https://cran.r-project.org/web/packages/metafolio/index.html
microPop Modelling microbial populations https://cran.r-project.org/web/packages/microPop/index.html
missDeaths Simulating and analysing time to event data in the presence of population mortality https://cran.r-project.org/web/packages/missDeaths/index.html
MortalityLaws Parametric mortality models, life tables, and HMD https://cran.r-project.org/web/packages/MortalityLaws/index.html
MortalityTables A framework for various types of mortality/life tables https://cran.r-project.org/web/packages/MortalityTables/index.html
morse Modelling tools for reproduction and survival data in ecotoxicology https://cran.r-project.org/web/packages/morse/index.html
mptools RAMAS metapop tools https://cran.r-project.org/web/packages/mptools/index.html
MRsurv A multiplicative-regression model for relative survival https://cran.r-project.org/web/packages/MRsurv/index.html
ncappc NCA calculations and population model diagnosis https://cran.r-project.org/web/packages/ncappc/index.html
NEff Calculating effective sizes based on known demographic parameters of a population https://cran.r-project.org/web/packages/NEff/index.html
LexisPlotR Plot Lexis diagrams for demographic purposes https://cran.r-project.org/web/packages/LexisPlotR/index.html
npsurv Nonparametric survival analysis https://cran.r-project.org/web/packages/npsurv/index.html
OBMbpkg Estimate the population size for the MB capture-recapture model https://cran.r-project.org/web/packages/OBMbpkg/index.html
openCR Open population capture-recapture https://cran.r-project.org/web/packages/openCR/index.html
pec Prediction error curves for risk prediction models in survival analysis https://cran.r-project.org/web/packages/pec/index.html
plethem Population life course exposure to health-effects modelling https://cran.r-project.org/web/packages/plethem/index.html
pop A flexible syntax for population dynamic modelling https://cran.r-project.org/web/packages/pop/index.html
popbio Construction and analysis of matrix population models https://cran.r-project.org/web/packages/popbio/index.html
popdemo Demographic modelling using projection matrices https://cran.r-project.org/web/packages/popdemo/index.html
POPdemog Plot population demographic history https://cran.r-project.org/web/packages/POPdemog/index.html
PopED Population and individual optimal experimental design https://cran.r-project.org/web/packages/PopED/index.html
popEpi Functions for epidemiological analysis using population data https://cran.r-project.org/web/packages/popEpi/index.html
popkin Estimate kinship and FST under arbitrary population structure https://cran.r-project.org/web/packages/popkin/index.html
poppr Genetic analysis of populations with mixed reproduction https://cran.r-project.org/web/packages/poppr/index.html
poptrend Estimate smooth and linear trends from population-count survey data https://cran.r-project.org/web/packages/poptrend/index.html
PROSPER Simulation of weed population dynamics https://cran.r-project.org/web/packages/PROSPER/index.html
PSPManalysis Analysis of physiologically structured population models https://cran.r-project.org/web/packages/PSPManalysis/index.html
PCAClone Population viability analysis with data cloning https://cran.r-project.org/web/packages/PVAClone/index.html
pyramid Draw population pyramid https://cran.r-project.org/web/packages/pyramid/index.html
rangeMapper A platform for the study of macro-ecology of life history traits https://cran.r-project.org/web/packages/rangeMapper/index.html
Rage Calculation of life history traits from matrix population models https://github.com/jonesor/Rage
RCOMPADRE Manipulation of data in the COMPADRE & COMADRE databases https://github.com/jonesor/Rcompadre
ribd Pedigree-based relatedness coefficients https://cran.r-project.org/web/packages/ribd/index.html
Interface to the software package MARK that constructs input files and extracts the output
RMark https://cran.r-project.org/web/packages/RMark/index.html
(from capture-mark-recapture data)
rmetasim An individual-based population genetic simulation environment https://cran.r-project.org/web/packages/rmetasim/index.html
Rpoppler PDF tools based on Poppler https://cran.r-project.org/web/packages/Rpoppler/index.html
Rramas Matrix population models https://cran.r-project.org/web/packages/Rramas/index.html
rSHAPE Simulated haploid asexual population evolution https://cran.r-project.org/web/packages/rSHAPE/index.html
SMITIDstruct Data structure and manipulations tool for host and viral population https://cran.r-project.org/web/packages/SMITIDstruct/index.html
sspse Estimating hidden population size using respondent-driven sampling data https://cran.r-project.org/web/packages/sspse/index.html
SSsimple State space models https://cran.r-project.org/web/packages/SSsimple/index.html
stagePop Modelling the population dynamics of a stage-structured species in continuous time https://cran.r-project.org/web/packages/stagePop/index.html
statepacer State space modelling in R https://cran.r-project.org/web/packages/statespacer/index.html
steps Spatially and temporally explicit population simulator https://cran.r-project.org/web/packages/steps/index.html
survival Survival analysis methods https://cran.r-project.org/web/packages/survival/index.html
survivalmodels Models for survival analysis https://cran.r-project.org/web/packages/survivalmodels/index.html
survParamSim Parametric survival simulation with parameter uncertainty https://cran.r-project.org/web/packages/survParamSim/index.html
timereg Flexible regression models for survival data https://cran.r-project.org/web/packages/timereg/index.html
continued
Table 2 Continued

Name of package
trackdem Particle tracking and demography https://cran.r-project.org/web/packages/trackdem/index.html

treestructure Detect population structure within phylogenetic trees https://cran.r-project.org/web/packages/TSSS/index.html
TSSS Time series analysis with state space models https://cran.r-project.org/web/packages/TSSS/index.html
wiqid Quick and dirty estimates for wildlife populations https://cran.r-project.org/web/packages/wiqid/index.html
wpp2019 World population prospects 2019 World population prospects in 2019
abiotic factors that researchers might want to consider in the field of population ecology, evolutionary
examining, together with the demographic data. demography, and conservation biology. Therefore,
Part II of this book contains an overview of the this last part is specifically dedicated to applications.
most commonly used methods in demography across Chapter 15 illustrates state-of-the-art methods in
different species, and it contains eight chapters. These spatial demography. Specifically, the author nicely
methodologies are data- and research question driven. applies some of the demographic tools described above
Each chapter clearly states the type of questions that (e.g. in Chapters 9, 10, 12, 13, and 14) to show that
can be tackled, and the data required to address them. demographic methods can inform the spatial response
Time series of population counts are commonly collect- of individuals and populations to changing environ-
ed in the field (see Chapter 5), and Chapter 7 provides ments. Chapter 16 showcases central ideas and meth-
precious methodological tools to analyse these types ods in evolutionary demography, a research field that
of data and get important insight on changes in pop- has grown rapidly in recent years. This includes the-
ulation size over time. Chapter 8 introduces life tables, ories of postreproductive life and models of mutation
the simplest kind of structured population model, and selection. Chapter 17 applies demographic theory
where individuals are organised and followed across to discuss reproductive value in a conservation biology
their lifetime based on their ages. Chapter 9 overviews context. Chapter 18 exploits the large amount of demo-
matrix population models, discrete time demographic graphic data across the Tree of Life to compare and
models where individuals are classified into ages, contrast life history traits and explore whether these
developmental stages, and/or discrete size ranges. are constrained by main ecological realms. Chapter
Chapter 10 introduces integral projection models 19 introduces the reader to adaptive management, a
(IPMs), discrete time structured population models specific form of decision-making in conservation biolo-
where individuals are classified along continuous gy and wildlife management. Chapter 20 brings a more
variables such as size. Common to life tables, matrix genetic perspective on demography by discussing her-
population models, and IPMs is the expectation that itability, polymorphisms, and population dynamics.
a population will go to stationary equilibrium if left It provides an eco-evolutionary framework and high-
under a constant environment. Of course, most pop- lights the strong relationship between changes in phe-
ulations are constantly affected by changes in a/biotic notypic traits and population dynamics. Chapter 21
conditions. The transient (short-term) responses that details applications of demographic tools for socially
result from these disturbances are the focus of Chapter structured populations. It nicely exemplifies the use of
11. Chapter 12 delves into individual-based models. In social data (Chapter 3) and demographic methods (e.g.
it, the reader is invited to move towards mechanistic Chapters 9, 10, and 12) to model the dynamics of social-
models, where demographic population trends can be ly structured populations. Finally, Chapter 22 focuses
explained by processes (e.g. local interactions, response on the interplay between demography and epidemiol-
to changing environments, and adaptive behaviour) ogy. It illustrates how demography shapes infectious
occurring at the individual level. Chapter 13 deals with disease dynamics and how, in turn, infectious disease
survival analyses. It provides a useful toolkit to esti- dynamics shape demography.
mate survival, a key vital rate in demographic studies,
for a diversity of species, ranging from lab organisms
to humans, birds, reptiles, and plants. Chapter 14 6 Target audience
introduces the reader to integrated population models This book is primarily targeted at researchers and prac-
(another kind of IPM!), where various sources of titioners. The authors have assumed that the readers of
demographic data (e.g. population counts, capture- this book will be at an upper-level undergraduate level
recapture data) are amalgamated, together with or higher. Our motivation to make this book as math-
biological knowledge on the life cycle of the species friendly as possible has guided the writing of all of the
of interest, to estimate vital rates and population chapters under the following six principles:
dynamics.
Part III in this book contains key applications of 1. Field/lab experience: We have made no assump-
demographic methods to ecology, evolution, epidemi- tions about the knowledge of the reader regarding
ology, and conservation biology. This part contains data collection. Instead, in Part I, we walk them step
eight chapters. Our way to engage with nonexperts by step through field methods to collect pertinent
on any one topic is by making explicit linkages to demographic data.
the questions that both managers and demographers 2. Programming: As mentioned above, demography
are interested in, and by providing tangible examples has quantitative inclinations, and these are best
satiated through programming (rather than point- this book nor the discipline as a cookbook. True, we
and-click software that prevents the user from truly have provided some utensils (data and methods), as
understanding what is under the hood, Farrell and well as some recipes (in the shape of commented R
Carey 2018). This book assumes that the user is scripts). We strongly encourage the readers to make
familiar with basic R programming at an introduc- these tools truly their own by moving towards inno-
tory level (e.g. how to load data, save outputs, and vative areas where none of us envisioned they could
interpret error messages), but all pertinent R scripts go. Readers of this book have a unique opportunity to
are accompanied with line-by-line explanations. contribute to the big picture: unless their questions and
3. Population ecology/formal demography: The book interests are taxonomically bound to a specific group
makes no assumptions about the background of species (and we emphasise, there is nothing wrong
knowledge of the reader regarding demography, so with that), we firmly support the idea of applying these
basic and advanced concepts are introduced and tools to areas of the planet and corners of the Tree of
discussed. Life where not much is yet known, demographically
4. Evolution: The book contains a series of chapters on speaking. After all, demography is everywhere, and
applications of demography in evolutionary biolo- demographic research goes well beyond the classical
gy (e.g. Chapters 16 and 20). Some assumptions are definition of a population (Griffith et al. 2016)
made regarding the evolutionary understanding of
the reader, but these are satisfied by pointing them
towards the relevant literature in each chapter. References
5. Statistics: Methodological and applied chapters that Barthold J. A., C. Packer, A. J. Loveridge, D. W. Macdonald,
rely on advanced statistics provide a gentle intro- and F. Colchero (2016). Dead or gone? Bayesian inference
duction to the concepts in use (e.g. Bayesian statis- on mortality for the dispersing sex. Ecology and Evolution
tics). When pertinent, the chapters (e.g. Chapter 9) 6(14), 4910–4923.
include conceptual boxes to better understand and Boddy L., M. E. Crockatt, and A. M. Ainsworth (2011). Ecol-
solidify these concepts. ogy of Hericium cirrhatum, H. coralloides and H. erinaceus in
6. Mathematics: Demographic analyses require a the UK. Fungal Ecology 4(2), 163–173.
good understanding of algebra. Chapters based on Bonnet T., M. B. Morrissey, A. Morris, et al. (2019). The role
of selection and evolution in changing parturition date in
advanced mathematics introduce the associated
a red deer population. PLOS Biology 17(11), 23.
topics with the relevant literature (Part II).
Carey J. R. (2001). Insect biodemography. Annual Review of
We see this book as a springboard that integrates Entomology 46, 79–110.
Carey J. R. (2002). Longevity minimalists: life table stud-
primary data collection and cutting-edge demograph-
ies of two species of northern Michigan adult mayflies.
ic analyses and their applications across a broad range
Experimental Gerontology 37(4), 567–570.
of topics. Naturally, one could have (and other authors Caswell H. (2001). Matrix Population Models: Construction,
have!) written a book on any of these topics (see refer- Analysis, and Interpretation. Sinauer Associates.
ences in this chapter). Instead, we encourage the reader Colchero F., O. R. Jones, and M. Rebke (2012). BaSTA:
to treat this book as a comprehensive introduction to an R package for Bayesian estimation of age-specific
the vast amount of demographic approaches in demog- survival from incomplete mark-recapture/recovery data
raphy. They are, of course, encouraged to deepen their with covariates. Methods in Ecology and Evolution 3(3),
understanding by searching into more specialised text- 466–470.
books and the pertinent peer-review literature. Conde D. A., J. Staerk, F. Colchero, et al. (2019). Data gaps
and opportunities for comparative and conservation biol-
ogy. Proceedings of the National Academy of Sciences of
7 What next? A plea for a stronger, the United States of America 116(19), 9658–9664.
Crowder M. J. (2001). Classical Competing Risks. Chapman
richer, broader, and more innovative & Hall.
demography Easterling M. R., S. Ellner, and P. Dixon (2000). Size-specific
sensitivity: applying a new structured population model.
And now that the reader is aware of our motivations
Ecology 81(3), 694–708.
for this book, the available online support, the structure
Ebert T. A. (1999). Plant and Animal Populations: Methods
of the parts, and the tools presented, we dare to make in Demography. Academic Press.
a plea. The vast availability of demographic methods Eikelboom J. A. J., J. Wind, E. Van De Ven, et al. (2019).
means that one might fall into the comfort of doing Improving the precision and accuracy of animal popula-
what others have done, with a slight tweak, and on a tion estimates with aerial image object detection. Methods
different system. We challenge the reader not to treat in Ecology and Evolution 10(11), 1875–1887.
Ellner S. P., D. Z. Childs, and M. Rees (2016). Data-Driven Metcalf C. J. E., S. M. McMahon, R. Salguero-Gómez, and
Modelling of Structured Populations. Springer. E. Jongejans (2013). IPMpack: an R package for integral
Elzinga C. L., D. W. Salzer, J. W. Willoughby, and J. P. projection models. Methods in Ecology and Evolution
Gibbs (2001). Monitoring Plant and Animal Populations: 4(2), 195–200.
A Handbook for Field Biologists. Blackwell. Morris W. F. and D. F. Doak (2002). Quantitative Conserva-
Farrell K. J. and C. C. Carey (2018). Power, pitfalls, tion Biology: Theory and Practice of Population Viability
and potential for integrating computational literacy into Analysis. Sinauer Associates.
undergraduate ecology courses. Ecology and Evolution Murray L. and B. K. Sandercock (2020). Population Ecology
8(16), 7744–7751. in Practice. Wiley.
Funston P. J., M. G. L. Mills, H. C. Biggs, and P. R. Pons J. and J. G. Pausas (2007). Acorn dispersal estimated by
K. Richardson (1998). Hunting by male lions: ecologi- radio-tracking. Oecologia 153(4), 903–911.
cal influences and socioecological implications. Animal Preston S., P. Heuveline, and M. Guillot (2000). Demog-
Behaviour 56, 1333–1345. raphy: Measuring and Modeling Population Processes.
Gibson D. (2002). Methods in Comparative Plant Population Wiley.
Ecology. Oxford University Press. Salguero-Gómez R., O. R. Jones, C. R. Archer, et al. (2014).
Gimenez O. , F. Abadi, J. Y. Barnagaud et al. (2013). How The COMPADRE plant matrix database: an open online
can quantitative ecology be attractive to young scientists? repository for plant demography. Journal of Ecology
Balancing computer/desk work with fieldwork. Animal 103(1), 202–218.
Conservation 16(2), 134–136. Schaub M. and M. Kéry (2021). Integrated Population Mod-
Goodman S. N., D. Fanelli, and J. P. A. Ioannidis (2016). What els: Theory and Ecological Applications with R and JAGS.
does research reproducibility mean? Science Translational Academic Press.
Medicine 8(341), 6. Shefferson R. P. and D. A. Roach (2013). Longitudinal anal-
Griffith A. B., R. Salguero-Gómez, C. Merow, and S. ysis in Plantago: strength of selection and reverse age
McMahon (2016). Demography beyond the population. analysis reveal age-indeterminate senescence. Journal of
Journal of Ecology 104(2), 271–280. Ecology 101(3), 577–584.
Harper J. L. (1977). Population Biology of Plants. Academic Shefferson R. P., S. Kurokawa, and J. Ehrlén (2021).
Press. LEFKO3: analysing individual history through size-
Heisey D. M. and B. R. Patterson (2006). A review of meth- classified matrix population models. Methods in Ecology
ods to estimate cause-specific mortality in presence of and Evolution 12(2), 378–382.
competing risks. Journal of Wildlife Management 70(6), Southey B. B., S. L. Rodriguez-Zas, and K. A. Leymaster
1544–1555. (2004). Competing risks analysis of lamb mortality in a
Hinde A. (1998). Demographic Methods. Routledge. terminal sire composite population. Journal of Animal
Hodgson J. C., R. Mott, S. M. Baylis, et al. (2018). Science 82(10), 2892–2899.
Drones count wildlife more accurately and precisely Sutherland W. J., R. P. Freckleton, H. C. J. Godfray, et al.
than humans. Methods in Ecology and Evolution 9(5), (2013). Identification of 100 fundamental ecological ques-
1160–1167. tions. Journal of Ecology 101(1), 58–67.
Jones O. R., A. Scheuerlein, R. Salguero-Gómez, et al. (2014). Team R. D. C. (2010). R: A Language and Environment for
Diversity of ageing across the tree of life. Nature 505(7482), Statistical Computing. R Foundation for Statistical Com-
169–173. puting.
Kéry M. and M. Schaub (2011). Bayesian Population Analy- Tomkiewicz S. M., M. R. Fuller, J. G. Kie, and K. K. Bates
sis Using WinBUGS. Academic Press. (2010). Global positioning system and associated tech-
Kettle H., G. Holtrop, P. Louis, and H. J. Flint (2018). microP- nologies in animal behaviour and ecological research.
op: modelling microbial populations and communities in Philosophical Transactions of the Royal Society B: Biolog-
R. Methods in Ecology and Evolution 9(2), 399–409. ical Sciences 365(1550), 2163–2176.
Lande R., E. Steinar, and B.-E. Saether (2003). Stochastic Pop- Vitousek P. M., C. M. Dantonio, L. L. Loope, M. Rejmanek,
ulation Dynamics in Ecology and Conservation. Oxford and R. Westbrooks (1997). Introduced species: a signif-
University Press. icant component of human-caused global change. New
Lanner R. M. and K. F. Connor (2001). Does bristlecone pine Zealand Journal of Ecology 21(1), 1–16.
senesce? Experimental Gerontology 36(4–6) 675–685. Williams B. K., J. D. Nichols, and M. J. Conroy (2002). Anal-
Levin B. R. (1990). Microbial ecology and population biol- ysis and Management of Animal Populations. Academic
ogy. In: J. A. Hansen (ed.), Environmental Concerns: An Press.
Inter-disciplinary Exercise, 177–189. Springer. Wilson A. J., D. Reale, M. N. Clements, et al. (2011). An ecol-
Lewis S. L. and M. A. Maslin (2015). Defining the Anthro- ogist’s guide to the animal model (vol 79, p. 13, 2010).
pocene. Nature 519(7542), 171–180. Journal of Animal Ecology 80(5), 1109–1109.
Metcalf C. J. E. and S. Pavard (2007). Why evolutionary Zhao P., S. M. Liu, Y. Zhou, et al. (2021). Estimating ani-
biologists should be demographers. Trends in Ecology & mal population size with very high-resolution satellite
Evolution 22(4), 205–212. imagery. Conservation Biology 35(1), 316–324.
PART 1
Demographic Data Collection:

From Genes to Environment
CHAPTER 1
Genetic data collection, pedigrees,

and phylogenies
Emily G. Simmonds, Alina K. Niskanen, Henrik Jensen, and Steven Smith
1.1 Introduction The second way genetic data can help inform
demographic analyses is through genetic phylogenies,
Demographic studies are highly diverse; they can focus
which address the challenge of identifying drivers of
on a single population (Ozgul et al. 2010) or span many
cross-taxon variation in life history. A key consid-
taxa (Salguero-Gómez et al. 2016). They can be geo-
eration when conducting comparative studies is the
graphically restricted (Simmonds and Coulson 2015) or
independence of data points, or the lack thereof. Taxa
global (Jones et al. 2013). Different types of demograph-
that are more closely related show a tendency to have
ic study have different challenges. In this chapter, we
similar life history traits (Blomberg et al. 2003). If we
cover the use of genetic data in demographic analyses.
aim to understand the driving forces behind life his-
There are two primary ways that genetic data enter into
tory variation across multiple taxa, we need to tease
demographic analyses, and each addresses a particu-
apart the variation resulting from shared evolution-
lar challenge. These challenges will be the focus of the
ary history from that caused by environmental drivers
remainder of this chapter. The first is estimating relat-
(Freckleton 2009). To control for evolutionary history,
edness between individuals in a population, and the
we require a model of how closely related different
second is identifying drivers of cross-taxon variation
taxa are—this is a phylogeny. Without phylogenetic
in life history. These are two very different challenges,
correction, the assumptions of statistical analyses typ-
but they are linked through their connection to genetic
ically used to assess drivers of variation are violated
data.
because data points remain nonindependent (Freckle-
The first way genetic data can help inform demo-
ton 2009).
graphic analyses is through genetic pedigrees. Pedi-
Pedigrees and phylogenies are two different mod-
grees address the challenge of estimating relatedness,
els which use different types of genetic data to address
which is needed to estimate reproductive success,
different problems, but they also have some common-
a core demographic rate. To accurately quantify the
alities: they share fundamental genetic data collection
reproductive success of an individual, we need to
methods, they are models of relatedness (although at
know exactly which offspring they produced. For life-
different scales), they support demographic analyses
time reproductive success, we require a record of
rather than being a demographic method themselves,
parentage spanning several reproductive cycles. To
and the quantities they estimate allow us to ask demo-
achieve this, we use a model of the relatedness of indi-
graphic questions in a more accurate and robust way.
viduals in the population: this is called a pedigree.
In addition, both pedigrees and phylogenies have a
Once relatedness is established it is possible to calcu-
history of being based on observational characteristics
late not only reproductive success (Robinson et al. 2006;
(e.g. social behaviour and morphology) with a more
López-Sepulcre et al. 2013; Pujol et al. 2014; Reid et al.
recent transition to using genetic data (Pemberton 2008;
2014) but also other quantities, which can help cap-
Patwardhan et al. 2014; Pyron 2015). Table 1.1 provides
ture change in life history traits, such as heritability
some examples of the use of pedigrees and phylogenies
(Charmantier and Gienapp 2014; Childs et al. 2016).
in demographic studies.
Emily G. Simmonds et al., Genetic data collection, pedigrees, and phylogenies. In: Demographic Methods Across the Tree of Life.
Edited by Roberto Salguero-Gómez and Marlène Gamelon, Oxford University Press.
© Oxford University Press (2021). DOI: 10.1093/oso/9780198838609.003.0001
Table 1.1 Literature examples of using pedigree and phylogenetic information in demographic analyses.
A. Pedigrees
Study Demographic question Taxa Genetic data? How the pedigree was used
Robinson Are there trade-offs between Soay sheep (Ovis Yes To assign parentage and estimate breeding
et al. 2006 fitness components and aries) success, which was later used to represent fitness
sexually antagonistic selection
on weaponry?
López- How does the fecundity Trinidadian guppies Yes To assign parentage and estimate breeding
Sepulcre of dead males influence (Poecilia reticulata) success, which was later used to estimate
et al. 2013 population growth? contribution to population growth
Pujol et al. Is there a quantitative genetic White campion Yes To assign parentage and estimate heritability with
2014 signal of senescence? (Silene latifolia) an animal model, this was later used to investigate
patterns of additive genetic variance with age
B. Phylogenies
Study Demographic question Genetic data? How the phylogeny was used
Tidière et al. Do animals live longer in Mammals Yes To control for nonindependence of data points
2016 zoos?
Salguero- What are the environmen- Plants Yes To control for phylogenetic nonindependence
Gómez et al. tal drivers of schedules of data points and to assess the strength of
2016 of survival, growth and phylogenetic influence on demographic rates
reproduction?
Jones et al. What are the patterns of mor- Cross-order Yes To assess the strength of phylogenetic influence
2013 tality and reproduction across on demographic rates and control for phylogenetic
the Tree of Life? nonindependence of data points
In recent decades, there has been a rapid increase in success of a species or a population. Prior to the
the availability of genetic data due to improvements in availability of vast amounts of genetic material, there
noninvasive collection methods, advances in sequenc- was a bias in our understanding of demography
ing technologies, and a reduction in processing costs across the Tree of Life, resulting in over-representation
(Ekblom and Wolf 2014; Patwardhan et al. 2014; Pyron of taxa for which we could obtain observational
2015). Large amounts of high-quality genetic data are data on reproductive events or morphological fea-
now more easily obtainable, from single genes to whole tures (e.g. humans and other vertebrates) and under-
genomes. These recent advances provide an opportuni- representation of others (e.g. arthropods, Pyron 2015).
ty for genetics to be used to support analyses across a Using genetic data in demographic analyses provides a
wider range of systems and taxa. Despite the potential means to begin closing this gap, extending our under-
gains, the use of genetic data is far from ubiquitous. standing to a wider range of taxa.
While genetic data could be used across all taxa, the This chapter gives an overview of genetic pedigree
magnitude of potential gains will be unequal across and genetic phylogeny construction, from the collec-
the Tree of Life. Some species and populations have tion of genetic data, through the processing of genetic
the potential for more dramatic advances than others, material, to the construction and optimisation of the
particularly those where observational information is models ready to be used in demographic analyses. This
either missing, hard to obtain, or compromised. Mode chapter provides a background for understanding how
of reproduction, pace of life, habitat, and conserva- pedigrees and phylogenies are created so that they can
tion status all influence how easy it is to obtain reli- be used appropriately in other analyses. It is also a
able morphological or observational data that allow starting point for how to make your own pedigrees and
you to estimate relatedness and calculate reproductive phylogenies.
G E N E T I C D ATA C O L L E C T I O N, P E D I G R E E S A N D P H Y L O G E N I E S 19
1.2 Genetic pedigrees can also be determined based on the expected pair-
wise relatedness between individuals. For example,
A pedigree is a record of ancestry that depicts rela-
two full siblings are expected to share 50% (pair-
tionships between individuals. Pedigrees are used for
wise relatedness r ≈ 0.5), and two half-siblings 25%
multiple purposes in various fields, including breed-
(r ≈ 0.25) of their genomes. Building a genetic pedigree
ing programs, disease mapping, quantitative genet-
requires polymorphic genetic markers and methods
ics, behavioural ecology, and wildlife management.
to determine whether two individuals’ shared alleles
Information provided by pedigrees is important for
match expectations from Mendelian inheritance. This
estimating quantities such as the relatedness between
is often combined with information about the year
individuals, inbreeding coefficients, individual repro-
of birth (and in some programs the year of death),
ductive success (Pemberton 2008), the heritability of
or a parental and an offspring cohort, and sex of the
quantitative traits (Wilson et al. 2010; Charmantier and
individuals, which is used to define a given indi-
Gienapp 2014), and effective population size (Engen
vidual’s potential mothers, fathers, or other types of
et al. 2007). Without pedigrees it is not possible to esti-
relatives.
mate some quantities, such as number of offspring, or
Even though genetic pedigrees are usually more cor-
to follow the inheritance of traits in families (heritabili-
rect than social pedigrees (Reid et al. 2014), they are
ty and its links to demography are discussed further in
not free from errors. Problems in pedigree construc-
Chapter 20).
tion are often caused by incomplete sampling of the
The primary purpose of pedigree building is to
population and low resolution of genetic markers; as
establish links between parents and offspring, which is
a result, parents will not be assigned to all nonfounder
called parentage analysis (Flanagan and Jones 2019). In
individuals (a founder is an individual with no known
addition, other genealogical relationships can be esti-
ancestors in the population). Another common source
mated, including sibship clustering or the assignment
of error is assigning a wrong parent, which is more
of grandparents and avuncular links (Huisman 2017).
frequent in building social than genetic pedigrees, but
Pedigrees can be built using observational (social)
may happen in either one. In species with high extra-
information from the study species, genetic mark-
pair mating, such as many bird species (Brouwer and
ers (e.g. microsatellites or single nucleotide polymor-
Griffith 2019), the usage of social pedigree instead of
phisms [SNPs]), or a combination of both methods.
genetic pedigree may for example give incorrect esti-
Microsatellites are short stretches of tandemly repeat-
mates of individual reproductive success and inbreed-
ed DNA, occurring primarily in noncoding regions,
ing (Reid et al. 2014) and bias heritability estimates
that differ in the length of these repeated elements
downwards (Charmantier and Réale 2005; Bourret and
between individuals. SNPs are simple point muta-
Garant 2017). In turn, this may for example result in
tions occurring anywhere throughout the genome.
biased predictions of relationships between ecological
Social pedigrees have traditionally been built for mam-
and evolutionary dynamics (see Chapter 20).
mal and bird species, where parental care of either
one (typically maternal) or both parents is used to
infer parentage. However, social pedigrees can solve
1.3 Genetic phylogenies
only parent–offspring and some sibling relationships, Phylogenies, or phylogenetic trees, are models that rep-
and are prone to observational errors in polygamous resent the evolutionary relatedness of taxa at various
species (Charmantier and Réale 2005). Furthermore, scales, from families to species, or even genes. They are
they are impossible to obtain when immediate evi- an essential component of demographic analyses that
dence of parenthood is not available, such as in plants, look for patterns in life history variation across taxa,
corals, and some fish species. a form of comparative analysis. As discussed in the
Genetic pedigrees can solve some of the prob- introduction of this chapter (section 1.1), the primary
lems related to social pedigrees, and have become use of phylogenies in comparative demographic anal-
increasingly common along with the development of yses is to control for phylogenetic pseudoreplication.
genetic methods (Jones et al. 2010). One major advan- Without the inclusion of phylogenetic information,
tage of genetic pedigrees is that they are useful in assumptions of most comparative statistical analyses
many taxonomic groups, such as plants (Ashley 2010) cannot be met. A failure to control for evolutionary
and fish (Yue 2014). Genetic pedigrees are usually relatedness can lead to incorrect conclusions (Garland
built by identifying parents and offspring across mul- et al. 1999; McKechnie et al. 2006). This could have
tiple generations. However, other types of relatives wide-reaching consequences as phylogenetic signal
has been detected in 92% of studies on greater than 20 specific loci. Different genetic markers are present
species (Blomberg et al. 2003). in different taxa and accumulate mutations at differ-
As stated above, phylogenetic trees are models. ent rates (Nadler 1995; Patwardhan et al. 2014), and
These models are based on the principle of identifying thus different genetic markers are useful for resolving
homologous (resulting from shared ancestry) charac- evolutionary relationships over different time scales.
ters. These characters can be based on morphological For example, the resolution of deep nodes require dif-
or genetic data. Phylogenetic trees are built of branches ferent markers compared to more recent nodes.
that represent evolutionary paths, with internal nodes There are several standard markers used in molec-
(where branches join), and end with taxa at the tips of ular phylogenetics, Patwardhan et al. (2014) provide a
the branches. They are considered working hypothe- good in-depth summary of these, we provide a brief
ses (Penny et al. 1992), i.e. best estimates based on overview here. Genetic information can come from pri-
current data. Some recent studies present a full distri- mary data collection or from secondary sources. There
bution of thousands of ‘best’ trees rather than selecting are a wide range of online repositories of genetic infor-
a single optimal topology (shape) in recognition of the mation. One of the biggest and most well-known is
uncertainty in estimation (Jetz et al. 2012). Phyloge- GenBank (Benson et al. 2013). Possible genetic markers
netic trees, and the statistical analyses associated with can be sourced from multiple parts of a cell (e.g.
them, are underpinned by several assumptions, one of the nucleus, mitochondria, and chloroplasts). Nuclear
which is a model of trait evolution (Freckleton 2009). material, often DNA encoding rRNA, is slow evolv-
This is distinct from models of sequence evolution dis- ing and useful for resolving distant relationships (Pat-
cussed in section 1.7.2. The most common model of trait wardhan et al. 2014). Some common examples of nucle-
evolution is Brownian motion, which assumes no trait ar DNA markers are Internal Transcribed Spacer (ITS)
optimum (Freckleton 2009). But others also exist, such Tippery and Les 2008), 16S (Vences et al. 2005), and
as the Ornstein–Uhlenbeck process, which does have 28S (Vences et al. 2005). Mitochondrial DNA is faster
an optimum and represents a trait evolving under sta- evolving in animals and most useful for species-level
bilising selection (Blomberg et al. 2003). All of these relationships. Some common mitochondrial markers
models give the expected trait covariance for a certain are Cyclooxygenase 1 (COX1) (Vences et al. 2005),
level of phylogenetic relatedness (Freckleton 2009). The 12S (Vences et al. 2005), and, for close relationships,
assumption of how traits have evolved can influence Cytochrome-b (Patwardhan et al. 2014). For plants
the amount of phylogenetic signal detected because only, it is also possible to use chloroplast DNA to
it influences the evolutionary distance between taxa resolve mainly species-level relationships. Some com-
(Blomberg et al. 2003). mon examples of chloroplast markers are ribulose
Phylogenetic signal is the strength of statistical bisphosphate carboxylase large chain (rbcL) (Morgan
dependence of species’ trait values as a result of their and Soltis 1993) and noncoding ribosomal protein L16
phylogenetic relationships (Revell et al. 2008). Beyond (rpl16) (Zhang 2000). These are just a few examples: in
the necessity of controlling for nonindependence of reality, there are many more.
data points, understanding the strength of phylogenet- Despite the gains that have been achieved through
ic signal in shaping life history variation can also be the incorporation of genetic data in phylogenetic anal-
of scientific interest (Jones et al. 2013; Salguero-Gómez yses, genetic material does not create limitation-free
et al. 2016). Testing for, or quantifying, phylogenetic phylogenies. Some of the same limitations of morpho-
signal is most commonly done through two methods: logical data are also present in genetic data: (1) homol-
phylogenetically independent contrasts (Felsenstein ogy of the characters can be hard to establish (Nadler
1985b; Blomberg et al. 2003; Freckleton 2009) and phy- 1995) because convergent evolution can still occur at
logenetic generalised least squares (PGLSs) (Blomberg the genetic level (Stern 2013), (2) within-species varia-
et al. 2003; Freckleton 2009) (discussed in more detail tion can lead to incorrect relationship attribution (Har-
in Chapter 18). ris and Crandall 2000), and (3) interpretation can be
Genetic phylogenies are the cornerstone of mod- challenging. The latter occurs because gene divergence
ern phylogenetic analyses. They provide many more often precedes species change, resulting in gene trees
characters than morphological data and are the pre- that do not match species trees (Nei and Saitou 1987;
dominant form of phylogenetic tree for extant taxa Nadler 1995), a result which can also result from intro-
(Nadler 1995). The genetic variation that can be used gression due to hybridisation. For comparative analy-
in phylogenetic analyses varies from whole genomes ses in particular, a drawback of existing phylogenies
to shorter DNA sequences, or a handful of SNPs at can be that they are limited in size. Finding a published
phylogenetic tree covering all the taxa in the compar- analyses but also, perhaps more importantly, what is
ative demographic analysis may not be possible. To available, physically and ethically, for collection. High-
overcome this, phylogenetic supertrees can be created quality DNA can be sourced from many origins (see
(Sanderson et al. 1998; Davies et al. 2004; Bininda- Table 1.2) but the choice varies with the requirements
Emonds et al. 2007). These supertrees are constructed of the project and the target taxa. Sequencing of mito-
from several smaller trees to create a larger tree that chondrial DNA typically requires the lowest starting
encompasses more taxa than have ever been included amount of material (e.g. trace material) whereas oth-
in a single analysis. er top-end high-throughput approaches such as next
In recognition of some of the limitations of genetic generation sequencing (NGS) will require higher quan-
and morphological data, there has been a more recent tity and quality starting material (blood/tissue). For
trend to revive ideas of combining both sources of many projects, the ability to collect the ideal sample
information into total-evidence phylogenetics (Pyron material is restricted due to target organism behaviour
2015; Gavryushkina et al. 2017). Combining the merits and/or ethical limitations. In these situations, nonin-
of each type of data can improve congruence of vasive sampling may be the only option (Morin and
results and permits incorporation of the fossil record. Woodruff 1996; Waits and Paetkau 2005), and this will
Inclusion of fossils can often be useful for calibrating require streamlined protocols to maximise the recovery
trees (Davies et al. 2004; Bininda-Emonds et al. 2007). of DNA that is in a suboptimal state (degraded and in a
Demographers working on cross-species comparative matrix of other DNA, and/or accompanied by analysis
questions should remain abreast of developments in inhibitors).
phylogenetics to ensure their inference is based on Sample transport and storage is clearly of
the latest working hypothesis. This is particularly paramount importance. Optimal methods include:
pertinent as phylogenies used in demographic analy- sample freezing immediately on dry ice or liquid
ses have been found to lag behind statistical advances nitrogen, storage in a buffering solution to protect
(Freckleton 2009). the DNA (e.g. Longmire et al. 1997), or immediate
processing to isolate DNA before freezing. However,
the ideal method of storage and transport are not
1.4 Collection of genetic samples always available in many field situations. Often,
When genetic samples are to be collected there are lack of electrical power or appropriate facilities will
two main considerations: what material to collect and preclude many preferred methods. Alternatives
how best to bring it to the lab. The first point will such as desiccation of scat samples (Roeder et al.
require multilayered decisions based upon the planned 2004) or drying blood or trace materials on specially
Table 1.2 Summary of potential source materials and handling considerations for DNA analysis.
Sampling technique Suitable taxa Suitable analyses Limitations Sample storage
Blood Vertebrates (especial- Microsatellites, SNPs, mtDNA, Often contains inhibitors, −20ºC, ethanol, buffer,
ly birds, reptiles, and whole genome, Amplicon Seq permission needed dried
amphibians)
Cuttings/biopsy Corals, plants, jellyfish, Microsatellites, SNPs, mtDNA, Often need ethics permission 4ºC, ethanol, buffer
animal organs Whole genome, Amplicon Seq
Faeces Vertebrates/intestinal Microsatellites, SNPs, mtDNA, Often contains inhibitors and −20◦ C, ethanol
parasites Amplicon Seq mixed species
Hair Mammals Microsatellites, SNPs, mtDNA, Low quantity and quality DNA 4◦ C, dried
Amplicon Seq
Shed skin Reptiles, whales Microsatellites, SNPs, mtDNA, Low quantity and quality DNA 4◦ C, dried
Amplicon Seq
Feathers Birds Microsatellites, SNPs, mtDNA, Low quantity and quality DNA 4◦ C, dried
Amplicon Seq
Swabs Bacteria/fungi Microsatellites, SNPs, mtDNA, Often contains inhibitors and 4◦ C, dried
Amplicon Seq mixed species
designed DNA preservation cards are an option for can be divided into three stages which are iterative-
such extreme situations (Smith and Burgoyne 2004). ly repeated in cycles. Each cycle consists of denat-
Typically, however, there is a trade off in these cases uration (when double stranded DNA is denatured
in terms of DNA recovery and price of collection. into two single stranded template strands), annealing
(short oligonucleotide primers targeting the region of
interest, bind to the complementary single stranded
1.5 Processing of samples (DNA template ends), and extension (elongation of oligonu-
extraction, amplification, cleotide primers by the activity of polymerase and the
addition of DNA building blocks called dNTPs, that
and sequencing) result in a DNA strand which is complementary to the
Nucleic acid extraction is the critical starting point in template).
every molecular genetic analysis as the recovery of For pedigree construction, relationships between
high-quality DNA is a key factor for the success of individuals are generally determined via comparisons
all downstream applications. A variety of extraction of genotypic data at microsatellite or SNP loci. SNPs
methods are available to maximise the possibility (single nucleotide polymorphisms) are mutations, or
of getting high-quality DNA and minimise degrada- polymorphisms, which can be compared between indi-
tion, but also to efficiently match requirements of viduals who will possess one (homozygous individu-
time, materials, and labour costs. Generally, extrac- als) or two (heterozygous individuals) of the generally
tion methods can be divided into two main groups: two possible variants at a given locus. At a given
solution-based or column-based protocols. Most pro- microsatellite locus, length variations can lead to more
tocols have been adopted into commercial kits that are than 10 distinct allelic variants. Microsatellite loci are
simple, streamlined methods for nucleic acid extrac- normally screened via some form of gel electrophore-
tion but are usually designed for specific starting sis to separate alleles based on their size compared to
materials. a known standard. Recently NGS techniques have also
In general, all successful nucleic acid extraction been developed to sequence the amplified loci in mul-
methods require a variation on the same important tiplex format and visualise length differences via an
steps (Sambrook et al. 1989). First is the effective dis- alignment (Vartia et al. 2016). There are a multitude of
ruption of cells and tissues, followed by the denatu- methods for SNP genotyping, but all revolve around
ration of nucleoprotein complexes. This process will interrogating which variant is present at the targeted
differ depending on sample type but basically involves polymorphic position. Some make use of different flu-
physical homogenisation of tissues and chemical cell orescent dyes for the two possible nucleotides that can
lysis. Often, detergents are used in cell lysis buffers but be incorporated at the site while others detect thermo-
also chloroform or EDTA may be suitable for certain kinetic differences in the amplified alleles based on the
cell types. The next step is inactivation of nucleas- contrasting nucleotide compositions. A recent NGS-
es and elimination of all contaminants (e.g. proteins, based method for SNP genotyping known as RAD
carbohydrates, or lipids), which can be achieved by sequencing (restriction associated DNA sequencing,
buffering with Tris-EDTA, enzymatic protein removal, Baird et al. 2008) uses restriction enzymes and DNA
and centrifugation. DNA is then isolated either by target capture to simultaneously discover and screen
precipitation with isopropanol or binding to silica large numbers of individuals at thousands of SNP loci
(often in a column filter) and washed with ethanol across a reduced representation of the entire genome.
to remove salts. The final step is the elution of DNA Sequencing options for phylogenetic tree recon-
into a storage buffer, most commonly Tris-EDTA, and struction range from the traditional Sanger sequencing
then storage until further processing (Smith and Morin (Sanger et al. 1977) to more sophisticated massively
2005). parallel sequencing (Metzker 2010). For Sanger
Typically, for downstream processing of DNA, it is sequencing, the skill set required is within the scope
desirable to investigate a particular gene or region of most government and academic laboratories.
of interest. This requires the amplification of the Throughput is scalable, but mainly in the medium
region of interest so that there are enough copies range, with 96 samples easily processed within a
to visualise via sequencing or genotyping meth- week for a sequence length of 700–1000 base pairs. A
ods. Amplification is achieved via polymerase chain much higher throughput can be achieved via the NGS
reaction (PCR). The idea behind PCR is simple and techniques. A popular application of NGS is amplicon
sequencing (Schirmer et al. 2015) that, in its simplest portions that relate to priming sequences, individual
form, involves standard PCR amplification of a target ID tags, or sequencing adaptors. Quality filtering can
genomic region using primers that have been modified be achieved in most cases by taking advantage of qual-
to include an individual ID tag and an adaptor used ity scores that are provided by the data provider or
to initiate the sequencing. Many individuals with instrument. Low-quality data can then be flagged and
different individual tags can be PCR amplified and excluded from further analysis.
then pooled in equimolar amounts for sequencing. For microsatellite and SNP data, allelic information
Subsequent data filtering steps separate out the indi- (or missing data values) need to be assigned at every
viduals according to their specific tag. Rather than locus across all individuals to allow comparison and
sequencing the entire genome of an organism, this inference of relatedness. Automation of this process is
approach targets one or more gene regions of interest often essential and is provided by software packages
for deep sequencing. As less of the genome is being that allow a user to assign alleles based on their con-
sequenced, more individuals can be pooled in each formity to certain size criteria or signal to noise ratios
sequencing run to greatly reduce the costs per sample. specific to the instrument generating the data (e.g.
Genemapper, Genemarker for microsatellites; ASSIsT,
Illumina GenCall for SNPs). The end result is a matrix
1.6 Preparation of data for analyses of multilocus genotypes for each individual which will
Once data have been generated for a project there are then be used as input for pedigree or phylogeny recon-
many steps that need to be performed before it is suit- struction.
able for analysis to construct pedigrees or phylogenies. For sequence data the researcher must make the data
These steps are similar for all data types but will vary in comparable across individuals and taxa. It is impor-
specific details according to the marker. Usually, these tant to align the sequences such that the corresponding
preparation steps can be incorporated into scripts or base-pair positions reflect homology and any differ-
bioinformatics pipelines, and often there is pre-existing ences relate to evolutionary events. In this way, it is
software that facilitates the process. The first step is clear where a base has mutated to create a polymor-
data cleaning and filtering for quality (Van Oosterhout phism or has been lost (deletion) or gained (inser-
et al. 2004; Teo 2008; Bokulich et al. 2013). Particular- tion). These are the important sequence differences
ly with sequencing data, there is a need to trim off any that will be used when constructing the phylogeny to
Table 1.3 Multiple sequence alignment software options for nucleotide sequence data.
Name Description Data set size Availability URL
CLC Main GUI (graphical user interface) allows Memory dependent Commercial https://www.qiagenbioinformatics.com/
Workbench integration with many add on tools for licence products/clc-main-workbench/
sequence analysis
Clustal2 Classic progressive alignment tool Memory dependent Free http://www.clustal.org/clustal2/
available as command line or GUI
Clustal Uses seeded guide trees and HMM >100,000 sequences Free http://www.clustal.org/omega/
Omega profile-profile techniques to generate
alignments
CodonCode Supports quality scores. Features include Memory dependent Commercial https://www.codoncode.com/aligner/
Aligner mutation detection and BLAST searching licence
Kalign Fast tool focuses on local region <2,000 sequences Free http://msa.cgb.ki.se/cgi-bin/msa.cgi
alignment
MAFFT Sequences clustered using Fast Fourier <30,000 sequences Free under https://mafft.cbrc.jp/alignment/software/
Transforms BSD licence
MUSCLE 3-stage progressive alignment Memory dependent, Free http://www.drive5.com/muscle/
recommended <10,000
sequences
determine evolutionary relationships. The alignment relationships. Increasing the number and/or polymor-
can be greatly facilitated by mapping to an existing phism of markers increases the power to establish
reference sequence if one is available. If not, multi- correct pedigree links (Harrison et al. 2013; Aykanat
ple sequence alignments will need to be made often et al. 2014). Genotyping errors and mutations may
with the help of a software package incorporating algo- cause non-Mendelian mismatches between parent and
rithms to reflect probabilities of various mutational offspring alleles. Ideally, a genotyping error rate is esti-
processes (see Table 1.3). The resulting output is a series mated from the data and used as a model parameter,
of aligned sequences with mutated positions highlight- or if that is not possible, a predefined number of
ed and ready for the extent of divergence to be analysed mismatches are allowed in the pedigree construction.
according to an appropriate evolutionary model. Some likelihood-based and Bayesian software estimate
the error rate simultaneously with the pedigree (see e.g.
Wang 2004).
1.7 Using the data to construct In addition to genotype data, various kinds of
pedigrees and phylogenies population-level ecological and individual phenotyp-
ic data can be used to reduce the number of model
1.7.1 Pedigrees assumptions one must make, and aid pedigree con-
After preparing the genetic data as described in section struction. The type of information used depends on the
1.6, a pedigree can be constructed using one of three software; the year of birth may be included as a cue
main methods: exclusion method, relatedness-based to solve which individual is an offspring and which
method, and likelihood-based method (Flanagan and a parent: for example, in Sequoia by Huisman (2017)
Jones 2019). The exclusion method excludes all poten- and FRANz by Riester et al. (2009). Other software,
tial parents that do not share at least one allele per such as Cervus (Kalinowski et al. 2007), MasterBayes
locus with an offspring, but allow for some mismatch- (Hadfield et al. 2006), and Colony (Wang 2004, 2012;
ing loci due to genotyping errors (Calus et al. 2011; Jones and Wang 2010), use cohort-specific parental lists
Hayes 2011). It is fast due to simplicity, but it may to assign parents. The group of potential parents is lim-
lead to assignment of incorrect parents when multi- ited by the year of death in some programs (FRANz),
ple potential parents share alleles with the offspring and some programs (e.g. Cervus, Colony, and FRANz)
(Huisman 2017). Relatedness-based methods estimate can make use of information on the proportion of the
pairwise relatedness (r) or kinship coefficients between adult population which is included in the list of can-
individuals. The relatedness estimates are then classi- didate parents. If a previous version of a pedigree
fied as first, and second-degree relatives, or unrelated exists for the study system (FRANz), or information
(Thompson 1975). When data contain overlapping gen- is available on expected relatedness between poten-
erations, other methods are needed to tell apart parent– tial parents or between parents and offspring (Cervus),
offspring pairs, full siblings, and other related pairs of this can be incorporated to aid parental inference.
individuals that share the same degree of relatedness. Finally, some programs can allow for focal popula-
Likelihood-based methods compare the likelihood of tions having various kinds of mating systems (e.g.
a specific relationship (e.g. parent–offspring relation- Colony). The correctness of a genetic pedigree can be
ship) compared to other potential relationships given estimated using multiple methods (Jones et al. 2010),
the genotypes of the individuals (Flanagan and Jones including: (1) simulation of genotype data sets for
2019). pedigree building to determine error rate in the sim-
Pedigree construction methods usually make vari- ulated pedigrees compared to the actual constructed
ous assumptions regarding the study population and pedigree (Kalinowski et al. 2007), (2) Bayesian poste-
genetic data, but all require polymorphic genetic mark- rior probabilities (Nielsen et al. 2001), and (3) simula-
ers, such as microsatellites or SNPs, that are in low tion of an experiment-wise expected error rate (Jones
linkage disequilibrium (e.g. Flanagan and Jones 2019). 2001).
Informative, highly heterozygous loci are required; Genetic data and software used to build pedi-
minor allele frequency (MAF) is often used to assess the grees do not provide a single answer for estimates of
selection of SNP loci. The number of markers needed relatedness between individuals in a population. Pedi-
depends on the species, mating system, and sampling grees can be constructed using a range of method-
effort (Aykanat et al. 2014). Typically, 10–20 microsatel- ological approaches and different sources of genetic
lites or a few hundred SNPs are needed to gain information. Each of these choices alters the degree
sufficient power to distinguish individuals and their of relatedness that can be established. The variety of
Table 1.4 Results from genetic parentage analyses carried out in two house sparrow (Passer domesticus) populations that are part of an
individual-based long-term study of an insular metapopulation at the coast of northern Norway (see e.g. Niskanen et al. 2020).
Population Relationship link/running time 50 SNPs 605 SNPs

Cervus Colony FRANz Sequoia Cervus Colony FRANz Sequoia
Aldra Maternal links 12 12 12 11 12 12 12 12

Paternal links 10 10 10 10 10 10 10 10
Fullsib links 0 0a 0 0a 0 0a 0 0a
Running time (s) 149.5 7.0 3.0 0.6 781.4 59.0 31.0 4.7
Hestmannøy Maternal links 57 56 56 51 54 54 54 54
Paternal links 57 58 58 53 59 57 57 57
Fullsib links 16 22b 16 14b 16 23b 16 22b
Running time (s) 1,750.1 412.2 3.0 4.6 20,409.4 1,332.0 30.0 30.7
a
Sibship clustering option included, but no dummy parents were found.
b
Fullsib links include link(s) due to dummy parents.
results that can be obtained is illustrated in Table 1.4 Pedigree information is stored as a simple three-
and Figure 1.1, which present a case study of parentage column file, which includes a focal individual, a
analyses using different amounts of genetic informa- first parent (either dam or sire), and a second par-
tion and different construction methods. In this case ent (either dam or sire). These parental relationships,
study, parentage was determined for house sparrow forming the pedigree, can be visualised using for
(Passer domesticus) recruits that hatched on two islands example programs Pedigree Viewer (Kinghorn 1994)
in northern Norway in 2008 (n = 13 on Aldra; n = 76 on or R package Pedantics (Morrissey and Wilson 2010).
Hestmannøy). All sampled adult birds present on the Pedigree statistics, including the number of parental
islands that year were included as putative parents. and other family links, depth of a pedigree (number of
The two populations reflect natural populations that generations), and number of founders, can be acquired
differ in demographic histories: Aldra was colonised using, for example, the Pedantics package. Pedigrees
in 1998 and has a relatively small population with are usually built for the specific study system to be able
a high level of inbreeding; the Hestmannøy popula- to answer research questions arising from that data.
tion is larger and more stable and has a low level Pedigree information can then be incorporated into
of inbreeding (Niskanen et al. 2020). The estimat- demographic analyses to estimate individual repro-
ed adult population size in 2008 was 43 individuals ductive success, inbreeding coefficients, and pairwise
on Aldra and 162 on Hestmannøy, of which 40 and relatedness of individuals (e.g. López-Sepulcre et al.
137 were sampled and genotyped, respectively. The 2013; Aykanat et al. 2014; Pujol et al. 2014).
birds were genotyped on a custom Affymetrix Axiom
SNP-array for house sparrows, including 200,000 SNPs
(Lundregan et al. 2018). In the parentage analyses 1.7.2 Phylogenies
we used either a selection of 605 independent SNPs Reconstruction of molecular phylogenies to use in
with a high level of heterozygosity in more than comparative demographic analyses takes several key
3000 birds present in the total metapopulation dur- steps: choosing an appropriate model of sequence
ing the years 1998–2013 (see Niskanen et al. 2020 evolution, establishing genetic characters, calculating
for details), or a random subset consisting of 50 pairwise distances between sequences, phylogenetic
of these 605 SNPs. Additional information used by tree reconstruction, and phylogenetic tree evaluation.
the different software was: sex of candidate parents Not all steps are used in all analyses or in the same
(all four software packages), estimated adult popula- way, but the general procedure is similar across all
tion sizes (i.e. proportion of candidate parents sam- phylogenetic analyses. Here we give an overview of
pled: Cervus, Colony, and FRANz), information on this general procedure and an introduction into some
hatch year of candidate parents (Sequoia and FRANz), of the variety available at each step. In this section we
estimates of relatedness between true parents and assume that phylogenetic reconstruction is being per-
other candidate parents (Cervus), and marker genotyp- formed at the species level, but different taxonomic
ing error rates (0.002: all four software packages). resolutions are also possible.
(a) distinguish homoplasy from homology and to estimate

2001 how much change has occurred between sequences,
it is necessary to assume a model of sequence evolu-
2002 tion (Zhang and Kumar 1997; Kalyaanamoorthy et al.
2017). Information on the amount of change between
2003
sequences forms the basis of estimating phylogenet-
2004 ic relatedness and is, thus, an integral part of tree
reconstruction. Models of sequence evolution provide
2005 a mathematical illustration of how often nucleotide or
amino acid replacements would be expected to occur
2006 (Whelan et al. 2001) and therefore influence the num-
ber of characters estimated from a particular set of
2007
sequences. Models of sequence evolution are used in
2008 all phylogenetic analyses whether they are formalised
and explicit, or not (Kelchner and Thomas 2007). Many
(b) different models of sequence evolution exist, and they
sex all include different assumptions. Two common exam-
female ples are the Jukes–Cantor one-parameter model (Jukes
male
Number of adult birds
30 and Cantor 1969), which assumes equal probability of

all substitutions and equal frequency of all nucleotides,
and the Kimura two-parameter model (Kimura 1980),
20 which gives independent probabilities to transversions
and transitions but still assumes equal frequency of
all nucleotides. A model of sequence evolution can
10 be selected using a program that runs statistical mod-
el selection for models of nucleotide substitution; two
examples are jModelTest (Posada 2008) and Partition-
0
Finder 2 (Lanfear et al. 2017), but many more exist.
0 1 2 3 4 Once a model of evolution has been selected, the
Number of offspring exact number of genetic characters can be established.
The next step is to use the characters to construct a phy-
Figure 1.1 (a) The pedigree constructed by Sequoia for the insular
house sparrow population on Hestmannøy using 605 SNPs as logenetic tree. Again, there are a multitude of options
explained in the text for Table 1.4. Individuals are presented as dots, available, implemented in different programs, with
and lines connect a parent to its offspring. Black dots and lines show different optimisation principles, benefits, and draw-
females and any links they have to offspring, and grey dots and links backs. These are summarised in Table 1.5. There is an
show males and any links they have to offspring. The y-axis position extensive literature on the different methods (see Yang
shows the hatch year of an individual. (b) The frequency distribution of and Rannala 2012, for a nice summary), we introduce
recruits in the 2008 cohort produced by adult females (black) and some of the most common methodologies here.
males (grey) present on Hestmannøy that year.
The first distinction in phylogenetic tree topology is
whether the tree is rooted or unrooted. Trees that are
When the data have been prepared using meth- rooted have a basal ancestor, whereas unrooted/free
ods outlined in section 1.6, they can then be used for forming trees do not include a known root. Trees can
phylogenetic tree reconstruction. Alignment suggests be rooted via an outgroup (dictating the most basal
homology, however, to reach an exact string of char- relationship in the tree) or via time, using assump-
acters, we require a model of how the sequences have tions about the rate of molecular evolution (Yang and
evolved. For instance, an observation of A (adenine) Rannala 2012).
at a particular locus in two aligned sequences could Once you have a final tree generated with any
have arisen from no evolutionary change (homology) approach, it is vital to evaluate the tree because the
or from multiple changes to the same locus, such as topology of a tree and its branch lengths are estimated
from A to G (guanine) and back to A. We call the lat- with uncertainty. There are several standard methods
ter homoplasy, when two characters look the same for estimating the uncertainty associated with nodes
but actually result from independent evolution. To (where branches join). The primary (non-Bayesian)
Table 1.5 Summary of phylogenetic reconstruction methods.
Name of approach: Optimisation principle:
Distance-matrix methods To minimise the tree length. The exact metric used to measure tree length and the method employed to minimise it vary.
Description Key benefits Key drawbacks Example programs
These methods work in two steps: Fast computation, particularly for cluster Sensitive to gaps or errors in sequences RapidNJ (Simonsen et al. 2008)
First, they take the whole sequences of genetic algorithms where fewer trees are compared (Yang and Rannala 2012; Xia 2018). NINJA (Wheeler 2009)
information and calculate pairwise differences (Yang and Rannala 2012). Ignores information in sequences beyond the PAUP* (Swofford 2002)
based on a model of evolution and use these to Popular for big data sets (Xia 2018). pairwise comparison. (But see Xia 2018
generate a distance matrix (Yang and Rannala Can be used with pairwise sequence for more recent solutions).
2012). It does not use specific characters. alignments, rather than needing multiple
Second, the distance matrix is used to construct a tree. sequence alignment (Thorne and Kishino
There are several different methods available to 1992, Xia 2018).
build trees from distance matrices, for example:
Clustering algorithm methods, such as UPGMA (Sneath
and Sokal 1973) and Neighbour-joining (Nei and
Saitou 1987).
Least squares (Cavalli-Sforza and Edwards 1967).
Minimum evolution (Rzhetsky and Nei 1992).
Maximum parsimony To minimise the number of character changes in the tree. There are various methods to minimise character changes.
This is a character-based method, so requires multiple Simplicity of philosophy and computationally Does not include an explicit model of PAUP* (Swofford 2002)
sequence alignment. efficient (Yang and Rannala 2012). evolution and therefore contains MEGA X (Kumar et al. 2018)
The principle is based on building all trees possible and Less biased than other methods when nonexplicit assumptions. TNT (Giribet 2005)
finding the one that minimises the number of heterogenous evolution occurs (Kolaczkowski Can suffer from long-branch attraction
character changes (Farris et al. 1970). In reality, this and Thornton 2004). (Felsenstein 1978). This is a phenomenon
would be too time consuming. Therefore, trees are where two taxa are incorrectly inferred to
built and searched using an algorithm developed by be closely related due to having long
Fitch (1970) to find the topology that minimises the branches (i.e. having a large number of
number of character changes. character changes since their common
Maximum parsimony does not include an explicit ancestor).
model of sequence evolution. The implicit model is
that all substitutions are equally likely and that
reversion does not occur (Yang and Rannala 2012).
continued
Table 1.5 Continued
Maximum likelihood (ML) Chooses the tree that maximises the likelihood given the data.
This is a character-based method, so requires multiple Heuristic searches so computationally efficient Can suffer from long-branch attraction if the RaxML (Stamatakis 2014) (heuristic
sequence alignment. for larger data sets. model of sequence evolution is incorrect search)
Uses an explicit model of sequence evolution. No long-branch attraction if the sequence (Yang and Rannala 2012). GARLI (Zwickl 2006)
This method is implemented in two steps. The first evolution model is sufficient. Can be biased in some scenarios PAUP* (Swofford 2002)
optimises the branches and the likelihood of the Robust to certain violations of assumptions and (Kolaczkowski and Thornton 2004). IQ-TREE (Nguyen et al. 2015)
tree calculated. The trees are then searched using errors (Kalyaanamoorthy et al. 2017). Can be computationally demanding.
iterative optimisation algorithms to find the tree Can include complex models of sequence
with the maximum likelihood value (Yang and evolution (Yang and Rannala 2012).
Rannala 2012).
Bayesian estimation Chooses the tree with best posterior support given a priori and the data.
This is a character-based method, so requires multiple No long-branch attraction if the model of Incorrect use of priors can result in incorrect MrBayes (Ronquist and Huelsenbeck
sequence alignment. sequence evolution is sufficient. trees (Nascimento et al. 2017). 2003)
Uses an explicit model of sequence evolution. Can include complex models of sequence Can be biased in some scenarios BEAST (Bouckaert et al. 2014) (a more
Bayesian methods also use the likelihood of the tree evolution (Yang and Rannala 2012). (Kolaczkowski and Thornton 2004). extensive list can be found in
but estimate parameters following Bayesian Uncertainty has a more intuitive interpretation Convergence can be difficult to achieve and Nascimento et al. (2017))
principles, using priors and producing a posterior than other methods (Yang and Rannala computationally demanding.
distribution for each parameter (see Rannala and 2012). Posterior probabilities can be overinflated
Yang 1996; Yang and Rannala 1997 for more (Yang and Rannala 2012).
details).
approach to evaluating phylogenetic trees is the and phylogenies are now among the fundamental tools
bootstrap (Felsenstein 1985a) analyses. Bootstrapping, of demographic analyses. These data are a comple-
the most popular method for determining branch sup- ment to the rest of the demographic toolkit, and, com-
port in phylogenetics (Holmes 2003), involves creating bined with other data sources, they can help create
a new data set the same size as the original, by ran- well-informed analyses and allow us to address new
domly sampling columns from the original data set, biological questions across the Tree of Life.
with replacement. This is done a set number of times
and a new tree constructed each time. The bootstrap
support of a node is defined as the percentage of boot-
strap trees in which the clade associated with that References
node was recovered. For trees created from Bayesian Ashley M. V. (2010). Plant parentage, pollination, and dis-
analyses, posterior probabilities are used to represent persal: how DNA microsatellites have altered the land-
uncertainty. scape. Critical Reviews in Plant Sciences 29(3), 148–161.
After the creation of a phylogeny using the meth- doi: 10.1080/07352689.2010.481167.
ods outlined above, the tree is ready to be used in a Aykanat T., Johnston, S. E., Cotter, D., et al. (2014). Molecular
pedigree reconstruction and estimation of evolutionary
comparative analysis. The phylogenetic construction
parameters in a wild Atlantic salmon river system with
methods detailed above are complex and require some
incomplete sampling: a power analysis. BMC Evolution-
level of expertise to implement them; however, recon- ary Biology 14(68), 1–17. doi: 10.1186/1471-2148-14-68.
structed phylogenies are also widely available. Trees Baird N. A., Etter P. D., Atwood T. S., et al. (2008). Rapid
can be obtained from published papers (see Chaw et al. SNP discovery and genetic mapping using sequenced
1997 for a gymnosperm phylogeny; Bininda-Emonds RAD markers. PLOS One 3(10), e3376. doi: 10.1371/jour-
et al. 1999 for a phylogeny of mammals; Qian and Jin nal.pone.0003376.
2016 for a vascular plant phylogeny) and from online Benson D. A., Cavanaugh M., Clark K., et al. (2013).
repositories such as PHYLACINE (Faurby 2019). Phy- GenBank. Nucleic Acids Research 41(D1), D36–D42. doi:
logenies that can be taken from these sources might 10.1093/nar/gkw1070.
Bininda-Emonds O. R. P., Gittleman J. L., and Purvis A.
need to be manipulated prior to use in a comparative
(1999). Building large trees by combining phylogenetic
analysis: for example, they might need pruning (some
information: a complete phylogeny of the extant Carnivo-
taxa removed) or change of species names to match ra (Mammalia). Biological Reviews 74(2), 143–175. doi:
the demographic data or a supertree may need to be 10.1111/j.1469-185X.1999.tb00184.x.
created so that the phylogeny covers all taxa in the Bininda-Emonds O. R. P., Cardillo M., Jones K. E., et al.
analysis (see Sanderson et al. 1998; Bininda-Emonds (2007). The delayed rise of present-day mammals.
et al. 2007 for examples). Lots of tools exist to manip- Nature 446(7135), 507–512. http//.doi.org.10.1038/nature
ulate phylogenetic trees to ensure they are in the right 05634.
format for a demographic analysis, see https://cran.r- Blomberg S. P., Garland T., and Ives A. R. (2003). Test-
project.org/web/views/phylogenetics.html for a list ing for phylogenetic signal in comparative data: behav-
ioral traits are more labile. Evolution 57(4), 717–745. doi:
of the options in R. More detail on comparative anal-
10.1111/j.0014-3820.2003.tb00285.x.
yses is available in Chapter 18.
Bokulich N. A., Subramanian S., Faith J. J., et al. (2013).
Quality-filtering vastly improves diversity estimates from
1.8 Conclusions Illumina amplicon sequencing. Nature Method 10, 57–59.
doi: 10.1038/nmeth.2276.
Over the past few decades there have been many Bouckaert R., Heled J., Kühnert D., et al. (2014). BEAST 2:
advances in genetic analyses that we can now exploit a software platform for Bayesian evolutionary analysis.
in demographic methods. Opportunities to use genet- PLOS Computational Biology 10(4), e1003537. doi:
ic information exist across the Tree of Life. While some 10.1371/journal.pcbi.1003537.
methodological practices differ by taxon, the benefits Bourret A. and Garant D. (2017). An assessment of the reli-
ability of quantitative genetics estimates in study systems
of using genetic data in demographic methods are
with high rate of extra-pair reproduction and low recruit-
truly cross-taxonomic. The types of models described
ment. Heredity 118, 229–238. doi: 10.1038/hdy.2016.92.
in this chapter provide a supporting role in demo- Brouwer L. and Griffith S. C. (2019). Extra-pair paternity in
graphic analyses. They are used to tackle two distinct birds. Molecular Ecology 28(22), 4864–4882. doi: 10.1111
challenges, but both can improve the accuracy and /mec.15259.
robustness of our estimates of demographic quantities Calus M. P. L., Mulder H. A., and Bastiaansen J. W. M. (2011).
and patterns. Genetic data and the resulting pedigrees Identification of Mendelian inconsistencies between SNP
and pedigree information of sibs. Genetics Selection Evo- Freckleton R. P. (2009). The seven deadly sins of
lution 43(34), 1–12. doi: 10.1186/1297-9686-43-34. comparative analysis. Journal of Evolutionary Biol-
Cavalli-Sforza L. L. and Edwards A. W. F. (1967). Phyloge- ogy 22(7), 1367–1375. doi: 10.1111/j.1420-9101.2009
netic analysis: models and estimation procedures. Society .01757.x.
for the Study of Evolution 21(3), 550–570. Garland T., Midford P. E., and Ives A. R. (1999). An
Charmantier A. and Gienapp P. (2014). Climate change introduction to phylogenetically based statistical meth-
and timing of avian breeding and migration: evolutionary ods, with a new method for confidence intervals on
versus plastic changes. Evolutionary Applications 7(1), ancestral values. American Zoologist 39(2), 374–388. doi:
15–28. doi: 10.1111/eva.12126. 10.1093/icb/39.2.374.
Charmantier A. and Réale D. (2005). How do misassigned Gavryushkina A., Heath T. A., Ksepka D. T., Stadler T.,
paternities affect the estimation of heritability in the wild? Welch D., and Drummond A. J. (2017). Bayesian total-
Molecular Ecology 14(9), 2839–2850. doi: 10.1111/j.1365- evidence dating reveals the recent crown radiation of
294X.2005.02619.x. penguins. Systematic Biology 66(1), 57–73. doi: 10.1093/
Chaw S. M., Zharkikh A., Sung H. M., Lau T. C., and Li W. sysbio/syw060.
H. (1997). Molecular phylogeny of extant gymnosperms Giribet G. (2005). TNT: tree analysis using new technol-
and seed plant evolution: analysis of nuclear 18s rRNA ogy. Systematic Biology 54(1), 176–178. doi: 10.1080/
sequences. Molecular Biology and Evolution 14(1), 56–68. 10635150590905830.
doi: 10.1093/oxfordjournals.molbev.a025702. Hadfield J. D., Richardson D. S., and Burke T. (2006).
Childs D. Z., Sheldon B. C., and Rees M. (2016). The evo- Towards unbiased parentage assignment: combining
lution of labile traits in sex- and age-structured popu- genetic, behavioural and spatial data in a Bayesian
lations. Journal of Animal Ecology 85(2), 329–342. doi: framework. Molecular Ecology 15(12), 3715–3730. doi:
10.1111/1365-2656.12483. 10.1111/j.1365-294X.2006.03050.x.
Davies T. J., Barraclough T. G., Chase M. W., Soltis P. S., Harris D. J. and Crandall K. A. (2000). Intragenomic
Soltis D. E., and Savolainen V. (2004). Darwin’s variation within ITS1 and ITS2 of freshwater cray-
abominable mystery: insights from a supertree of the fishes (Decapoda: cambaridae): implications for phylo-
angiosperms. Proceedings of the National Academy genetic and microsatellite studies. Molecular Biology
of Sciences of the United States of America 101(7), and Evolution 17(2), 284–291. doi: 10.1093/oxfordjour-
1904–1909. doi: 10.1073/pnas.0308127100. nals.molbev.a026308.
Ekblom R. and Wolf J. B. W. (2014). A field guide to Harrison, H. B., Saenz-Agudelo, P., Planes, S., Jones,
whole-genome sequencing, assembly and annotation. G. P., and Berumen, M. L. (2013). Relative accuracy of
Evolutionary Applications 7, 1026–1042. doi: 10.1111/eva three common methods of parentage analysis in natu-
.12178. ral populations. Molecular Ecology 22(4), 1158–1170. doi:
Engen S., Ringsby T. H., Sæther B. E., et al. (2007). Effec- 10.1111/mec.12138.
tive size of fluctuating populations with two sexes and Hayes B. J. (2011). Technical note: efficient parentage assign-
overlapping generations. Evolution 61(8), 1873–1885. doi: ment and pedigree reconstruction with dense single
10.1111/j.1558-5646.2007.00155.x. nucleotide polymorphism data. Journal of Dairy Science
Farris J. S., Arnold K. G., and Eckardt M. J. (1970). A numeri- 94(4), 2114–2117. doi: 10.3168/jds.2010-3896.
cal approach to phylogenetic systematics. Systematic Biol- Holmes S. (2003). Bootstrapping phylogenetic trees: theo-
ogy 19(2), 172–189. ry and methods. Statistical Science 18(2), 241–255. doi:
Faurby S. (2019). Data from PHYLACINE 1.2: the phyloge- 10.1214/ss/1063994979.
netic atlas of mammal macroecology. Journal Ecology 99, Huisman J. (2017). Pedigree reconstruction from SNP data:
2626. doi: https://doi.org/10.5061/dryad.bp26v20. parentage assignment, sibship clustering and beyond.
Felsenstein J. (1978). Cases in which parsimony or compat- Molecular Ecology Resources 17(5), 1009–1024. doi:
ibility methods will be positively misleading. Systematic 10.1111/1755-0998.12665.
Zoology 27(4) 401–410. Jetz W., Thomas G. H., Joy J. B., Hartmann K., and
Felsenstein J. (1985a). Confidence limits on phylogenies: an Mooers A. O. (2012). The global diversity of birds in
approach using the bootstrap. Evolution: International space and time. Nature 491(7424), 444–448. doi: 10.1038
Journal of Organic Evolution 39, 783–791. doi: 10.1111/ /nature11631.
j.1558-5646.1985.tb00420.x. Jones A. G. (2001). GERUD1.0: a computer program for
Felsenstein J. (1985b). Phylogenies and the comparative the reconstruction of parental genotypes from proge-
method. The American Naturalist 125(1), 1–5. doi: ny arrays using multilocus DNA data. Molecular Ecol-
10.1086/284325. ogy Notes 1, 215–218. doi: 10.1046/j.1471-8278.2001.
Fitch W. (1970). Distinguishing homologous from analogous 00062.x.
proteins. Systematic Biology 19(2), 99–113. Jones A. G., Small C. M., Paczolt K. A., and Ratterman
Flanagan S. P. and Jones A. G. (2019) The future of parent- N. L. (2010). A practical guide to methods of parentage
age analysis: from microsatellites to SNPs and beyond. analysis. Molecular Ecology Resources 10(1), 6–30. doi:
Molecular Ecology 28(3), 544–567. 10.1111/j.1755-0998.2009.02778.x.
Jones O. R. and Wang J. (2010). Molecular marker- polygenic basis. Molecular Ecology 27(17), 3498–3514.
based pedigrees for animal conservation biologists. Ani- doi:10.1111/mec.14811.
mal Conservation 13(1), 26–34. doi: 10.1111/j.1469-1795 McKechnie A. E., Freckleton R. P., and Jetz W. (2006). Pheno-
.2009.00324.x. typic plasticity in the scaling of avian basal metabolic rate.
Jones O. R., Scheuerlein A., Salguero-Gómez R., et al. (2013). Proceedings of the Royal Society B: Biological Sciences
Diversity of ageing across the tree of life. Nature 505(7482), 273(1589), 931–937. doi:10.1098/rspb.2005.3415.
169–173. doi: 10.1038/nature12789. Metzker M. L. (2010). Sequencing technologies–the next
Jukes T. H. and Cantor, C. R. (1969). Evolution of protein generation. Nature Reviews Genetics 11, 31–46. doi:
molecules. In: H. N. Munro (ed.), Mammalian Protein 10.1038/nrg2626.
Metabolism, 21–123. Academic Press. Morgan D. R. and Soltis D. E. (1993). Phylogenetic relation-
Kalinowski S. T., Taper M. L., and Marshall T. C. (2007). ships among members of Saxifragaceae sensu lato based
Revising how the computer program CERVUS accom- on rbcL sequence data. Annals of the Missouri Botanical
modates genotyping error increases success in paterni- Garden 80(3), 631–660.
ty assignment. Molecular Ecology 16(5), 1099–1106. doi: Morin P. A. and Woodruff D. S. (1996). Non-invasive sam-
10.1111/j.1365-294X.2007.03089.x. pling for vertebrate conservation. In: R. Wayne and T.
Kalyaanamoorthy S., Minh B. Q., Wong T. K. F., Von Smith (eds.), Molecular Approaches in Conservation, 298–
Haeseler A., and Jermiin L. S. (2017). ModelFinder: 313. Oxford University Press.
fast model selection for accurate phylogenetic estimates. Morrissey M. B. and Wilson A. J. (2010). Pedantics: an r
Nature Methods 14(6), 587–589. doi: 10.1038/nmeth.4285. package for pedigree-based genetic simulation and pedi-
Kelchner S. A. and Thomas M. A. (2007). Model use in gree manipulation, characterization and viewing. Molec-
phylogenetics: nine key questions. Trends in Ecology and ular Ecology Resources 10(4), 711–719. doi: 10.1111/j.1755
Evolution 22(2), 87–94. doi: 10.1016/j.tree.2006.10.004. -0998.2009.02817.x.
Kimura M. (1980). A simple method for estimating evo- Nadler S. A. (1995). Advantages and disadvantages of
lutionary rate of base substitution through comparative molecular phylogenetics: a case study of ascaridoid nema-
studies of nucleotide sequences. Evolution 16, 111–120. todes. Journal of Nematology 27(4), 423–432.
Kinghorn (1994). Pedigree Viewer—a graphical utility for Nascimento F. F., dos Reis M., and Yang Z. (2017). A biol-
browsing pedigreed data sets. In: C. Smith, J. S. Gavora, B. ogist’s guide to Bayesian phylogenetic analysis. Nature
Benkel, J. Chesuais, W. Fairfull, and J. P. Gibson (eds), Pro- Ecology and Evolution 1(10), 1446–1454. doi:10.1038/
ceedings of the 5th World Congress on Genetics Applied s41559-017-0280-x.
to Livestock Production, 85–86. University of Guelph. Nei M. and Saitou N. (1987) The neighbor-joining method:
Kolaczkowski B. and Thornton J. W. (2004). Performance a new method for reconstructing phylogenetic trees.
of maximum parsimony and likelihood phylogenet- Molecular Biology and Evolution 4(4), 406–425. doi:
ics. Nature 431(October), 980–984. doi: 10.1038/nature 10.1093/oxfordjournals.molbev.a040454.
02928.1. Nguyen L. T., Schmidt H. A., Von Haeseler A., and Minh
Kumar S., Stecher G., Li M., Knyaz C., and Tamura K. (2018). B. Q. (2015). IQ-TREE: a fast and effective stochastic
MEGA X: molecular evolutionary genetics analysis across algorithm for estimating maximum-likelihood phyloge-
computing platforms. Molecular Biology and Evolution nies. Molecular Biology and Evolution 32(1), 268–274.
35(6), 1547–1549. doi: 10.1093/molbev/msy096. doi:10.1093/molbev/msu300.
Lanfear R., Frandsen P. B., Wright A. M., Senfeld T., and Nielsen R., Mattila D. K., Clapham P. J., and Palsbøl, P. J.
Calcott B. (2017). Partitionfinder 2: new methods (2001). Statistical approaches to paternity analysis in nat-
for selecting partitioned models of evolution for ural populations and applications to the North Atlantic
molecular and morphological phylogenetic analy- humpback whale. Genetics 157(4), 1673–1682.
ses. Molecular Biology and Evolution 34(3), 772–773. Niskanen, A., Billing, A., Holand, H., et al. (2020) Consis-
doi:10.1093/molbev/msw260. tent scaling of inbreeding depression in space and time.
Longmire J., Baker R., and Maltbie M. (1997). Use of ‘lysis Proceedings of the National Academy of Sciences 117(25),
buffer’ in DNA isolation and its implication for museum 14584–14592.
collections. Museum of Texas Tech University 163, 1–4. Ozgul A., Childs D. Z., Oli M. K., et al. (2010). Cou-
López-Sepulcre A., Gordon S. P., Paterson I. G., Bentzen pled dynamics of body mass and population growth
P., and Reznick D. N. (2013). Beyond lifetime repro- in response to environmental change. Nature 466(7305),
ductive success: the posthumous reproductive 482–485. doi: 10.1038/nature09210.
dynamics of male Trinidadian guppies. Proceedings Patwardhan A., Ray S., and Roy A. (2014). Molecular
of the Royal Society B: Biological Sciences 280(1763). markers in phylogenetic studies-A review. Journal of
doi:10.1098/rspb.2013.1116. Phylogenetics & Evolutionary Biology 2(2). doi: 10.4172/
Lundregan S. L., Hagen I. J., Gohli J., et al. (2018). Infer- 2329-9002.1000131.
ences of genetic architecture of bill morphology in house Pemberton J. M. (2008). Wild pedigrees: the way forward.
sparrow using a high-density SNP array point to a Proceedings of the Royal Society B: Biological Sciences
275, 613–621. doi: 10.1098/rspb.2007.1531.
Penny D., Hendy M. D., and Steel M. A. (1992). Progress Sambrook J., Fritsch E. F., and Maniatis T. (1989). Molecular
with methods for constructing evolutionary trees. Trends Cloning: A Laboratory Manual. Spring Harbour Labora-
in Ecology and Evolution 7(3), 73–79. doi: 10.1016/0169- tory Press.
5347(92)90244-6. Sanderson M. J., Purvis A., and Henze C. (1998). Phyloge-
Posada D. (2008). jModelTest: phylogenetic model averag- netic supertrees: assembling the trees of life. Trends in
ing. Molecular Biology and Evolution 25(7), 1253–1256. Ecology and Evolution 13(3), 105–109. doi: 10.1016/S0169-
doi: 10.1093/molbev/msn083. 5347(97)01242-1.
Pujol B., Marrot P., and Pannell J. R. (2014). A quanti- Sanger F., Nicklen S., and Coulson A. R. (1977). DNA
tative genetic signature of senescence in a short-lived sequencing with chain-terminating inhibitors. Proceed-
perennial plant. Current Biology 24(7), 744–747. doi: ings of the National Academy of Sciences of the United
10.1016/j.cub.2014.02.012. States of America 74(12), 5463–5467.
Pyron R. A. (2015). Post-molecular systematics and the Schirmer M., Ijaz U. Z., D’Amore R., Hall N., Sloan W. T.,
future of phylogenetics. Trends in Ecology and Evolution and Quince C. (2015). Insight into biases and sequenc-
30(7), 384–389. doi: 10.1016/j.tree.2015.04.016. ing errors for amplicon sequencing with the Illumina
Qian H. and Jin Y. (2016). An updated megaphylogeny of MiSeq platform. Nucleic Acids Research 43(6), e37. doi:
plants, a tool for generating plant phylogenies and an 10.1093/nar/gku1341.
analysis of phylogenetic community structure. Journal of Simmonds E. G. and Coulson T. (2015). Analysis of phe-
Plant Ecology 9(2), 233–239. doi: 10.1093/jpe/rtv047. notypic change in relation to climatic drivers in a pop-
Rannala B. and Yang Z. (1996). Probability distribution ulation of Soay sheep Ovis aries. Oikos 124(5). doi:
of molecular evolutionary trees: a new method of phy- 10.1111/oik.01727.
logenetic inference. Journal of Molecular Evolution 43, Simonsen M., Mailund T., and Pedersen C. N. S. (2008).
304–311. Rapid neighbour-joining. In Crandall, K. A. and Lager-
Reid J. M., Keller L. F., Marr A. B., Nietlisbach P., Sardell gren, J. (eds) Algorithms in Bioinformatics, vol 5251.
R. J., and Arcese P. (2014). Pedigree error due to Springer, Berlin, pp. 113–22. doi: 10.1007/978-3-540-
extra-pair reproduction substantially biases estimates of 87361-7_10.
inbreeding depression. Evolution 68(3), 802–815. doi: Smith L. M. and Burgoyne L. A. (2004). Collecting, archiv-
10.1111/evo.12305. ing and processing DNA from wildlife samples using
Revell L. J., Harmon L. J., and Collar D. C. (2008). Phylo- FTA® databasing paper. BMC Ecology 4(4), 1–11. doi:
genetic signal, evolutionary process, and rate. Systematic 10.1186/1472-6785-4-4.
Biology 57(4), 591–601. doi: 10.1080/10635150802302427. Smith S. and Morin P. A. (2005). Optimal storage conditions
Riester M., Stadler P. F., and Klemm K. (2009). FRANz: for highly dilute DNA samples: a role for trehalose as a
reconstruction of wild multi-generation pedigrees. Bioin- preserving agent. Journal of Forensic Sciences 50(5), 1101–
formatics. doi: 10.1093/bioinformatics/btp064. 1108. doi: 10.1520/jfs2004411.
Robinson M. R., Pilkington J. G., Clutton-Brock T. H., Pem- Sneath P. H. and Sokal R. R. (1973). Numerical Taxonomy:
berton J. M., and Kruuk L. E. B. (2006). Live fast, die The Principles and Practice of Numerical Classification.
young: trade-offs between fitness components and sex- Freeman.
ually antagonistic selection on weaponry in Soay sheep. Stamatakis A. (2014). RAxML version 8: a tool for phylo-
Evolution 60(10), 2168. doi: 10.1554/06-128.1. genetic analysis and post-analysis of large phylogenies.
Roeder A. D., Archer F. I., Poinar H. N., and Morin P. A. Bioinformatics 30(9), 1312–1313. doi: 10.1093/bioinfor-
(2004). A novel method for collection and preservation matics/btu033.
of faeces for genetic studies. Molecular Ecology Notes 4, Stern D. L. (2013). The genetic causes of convergent evo-
761–764. doi: 10.1111/j.1471-8286.2004.00737.x. lution. Nature Reviews Genetics. 14(11), 751–764. doi:
Ronquist F. and Huelsenbeck J. P. (2003). MrBayes 3: 10.1038/nrg3483.
Bayesian phylogenetic inference under mixed models. Swofford D. L. (2002). PAUP*. Phylogenetic Analysis Using
Bioinformatics 19(12), 1572–1574. doi: 10.1093/bioinfor- Parsimony (*and Other Methods). Version 4. Sinnauer
matics/btg180. Associates. doi: 10.1007/BF02198856.
Rzhetsky A. and Nei M. (1992). A simple method Teo Y. Y. (2008). Common statistical issues in genome-wide
for estimating and testing minimum-evolution trees. association studies: a review on power, data quality con-
Molecular Biology and Evolution 9(5), 945–967. doi: trol, genotype calling and population structure. Current
10.1093/oxfordjournals.molbev.a040771. Opinion in Lipidology 19(2), 133–143.
Salguero-Gómez R., Jones O. R., Jongejans E., et al. (2016). Thompson E. A. (1975). The estimation of pairwise rela-
Fast–slow continuum and reproductive strategies struc- tionships. Annals of Human Genetics 39(2), 173–188. doi:
ture plant life-history variation worldwide. Proceedings 10.1111/j.1469-1809.1975.tb00120.x.
of the National Academy of Sciences 113(1), 230–235. doi: Thorne J. L. and Kishino H. (1992). Freeing phyloge-
10.1073/pnas.1506215112. nies from artifacts of alignment. Molecular Biology
and Evolution 9(6), 1148–1162. doi: 10.1093/oxfordjour-
nals.molbev.a040783.
Tidière M., Gaillard J. M., Berger V., et al. (2016). Compara- in Bioinformatics, vol 5724. Springer, Berlin, pp. 375–389.
tive analyses of longevity and senescence reveal variable doi: 10.1007/978-3-642-04241-6_31.
survival benefits of living in zoos across mammals. Scien- Whelan S., Liò P., and Goldman N. (2001). Molecular phylo-
tific Reports 6(June), 1–7. doi: 10.1038/srep36361. genetics: state-of-the-art methods for looking into the past.
Tippery N. P. and Les D. H. (2008). Phylogenetic analysis Trends in Genetics 17(5), 262–272.
of the internal transcribed spacer (ITS) region in Menyan- Wilson A. J., Réale D., Clements M. N., et al. (2010). An ecol-
thaceae using predicted secondary structure. Molecu- ogist’s guide to the animal model. Journal of Animal Ecol-
lar Phylogenetics and Evolution 49(2), 526–537. doi: ogy 79(1), 13–26. doi: 10.1111/j.1365-2656.2009.01639.x.
10.1016/j.ympev.2008.07.019. Xia X. (2018). Distance-based phylogenetic methods. In: X.
Van Oosterhout C., Hutchinson W. F., Wills D. P., and Xia, Bioinformatics and the Cell, 343–379. Springer. doi:
Shipley P. (2004). MICRO-CHECKER: software for iden- 10.1007/978-3-319-90684-3_15.
tifying and correcting genotyping errors in microsatellite Yang Z. and Rannala B. (1997). Bayesian phylogenetic
data. Molecular Ecology Notes 4(3), 535–538. inference using DNA sequences: a Markov chain Monte
Vartia S., Villanueva-Cañas, J. L., Finarelli, J., et al. (2016). A Carlo method. Molecular Biology and Evolution 14(7),
novel method of microsatellite genotyping-by-sequencing 717–724.
using individual combinatorial barcoding. Royal Society Yang Z. and Rannala B. (2012). Molecular phylogenetics:
Open Science 3(1), 150565. principles and practice. Nature Reviews Genetics 13(5),
Vences M., Thomas M., Bonett R. M., and Vieites D. R. (2005). 303–314. doi: 10.1038/nrg3186.
Deciphering amphibian diversity through DNA barcod- Yue G. H. and Xia J. H. (2014). Practical Considerations
ing: chances and challenges. Philosophical Transactions of Molecular Parentage Analysis in Fish. Journal of the
of the Royal Society B: Biological Sciences 360(1462), World Aquaculture Society 45, 89–103. doi: 10.1111/
1859–1868. doi: 10.1098/rstb.2005.1717. jwas.12107.
Waits L. P. and Paetkau D. (2005). Noninvasive genet- Zhang J. and Kumar S. (1997). Detection of convergent
ic sampling tools for wildlife biologists: a review and parallel evolution at the amino acid sequence lev-
of applications and recommendations for accurate el. Molecular Biology and Evolution 14(5), 527–536. doi:
data collection. Journal of Wildlife Management 69(4) 10.1093/oxfordjournals.molbev.a025789.
1419–1433. doi:10.2193/0022-541x(2005)69[1419:ngstfw] Zhang W. (2000). Phylogeny of the grass family (Poaceae)
2.0.co;2. from rpl16 intron sequence data. Molecular Phylo-
Wang J. (2004). Sibship reconstruction from genetic data genetics and Evolution 15(1), 135–146. doi: 10.1006/
with typing errors. Genetics 166(4), 1963–1979. doi: mpev.1999.0729.
10.1534/genetics.166.4.1963. Zwickl D. J. (2006). Genetic algorithm approaches for
Wang J. (2012). Computationally efficient sibship and the phylogenetic analysis of large biological sequence
parentage assignment from multilocus marker data. data sets under the maximum likelihood criterion.
Genetics 191(1), 183–194. doi: 10.1534/genetics.111 University of Texas at Austin. (http://www.zo.utexas.
.138149. edu/faculty/antisense/garli/garli.html).
Wheeler T. J. (2009). Large-scale neighbor-joining with NIN-
JA. In Salzberg S. L. and Warnow T. (eds) Algorithms
CHAPTER 2
Biochemical and physiological

data collection
Oldřich Tomášek, Alan A. Cohen, Erola Fenollosa, Maurizio Mencuccini,
Sergi Munné-Bosch, and Fanie Pelletier
2.1 Introduction in functional ecology (Salguero-Gómez et al. 2018) or

ageing and medical research (Cohen et al. 2014).
The past three decades have witnessed increasing
recognition that incorporation of physiological and
biochemical data into demographic and population
studies can provide insights into drivers of popula-
tion dynamics and species distribution, as well as refine Box 2.1 Biomarker and functional trait
estimation of vital rates, thereby increasing predictive definition
power of population models (Cooke et al. 2013; Merow
et al. 2014; Violle et al. 2014; Salguero-Gómez et al. 2018; We use the term biomarker to describe any biochemical or
see Chapter 9). An emergence of an integrative field physiological trait, or their combination measured at the
of conservation physiology is one of the most tangi- individual level (from the cell to the whole-organism level)
ble results of such a recognition (Cooke et al. 2013). that can be used as a proxy for intrinsic or environmental
One of the major promises of integrating physiolog- states, processes, features, or conditions. These variables
ical data into population and conservation research may be linked to individual fitness or demographic rates
is the possibility to characterise population structure or may otherwise be of interest in demographic stud-
and dynamics on much shorter time-scales compared ies, but are often difficult or even impossible to measure
to classical demographic approaches (Salguero-Gómez directly (for examples, see the first paragraph of this
et al. 2018). This possibility is based on the assump- chapter). For the term functional trait, we follow the
tion that physiological and biochemical traits or their definition proposed by Violle et al. (2007), who defined
combinations can be used as proxies (biomarkers; see it as any morphological, physiological, or phenological
Box 2.1) for individual ontogenetic stage (Cattet et al. feature measured at the individual level that impacts fit-
2018), age (Horvath 2013), sex (Cattet et al. 2017), ness indirectly via its effects on performance traits. Hence,
behavioural or life history strategy (Réale et al. 2010), functional traits may be used as biomarkers of fitness
body condition and health state (Milot et al. 2014), fit- components, but other biomarkers need not fulfil the
ness components (Violle et al. 2007), or environmental definition of functional traits. Note that, unlike Violle et al.
conditions (Dantzer et al. 2014). Apart from their use (2007), we do not consider morphological or phenological
as biomarkers, physiological and biochemical traits, traits among functional traits here due to the focus of this
together with other functional traits, also emerge as
chapter on biochemical and physiological data. Nonethe-
less, morphological (e.g. body mass) or phenological traits
direct targets of conservation science since function-
can also be used as biomarkers and then the same cri-
al diversity may be as important to protect as the
teria for biomarker validation apply as presented in this
rarity of species (Violle et al. 2017). In addition, demo-
chapter.
graphic approaches can help answer various questions
Oldřich Tomášek et al., Biochemical and physiological data collection. In: Demographic Methods Across the Tree of Life.
Despite the clear benefits of such an integration, bio- environment can lead to adaptive antioxidant upreg-
chemical and physiological data collection has rarely ulation (Hõrak and Cohen 2010).
been discussed in demographic textbooks, indicating Another factor to consider in biochemical biomark-
that the potential benefits still await wider recogni- ers is their stability. In some cases, biomarkers may
tion. Slow blending of physiological and demographic degrade or form during storage, biasing estimates, par-
approaches may stem not only from disciplinary ticularly if not all samples are analysed simultaneously.
cloistering, but also from the difficulties of match- Biomarkers that are less stable over time need to
ing ecological replication numbers with biochemical be stored in –80◦ C or even in liquid nitrogen or its
replication possibilities due to field and laboratory vapours, which poses logistic complications in the field
limitations, such as laboriousness, high costs, limited and increases study costs. Nevertheless, dry-shipper
sample volume, and storage difficulties. Nevertheless, containers are available, in which liquid nitrogen is
overcoming those difficulties promises to bring signifi- bound in a porous material to prevent spilling. These
cant advances in both ecological and medical research. containers can be carried safely to the field and can
Luckily, new demographic approaches, such as inte- even be transported by air. When storage in −20◦ C
gral projection models (see chapter 10), are robust to is sufficient, non-frost-free freezers should be used
low sample sizes, rendering such efforts more feasible because temperature rises briefly to 0◦ C every 24
(Salguero-Gómez et al. 2018). hours in frost-free freezers (Sheriff et al. 2011). Despite
In this chapter we briefly review most prominent these considerations, caution should be exercised when
physiological traits and biomarkers that may be inter- analysing/interpreting samples stored over long peri-
esting within demographic framework, including the ods (years).
methods of their collection, storage, and analysis, and Prior to the analysis of the actual samples, the analyt-
the criteria to be met before the trait is validated as ic methods for biomarker quantification must be vali-
a biomarker. We hope that this effort will stimulate dated in each species and sample matrix used (Hõrak
further integration of physiological and demographic and Cohen 2010). The criteria evaluated during the
approaches. validation process include method specificity (or selec-
tivity), sensitivity, accuracy, range of linearity, reagent
and sample stability, and measurement precision
2.2 General criteria for selecting
biomarkers in demographic studies
Selection of reliable biomarkers can be a difficult Method validaton
task, and an array of criteria should be met during
biomarker selection and validation. One of the most Specificity
important factors to consider is whether the biomark-
Sensitivity
er is intended as a marker of stable (e.g. genetic)
differences between individuals (trait biomarker) or Accuracy
as a marker of transient condition or environmental
stress (state biomarker). Good trait biomarkers must Range of linearity
show high within-individual repeatability over the Reagent and sample stability
long term, whereas state biomarkers should respond
to environmental conditions. The best way to assess Precision
biomarker consistency is analysing short- and long-
term within-individual repeatabilities based on intra- Biomarker validation
class correlation coefficients, which take into account
the between-individual variation within the popula- Variation in the population
tion (Hõrak and Cohen 2010). Importantly, interpre-
Confounding factors
tation of state markers may be context dependent
and regulated and unregulated processes must be dis- Link to fitness components
tinguished. For example, an exposure to a stressful or state/process of interest
environment can result in unregulated depletion of
an antioxidant; however, long-term exposure to that Figure 2.1 Biomarker validation process and criteria.
B I O C H E M I C A L A N D P H YS I O L O G I C A L D ATA C O L L E C T I O N 37
(Figure 2.1). Measurement precision is the ultimate and their repeated sampling over time, ideally over
criterion determining reliability and usefulness of the their lifetimes, being the preferred sampling methods
method. Although coefficient of variation (i.e. relative (Hõrak and Cohen 2010; Violle et al. 2012).
standard deviation) is usually recommended as a preci- In plants, repeated sample collection is usually pos-
sion measure in the methodological literature, repeata- sible even in the case of whole organs (e.g. leaves),
bility calculated as intraclass correlation coefficient although the extent of the damage and an herbivory-
should be preferred as it takes into account biological like response should be considered. In animals includ-
context (i.e. actual biomarker variation in the pop- ing humans, however, the requirement of repeated
ulation; Hõrak and Cohen 2010). To this end, each sampling largely constrains the spectrum to nonde-
sample is aliquoted, and the assay is replicated in at structive and noninvasive sample types, such as bod-
least two aliquots per sample, ideally including all the ily fluids (e.g. blood), secretions (e.g. saliva), excreta,
sample processing, not only the final measurement. and integumentary structures (e.g. hairs or feathers).
In addition, repeatability among measurers should be Another limitation is posed by the need to observe both
assessed for manual methods (e.g. microscopy). As general and country- and taxon-specific ethical norms
method validation is of utmost importance, we encour- and recommendations (Lindsjö et al. 2016; Drinkwater
age readers to familiarise themselves with this process et al. 2019). In any case, sample collection should only
from existing reviews on this topic (e.g. Green 1996), minimally interfere with normal life activities of the
especially if the analysis will not be done by a spe- studied organism. Noninvasive samples, such as hairs
cialised laboratory. or faeces, are ideal in this regard as the information can
Biomarker validation is the next stage of the be obtained without even approaching the individual.
validation process (Hõrak and Cohen 2010). This When collecting blood, the volume should not exceed
stage involves exploration of within- and between- 1% of body mass per sampling event (Owen 2011). The
individual variation in the population and determi- site and means of blood collection are taxon specific
nation of confounding factors such as sex, age, body and can be found in relevant literature (Sheriff et al.
size, reproductive state, temperature, season, or time 2011). Blood-sucking bugs can be used to minimise
of sample collection. Ultimately, the link between stress and interference with normal activities, although
the biomarker and fitness component (or any other applicability of this method is taxon and biomarker
state/process of interest) must be established. This is dependent (Bauch et al. 2013). Chelating anticoagu-
probably the most challenging task of the biomark- lants (e.g. EDTA) must be avoided during blood col-
er validation process, however, as the strength and lection for assays requiring free metal ions (Harr 2005).
direction of correlation with fitness components may In arthropods, several protocols for haemolymph sam-
vary among contexts (e.g. Breuner et al. 2008). Hence, pling have been developed, although in small-bodied
understanding this variation is undoubtedly one of the species, samples need to be pooled (Tabunoki et al.
most important tasks of current demographic research 2019).
(Shipley et al. 2016). At the population or species level, many physio-
logical and biochemical trait values can be obtained
from published data sets and databases. We provide
2.3 Designing the data collection
a (definitely not exhaustive) list of databases contain-
The physiological and biochemical data can enter ing physiological and biochemical traits and related
demographic models either at the individual level or sources in Table 2.1. Additional data sets can certainly
at the level of populations or species (Edmunds et al. be found in public data repositories.
2014). The design of the data collection depends on
the data type needed, and this in turn depends on
the question at hand. In many cases, population- or 2.4 Physiological traits
species-level data will be sufficient. Nevertheless, both
2.4.1 Animal metabolism
between- and within-individual variation may affect
vital rates and individual-level data are needed if Metabolic traits belong among the most promising
those effects are of interest (Edmunds et al. 2014). In physiological traits for demographic and large-scale
such a case, the sampling design should aim to accu- comparative studies because energy is a common cur-
rately quantify within-species biomarker variability in rency across the Tree of Life (Salguero-Gómez et al.
space and time, with random selection of individuals 2018). It is tightly linked to fitness due to its importance
Table 2.1 Databases containing biochemical and physiological traits and other related sources.
Database name Description Link
AnAge Database of animal life history traits including metabolic http://genomics.senescence.info/species/

rates
Biofuel Ecophysiological Open-access database of plant traits, biomass yields, www.betydb.org
Traits and Yields database and observations of ecosystem dynamics oriented on
(BETYdb) bioenergy production
Coral Trait Database Database of coral traits coraltraits.org
Encyclopedia of Life Global trait database for all organisms eol.org/docs/what-is-eol/traitbank
TraitBank
GlobTherm Global database on thermal tolerances for aquatic and datadryad.org/resource/doi:10.5061/dryad.1cv08
terrestrial organisms; maintained on iDiv Data
HormoneBase Population-level database of steroid hormone levels from hormonebase.org
free-living, adult vertebrates
iDiv Data Data repository of German Centre for Integrative idata.idiv.de
Biodiversity Research curating multiple data sets
Planteome Project developing common annotation standards and planteome.org
reference ontologies for plant genes and traits
Registry of Research Data Comprehensive registry of research data repositories re3data.org
Repositories
Thesaurus Of Plant Char- Project standardising terminology for plant traits and top-thesaurus.org
acteristics for Ecology and environmental associations
Evolution (TOP)
TraitNet Collaborative network website for plant trait-based traitnet.ecoinformatics.org
research including a list of external trait databases
traits R package enabling accessing multiple trait databases cran.r-project.org/web/packages/traits/index.html
TRY Plant Trait Database Global plant trait database www.try-db.org/
for growth, reproduction, and survival (Burton et al. can be uncoupled and subject to different selection
2011; Pettersen et al. 2018). The field of energetics has pressures (Norin and Metcalfe 2019). Despite pro-
therefore developed several tools to measure the rates nounced plasticity, metabolic rates are repeatable
at which an individual uses energy. within individuals and there is typically up to three-
Minimum metabolic rates, that is, basal metabolic fold intraspecific variation among individuals (Burton
rate in endotherms, standard metabolic rate in et al. 2011; Norin and Metcalfe 2019).
ectotherms, and their less rigorous version, resting Except for FMR, metabolic rates are usually esti-
metabolic rate, are defined as the minimum rates of mated using respirometry, which measures oxygen
energy expenditure per unit time needed to maintain consumption and/or carbon dioxide production of
vital body functions (Chabot et al. 2016; Norin and an individual in a metabolic chamber (Lighton 2008).
Metcalfe 2019). In contrast, peak (maximum; summit) This requires establishing a field laboratory, which
metabolic rate expresses the maximum rate of aero- renders respirometry difficult to perform in the field
bic energy utilisation (Norin and Metcalfe 2019). In and only feasible in small animals (perhaps up to
between lies field metabolic rate (FMR; or daily energy the size of a rabbit). In animals that can easily be
expenditure) that gives the free-living energy demands captured repeatedly, FMR estimation using doubly
of the individual (Speakman 1997). In addition, routine labelled water (DLW) to estimate carbon dioxide pro-
metabolic rate has frequently been measured in fishes, duction may be a more convenient, but expensive,
involving nonexhaustive swimming in laboratory alternative (Speakman 1997; Speakman and Hambly
conditions (Chabot et al. 2016). Importantly, mini- 2016). The classical procedure involves DLW injec-
mum, maximum, and field/routine metabolic rates tion upon capture, collection of a first blood sample,
release of an individual, and its recapture after one molecules. Depending on the protocol, the intensity
or more days, followed by the collection of a second and spectrum of emitted fluorescence give informa-
blood sample. Alternatively, DLW application through tion on the photochemistry of photosystem II, such
a food item followed by urine or faeces collection or as Fv /Fm , the maximum efficiency of photosystem II
blood collection using blood-sucking bugs could theo- (Murchie and Lawson 2013). Many fluorometers are
retically enable DLW method to be performed without suitable for rapid and large demographic surveys
capturing the animal (Speakman and Hambly 2016). (Mencuccini et al. 2014; Semerci et al. 2017).
Alternative methods of FMR estimation more suitable Measurement of vertical xylem sap transport in
for large animals or animals that are difficult to recap- woody plant stems obtain whole-plant sap flow (close-
ture include heart rate telemetry or accelerometry (see ly related to photosynthesis). Implanted resistors heat
section 2.9). the xylem, while highly sensitive thermocouples mea-
sure temperature changes as sap moves in the wood.
Once installed, these sensors can monitor plant water
2.4.2 Plant metabolism fluxes for months to years at high time resolution (e.g.
Plant energy metabolism is centred on photosynthesis Martínez-Vilalta et al. 2007). Well-tested protocols for
and respiration. In demographic studies, measure- in-house sensor construction and calibration facilitate
ments of photosynthesis and respiration are almost the task (e.g. Fuchs et al. 2017).
universally carried out on organs, such as leaves, flow-
ers, fruits, or photosynthetic stems, while they are still
2.4.3 Semen physiology
attached to the plant. Several types of equipment exist,
and users need to be aware of technical characteris- Semen characteristics, especially sperm phenotypic
tics and potential sources of error and bias (Long and and performance traits, represent important physio-
Bernacchi 2003; cf. http://prometheuswiki.org/tiki- logical features that may either be used as biomarkers
custom_home.php). The most common instruments or be of interest per se (Reinhardt et al. 2015; Lemaître
are infra-red gas analysers (IRGAs). IRGAs consist of and Gaillard 2017). Given their fundamental role in
an environmentally controlled cuvette, inside which male reproductive success, sperm traits can serve as
the leaf is enclosed, and of computer-controlled units biomarkers of male fertility and reproductive fitness
transporting and measuring gas concentrations and (Gomendio et al. 2006), or predict population and
flows. Ultimately, photosynthetic rates in units of species differences in extrapair paternity rate (Lifjeld
moles of carbon dioxide exchanged with the atmo- et al. 2010; Laskemoen et al. 2013; Šandera et al. 2013).
sphere per unit of time and leaf area are measured Sperm traits further show pronounced environment-
(the corresponding water vapour flux being tran- driven plasticity, implying their potential as biomark-
spiration). Once organ-scale fluxes are determined, ers of environmental conditions (Reinhardt et al. 2015).
whole-organism values can be obtained using total Sperm traits and other semen characteristics also
leaf areas, sapwood volumes, or stem surface areas. appear as main objects of interest in demographic stud-
Simultaneous measurement of environmental vari- ies, such as those focused on reproductive senescence
ables is essential to obtain meaningful estimates (Johnson et al. 2015; Lemaître and Gaillard 2017).
of traits, such as light-saturated photosynthesis or The feasibility of repeated semen collection greatly
temperature-corrected dark respiration (Abdul-Hamid varies across animal taxa. In some vertebrates, includ-
and Mencuccini 2009; Pérez-Harguindeguy et al. 2013). ing humans, semen can be collected by massaging,
Fast-response systems suitable for demographic applying pressure, either with or without anaesthe-
surveys also exist to measure other metabolic quanti- sia, or using an artificial vagina or a dummy female
ties. Leaf stomatal conductance (related to the stomatal (Birkhead et al. 2009; Lifjeld et al. 2010; Wasden et al.
aperture to CO2 and water exchange) can be obtained 2017; Humann-Guilleminot et al. 2018). In other verte-
with relatively cheap equipment with short equilibra- brate species, electroejaculation can be feasible (Lierz
tion times (60–120 s; e.g. Vanderklein et al. 2007). The et al. 2013), though it can affect semen characteristics
fluorescence naturally emitted by leaves upon illu- (Birkhead et al. 2009). Nonetheless, surgical or destruc-
mination by a specific light source can be measured tive semen collection is the only option in many
with modulated chlorophyll fluorometers. Modu- taxa, rendering them unsuitable for longitudinal
lated fluorometers consist of high-frequency light- studies.
emitting diodes and receiving transducers measuring Semen characteristics commonly measured include
the fluorescence re-emitted from excited chlorophyll semen volume, sperm concentration, sperm motility,
sperm velocity, sperm morphometric traits, and abnor- contexts. For example, uric acid is an antioxidant but
mal sperm proportion (Fitzpatrick and Lüpold 2014). is also related to protein catabolism during fasting
In animal field studies, however, sperm motility and (Hollmen et al. 2001), so high uric acid levels may
velocity analysis may be challenging as it requires indicate different things depending on food availabili-
video recording of fresh samples using a microscope ty. For these reasons, while many ecophysiology stud-
with a heated table, whereas, for the other analyses, ies do use specific markers in well-validated contexts
samples can be stored in preservative media, such (Podlesak et al. 2005), broad-scale or comparative use
as formalin (Tomášek et al. 2017). Increasing atten- can be more challenging, and care should be exercised
tion has recently been devoted to more biochemi- in a demographic context.
cal traits, including sperm ATP content or seminal Stress and recent feeding affect concentrations of
plasma composition; however, those usually require some metabolites such as glucose, uric acid, or lipids
immediate storage in liquid nitrogen (Poiani 2006; Fitz- (Harr 2005). Fasting is usually unattainable in the
patrick and Lüpold 2014). It should be emphasised wild, but stress effects can be minimised by taking
that the importance of semen characteristics for male blood samples within three minutes from the ini-
reproductive success varies greatly among taxa or even tial capture (Tomášek et al. 2019). Blood should be
among studies on the same species (Fitzpatrick and centrifuged immediately upon collection to prevent
Lüpold 2014). Hence, the link to fitness components glucose consumption by blood cells. Plasma or serum
must always be validated. can be stored up to 6 hours in room temperature for
standard biochemistry arrays (Monneret et al. 2016).
Analyses are usually done in specialised human or
2.5 Biochemical traits veterinary laboratories, although analysers are com-
2.5.1 General biochemistry mercially available for in-house use. As always, all the
analytic methods must be validated when used for
Animal and human biochemistry nonmodel species.
Plasma and serum biochemistry (i.e. arrays quanti- Biochemistry can also provide useful information
fying activity of various tissue enzymes and con- in invertebrates. In corals, for example, biochemical
centrations of electrolytes, minerals, proteins, lipids, data, including calcification and content of chlorophyll,
and other metabolites in blood plasma or serum) are lipids, and proteins, have been proposed to inform
commonly used in human and veterinary medicine demographic modelling of ecological performance of
to assess individual health, nutritional status, and coral reefs (Edmunds et al. 2014).
metabolic functions (Harr 2005; Cohen et al. 2014).
Because clinical biochemistry is usually available in
large longitudinal human studies, these tests are Plant biochemistry and element content
increasingly used in demographic studies of human Plant biochemistry deals with the structure and func-
ageing and age-related diseases (Cohen et al. 2014). tion of compounds such as carbohydrates, proteins,
Unlike human studies, plasma/serum biochemistry is lipids, and other biomolecules. The most abundant
underutilised in animal ecological research, despite plant molecule, however, is water. Relative water con-
its potential to provide valuable information about tent and water potential Ψ are frequently employed
individual body condition and biological ageing rate to quantify abundance and chemical energy of water
(Cohen et al. 2014; Milot et al. 2014). One reason for in plant tissues. Plant Ψ measures the unit-volume
the underutilisation is that the interpretation of indi- potential energy of water molecules relative to unit-
vidual markers can vary across species. For example, volume free water under reference conditions. It quan-
glucose levels that would be toxic in mammals are tifies the tendency of water to move from areas of
standard in birds (Tomášek et al. 2019). Similarly, high to low potential energy. Hence, the difference
bird species with carotenoid-based colouration toler- in Ψ between, for example, roots and leaves, drives
ate and even seem to benefit from carotenoid levels water flow. Plant Ψ is quantified using Scholander
that are known to cause pathologies in species lacking pressure chambers (Turner 1988) or one of several
carotenoid-based displays, implying that physiological psychrometric techniques. In the field, plant Ψ is mea-
consequences of carotenoid levels are highly species sured prior to dawn (after night-time rehydration)
specific (Cohen and McGraw 2009). Likewise, a sin- and at midday (minimum water content). A mean-
gle biomarker may indicate different things in different ingful trait for demographic studies is the seasonal
absolute minimum Ψmin (Ψ is negative), which mea- hormones in vertebrates are sex (e.g. testosterone or
sures the maximum level of water stress a population oestrogen) and stress steroids (glucocorticoids). Gluco-
of plants can typically tolerate at a site. Ψmin correlates corticoids mediate physiological stress response which
with several other water relations and hydraulic traits, evolved as a mechanism protecting internal stability
such as osmotic potential at saturation, turgor loss (homeostasis) in the face of predictable and unpre-
point, and resistance to embolism (e.g. Fu and Meinzer dictable environmental stimuli (‘stressors’). Hence,
2019). glucocorticoid levels can provide information about
Carbohydrates and lipids are important storage environmental stress, including stress from predation
and communication compounds strictly related to risk or social conflict, as well as psychological or
plant metabolic status. Among the first class, starch, anticipatory stress (Dantzer et al. 2014). Their con-
sucrose, glucose, fructose, and saccharose are often centrations are also linked to metabolic rate (Jimeno
determined in demographic studies. Lipids (e.g. et al. 2018). In addition, sex steroids mediate sex-
glyceride-glycerols) can be important in some species specific life history trade-offs (Brooks and Garratt
(e.g. Sala and Hoch 2009). Beside extraction, the choice 2017), and a combination of multiple sex and stress
of quantification method, either enzymatic, acid diges- hormones can be used as markers of sex, develop-
tion, or ion chromatography, can affect the results mental stage, reproductive state, or individual age-
(Landhäusser et al. 2018). Generalised correlations ing rate (Walther et al. 2016, Cattet et al. 2017; Cattet
between high-throughput near-infrared spectroscopy et al. 2018). Nonsteroid hormones that might be
spectra and wet-chemistry concentrations have been of interest in this context include human chorionic
determined (Ramirez et al. 2015). Carbohydrate and gonadotropin, follicle-stimulating hormone, luteinis-
lipid contents can vary at diurnal and seasonal time- ing hormone, and prolactin (Valeggia 2007).
scales and standardised sampling is required. To obtain baseline concentrations, it is necessary
Pigments such as chlorophyll and carotenoids (light- to collect blood samples within 2 or 3 minutes from
absorption and photo-protection) and vitamins such the initial capture because glucocorticoids increase
as tocopherols (vitamin E) and ascorbate (vitamin C), rapidly following the capture. The concentrations peak
which are both antioxidants, can provide valuable between 15 and 30 minutes from the capture, when
information in demographic studies (Asensi-Fabado a second sample should be collected if acute stress
and Munné-Bosch 2010; Mencuccini et al. 2014; response is of interest (Sheriff et al. 2011). Adreno-
Mencuccini and Munné-Bosch 2017). Similarly, macro- corticotropic hormone (ACTH) may be injected to
(N, P, K) and micro-nutrient concentrations can be standardise the stress response (Sheriff et al. 2011).
strong indicators of different life history strategies Chronic stress can also be assessed. It should be
(Pérez-Harguindeguy et al. 2013). Intrinsic water use noted that blood plasma glucocorticoids are present in
efficiency (ratio of photosynthesis to stomatal con- two states: free, and bound to corticosteroid-binding
ductance) integrated over the lifetime of a tissue can globulin (for discussion about biological importance,
be estimated via 13 C/12 C stable isotope composition see Breuner et al. 2013).
(Dawson et al. 2002). This and other stable isotope Both sex and stress hormones can be analysed in
ratios (1 H/2 H, 18 O/16 O, 15 N/14 N) are determined by plasma, serum, and dry blood spot samples, but also
mass spectrometry of destructively sampled plant tis- noninvasively in saliva, faeces, urine, feathers, or hairs
sues and constitute traits useful for demographic stud- (Valeggia 2007; Sheriff et al. 2011; Cattet et al. 2017;
ies (Pérez-Harguindeguy et al. 2013). Bílková et al. 2019). While saliva steroid concentra-
tions reflect immediate blood levels, samples of excreta
and integumentary structures provide an integrated
2.5.2 Hormones measurement reflecting endocrine status during an
extended period of time from hours (excreta) to weeks
Animal and human hormones (integumentary structures), rendering them suitable
Hormones are signalling molecules that, based on for analysis of chronic stress, which may be of inter-
information about the internal body state and envi- est in demographic studies for its fitness consequences
ronmental conditions, regulate physiological process- (Dantzer et al. 2014).
es, phenotype, development, and other life history Anticoagulants such as heparin or EDTA can be used
traits in order to promote individual fitness (Finch for blood plasma collection. Samples can be stored up
and Rose 1995; Bonier et al. 2009). The most studied to 72 hours at 4◦ C or for decades at −20 or −80◦ C
(Valeggia 2007; Sheriff et al. 2011). Dry blood spots importance may depend on the research question and
can even be stored up to 8 weeks at room tempera- a completely different hormonal profiling with con-
ture or 1 week at tropical temperature (37◦ C) (Valeggia trasting endogenous contents in some of the phytohor-
2007). Radio- or enzyme-immunoassays are available mones may be obtained if either leaves, roots, fruits,
for quantification of both steroid (Sheriff et al. 2011) flowers, or seeds are sampled.
and nonsteroid hormones (Sturgeon and McAllister
1998; Wheeler 2006). Alternatively, LC-MS/MS meth-
2.5.3 Oxidative stress
ods exists that allow quantification of multiple steroid
hormones in one analysis (Bílková et al. 2019). Oxidative stress is an ancient stress response shared
by both animals and plants (Costantini et al. 2010;
Demidchik 2015). It results from excessive formation
Plant hormones of free radicals and other reactive species (RSs) that are
Phytohormones are molecules produced in very low insufficiently eliminated by antioxidant mechanisms.
concentrations that regulate all developmental pro- Accumulating RSs induce oxidative damage to cell
cesses throughout the life cycle in plants, from seed components causing their dysfunction (Halliwell and
germination to plant senescence. Although no study Gutteridge 2015). For this reason, oxidative stress has
has directly linked hormones with demographic plant long been thought to be involved in ageing and devel-
projection, the role of plant hormones in relation to opment of numerous pathological conditions, though
life history traits is well understood. Phytohormones it has fallen out of favour as a primary ageing cause
not only include the five ‘classical’ hormones, auxin, (Halliwell and Gutteridge 2015). Nonetheless, the need
gibberellins (GAs), cytokinins, abscisic acid (ABA) and to maintain redox homeostasis is believed to be a major
ethylene, but also jasmonates, salicylates, strigolac- constraint in life history, possibly partially underpin-
tones, brassinosteroids, and polyamines, among other ning the trade-off between reproduction and longevity
newly emerging organic compounds with important (Costantini et al. 2010; Isaksson et al. 2011). Elevated
regulatory function and hormone-like activity, such oxidative stress can be associated with ageing or any
as melatonin and karrikins. Although some of them biotic or abiotic stressor (Costantini et al. 2010; Demid-
have been classically defined as stress hormones (e.g. chik 2015). Redox state markers may hence serve as
ABA), this is far too simplistic, and it is now clear proxies of biological age or body condition. More-
that all physiological processes are regulated by the over, their integration into demographic models may
so-called hormonal balance rather than by a single advance our mechanistic understanding of ageing pro-
compound (Müller and Munné-Bosch 2011). Hormon- cess and disease development, provide insights into
al cross-talk is therefore essential to study hormone- the roles of senescence or environmental stress in popu-
related processes in plants and the understanding of lation dynamics, and shed light on mechanisms under-
plant hormonal response requires a complete hormon- pinning life history evolution.
al profiling analysis. Moreover, hormonal action does Redox state analysis represents a major challenge
not only depend on hormone contents (which in turn since redox state integrates a complex system of
are not only influenced by biosynthesis but also con- oxidation-reduction reactions and pathways (Jones
jugation and breakdown), but on hormonal transport, 2006; Monaghan et al. 2009). Ideally, all components
hormone sensitivity, and hormonal signalling. determining redox state should be measured, namely
A simultaneous quantification of multiple plant RS production, antioxidant defences, oxidative dam-
hormones is the best approximation for an accurate age, and repair mechanisms (Dotan et al. 2004). This
hormonal profiling. In that way, multiplex gas chro- is, however, rarely done in practice, and quantifica-
matography tandem mass spectrometry (GC-MS/MS) tion of oxidative damage together with some measure
or liquid chromatography coupled to tandem mass of antioxidant defences is the approach that is most
spectrometry (LC-MS/MS) are some of the used meth- commonly taken (Monaghan et al. 2009). Nonetheless,
ods to quantify phytohormones. Although plant mate- the view of redox homeostasis as a single equilibrium
rial needed is below 0.5 g of fresh weight, samples must between RSs and antioxidants is an oversimplification
be strictly collected and preserved in ultracold condi- of reality as dysregulation in any single redox pathway
tions, using liquid nitrogen for sampling followed by can trigger tissue-specific damage (Jones 2006).
storage at −80ºC. All organs can provide information For this reason, it is highly advisable to employ
regarding plant physiological status which is potential- multiple assays and biomarkers reflecting oxidative
ly linkable with life history traits. Nevertheless, their damage to all major biomolecule types, namely lipids
(e.g. d-ROMs test, isoprostanes, and malondialde- recommended to analyse samples within 1 month (Bar-
hyde), proteins (e.g. carbonyl assay, ortho-tyrosine, den et al. 2014), but samples stored in −80◦ C for up
3-nitrotyrosine, or halogenated tyrosines) and DNA to 60 months give reliable results for several mark-
(e.g. comet assay, 8-hydroxy-2′deoxyguanosine, 8- ers and antioxidants (Piłacik et al. 2002; Jansen et al.
hydroxyguanosine). Mass spectrometry-based meth- 2017). Addition of EDTA/BHT/GSH may help pre-
ods are generally preferred over spectrophotometric vent sample autoxidation during storage or processing
methods due to their higher specificity and precision (Barden et al. 2014), but one must make sure that these
(Mateos and Bravo 2007; Halliwell and Gutteridge chemicals do not interfere with the assays.
2015; Morales and Munné-Bosch 2019).
Detection of RSs is difficult due to their very short
half-lives and is rarely done in clinical and field studies,
2.6 Telomeres
although fluorescent probes have been used success- Telomeres are complexes of tandem repeats of a short
fully to detect superoxide in ecological studies (e.g. DNA sequence and associated proteins that cap the
Olsson et al. 2008). ends of linear chromosomes. They identify and protect
The most important antioxidant defences include chromosome ends, thus maintaining integrity of the
antioxidant enzymes and nonenzymatic hydrophilic coding sequences (Nussey et al. 2014). Importantly, in
and lipophilic antioxidants (Valko et al. 2007; the absence of restoration processes such as telomerase
Pamplona and Costantini 2011). The hydrophilic activity, telomeres shorten during each cell division
antioxidants glutathione and ascorbic acid are active due to the inability of DNA polymerase to complete-
primarily in aqueous phases of cells and in the ly replicate the end of the lagging DNA strand (the
circulatory system (Yeum et al. 2004). The ratios of their ‘end replication problem’). When telomeres shorten to
reduced and oxidised forms are often used as mark- a critical limit, cell division stops, and this state termed
ers of whole-organ or whole-body oxidative stress cellular replicative senescence is usually followed by cell
(Queval and Noctor 2007). In contrast, lipophilic death. This mechanism limits the number of poten-
antioxidants (e.g. tocopherols or carotenoids) are tial cell divisions and is believed to have evolved to
related to oxidative stress in cell and organelle mem- prevent cancer development (Young 2018).
branes (Yeum et al. 2004). Spectrophotometric assays The possible causal involvement of telomere short-
are available for important antioxidant enzymes, ening in ageing is the focus of a considerable body
glutathione, and ascorbic acid (Queval and Noctor of research, which greatly benefits from longitudinal
2007; Halliwell and Gutteridge 2015). In addition, demographic approaches (Blackburn et al. 2015; Young
all nonenzymatic antioxidants and chloroplastic pig- 2018). In demographic studies, telomere length also
ments (related to oxidative stress in chloroplasts) can emerges as a promising biomarker of ageing, survival,
be measured using chromatography-based techniques remaining lifespan, or stress accumulated over life.
(Esteban et al. 2015; Halliwell and Gutteridge 2015). This interest stems from the presumable association
A variety of in vitro assays have been developed between cellular replicative senescence and mortality,
intended to provide an integrated measure of total and from observations that stressors, including oxida-
antioxidant capacity (TAC) (reviewed in Somogyi et al. tive stress, increase telomere attrition during each cell
2007; Monaghan et al. 2009; Halliwell and Gutteridge division (Blackburn et al. 2015; Eastwood et al. 2018;
2015). Recently, a ratio of stereoisomers of hydroxy- Wilbourn et al. 2018). Moreover, telomere shortening
octadecadienoic acid, an oxidation product of linoleic rate, rather than telomere length itself, has been report-
acid, has been proposed as a promising marker of ed to predict species lifespan in birds and mammals
in vivo lipophilic antioxidant capacity (Yoshida et al. in comparative studies (Tricola et al. 2018; Whittemore
2007; Tomášek et al. 2016). et al. 2019; Udroiu 2020).
Oxidative stress can be measured in any tissue or We also note here that, while telomeres are of contin-
body fluid, although nondestructive or noninvasive ued interest in the physiological ecology community,
sample types are preferable in longitudinal studies. they are increasingly being questioned as a biomarker
In plants, light conditions during sampling need to of human ageing (Sanders and Newman 2013; Ferrucci
be standardised as oxidative stress may increase with et al. 2020). Several studies have shown that telom-
the accumulation of intense light hours. Unless intact ere length correlates poorly with other measures of the
cells are required, samples must immediately be frozen ageing process (Belsky et al. 2018). It is increasingly
in liquid nitrogen and stored at −80◦ C or preferably viewed as a manifestation of psychological stress and
in liquid nitrogen (Barden et al. 2014). It is usually allostatic load (Meier et al. 2019), but not of ageing
per se, though there is still some debate. The extent to changes in the balance between innate and acquired
which these human findings translate into the phys- immune function that were previously considered to
iological ecology literature is not clear, but given the be ‘immunosenescence’ have recently been proposed
history of physiological ecologists jumping naively on to be adaptive changes in the immune system based
physiological bandwagons emerging from human or on changing needs across ages and on ageing in oth-
biomedical literature (Viney et al. 2005; Bonier et al. er aspects of physiology (Fulop et al. 2017). While
2009; Isaksson et al. 2011; Cohen et al. 2012), we urge the truth may be somewhere in the middle, it is clear
some caution. that we do not fully understand how to distinguish
Multiple methods of telomere length measurement adaptive versus pathological changes in immune state,
have been developed, including terminal fragment nor indications of poor condition versus alternative
(TRF) analysis, real-time qPCR, quantitative fluores- strategies.
cence in situ hybridisation (Q-FISH), single telomere In vertebrates, including humans, the most basic
length analysis (STELA), and dot blot (reviewed in measures are cell counts from slides quantifying sev-
Aubert et al. 2012; Montpetit et al. 2014; Nussey et al. eral key cell types. Beyond this, there are numerous
2014). Nonetheless, none of the methods is ideal, and specific assays for cytokines implicated in regulating
the choice of an appropriate method for a given study various aspects of immune response, as well as flow
depends on whether the desired qualities include anal- cytometry approaches to get cell counts for precise cell
ysis of not only the mean, but also the whole distribu- types based on surface markers (Lu et al. 2016). Under
tion of telomere lengths, high throughput, low amount field conditions, a series of field-adapted immune mea-
of DNA needed, a lack of interference by interstitial sures have been developed. In addition to PHA and
telomeric (e.g. centromeric) sequences, or the ability to slide-based cell counts, key assays include haptoglobin
detect very short telomeres that are the most relevant (an acute phase protein), immunoglobulin concen-
to fitness (Nussey et al. 2014). Other factors that need trations and globulin concentrations more generally
to be considered include comparability between stud- (Cheynel et al. 2017), hemagglutination and lysis (indi-
ies (Verhulst 2020), methods of long-term storage and cators of natural antibody response in the innate sys-
extraction that ensure the highest DNA quality (East- tem; Millet et al. 2007), and bacterial killing assays
wood et al. 2018), or specialised equipment required. (Millet et al. 2007). Nevertheless, there are notable
Tissue type collected is yet another important factor to challenges. First, the correlation structure among these
consider as different cell types (even different leuco- assays is unstable, such that there is no agreed-upon
cytes, for example) can vary greatly in proliferation rate way to measure overall immune status, nor several key
or telomerase activity, and thus in telomere dynamics axes (Matson et al. 2006; Buehler et al. 2011; Pigeon
over lifetime (Nussey et al. 2014). et al. 2013). Second, a key aspect of immune function,
the acquired immune response, is notably absent from
field assays, yielding uncertainty whether low levels of
2.7 Immune markers innate immunity (the primary subject of most assays)
Immune function is a crucial aspect of physiology simply reflect investments in adaptive immunity, or
linking various physiological systems (endocrine, neu- insufficient levels of protection. Third, while bacteri-
robiological, metabolic, etc.), psychology, stress, and al killing assays may seem to be a promising way to
various aspects of the environment (pathogen pres- integrate overall function, in practice, results can be
sure, nutrition, etc.). Accordingly, it is a potent signal hard to interpret: scores for different strains of bac-
of organismal state, but also challenging to quantify teria/yeast are often uncorrelated (Millet et al. 2007),
and interpret. The immune system is immensely com- and it is unclear how increased investment in protec-
plex, with innate and acquired components, cellular tion against bacteria might trade off with other kinds of
and humoral components, and dozens if not hundreds protections, such as against viruses or macro-parasites.
of specific cell types, all linked in complex regulato- The most promising uses of immune markers in
ry networks. Therefore, it is increasingly accepted that demography will thus likely rely on specific vali-
no single measure is sufficient to understand immune dation of particular assays/approaches in a particu-
variation (Matson et al. 2006; Buehler et al. 2011). lar context. For example, in modern humans, chronic
Moreover, differential investment in different parts of low-grade inflammation is a major cause of chron-
the immune system may reflect alternative strategies, ic diseases and generally accompanies the ageing
either within individuals over time or across individ- process (Franceschi et al. 2000). Individual cytokines
uals (Pulendran 2004; Viney et al. 2005). For example, involved in the process may also vary for other reasons
(e.g. with acute infection), such that a more reliable identified using a machine learning algorithm that
signal of this ‘inflamm-ageing’ can be obtained by sta- together can generate an index of biological age that
tistical integration of several key cytokines via prin- correlates with chronological age at r = 0.96 across a
cipal components analysis (Morrisette-Thomas et al. wide array of tissues (Horvath 2013). While the sta-
2014; Cohen et al. 2018). This analysis shows that the bility of the epigenetic clock is a striking biological
process involves upregulation of both pro- and anti- phenomenon, caution is also warranted when inter-
inflammatory cytokines, indicating activation of the preting such indices. It is not clear to what extent the
system rather than pro-anti balance. Accordingly, it is clock is measuring biological ageing per se, versus
possible to obtain precise measurement of a specific methylation sites that change with age-correlated phe-
immune process that is of particular demographic nomena despite having little link to the ageing process
importance in modern human populations; however, (Hertel et al. 2019). While it is clear that the omics rev-
this index would likely be of little use in other species olution holds great promise for demography, we are of
or even premodern humans. A second example is the the opinion that application in most cases would still
use of antihelminth-specific antibodies in Soay sheep be premature, except when a precise question is asked
(Ovis aries), levels of which predict survival and body using a well-validated method. One example could be
mass in the study population, though with relation- using heat shock proteins (i.e. targeted panel of select-
ships highly specific to age and isotype (Sparks et al. ed protein markers) as biomarkers of stress (Feder and
2018). This study shows that detailed natural histo- Hofmann 1999).
ry can provide highly tailored approaches to specific
contexts, even if they are not more broadly applica-
ble across species and populations. In fact, we should
2.9 Biologging
expect such specificity to be the norm in most cases. With the development of new technologies, new sets
of tools are now available to measure physiologi-
cal traits outside laboratory environment. The term
2.8 Epigenetic and omics data biologging has been proposed to describe a process by
The recent explosion in capacity to measure large quan- which researchers gain information (typically position,
tities of biological information (‘omics’) cheaply has behaviour, movement, or physiological status) from an
led to a revolution in our ability to measure genotype individual remotely (Payne et al. 2014). In animals and
and phenotype. Roughly speaking, the main types of humans, biologging units are now widely available
omics data currently available include (1) transcrip- and consist of miniaturised electronic data recorders
tomics, quantifying messenger mRNA levels (i.e. gene fitted on tags, badges, or collars. These devices allow
expression); (2) epigenomics, usually DNA methyla- the measurement of individual vital function, includ-
tion data, which influences the likelihood of gene ing body position, activity, and physiology on prolong
expression; (3) metabolomics, measuring large num- period (Kavanagh and Menz 2008). Biological data
bers of circulating small molecules, often intermediates generally recorded on biologgers, include body tem-
of various metabolic pathways; and (4) proteomics, perature, heart rate and 3D body acceleration. A depth
quantifying the levels of many proteins. While many sensor and captor for dissolved oxygen levels can also
of these methods are still too expensive for routine use be included on device used in aquatic environment
in field contexts, that may be changing as technology (Humphries et al. 2010). Coupled with data on envi-
improves and prices fall. Luckily, our experience with ronmental variables, body temperature and heart rate
metabolomics suggests that price per sample may be data can be used to reconstruct energetic landscape
reduced considerably when large numbers of samples (Steiger et al. 2009; Scharf et al. 2016). Accelerometers
are analysed. record whole- or partial-body acceleration in one, two
The quantity of data generated has not been fully or three spatial axes simultaneously (Payne et al. 2014).
accompanied by ways to make sense of it. Multiple These data can then be used to determine individual
testing issues become daunting when assessing thou- behaviour and derive metric for energy effort (Halsey
sands of molecules simultaneously, particularly with et al. 2009). More recently, it had been suggested that
the limited sample sizes often available in field stud- body-acceleration data can also be used to assess indi-
ies. In some cases, algorithms have been generated that vidual’s internal states including health states (Wilson
can simplify large numbers of molecules for specif- et al. 2014). In plants, biologging approach include
ic purposes. Perhaps the best example is the human (hyperspectral) cameras and eddy covariance tow-
epigenetic clock, a set of 353 DNA methylation sites ers used to analyse individual colour/multispectral
changes and gas exchange dynamics at the communi- the sources of variation yields understanding of how
ty level, respectively (Jalota et al. 2018; Khanna et al. the system works. Hence, methodological research will
2019). often answer substantive ecological, evolutionary, or
medical questions. The most challenging task will cer-
tainly be to validate the link between a biomarker and
2.10 Combining biomarkers fitness components, as their relationship may often
It is rare that a single biomarker or physiological be context dependent. As a result, the potential of
measure can provide a clear, stable signature of biomarkers to predict fitness remains largely unknown
demography-relevant processes, and when it is pos- (Salguero-Gómez et al. 2018). Nevertheless, exploring
sible, it usually requires extensive validation in the this potential is of ultimate importance because func-
study population (Sparks et al. 2018). One approach tional traits, including physiological and biochemical
to circumvent this problem has been to measure mul- traits, are defined as features that impact fitness (Violle
tiple markers simultaneously and to use statistical et al. 2007). We also advocate efforts to standardise pro-
approaches to integrate the markers. Principal com- tocols of physiological and biochemical data collection
ponents analysis and similar methods can elegantly and analysis across populations and species, and to
extract a joint signal from many variables (Matson et al. collect these data in global, open-source repositories.
2006; Buehler et al. 2011, Renaud et al. 2019), and sta- Such endeavours are crucial for addressing questions
tistical distance techniques can quantify homeostatic on large spatial and taxonomical scales.
dysregulation (Cohen et al. 2014; Milot et al. 2014).
The challenge, however, is that both these methods
rely heavily on the correlation structure among the Acknowledgements
markers, and this structure can often change marked- We thank Lumír Gvoždík for constructive com-
ly across time and space, depending on conditions ments on animal metabolic rate measurement, and
(Matson et al. 2006; Cohen and McGraw 2009; Buehler Jean-François Lemaître and one anonymous review-
et al. 2011; Pigeon et al. 2013). It appears that such er for valuable comments on the early version of the
methods may thus work better under relatively homo- manuscript. OT was supported by three grants from
geneous conditions (e.g. captivity) than in highly var- the Czech Science Foundation (GA17–24782S; GA19–
ied conditions typical in the wild (Fowler et al. 2018). 22538S; GA21-22160S). This work was further funded
An alternative approach is to decompose the correla- by an NSERC Discovery Grant (402079–2011) to AAC.
tion structure at different ecologically relevant levels AAC is supported by a CIHR New Investigator Salary
of variation (Renaud et al. 2019); while promising, Award and is a member of the FRQ-S funded Cen-
this approach has yet to be widely validated. Similar tre de recherche du CHUS and Centre de recherche sur le
challenges are likely to be present with omics data. vieillissement.
2.11 Conclusions and perspectives

References
Integration of physiological and biochemical data
Abdul-Hamid H. and Mencuccini M. (2009). Age- and
into demographic and population ecology research is size-related changes in physiological characteristics and
emerging as a promising approach that may bring chemical composition of Acer pseudoplatanus and Fraxinus
advances in multiple branches of ecology, conserva- excelsior trees. Tree Physiology 29, 27–38.
tion, and human health and ageing research (Violle Asensi-Fabado M. A. and Munné-Bosch S. (2010). Vitamins
et al. 2007, 2014; Cooke et al. 2013; Cohen et al. 2014; in plants: occurrence, biosynthesis and antioxidant func-
Cohen and Milot 2015; Salguero-Gómez et al. 2018). tion. Trends in Plant Science 15, 582–592.
Such an approach also has its challenges, however. Aubert G., Hills M., and Lansdorp P. M. (2012). Telom-
We must emphasise the need to stick to stringent val- ere length measurement—caveats and a critical assess-
ment of the available technologies and tools. Mutation
idation procedures, including validation of the ana-
Research/Fundamental and Molecular Mechanisms of
lytical methods, exploration of the extent and sources
Mutagenesis 730, 59–67.
of biomarker variation in a population and ultimate- Barden A. E., Mas E., Croft K. D., Phillips M., and Mori
ly validation of biomarker importance to fitness. This T. A. (2014). Minimizing artifactual elevation of lipid
will always be a tedious and challenging task but, as peroxidation products (F2-isoprostanes) in plasma dur-
recognised by Hõrak and Cohen (2010), understanding collection and storage. Analytical Biochemistry 449,
ing how to properly measure the system and what are 129–131.
Bauch C., Wellbrock A. H. J., Nagel R., Rozman J., and Witte but not between sexes in a long-lived mammal. Scientific
K. (2013). ‘Bug-eggs’ for common swifts and other small Reports 7, 13700.
birds: minimally-invasive and stress-free blood sampling Cohen A. A. and McGraw K. J. (2009). No simple measures
during incubation. Journal of Ornithology 154, 581–585. for antioxidant status in birds: complexity in inter- and
Belsky D. W., Moffitt T. E., Cohen A. A., et al. (2018). Eleven intraspecific correlations among circulating antioxidant
telomere, epigenetic clock, and biomarker-composite types. Functional Ecology 23, 310–320.
quantifications of biological aging: do they measure the Cohen A. A., Martin L. B., Wingfield J. C., McWilliams
same thing? American Journal of Epidemiology 187, S. R., and Dunne J. A. (2012). Physiological regulatory
1220–1230. networks: ecological roles and evolutionary constraints.
Bílková, Z., Adámková M., Albrecht T., and Šimek Z. Trends in Ecology & Evolution 27, 428–435.
(2019). Determination of testosterone and corticosterone Cohen A. A., Milot E., Li Q., Legault V., Fried L. P.,
in feathers using liquid chromatography-mass spectrom- and Ferrucci L. (2014). Cross-population validation of
etry. Journal of Chromatography A 1590, 96–103. statistical distance as a measure of physiological dys-
Birkhead T. R., Hosken D. J., and Pitnick S. (2009). Sperm regulation during aging. Experimental Gerontology 57,
Biology: An Evolutionary Perspective. Academic Press. 203–210.
Blackburn E. H., Epel E. S., and Lin J. (2015). Human Cohen A. A., Milot E., Li Q., et al. (2015). Detection of a novel,
telomere biology: a contributory and interactive fac- integrative aging process suggests complex physiological
tor in aging, disease risks, and protection. Science 350, integration. PLOS One 10, e0116489.
1193–1198. Cohen A. A., Bandeen-Roche K., Morissette-Thomas V., and
Bonier F., Martin P. R., Moore I. T., and Wingfield J. C. Fülöp T. (2018). A robust characterization of inflamm-
(2009). Do baseline glucocorticoids predict fitness? Trends aging and other immune processes through multivariate
in Ecology & Evolution 24, 634–642. analysis of cytokines from longitudinal studies. In: T.
Breuner C. W., Patterson S. H., and Hahn T. P. (2008). Fülöp, C., Franceschi, K. Hirokawa, and G. Pawelec (eds.),
In search of relationships between the acute adreno- Handbook of Immunosenescence: Basic Understanding
cortical response and fitness. General and Comparative and Clinical Implications, 1–16. Springer.
Endocrinology 157, 288–295. Cooke S. J., Sack L., Franklin C. E. et al. (2013). What is
Breuner C. W., Delehanty B., and Boonstra R. (2013). conservation physiology? Perspectives on an increasingly
Evaluating stress in natural populations of vertebrates: integrated and essential science. Conservation Physiology
total CORT is not good enough. Functional Ecology 27, 1, cot001.
24–36. Costantini D., Rowe M., Butler M. W., and McGraw K. J.
Brooks R. C. and Garratt M. G. (2017). Life history evolution, (2010). From molecules to living systems: historical and
reproduction, and the origins of sex-dependent aging and contemporary issues in oxidative stress and antioxidant
longevity. Annals of the New York Academy of Sciences ecology. Functional Ecology 24, 950–959.
1389, 92–107. Dantzer B., Fletcher Q. E., Boonstra R., and Sheriff M.
Buehler D. M., Versteegh M. A., Matson K. D., and Tiele- J. (2014). Measures of physiological stress: a transpar-
man B. I. (2011). One problem, many solutions: simple ent or opaque window into the status, management
statistical approaches help unravel the complexity of the and conservation of species? Conservation Physiology 2,
immune system in an ecological context. PLOS One 6, cou023.
e18592. Dawson T. E., Mambelli S., Plamboeck A. H., Templer
Burton T., Killen S. S., Armstrong J. D., and Metcalfe P. H., and Tu K. P. (2002). Stable isotopes in plant ecol-
N. B. (2011). What causes intraspecific variation in resting ogy. Annual Review of Ecology and Systematics 33,
metabolic rate and what are its ecological consequences? 507–559.
Proceedings of the Royal Society B: Biological Sciences Demidchik V. (2015). Mechanisms of oxidative stress in
278, 3465–3473. plants: from classical chemistry to cell biology. Environ-
Cattet M., Stenhouse G. B., Janz D. M. et al. (2017). The quan- mental and Experimental Botany 109, 212–228.
tification of reproductive hormones in the hair of captive Dotan Y., Lichtenberg D., and Pinchuk I. (2004). Lipid
adult brown bears and their application as indicators of peroxidation cannot be used as a universal criteri-
sex and reproductive state. Conservation Physiology 5, on of oxidative stress. Progress in Lipid Research 43,
cox032. 200–227.
Cattet M., Stenhouse G. B., Boulanger J. et al. (2018). Can con- Drinkwater E., Robinson E. J. H., and Hart A. G. (2019).
centrations of steroid hormones in brown bear hair reveal Keeping invertebrate research ethical in a landscape of
age class? Conservation Physiology 6, coy001. shifting public opinion. Methods in Ecology and Evolu-
Chabot, D., Steffensen, J. F., and Farrell, A. P. (2016). The tion 10, 1265–1273.
determination of standard metabolic rate in fishes. Journal Eastwood J. R., Mulder E., Verhulst S., and Peters A. (2018).
of Fish Biology 88, 81–121. Increasing the accuracy and precision of relative telom-
Cheynel L., Lemaître J.-F., Gaillard J.-M. et al. (2017). ere length estimates by RT qPCR. Molecular Ecology
Immunosenescence patterns differ between populations Resources 18, 68–78.
Edmunds P. J., Burgess S. C., Putnam H. M. et al. (2014). Hertel J., Frenzel S., König, J. et al. (2019). The informative
Evaluating the causal basis of ecological success within error: a framework for the construction of individualized
the scleractinia: an integral projection model approach. phenotypes. Statistical Methods in Medical Research 28,
Marine Biology 161, 2719–2734. 1427–1438.
Esteban R., Barrutia O., Artetxe U., Fernández-Marín B., Hollmen T., Franson J., Hario M., Sankari S., Kilpi M.,
Hernández A., and García-Plazaola J. I. (2015). Internal and Lindstrom K. (2001). Use of serum biochemistry to
and external factors affecting photosynthetic pigment evaluate nutritional status and health of incubating com-
composition in plants: a meta-analytical approach. The mon eiders (Somateria mollissima) in Finland. Physiological
New Phytologist 206, 268–280. and Biochemical Zoology 74, 333–342.
Feder M. E. and Hofmann G. E. (1999). Heat-shock pro- Hõrak P. and Cohen A. A. (2010). How to measure
teins, molecular chaperones, and the stress response— oxidative stress in an ecological context: method-
evolutionary and ecological physiology. Annual Review ological and statistical issues. Functional Ecology 24,
of Physiology 61, 243–282. 960–970.
Ferrucci L., Gonzalez-Freire M., Fabbri E. et al. (2020). Mea- Horvath S. (2013). DNA methylation age of human tissues
suring biological aging in humans: a quest. Aging Cell 19, and cell types. Genome Biology 14, R115.
e13080. Humann-Guilleminot S., Blévi P., Azou-Barré A. et al.
Finch C. and Rose M. (1995). Hormones and the (2018). Sperm collection in black-legged kittiwakes and
physiological architecture of life-history evolution. characterization of sperm velocity and morphology.
Quarterly Review of Biology 70, 1–52. Avian Research 9, 24.
Fitzpatrick J. L. and Lüpold S. (2014). Sexual selection and Humphries N. E., Queiroz N., Dyer J. R. M. et al. (2010).
the evolution of sperm quality. Molecular Human Repro- Environmental context explains Lévy and Brownian
duction 20, 1180–1189. movement patterns of marine predators. Nature 465,
Fowler M. A., Paquet M., Legault V., Cohen A. A., and 1066–1069.
Williams T. D. (2018). Physiological predictors of repro- Isaksson C., Sheldon B. C., and Uller T. (2011). The chal-
ductive performance in the European starling (Sturnus lenges of integrating oxidative stress into life-history biol-
vulgaris). Frontiers in Zoology 15, 45. ogy. BioScience 61, 194–202.
Franceschi C., Bonafè M., Valensin S., et al. (2000). Inflamm- Jalota S. K., Vashisht B. B., Sharma S., and Kaur S. (2018).
aging. An evolutionary perspective on immunosenes- Emission of greenhouse gases and their warming effect.
cence. Annals of the New York Academy of Sciences 908, In: S. K. Jalota, B. B. Vashisht, S. Sharma,and S. Kaur
244–254. (eds.), Understanding Climate Change Impacts on Crop
Fu X. and Meinzer F. C. (2019). Metrics and proxies Productivity and Water Balance, 1–53. Academic Press.
for stringency of regulation of plant water status Jansen E. H. J. M., Beekhof P. K., Viezeliene D., Muzakova
(iso/anisohydry): a global data set reveals coordina- V., and Skalicky J. (2017). Long-term stability of oxidative
tion and trade-offs among water transport traits. Tree stress biomarkers in human serum. Free Radical Research
Physiology 39, 122–134. 51, 970–977.
Fuchs S., Leuschner C., Link R., Coners H., and Schuldt Jimeno B., Hau M., and Verhulst S. (2018). Corticosterone
B. (2017). Calibration and comparison of thermal dissi- levels reflect variation in metabolic rate, independent of
pation, heat ratio and heat field deformation sap flow ‘stress.’ Scientific Reports 8, 13020.
probes for diffuse-porous trees. Agricultural and Forest Johnson S. L., Dunleavy J., Gemmell N. J., and Nakagawa
Meteorology 244–245, 151–161. S. (2015). Consistent age-dependent declines in human
Fulop T., Larbi A., Dupuis G. et al. (2017). Immunosenes- semen quality: a systematic review and meta-analysis.
cence and inflamm-aging as two sides of the same coin: Ageing Research Reviews 19, 22–33.
friends or foes? Frontiers in Immunology 8, 1960. Jones D. P. (2006). Redefining oxidative stress. Antioxi
Gomendio M., Malo A. F., Soler A. J. et al. (2006). Male dants & Redox Signaling 8, 1865–1879.
fertility and sex ratio at birth in red deer. Science 314, Kavanagh J. J. and Menz H. B. (2008). Accelerometry: a tech-
1445–1447. nique for quantifying movement patterns during walking.
Green J. M. (1996). A practical guide to analytical method Gait & Posture 28, 1–15.
validation. Analytical Chemistry 68, 305A–309A. Khanna R., Schmid L., Walter A., Nieto J., Siegwart R., and
Halliwell B. and Gutteridg, J. M. C. (2015). Free Radicals in Liebisch F. (2019). A spatio temporal spectral framework
Biology and Medicine. Oxford University Press. for plant stress phenotyping. Plant Methods 15, 13.
Halsey L. G., Portugal S. J., Smith J. A., Murn C. P., and Landhäusser S. M., Chow P. S., Dickman L. T. et al. (2018).
Wilson R. P. (2009). Recording raptor behavior on the Standardized protocols and procedures can precisely and
wing via accelerometry. Journal of Field Ornithology 80, accurately quantify non-structural carbohydrates. Tree
171–177. Physiology 38, 1764–1778.
Harr K. E. (2005). Diagnostic value of biochemistry. In: Laskemoen T., Albrecht T., Bonisoli-Alquati A. et al.
G. Harrison and T. Lightfoot (eds.), Clinical Avian (2013). Variation in sperm morphometry and sperm
Medicine, 611–603. Spix Publishing. competition among barn swallow (Hirundo rustica)
populations. Behavioral Ecology and Sociobiology 67, models: a practical guide. Methods in Ecology and Evo-
301–309. lution 5, 99–110.
Lemaître J.-F. and Gaillard J.-M. (2017). Reproductive senes- Millet S., Bennett J., Lee K. A., Hau M., and Klasing K. C.
cence: new perspectives in the wild. Biological Reviews (2007). Quantifying and comparing constitutive immuni-
92, 2182–2199. ty across avian species. Developmental and Comparative
Lierz M., Reinschmidt M., Müller H., Wink M., and Immunology 31, 188–201.
Neumann D. (2013). A novel method for semen collection Milot E., Cohen A. A., Vezina F., Buehler D. M., Matson K.
and artificial insemination in large parrots (Psittaciformes). D., and Piersma T. (2014). A novel integrative method
Scientific Reports 3, 1–8. for measuring body condition in ecological studies based
Lifjeld J., Laskemoen T., Kleven O., Albrecht T., and Robert- on physiological dysregulation. Methods in Ecology and
son R. (2010). Sperm length variation as a predictor Evolution 5, 146–155.
of extrapair paternity in passerine birds. PLOS One 5, Monaghan P., Metcalfe N. B., and Torres R. (2009). Oxida-
e13456. tive stress as a mediator of life history trade-offs: mecha-
Lighton J. R. B. (2019). Measuring Metabolic Rates: A Man- nisms, measurements and interpretation. Ecology Letters
ual for Scientists, 2nd edition. Oxford University Press. 12, 75–92.
Lindsjö J., Fahlman Å., and Törnqvist E. (2016). Animal wel- Monneret D., Godmer A., Le Guen R. et al. (2016). Stability
fare from mouse to moose—implementing the principles of routine biochemical analytes in whole blood and plas-
of the 3Rs in wildlife research. Journal of Wildlife Diseases ma from lithium heparin gel tubes during 6-hr storage.
52, S65–S77. Journal of Clinical Laboratory Analysis 30, 602–609.
Long S. P. and Bernacchi C. J. (2003). Gas exchange Montpetit A., Alhareeri A., Montpetit M. et al. (2014). Telom-
measurements, what can they tell us about the under- ere length: a review of methods for measurement. Nursing
lying limitations to photosynthesis? Procedures and Research 63, 289–299.
sources of error. Journal of Experimental Botany 54, Morales M. and Munné-Bosch S. (2019). Malondialdehyde:
2393–2401. facts and artifacts. Plant Physiology 180, 1246–1250.
Lu Y., Tan C. T. Y., Nyunt M. S. Z. et al. (2016). Inflammato- Morrisette-Thomas V., Cohen A. A., Fülöp T. et al. (2014).
ry and immune markers associated with physical frailty Inflamm-aging does not simply reflect increases in pro-
syndrome: findings from Singapore longitudinal aging inflammatory markers. Mechanisms of Ageing and Devel-
studies. Oncotarget 7, 28783–28795. opment 139, 49–57.
Martínez-Vilalta, J. Vanderklein, D. and Mencuccini M. Müller M. and Munné-Bosch S. (2011). Rapid and sensitive
(2007). Tree height and age-related decline in growth hormonal profiling of complex plant samples by liquid
in Scots pine (Pinus sylvestris L.). Oecologia 150, chromatography coupled to electrospray ionization tan-
529–544. dem mass spectrometry. Plant Methods 7, 37.
Mateos R. and Bravo L. (2007). Chromatographic and elec- Murchie E. H. and Lawson T. (2013). Chlorophyll fluo-
trophoretic methods for the analysis of biomarkers of rescence analysis: a guide to good practice and under-
oxidative damage to macromolecules (DNA, lipids, and standing some new applications. Journal of Experimental
proteins). Journal of Separation Science 30, 175–191. Botany 64, 3983–3998.
Matson K. D., Cohen A. A., Klasing K. C., Ricklef R. E., Norin T. and Metcalfe N. B. (2019). Ecological and evolution-
and Scheuerlein A. (2006). No simple answers for ecolog- ary consequences of metabolic rate plasticity in response
ical immunology: relationships among immune indices at to environmental change. Philosophical Transactions of
the individual level break down at the species level in the Royal Society B: Biological Sciences 374, 20180180.
waterfowl. Proceedings of the Royal Society B: Biological Nussey D. H., Baird D., Barrett E., et al. (2014). Measuring
Sciences 273, 815–822. telomere length and telomere dynamics in evolutionary
Meier H. C. S., Hussein M., Needham B. et al. (2019). Cellular biology and ecology. Methods in Ecology and Evolution
response to chronic psychosocial stress: ten-year longi- 5, 299–310.
tudinal changes in telomere length in the Multi-Ethnic Olsson M., Wilson M., Uller T. et al. (2008). Free radicals run
Study of Atherosclerosis. Psychoneuroendocrinology 107, in lizard families. Biology Letters 4, 186–188.
70–81. Owen J. C. (2011). Collecting, processing, and storing
Mencuccini M. and Munné-Bosch S. (2017). Physiological avian blood: a review. Journal of Field Ornithology 82,
and biochemical processes related to ageing and senes- 339–354.
cence in plants. In: R. P. Shefferson, O. R. Jones, and R. Pamplona R. and Costantini D. (2011). Molecular and struc-
Salguero-Gómez (eds.), The Evolution of Senescence in tural antioxidant defenses against oxidative stress in
the Tree of Life, 257–283. Cambridge University Press. animals. American Journal of Physiology—Regulatory,
Mencuccini M., Oñate M., Peñuelas J., Rico L., and Integrative and Comparative Physiology 301, R843–R863.
Munné-Bosch S. (2014). No signs of meristem senescence Payne N. L., Taylor M. D., Watanabe Y. Y., and Semmens
in old Scots pine. Journal of Ecology 102, 555–565. J. M. (2014). From physiology to physics: are we rec-
Merow C., Dahlgren J. P., Metcalf C. J. E. et al. (2014). ognizing the flexibility of biologging tools? Journal of
Advancing population ecology with integral projection Experimental Biology 217, 317–322.
Pérez-Harguindeguy N., Díaz S., Garnier E. et al. (2013). Šandera M., Albrecht T., and Stopka P. (2013). Variation in
New handbook for standardised measurement of plant apical hook length reflects the intensity of sperm compe-
functional traits worldwide. Australian Journal of Botany tition in murine rodents. PLOS One 8, e68427.
61, 167–234. Sanders J. L. and Newman A. B. (2013). Telomere length in
Pettersen A. K., Marshall D. J., and White C. R. (2018). epidemiology: a biomarker of aging, age-related disease,
Understanding variation in metabolic rate. Journal of both, or neither? Epidemiologic Reviews 35, 112–131.
Experimental Biology 221, jeb166876. Scharf A. K., LaPoint S., Wikelski M., and Safi K. (2016).
Pigeon G., Bélisle M., Garant D., Cohen A. A., and Pel- Acceleration data reveal highly individually structured
letier F. (2013). Ecological immunology in a fluctuating energetic landscapes in free-ranging fishers (Pekania pen-
environment: an integrative analysis of tree swallow nanti). PLOS One 11, e0145732.
nestling immune defense. Ecology and Evolution 3, Semerci A., Semerci H., Çalişkan B., Çiçek N., Ekmekçi Y.,
1091–1103. and Mencuccini M. (2017). Morphological and physiolog-
Piłacik, B., Nofer, T. W., and Wasowicz, W. (2002). F2- ical responses to drought stress of European provenances
isoprostanes biomarkers of lipid peroxidation: their utility of Scots pine. European Journal of Forest Research 136,
in evaluation of oxidative stress induced by toxic agents. 91–104.
International Journal of Occupational Medicine and Envi- Sheriff M. J., Dantzer B., Delehanty B., Palme R., and
ronmental Health 15, 19–27. Boonstra R. (2011). Measuring stress in wildlife: tech-
Podlesak D. W., McWilliams S. R., and Hatch K. A. (2005). niques for quantifying glucocorticoids. Oecologia 166,
Stable isotopes in breath, blood, feces and feathers can 869–887.
indicate intra-individual changes in the diet of migratory Shipley B., De Bello F., Cornelissen J. H. C., Laliberté E.,
songbirds. Oecologia 142, 501–510. Laughlin D. C., and Reich P. B. (2016). Reinforcing loose
Poiani A. (2006). Complexity of seminal fluid: a review. foundation stones in trait-based plant ecology. Oecologia
Behavioral Ecology and Sociobiology 60, 289–310. 180, 923–931.
Pulendran B. (2004). Modulating Th1/Th2 responses with Somogyi A., Rosta K., Puszta, P., Tulassay Z., and Nagy
microbes, dendritic cells, and pathogen recognition recep- G. (2007). Antioxidant measurements. Physiological Mea-
tors. Immunologic Research 29, 187–196. surement 28, R41–R55.
Queval G. and Noctor G. (2007). A plate reader method Sparks A. M., Watt K., Sinclair R. et al. (2018). Natural
for the measurement of NAD, NADP, glutathione, and selection on antihelminth antibodies in a wild mammal
ascorbate in tissue extracts: application to redox profil- population. The American Naturalist 192, 745–760.
ing during Arabidopsis rosette development. Analytical Speakman J. (1997). Doubly Labelled Water: Theory and
Biochemistry 363, 58–69. Practice. Springer.
Ramirez J. A., Posada J. M., Handa I. T. et al. (2015). Near- Speakman J. R. and Hambly C. (2016). Using doubly-
infrared spectroscopy (NIRS) predicts non-structural car- labelled water to measure free-living energy expenditure:
bohydrate concentrations in different tissue types of a some old things to remember and some new things to
broad range of tree species. Methods in Ecology and consider. Comparative Biochemistry and Physiology. Part
Evolution 6, 1018–1025. A, Molecular & Integrative Physiology 202, 3–9.
Réale D., Garant D., Humphries M. M., Bergeron P., Steiger S. S., Kelley J. P., Cochran W. W., and Wikelski M.
Careau V., and Montiglio P.-O. (2010). Personality and (2009). Low metabolism and inactive lifestyle of a tropical
the emergence of the pace-of-life syndrome concept at the rain forest bird investigated via heart-rate telemetry.
population level. Philosophical Transactions of the Royal Physiological and Biochemical Zoology 82, 580–589.
Society of London B: Biological Sciences 365, 4051–4063. Sturgeon C. M. and McAllister E. J. (1998). Analysis of hCG:
Reinhardt K., Dobler R., and Abbott J. (2015). An ecolo- clinical applications and assay requirements. Annals of
gy of sperm: sperm diversification by natural selection. Clinical Biochemistry 35, 460–491.
Annual Review of Ecology, Evolution, and Systematics 46, Tabunoki H., Dittmer N. T., Gorman M. J., and Kanost
435–459. M. R. (2019). Development of a new method for collect-
Renaud L.-A., Blanchet F. G., Cohen A. A., and Pelletier ing hemolymph and measuring phenoloxidase activity in
F. (2019). Causes and short-term consequences of varia- Tribolium castaneum. BMC Research Notes 12, 7.
tion in milk composition in wild sheep. Journal of Animal Tomášek O., Gabrielova B., Kacer P. et al. (2016). Opposing
Ecology, 88, 857–869. effects of oxidative challenge and carotenoids on antiox-
Sala A. and Hoch G. (2009). Height-related growth declines idant status and condition-dependent sexual signalling.
in ponderosa pine are not due to carbon limitation. Plant, Scientific Reports 6, 23546.
Cell & Environment 32, 22–30. Tomášek O., Albrechtova J., Nemcova M., Opatova P., and
Salguero-Gómez R., Violle C., Gimenez O., and Childs D. Albrecht T. (2017). Trade-off between carotenoid-based
(2018). Delivering the promises of trait-based approaches sexual ornamentation and sperm resistance to oxidative
to the needs of demographic approaches, and vice versa. challenge. Proceedings of the Royal Society B: Biological
Functional Ecology 32, 1424–1435. Sciences 284, 20162444.
Tomášek O., Bobek L., Kralova T., Adamkova M., and Violle C., Thuiller W., Mouquet N. et al. (2017). Functional
Albrecht T. (2019). Fuel for the pace of life: baseline blood rarity: the ecology of outliers. Trends in Ecology &
glucose concentration co-evolves with life-history traits in Evolution 32, 356–367.
songbirds. Functional Ecology 33, 239–249. Walther A., Philipp M., Lozza N., and Ehlert U. (2016).
Tricola G.M., Simons M.J.P., Atema E. et al. (2018). The The rate of change in declining steroid hormones: a
rate of telomere loss is related to maximum lifespan in new parameter of healthy aging in men? Oncotarget 7,
birds. Philosophical Transactions of the Royal Society B: 60844–60857.
Biological Sciences 373, 20160445. Wasden M. B., Roberts R. L., and DeLaurier A. (2017).
Turner N. C. (1988). Measurement of plant water status Optimizing sperm collection procedures in zebrafish.
by the pressure chamber technique. Irrigation Science 9, Journal of the South Carolina Academy of Science
289–308. 15, 7.
Udroiu I. (2020). On the correlation between telomere short- Wheeler M. J. (2006). Assays for LH, FSH, and pro-
ening rate and life span. Proceedings of the National lactin. Methods in Molecular Biology (Clifton, NJ) 324,
Academy of Sciences 117, 2248–2249. 109–124.
Valeggia C. R. (2007). Taking the lab to the field: monitoring Whittemore K., Vera E., Martínez-Nevado E., Sanpera C.,
reproductive hormones in population research. Popula- and Blasco M. A. (2019). Telomere shortening rate predicts
tion and Development Review 33, 525–542. species life span. Proceedings of the National Academy of
Valko M., Leibfritz D., Moncol J., Cronin M. T. D., Mazur Sciences of the United States of America 116, 15122–15127.
M., and Telser J. (2007). Free radicals and antioxidants in Wilbourn R. V., Moatt J. P., Froy H., Walling C. A.,
normal physiological functions and human disease. Inter- Nussey D. H., and Boonekamp J. J. (2018). The rela-
national Journal of Biochemistry & Cell Biology 39, 44–84. tionship between telomere length and mortality risk in
Vanderklein D., Martínez-Vilalta J., Lee, S., and Mencuccini non-model vertebrate systems: a meta-analysis. Philo-
M. (2007). Plant size, not age, regulates growth and gas sophical Transactions of the Royal Society B: Biological
exchange in grafted Scots pine trees. Tree Physiology 27, Sciences 373, 20160447.
71–79. Wilson R. P., Grundy E., Massy R. et al. (2014). Wild state
Verhulst S. (2020). Improving comparability between qPCR- secrets: ultra-sensitive measurement of micro-movement
based telomere studies. Molecular Ecology Resources 20, can reveal internal processes in animals. Frontiers in
11–13. Ecology and the Environment 12, 582–587.
Viney M., Riley E., and Buchanan K. (2005). Optimal Yeum K. J., Russell R. M., Krinsky N. I., and Aldini G. (2004).
immune responses: immunocompetence revisited. Trends Biomarkers of antioxidant capacity in the hydrophilic
in Ecology & Evolution 20 665–669. and lipophilic compartments of human plasma. Archives
Violle C., Navas M.-L., Vile, D. et al. (2007). Let the concept of Biochemistry and Biophysics 430, 97–103.
of trait be functional! Oikos 116, 882–892. Yoshida Y., Hayakawa M., Habuchi Y., Itoh N., and Niki
Violle C., Enquist B. J., McGill B. J. et al. (2012). The return E. (2007). Evaluation of lipophilic antioxidant efficacy in
of the variance: intraspecific variability in community vivo by the biomarkers hydroxyoctadecadienoic acid and
ecology. Trends in Ecology & Evolution 27, 244–252. isoprostane. Lipids 42, 463–472.
Violle C., Reich P. B., Pacala S. W., Enquist B. J., and Young A. J. (2018). The role of telomeres in the mechanisms
Kattge J. (2014). The emergence and promise of functional and evolution of life-history trade-offs and ageing. Philo-
biogeography. Proceedings of the National Academy of sophical Transactions of the Royal Society B: Biological
Sciences 111, 13690–13696. Sciences 373, 20160452.
CHAPTER 3
Social data collection and analyses

Marie J. E. Charpentier, Marie Pelé, Julien P. Renoult, and Cédric Sueur
3.1 Introduction practically infeasible (Castles et al. 2014). Collecting

social behaviours is also time-consuming and labour-
Understanding how and why animals interact non-
intensive because of the very nature of this type of
randomly with conspecifics is at the core of evolu-
data: in several social species, individuals dedicate
tionary questions related to a broad range of topics
only a small fraction of their time socialising. Spend-
(e.g. sexual selection, parental care, and kin selec-
ing too little time observing animals in their social
tion). Individuals living in complex social systems with
environment can be misleading (Whitehead 2008)
well-differentiated social bonds (i.e. social bonds differ
(Figure 3.1), and a thoughtful description of animal
quantitatively or qualitatively between pairs of indi-
behaviours generally requires years or decades of
viduals) generally benefit (i.e. have an increased fit-
monitoring, especially when working with long-living
ness) from positive social interactions1 (Silk 2007). For
animals that need years to grow. Although long-term
example, in baboons, females that are socially inte-
monitoring introduces new difficulties, such as coping
grated, connected, or central into their group show
with interobserver variations, it is currently consid-
improved offspring survival (Silk et al. 2003, 2009;
ered the gold standard for behavioural ecologists,
Cheney et al. 2016) and they live longer (Archie et al.
ethologists, and demographists.
2014). Conversely, social isolation is related to reduced
In section 3.2, we propose a brief history of animal
life span in humans (House et al. 1988; Holt-Lunstad
social studies and an overview of the classic methods
et al. 2010) and in nonprimate species (Silk 2007;
to sample social interactions and association patterns.
Formica et al. 2012; Stanton and Mann 2012; Kappeler
Studying social behaviours has a long tradition with
et al. 2015; Busson et al. 2019), highlighting further the
well-established methodologies; however, newly dev-
strong and ubiquitous adaptive value of sociality.
eloped tools have the potential to circumvent some of
Studying the functional significance of sociality
the difficulties of describing social interactions and/or
requires to accurately describe social interactions
spatial associations. In section 3.3, we present several
and/or spatial associations, which has proven to be
new technologies that are increasingly used for collect-
highly challenging. Difficulties may arise from animal
ing behavioural data. These technologies can generate
behaviours themselves, for example the elusiveness of
a large amount of data, and thus require specific ana-
lower-ranked individuals that bias sampling towards
lytical tools. In section 3.4, we present approaches to
higher-ranked, more easily observable individuals
analyse social networks using large data sets collect-
(Gimenez et al. 2019). Methods can be problematic
ed when using new technologies. We also discuss how
too, as it is generally impossible to compare data
these approaches can increase our understanding of
collected with different approaches. Using a similar
the structure of social networks, of the link between
methodology to monitor different populations, or the
behaviour and sociality (e.g. the emergence of collec-
same population in the long term, may be, however,
tive decisions) and between sociality and other param-
eters (e.g. disease diffusion). We further propose new
1
‘Positive interactions’ are defined here as all types of social analytical tools, such as the use of artificial intelligence
interactions that involve cooperative or mutualistic behavior,
such as grooming relationships, coalitions, and food sharing,
algorithms to characterise individual behavioural
by contrast with, for example, aggression and submission. patterns.
Marie J. E. Charpentier et al., Social data collection and analyses. In: Demographic Methods Across the Tree of Life.
the descriptive nature of behavioural studies isolat-

0.8 ed the field of ethology from other scientific domains.
Since the 1960s, though, behavioural ecologists have
Social network density
started to systematically record social behaviours using

0.6 detailed and repeatable protocols. Quantitative studies
in a large range of species measuring fine-grain traits
of animal social systems appeared (Carpenter 1942a;
Kruuk 1972; Sade 1972; Barash 1973; Jarman 1974;
0.4
Emlen and Oring 1977; Saitou 1978; Ekman 1979).
This period also saw the development of theoretical
works on the ultimate and proximate mechanisms
0.2
of group living (Carpenter 1942b; Crook 1964, 1970;
10 30 50 70 90 McBride 1964; Crook and Gartlan 1966; Clutton-Brock
Number of interactions 1974; Wilson 1975). The first long-term, individually
centred research on animal societies living in ecolog-
Figure 3.1 Simulated relationship between the number of ically realistic conditions appeared in this stimulat-
interactions recorded and the density of the social network. The ing context (e.g. in insects, Trivers and Hare 1976;
boxplot represents the median density (± 95% confidence intervals) birds, Stacey and Koenig 1990; ungulates, Boyd et al.
obtained from 100 networks of different sizes (90–500 individuals). 1964; Clutton-Brock 1982; cetaceans, Siebenaler and
Caldwell 1956; Saayman and Tayler 1973; nonhuman
primates, Seyfarth 1976; Altmann et al. 1985; and for a
3.2 Classic field-based methods for review see Kappeler and Watts 2012).
sampling social behaviours In 1974, Jeanne Altmann published an article in the
journal Behaviour that would soon become a reference
3.2.1 A brief history of social studies, Jeanne
in behavioural ecology. In her article entitled ‘Obser-
Altmann, and the sampling of social behaviour vational study of behavior: sampling methods’, Alt-
The first studies on animal societies arose in the nine- mann (1974) proposed different methods to sample
teenth century. Emile Durkheim, Herbert Comte, and behaviours in field conditions, most of the techniques
Herbert Spencer first linked natural sciences perfor- described being still largely in use. For example, the
med on animals to social science studies on humans. ad libitum method allows to record every behaviour in
As well, Charles Darwin scrutinised human societies all occurrences. It is particularly useful in conditions of
under the prism of his Theory of Evolution. In 1877, poor visibility or for preliminary studies. The ad libitum
Alfred Espinas (1887) published ‘Des Sociétés ani- sampling method is generally not used alone, however,
males, étude de psychologie comparée’ (‘Animal soci- because it does not allow to control for several bias-
eties, a comparative psychology study’), in which a parallel es, including oversampling conspicuous behaviours or
was drawn between human and animal social sys- individuals (Altmann 1974). Focal-animal sampling,
tems. Later on, Jacob Uhrich (1938) performed prob- in which the observer focuses on one particular focal
ably the first individually centred observations and individual during a predefined duration, helps to cir-
used hierarchical social networks to understand the cumvent these observational biases. One at a time, the
dominance hierarchy in mice. In the mid-century, the observer collects data on each individual in the group,
anthropologist Kinji Imanishi (1952) tried to under- randomly selecting focal animals. Another method
stand human’s aggressive and pacific behaviours by consists of selecting a specific behaviour and record-
studying Japanese macaques from Koshima Island ing it every time it occurs. This behavioural sampling
(Japan) who were provisioned with sweat potatoes. is particularly indicated if the behaviour of interest is
He demonstrated the first case of culture and social rare. In social studies, for example, grooming in pri-
learning in animals. The first case of social facilita- mates or ‘petting’ in cetaceans are typically recorded
tion was reported in birds from Great Britain, with through behavioural sampling. Instantaneous and scan
tits learning from conspecifics how to open milk bot- samplings score states rather than events as in the pre-
tles (Fisher and Hinde 1949; Hinde and Fisher 1951), vious methods. A state is a sequence of different events;
when Jane Goodall documented for the first time tool for example, locomotion is a state defined by differ-
use and tool making in chimpanzees at Gombe Stream ent steps (Altmann 1974). Here, the observer records at
National Park (Tanzania; Goodall 1986). At this time, predefined time intervals different information about
S O C I A L D ATA C O L L E C T I O N A N D A N A LYS E S 55
group members (e.g. activity, spatial distance across tions with limited disturbance. For example, RFID
individuals). Instantaneous and scan samplings can technologies were used to identify great tits (Parus
simultaneously apply to a large number of individuals major) in the wild and to monitor individuals’ per-
and thus are useful, for example, to reconstruct social formance on novel feeding devices in order to better
networks. Altmann’s sampling methods have gone far understand innovation and social diffusion (Aplin
beyond the boundaries of primatology and are still et al. 2015). Recently, artificial intelligence was also
widely used in a large panel of species, making her arti- applied to individual recognition of wild primates
cle the most quoted in the field of behavioural studies (Schofield et al. 2019) and birds (Ferreira et al. 2019)
(>8,400 citations, WoS, December 2019). using photographs, with possible applications in the
long-term study of cognition (e.g. tool use learning),
social networks, and kin selection (Charpentier et al.
2020).
3.2.2 The interplay between field
and lab studies
Although long-term field sites are invaluable for
3.3 New technologies to sample social
behavioural ecology studies, experimental manipula- behaviour in the wild
tions, mostly carried out in captive and controllable Although direct observations are sometimes mandato-
environments, may provide valuable and complemen- ry, such as in species that cannot be captured, animal-
tary inputs on animal social systems. Notably, lab attached remote sensing—necessitating the capture of
studies have allowed researchers to highlight the cog- animals—offers new perspectives for the study of ani-
nitive mechanisms behind decision-making, such as mal behaviour. Indeed, biologging has the potential
delayed gratification (Dufour et al. 2007; Pelé et al. to revolutionise the sampling of social relationships
2010, 2011) or risk calculation, potentially underly- because of the tremendous quantity of data sam-
ing the reciprocity of services observed between non- pled, the quality of which is also improving rapid-
human primates in the wild (Pelé et al. 2014; Broi- ly. The diversity of biologgers available allows the
hanne et al. 2018). Similarly, elaborated cognitive per- sampling of a large array of biological and physi-
formances in keas or ravens have been largely evi- ological data, bringing undiscovered and sometimes
denced in controlled environments (Heaney et al. 2017; unexpected outputs, and consistently nurturing new
Boucherie et al. 2019). Great apes were also shown to research questions (Chapter 5). Here, we propose
anticipate conspecifics’ behaviour according to false an overview of the main devices currently used to
beliefs (Krupenye et al. 2016) using a technology (eye sample social behaviours and spatial associations in
tracker) not yet transposable to the field. animals.
Initiated in primatology, the combination of
field and lab experiments has increased our under-
3.3.1 A brief overview of animal’s biologging
standing of the social lives of many animal species
(Bueno-Guerra and Amici 2018; in canids, Miklosi Starting 60 years ago, biologging technologies develop
2007; elephants, Garstang 2015; birds, Firth and from simple, although often cumbersome devices col-
Sheldon 2015; Emery 2016; fish, Filer 2008; and lecting a few data per hour (e.g. Kooyman 1965) to
cephalopods, Godfrey-Smith 2016). Yet, the method- miniaturised devices capable of continuously record-
ological approaches of lab versus field studies differ ing and archiving data for several years (Kays et al.
markedly because experimenters often seek to estab- 2015). Biologging studies initially flourished in marine
lish special relationships with their study subjects, (Hazen et al. 2012; Hussey et al. 2015) and aerial envi-
whereas field observers try to minimise their impact ronments (Wilson et al. 1997) because of challenging
on them. Observers are, however, rarely undetectable, visual conditions. These studies allowed researchers to
and disturbances are sometimes unavoidable in the decipher, for example, the large-scale migration pat-
field. Indeed, labour-intensive habituation processes terns of birds (McKinnon and Love 2018) and fishes
are often mandatory before starting any field studies, (Block et al. 2005; Bonfil et al. 2005), and the diving
and, if incomplete, animals may react to the pres- depths of marine mammals (Schreer and Testa 1996).
ence of observers. When habituation is not possible, Our understanding about life in the sea and in the
indirect observations recorded from camera traps or air has therefore greatly increased thanks to various
technologies (e.g. accelerometers, radio-frequency electronic devices.
identification (RFID) tags, and GPS) allow researchers In general, radio-tracking involves two devices.
to sample behaviours and data on spatial associa- First, a transmitter is generally attached to an animal’s
(a)
(b) (c) Figure 3.2 Examples of different

biologgers. (a) Female vervet monkey
(Chlorocebus pygerythrus; GPS) in South
Africa (credit: Erica van de Waal, University
of Lausanne, Switzerland); (b) female
elephant seal (Mirouga leonina; GPS) on
Kerguelen Island (credit: IPEV/CNRS/IPHC/
DEPE); (c) Adélie penguins (Pygoscelis
adeliae; temperature, hydrostatic pressure,
and light intensity sensors) in Antarctica
(credit: IPEV/CNRS/IPHC/DEPE).
body, and it sends signals at pre-set intervals. If collect positional information sent with transmitters
mounted on a collar, it can be placed around the neck placed on earth-orbiting satellites using, for example,
(Figure 3.2a), the ankle, or the leg of the study model. GPS technology (for a review see Robertson et al. 2012).
Alternatively, the transmitter may be embedded into GPS systems allow remote detection and record highly
a protective shell and stick on the carapace or direct- accurately an animal’s location; this information may
ly in contact with the skin (e.g. on the head or on the be then stored in the device’s memory associated to the
back; Figure 3.2b,c). Transmitters may also be surgi- receiver until the latter is retrieved. This necessitates
cally implanted into an animal’s body, generally in recapturing the equipped animal. More conveniently,
the abdomen. Second, a radio-receiver collects brief data may also be downloaded remotely via short-range
pulses sent by the transmitter, typically using a cus- radio signals and decoded using a custom receiver or
tomisable frequency in the VHF band (30–300 MHz), using the GSM mobile phone network or the scien-
with a mobile antenna that can be used manually or tific satellite system ARGOS (advanced research and
equipped on a vehicle such as an airplane or a heli- global observation satellite). GPS devices are, howev-
copter. The advantages of this system include reliable er, generally costly (Robertson et al. 2012), although
and versatile ease of operation. It is also generally prices are rapidly dropping. The energy consumption
affordable. Disadvantages are related to a limited sig- of GPS devices is also generally high, and this adds
nal range (less than a few kilometres): animals may to the total weight of the device fitted on the animal
therefore be disturbed by human presence, and the and shortens the device’s life span. Very recently, how-
localisation of the animal, generally obtained through ever, massive progress has been made through the
triangulation, is normally coarse. Alternatively, pas- ICARUS project (https://www.icarus.mpg.de/28874/
sive receivers equipped on the animal’s body may sensor-animals-tracking) to reduce GPS dimensions,
with many new applications in lightweight species receivers located in relevant places in an animal’s com-
such as black birds (Turdus merula). mon environment, such as a sleeping tree or a water
Biologging is no longer restricted to localising hole. When tagged animals come close to the receiver,
animals. Multiple sensors may additionally record the information is recorded through, for example, RFID
environmental and/or biological variables, helping (Figure 3.3a). Coded nanotags, internal or glued pas-
researchers to evaluate the physiological state of their sive integrated transponder (PIT) tags, acoustic trans-
study model (Chapter 2). These sensors may include, mitters, or simple QR codes stuck on small animals,
for example, heartbeat frequency, skin humidity, brea- such as insects (Figure 3.3b), provide individual life-
thing rates, outer and inner body temperatures, EEG, time barcodes. The main advantage here is that these
3D body acceleration, and wing-flap rate, as well tags do not require much power (e.g. a PIT tag has
as light-level information, salinity, or conductivity of an internal microchip that is activated when coming
the environment (Ropert-Coudert and Wilson 2005; close to the receiver, and QR codes may be read with
Chapter 6). These different options help in moni- special cameras). Applications are multiple (Table 3.1).
toring different characteristics of an animal’s envi- For example, Scardamaglia et al. (2017) studied brood
ronment and biology, including its social relation- parasitism in birds. They equipped 74 females of two
ships and behaviour. One of the first evidence of species of cowbirds (parasites) with 1.0g coded radio-
the use of biologging in social studies came from tags (transmitters) for 2.5 months and they positioned
two female penguins (Eudyptes chrysocome moseleyi) 45 fixed data loggers (receivers) just below the nests of
that performed 286 synchronous foraging dives over the two parasitised species. The authors demonstrated
7 consecutive hours (Tremblay and Cherel 1999). In that during prelaying visits, females from both parasite
this study, the authors compared diving patterns species identified nests to parasitise. Female parasites
using electronic time-depth recorders. Using the same appear to form a dynamic memory library of potential
type of recorders supplemented with cameras, oth- laying opportunities that they use later to lay their own
ers showed that Weddell seal (Leptonychotes weddellii) eggs (Scardamaglia et al. 2017).
mothers modified their diving behaviour in the pres- Alternatively, transmission and reception may be
ence of their pups, preferring shallower dives than conveniently combined into the same interactive
when diving alone, probably as a learning strategy talkative tag. This promising technology emerged 20
(Sato et al. 2003). It is also now possible to decipher fine- years ago and is currently bringing new perspectives
grained motions thanks to 3D accelerometers (Wilson in the study of animal’s social networks because it
et al. 2008; Brown et al. 2013), with possible applica- allows researchers to quantify, with a tremendous
tions in the study of social behaviour; for example, a amount of data, fine-scale spatial associations across
mutual clash/headbutt in rutting male mountain goats individuals, in a multitude of animal species (Krause
produces typical abrupt peaks in all three accelera- et al. 2011). These tags record contacts between individ-
tion axes (Shepard 2008). When precisely calibrated, uals when they come close to each other (a few meters),
all these devices may therefore give precious indica- using, for example, wireless technology. The informa-
tions about an animal’s behaviour, including its inter- tion is then stored in the memory of the transmitters
actions and associations with members of its social and downloaded with mobile or fixed receivers. This
environment. device constitutes probably the gold standard in the
study of animals’ spatial associations because they
do not depend on contacts with fixed receivers. The
number of studies using such devices has marked-
3.3.2 Biologging to determine individual social
ly increased for the past 10 years, but the studied
interactions and spatial associations species are still largely dominated by medium-sized
By providing high-resolution positions in space and mammals (Table 3.1). Studies may be classified into
time, often in fractions of seconds, lightweight GPS three main categories. A significant number are ded-
devices may capture fine-grain association patterns icated to methodological developments, with several
inferred from positional data (Davis et al. 2013). For studies involving comparisons of the efficacy of dif-
example, the average spatial position of a pigeon in ferent devices in recording spatial associations or con-
its flock depends on its position in the social hier- tacts (Lavelle et al. 2014; Tosa et al. 2015); others have
archy (Nagy et al. 2010). Other tags resulting in tested different possible biases such as nonreciprocal
individual-based locations in time may provide similar agreements across tags or inter-logger variation in
information by transmitting a unique code to fixed signal strength (Boyland et al. 2013; Meise et al. 2013);
(a)
(b)
Figure 3.3 Biologging on ants.

(a) Odontomachus hastatus equipped
with an RFID tag (credit: Cédric Sueur,
IPHC, Strasbourg); (b) Lasius niger
performing trophallaxis (mouth-to-mouth
food exchange) and equipped with an
Aruco tag (https://www.uco.es/investiga/
grupos/ava/node/26). The software
Usetracker is used to identify individuals
(http://www.usetracker.org/). (credit: Guy
Theraulaz, CNRS, CRCA, CBI, Toulouse.)
and a few others have proposed analytical develop- grey reef sharks from the Pacific Ocean, group leaders
ments (Cross et al. 2012; Duncan et al. 2012). Finally, associate the most with conspecifics.
ecological studies on natural species have increased When compared with other techniques, proximity
in number and in the diversity of questions asked, loggers generally outperformed tags using individual-
going beyond the simple description of spatial associ- based locations in time (Walrath et al. 2011; Lavelle
ations. For example, by taking advantage of wildfires et al. 2014; Tosa et al. 2015). Using proximity loggers
that occurred in 2009 in Victoria, Australia, Banks et al. has also proved better than using observational data
(2011) studied the proximate effects of these environ- (Meise et al. 2013) to estimate spatial associations
mental disturbances on mountain brushtail possums across animals (see also Ryder et al. 2012). Yet, prox-
equipped with proximity data loggers. They showed imity data logging technology, in addition to being
that possums adapted to a decline in the number of quite expensive, has some important pitfalls chal-
suitable shelters by being more flexible in their selec- lenging both the analysis and the interpretation of
tion and not by increasing shelter sharing (Banks et al. results. First, proximity loggers, like GPS, do not pro-
2011). Others demonstrated that social thermoregula- vide any indication about the nature of the record-
tion was driven by kin relationships in wild raccoons ed contacts (e.g. aggression, affiliation, or passive
(Robert et al. 2013). Using underwater acoustic loggers, spatial associations). Deciphering an animal’s social
Jacoby et al. (2016) recently showed also that, among structure using proximity logging technology alone
Table 3.1 A web survey of research that used proximity biologging technologies to study spatial (and social) associations across conspecifics. ‘Proximity data-logger’ or ‘proximity logger’ were the two
keywords used on Web of Sciences; the survey resulted in 50 articles matching our criteria. NB: depending on the study, the sample size either indicates the total number of individuals equipped with
loggers (including those that failed) or the total number of loggers that successfully recovered data (‘ids’ stands for individuals).
Species Sample size Device (company) Aim/main finding Reference
Fish Grey reef shark 44 ids; 3 years Acoustic transmitters Leadership is predicted by the duration of Jacoby et al. 2016
Carcharhinus amblyrhynchos associated to external co-occurrences between conspecifics
receivers (VEMCO)
Lemon sharks 15 ids; 17 days Internal acoustic transmitters Methodological (loggers’ testing) Guttridge et al. 2010
Negaprion brevirostris and receivers
(Sonotronics Inc)
Brown trout 9 ids; 1 week Proximity talkative tags Males spending the most time with Tentelier et al. 2016
Salmo trutta (Encounternet) females mate more; temporal
distribution of encounters reflects
shifts in dominance status
Birds Wire-tailed manakin 18 ids; 1 month Coded nanotags with an Methodological (SNA) Ryder et al. 2012
Pipra filicauda external digitally encoded
data-logger (Lotek Wireless)
Cowbird 74 ids; 79 days Coded tags with an external Females form a dynamic memory library Scardamaglia et al. 2017
Molothrus bonariensis and digitally encoded of laying opportunities
Molothrus rufoaxillaris data-logger (Lotek Wireless
and Biotrack)
Barn swallow 21 ids; 13 days Proximity talkative tags Methodological (loggers’ testing) Levin et al. 2015
Hirundo rustica erythrogaster (Encounternet)
New Caledonian crows 33 ids; 19 days Proximity talkative tags Methodological (loggers’ testing) Bettaney et al. 2015;
Corvus moneduloides (Encounternet) Network structure responds quickly to St Clair et al. 2015
environmental change
New Caledonian crows 34 ids; 7 days Proximity talkative tags Close-range association between Rutz et al. 2012
Corvus moneduloides (Encounternet) nonfamily birds
Reptiles Agassiz’s desert tortoise ~100 ids; 6–7 Proximity talkative tags Sex and season-biased contact Aiello et al. 2018
Gopherus agassizii months (Sirtrack; Encounternet) interaction
Mammals Sheep 94 ids; 23 days Proximity talkative tags Age difference, temperature and rain, Doyle et al. 2016
Ovis aries (Sirtrack) and acoustic similarity impact daily
contact time
continued
Table 3.1 Continued
Holstein-Friesian calf 40 ids; 1 month Proximity talkative tags Familiarity during rearing allows social Bolt et al. 2017
Bos taurus (Sirtrack) association
Cattle 20 ids; 3 weeks Proximity talkative tags Methodological (inter-logger variation) Boyland et al. 2013
Bos taurus (Sirtrack)
Belmont and Brahman Red 33 ids; 29 days Proximity talkative tags Increased contact with the bull when Corbet et al. 2018
cattle (Sirtrack) cows are in oestrus
Bos taurus
Belmont Red cattle 48 ids; 48 days Proximity talkative tags Number and duration of bull-cow O’Neill et al. 2014
Bos taurus (Sirtrack) affiliations greater in oestrous cows
compared to anoestrus cows
Cattle 58 ids; 11 weeks Proximity talkative tags More frequent association among cows Swain et al. 2015
Bos taurus (Sirtrack) of similar maternal status (pregnant or
calved) and among cows with calves
of similar ages
Angus/Limousin and Luing 30 ids; 2 months Proximity talkative tags Methodological (disease modelisation) Duncan et al. 2012
cattle (Sirtrack)
Bos taurus
Cattle 198 ids; 16 months Coded tags with an external Indirect contacts between wildlife and Lavelle et al. 2016
Bos taurus stationary data-logger cattle; highest contact rates between
White-tailed deer (Sirtrack) raccoons and cattle during summer
Odocoileus virginianus and fall
Raccoon
Procyon lotor
Virginia opossum
Didelphis virginiana
Cattle 25 ids; 4 months Proximity talkative tags Specific individuals with high contact Böhm et al. 2009
Bos taurus (Sirtrack) rates in both livestock and wildlife
Badger populations
Meles meles
Cattle 109 ids; 17 months Proximity talkative tags Methodological (loggers’ testing) Drewe et al. 2012
Bos taurus (Sirtrack)
Badger
Meles meles
Cattle 94 ids; 12 months Proximity talkative tags Heterogeneity in contact patterns Drewe et al. 2013
Bos taurus (Sirtrack) between cattle and wildlife; more
Badger indirect than direct contacts
Meles meles
Brushtail possums 4 ids; 2 days Proximity talkative tags Methodological (loggers’ testing) Douglas et al. 2006
Trichosurus vulpecular (MateID, Electronics
Laboratory)
Brushtail possum 40 ids; 6 months Proximity talkative tags Similar female–male and male–male Rouco et al. 2018
Trichosurus vulpecula (Sirtrack) connection rates
Brushtail possum 22 ids; 3 years Proximity talkative tags Polygamous mating system, with Ji et al. 2005
Trichosurus vulpecula (Sirtrack) polygyny and polyandry
Mountain brushtail possums 26 ids; 3–10 Proximity talkative tags Lower rate of den sharing in burnt areas Banks et al. 2011
Trichosurus cunninghami months (Sirtrack) compared to nonburnt areas
Raccoon 42 ids; 12 months Proximity talkative tags Male–male pairs exhibited higher Prange et al. 2011
Procyon lotor (Sirtrack) contact values than male–female or
female–female pairs, especially during
winter
Raccoon 35 ids; 12 months Proximity talkative tags Methodological (loggers’ testing) Prange et al. 2006
Procyon lotor (Sirtrack)
Raccoon 15 ids; 4–6 months Proximity talkative tags Correlation between home-range overlap Robert et al. 2012
Procyon lotor (Sirtrack) and intraspecific contact rate (weaker
in winter)
Raccoon 15 ids; 17 months Proximity talkative tags Higher proximity rate in winter, with cold Robert et al. 2013
Procyon lotor (Sirtrack) ambient temperatures (social
thermoregulation) and among related
individuals
Raccoon 25 ids; 2 years Proximity talkative tags Several latrine sites visited by multiple Hirsch et al. 2014
Procyon lotor (Sirtrack) raccoons on short time-periods
Raccoon 42 ids; 13 months Proximity talkative tags Elevated likelihood of a rabies’ outbreak Reynolds et al. 2015
Procyon lotor (Sirtrack) when entering a suburban raccoon
population (network modelling
approach)
River otter 21 ids; 1–3 months Proximity talkative tags Timing of group fissions and fusions Barocas et al. 2016
Lontra canadensis (Encounternet) coincides with latrine visits; spatial
overlap is a good predictor of social
interactions
continued
Table 3.1 Continued
European badger 51 ids; 1 year Proximity talkative tags Seasonality in daily patterns of contact Silk et al. 2017
Meles meles (Sirtrack) frequency and duration
Island fox 48 ids; 4–12 Proximity talkative tags Mated females are in contact with other Ralls et al. 2013
Urocyon littoralis months (Sirtrack) males during oestrus; mated males
are in contact with other females than
their mate
European wild rabbit 126 ids; 8 months Proximity talkative tags Strong and highly stable intra-group Marsh et al. 2011a
Oryctolagus cuniculus (Sirtrack) associations, rare and transient
inter-group associations
European wild rabbit 126 ids; 6 months Proximity talkative tags Spatial and temporal heterogeneities in Marsh et al. 2011b
Oryctolagus cuniculus (Sirtrack) contact between populations and
between and within social groups in
the same population
Tasmanian devil 46 ids; 5 months Proximity talkative tags All individuals connected in a single large Hamede et al. 2009
Sarcophilus harrisii (Sirtrack) component
Elk 24 ids; 2-3 months Proximity talkative tags Higher interaction rate but of lower Vander Wal et al. 2012a
Cervus elaphus (Sirtrack) length in males at high densities;
longer interactions in females at
medium densities
Elk 106 ids; 1 year Proximity talkative tags Higher interaction rate but of shorter Vander Wal et al. 2013
Cervus canadensis (Sirtrack) duration among females than among
males
Elk 104 ids; 1 year Proximity talkative tags Close-contact interaction rate and Vander Wal et al. 2012b
Cervus canadensis (Sirtrack) duration are independent from genetic
relatedness
Elk about 180 ids Proximity talkative tags (?) Methodological (modelling approach) Cross et al. 2012
Cervus canadensis (unclear); 6 months
Elk 149 ids; 3 years Proximity talkative tags Pairwise contact rates decrease when Cross et al. 2013
Cervus canadensis (Sirtrack) group size increases
Elk 30 ids; 2 weeks Proximity talkative tags Low-density feeding led to reductions in Creech et al. 2012
Cervus canadensis (Sirtrack) total number of contacts and number
of individuals contacting a source of
brucellosis transmission
White-tailed deer 22 ids; 5 months Proximity talkative tags Methodological (comparison of Tosa et al. 2015
Odocoileus virginianus (Sirtrack) technologies)
White-tailed deer 20 ids; 8 months Proximity talkative tags Methodological (comparison of methods) Walrath et al. 2011
Odocoileus virginianus (Sirtrack)
White-tailed deer 26 ids; 2 weeks Proximity talkative tags Methodological (comparison of Lavelle et al. 2014
Odocoileus virginianus (Sirtrack) technologies)
Koala 39 ids; 13 months Proximity talkative tags More frequent male–female, but not Ellis et al. 2015
Phascolarctos cinereus (Sirtrack) male–male, encounters during the
breeding season
Mandrill 13 ids; 1 year Proximity talkative tags (ELA No correlation between spatial Poirotte et al. 2017
Mandrillus sphinx Innovation) associations and animal’s parasitic
status
Galápagos sea lion 23 ids; 2 months Proximity talkative tags Methodological (comparison of Meise et al. 2013
Zalophus wollebaeki (Encounternet) methods); extended association
between nonterritorial males
would probably not yield biologically relevant out- huge quantity of data generated by these new technolo-
puts, and researchers should always complement these gies may be complex, and new methodologies (Box 3.1)
logging data with basic information about an ani- and software have been developed for this purpose.
mal’s life and ecology. Second, proximity logging These methods are unified under the name social net-
devices may produce spurious data. For example, work analysis (SNA). SNA originated from mathemat-
signal detection ranges may vary with the antenna ical graph theory and was soon applied to the social
angle; unique contact may be recorded as multiple sciences (for a review, see Newman 2010). While ear-
events; and nonexistent ‘phantom’ codes are some- ly on SNA was employed to study social hierarchy in
times recorded (Prange et al. 2006; Böhm et al. 2009; mice (Uhrich 1938), its development arose with studies
Drewe et al. 2012; Meise et al. 2013). Signal intensity on social relationships in macaques (Chepko and Sade
that indicates the distance between equipped individ- 1979; Sade et al. 1988; Chepko-Sade et al. 1989; Sade
uals is further subject to environmentally related vari- 1989).
ations due to weather conditions, for example (Böhm SNA is used to grasp several aspects of sociality,
et al. 2009). Interactions that occur at the edge of the such as the role of individuals in their group (‘node’
receiving range and the nature of animal behaviours level), the clustering of group members (‘intermedi-
may further affect the probability to record a con- ary’ levels) as a function of sociodemographic factors,
tact. In general, researchers tend to find higher agree- or how these individuals also structure the network
ment in the duration of contacts rather than in their (‘group’ level). SNA has already proved to be an effec-
frequency (Drewe et al. 2009; Meise et al. 2013). Pre- tive tool for modelling the effects of ecological factors
liminary testing, under various conditions, is there- on social relationships and association patterns such
fore mandatory to calibrate each tag individually and as food distribution (Foster et al. 2012), disease risk
to test tags’ performance and reciprocity when com- (Romano et al. 2016), and predation (Heathcote et al.
municating with each other. Data obtained should 2017). SNA also provides new approaches such as
also be filtered using automated programs to detect multilayer, temporal, or modelling analyses to disen-
spurious records. Third, and by contrast with GPS- tangle the effect of these different factors on animals’
based data, proximity loggers only provide informa- social relationships. Here, we provide a short summa-
tion when two tags are within a certain distance and ry of the main metrics currently used in SNA with
do not ground data points in true space, possibly limit- possible applications in the study of animal social
ing interpretations. Fourth, the equipment in itself may systems.
be labour-intensive and expensive, although not pre-
venting from high probability of technological failure
(Matthews et al. 2013). Capturing and fitting individ-
uals is therefore generally limited to a fraction of the
3.4.1 From data collection to social
population, which inevitably generates sampling bias- network analyses
es, especially when studying social relationships. Final- A social network is made of nodes representing, for
ly, equipping wild, often threatened animal species example, individuals, groups, populations, or species,
poses ethical issues that need to be carefully addressed and edges representing, for example, spatial associa-
and considered well in advance (Ropert-Coudert and tions or any types of relationships such as affiliative
Wilson 2005; Wilson and McMahon 2006; Forin-Wiart or agonistic interactions. Nodes are characterised by
et al. 2019): loggers may affect animal behaviour or, different sociodemographic factors, such as sex, age,
worse, reproduction (Paton et al. 1991; Ackerman et al. personality, hierarchical rank, or body weight. Edges
2004) or survival (Paton et al. 1991; Saraux et al. may take different forms depending on the research
2011). question and the sociality of the studied species. For
example, grooming is interesting for studying the
topology of networks in chimpanzees, macaques, or
3.4 Analysing social behaviour parrots in which this behaviour is frequent. However,
Biologging data may be analysed at the individual lev- relying on grooming in species where this behaviour
el and at the group or population levels. For example, is rare, like in gorillas, cows, or lizards, would pro-
biologging allowed for measuring individual diving duce a not well-resolved network. In such condi-
patterns in penguins that showed synchronisation tions, it may be preferable to analyse spatial dis-
between individuals and group foraging (Daniel et al. tances, which often indicate social preferences, through
2007; McInnes et al. 2017), or for studying group instantaneous sampling or using loggers (Haddadi
hunting in bats (Dechmann et al. 2009). Analysing the et al. 2011).
Box 3.1 Deep learning in behavioural ecology
With increasingly inexpensive and miniaturised sensing tech- are the most relevant for prediction. A CNN relies on super-
nologies, collecting data on individual trajectories, pose, vised learning and needs to be trained. During this training
behaviour, physiology, and other phenotypic traits in ani- phase, the algorithm is told when it fails and when it suc-
mals is more and more feasible (Krause et al. 2013). The ceeds. The training data set contains labelled data consisting
main challenge is now to make sense of all this data to of, for example, images or sound recordings of individu-
extract meaningful information. In classical data analysis, als with known identity, reproductive success, or longevity.
researchers select explanatory variables, manually extract The algorithm then predicts the response variable for a test-
these variables from the raw data and correlate them with ing data set, for which the algorithm was blind to during
one or several response variables. For example, to study how training.
phenotypic variation in a bird population is associated with CNNs all present the same overall design, being com-
reproductive success, one would typically first hypothesise posed of two main parts: the encoder that extracts features
which trait is relevant (e.g. beak stoutness), then convert this and the classifier that makes predictions (Boxed Figure 3.1).
trait into an explanatory variable (e.g. a product between The encoder extracts features by successively repeating three
beak length and depth) and eventually measure it (e.g. in elementary operations—each repetition is called a layer—
pixels on photographs). Although this approach has provid- with the input of a given layer being the output of the
ed major insights on animal behaviour, it presents at least preceding one. Convolution is the central operation that
five major limitations: (1) it does not guarantee that the extracts features. It is followed by activation that adds
selected trait is relevant, which limits the chance to detect nonlinearity to features. Activation functions allow CNNs
an association; (2) the conversion from a trait to a vari- to solve highly nonlinear problems such as differentiat-
able may not be optimal; (3) reliable measurements require ing between two species at all ages, even if juveniles of
highly standardised procedures for data collection (e.g. head either species resemble more to each other than conspecific
position perfectly consistent); (4) it is prone to human error; adults do. The third operation, pooling, merges informa-
and (5) it is time-consuming and thus intractable for very tion that is spatially or temporally close and redundant.
large data sets. Deep learning algorithms have become This operation makes features invariant to uninformative
explosively popular in computer science over the last 5 variations. For example, it allows to accurately estimate
years, precisely because they can simultaneously overcome beak stoutness from a picture of a head even if not per-
these limitations (LeCun et al. 2015). To understand how fectly framed. While embedding raw data into a feature
this is possible, we briefly present the convolutional neu- space, the encoder reduces the dimensionality of the data
ral network (CNN), the most commonly used deep learning set, from hundred thousands of variables (typical for an
algorithm. image) to a few thousand features. The low-dimensional,
A CNN takes raw data as inputs (e.g. images of birds), highly informative output of the encoder eventually feeds
and directly predicts a response variable (e.g. individual into the classifier (which can also be a regression mod-
identity) as output, thus automatically selecting those el for predicting continuous variables; Lathuiliére et al.
features (i.e. explanatory variables in computer science) that 2019).
Encoder Classifier
Features Output
Input
(explanatory (response
(raw data)
variables) variable)
Layer 1 Layer 2 Layer n
Boxed Figure 3.1 General design of a deep CNN. Each layer performs three elementary operations: convolution (blue), nonlinear
activation (red), and pooling (green).
Among artificial intelligence methods, a particularity of large data set is retrained (both the encoder and the clas-
CNNs is that the encoder and the classifier work hand-to- sifier) on a smaller data set (Weiss et al. 2016). However, it
hand. During training, at each iteration (i.e. when a subsam- is important to bear in mind that using artificial intelligence is
ple of the raw training data is evaluated), the performance a test and trial process and that multiple approaches should
of the classifier is estimated and its error back-propagated be tested to obtain the best performances. For example, in
into the encoder such that the encoder becomes increasingly another study based on 1,550 pictures of pigs, represent-
sensitive to features relevant to the classifier over each iter- ing 10 different individuals, training a CNN from scratch was
ation. Nevertheless, the encoder can be used in isolation if more successful (97% in individual recognition) than transfer
the goal is to describe the data informatively and not to pre- learning (Hansen et al. 2018).
dict a response variable. For example, the similarity between In addition to identifying individuals based on images
two phenotypes can be estimated from the Euclidean dis- (Deb et al. 2018; Hansen et al. 2018) or motion patterns
tance in the low-dimensional feature space (Schroff et al. (Neverova et al. 2016), and to estimating similarity between
2015). Interestingly, one psychophysical experiment showed phenotypes (Schroff et al. 2015), CNNs can predict the pose
that such distances accurately predict the visual similarity (Mathis et al. 2018) and behaviour (Nguyen et al. 2019) of
perceived by humans (Zhang et al. 2018). In that study, animals. Interestingly, transfer learning allows to train algo-
the only human-specific data was the input: RGB images rithms with very small data sets: 200 labelled images can
tuned to human colour vision. It is therefore possible that be enough to recognise pose (i.e. the spatial configuration
with appropriate input data, CNNs could also model the of body parts) in various animal species when transfer-
visual perceptual space of other vertebrates (Renoult et al. ring a network pretrained with human data (Mathis et al.
2019). 2018). Behavioural ecologists will therefore certainly soon
The encoder can be used in isolation also when per- benefit from the extensive amount of research on human
forming a classification task with a limited data set. CNNs behaviours.
typically need very large data sets for training, which are For behavioural ecologists willing to engage with CNNs,
often unavailable in behavioural studies. For small data sets, we recommend using either the Python or R programming
it is usually recommended to extract features using the languages. Over the last decade, Python has become a
encoder of a trained CNN obtained from a public reposito- principal language of scientific programming. Most CNN
ry, and to feed an independent classifier with these features tutorials and code available on public repositories are in
(e.g. a discriminant analysis or a support vector machine Python. Tensorflow (developed by Google) and PyTorch
SVM). For example, in a study with a training data set con- (developed by Facebook) are two of the most popular
taining only 1,550 images of 276 different wild elephants Python-based frameworks, allowing the design of new
(8 images per individual on average), a SVM fed with fea- models or the transfer of existing ones for all kinds of
tures extracted from the ResNet50 algorithm pretrained on input data and applications. To those with a limited back-
the ImageNet data set reached 74% accuracy in individual ground in programming, Keras is an easy-to-use frame-
recognition (Körschens et al. 2018). This is remarkable given work running on top of Tensorflow. It abstracts CNN ele-
that ImageNet contains only six images of elephants out of mentary operations, which are then arranged like Lego
millions. Alternatively, with small data sets one can apply bricks. Recently, Tensorflow and Keras have been made
transfer learning, in which a CNN pretrained with a very accessible in R.
To ensure repeatability and consistency of results Socprog, Whitehead 2009; and Asnipe, Farine 2013),
obtained from SNA, the following questions should others have proved useful for analysing animal soci-
therefore be asked when collecting data in the field: eties even though they were firstly developed for other
(1) Are all group members observed equally and for applications (e.g. Ucinet, Borgatti et al. 2002; igraph,
a comprehensive period of time? (2) Are the results Csardi and Nepusz 2006).
obtained by studying one specific group representative
of the species sociality? (3) How might uncertainty
3.4.2 Individual level: the roles of nodes
be reduced? Different software or R packages are
dedicated to SNA, some specifically designed to take Figure 3.4a, which clearly identifies two clusters,
into account observational biases (for a discussion on describes a typical social network generated from
the different packages see Sosa et al. 2018a). While Table 3.2, which presents the main metrics currently
some software were specifically developed for in use (for comprehensive reviews of SNA metrics see
behavioural ecologists (e.g. ANTs, Sosa et al. 2018a; Whitehead 2008; Sueur et al. 2011a; Krause et al. 2015;
Sosa et al. 2020). Centrality refers to the position of an across categories (E, for ‘external’, such as across sexes)
individual in its group and may have different def- with the ones observed within a given category (I, for
initions (in Table 3.2, degree, strength, betweenness, ‘internal’, such as among individuals of the same sex).
eigenvector coefficient, and clustering coefficient are This metric therefore reports preferential interaction
the main centralities used in SNA). The degree is the between individuals with similar (homophily) or dif-
number of individuals connected to (or having a rela- ferent (heterophily) attributes. These attributes can
tionship with) a node. It can be interpreted as the even take the form of node metrics such as the degree
popularity of an individual. The strength is the sum (i.e. the number of links to a node) to measure assor-
of interactions involving the node and it represents tativity. Assortativity is a special case of homophily
the social activity of an individual with group mem- based on centralities: nodes that share similar centrali-
bers. The betweenness expresses the number of times a ties are more associated. For example, individuals with
node links to other nodes. Individuals g and j have, for high degrees associate more with each other. MRQAP
example, higher betweenness because they are linked (multiple regression quadratic assignment procedure)
to all other nodes (Figure 3.4a). Diseases or social infor- and matrices of correlations are particularly useful
mation should go through these two nodes (i.e. g and j) when studying homophily because a matrix of social
to spread rapidly into the group. By contrast, if g and j interactions or of spatial associations may be compared
disappear, the group may split: group fission happens, to a matrix of genetic relatedness or age differences, for
for example, when an old individual linking two clus- example. Dominant individuals are generally highly
ters (e.g. matrilines) dies (Lefebvre et al. 2003; Sueur assorted because they have differentiated social rela-
and Maire 2014). Betweenness therefore matters when tionships and share high degrees (Sueur et al. 2011b;
studying the importance of a node regarding group Sosa et al. 2018b).
cohesion or social/disease diffusion. The eigenvector
provides information about the connection between a
node and its neighbours. Elevated eigenvectors indi-
cate the high popularity of both an individual and its 3.4.4 Group level and group comparison
partners. The clustering coefficient represents an individ- Density is one of the primary index used to describe
ual measure of cohesiveness: it estimates how partners social networks at the group level. It represents the
of a node are connected. For example, a node with number of observed social relationships or spatial asso-
unconnected partners shows a clustering coefficient ciations divided by the total number of all possibilities.
equal to 0, whatever the number of partners involved. Density is thus a measure of group cohesion and
Although other metrics are available to study social further characterises social style (Sueur et al. 2011a).
centrality, these five measurements (degree, strength, Modularity is another index that represents the prob-
betweenness, eigenvector, and clustering coefficient) ability for a group to split into several clusters
can all be explained by relatively simple social dif- (Figure 3.4b). With this metric, the total number of
ferences between individuals (e.g. Firth et al. 2017a) clusters in a group may be compared to other vari-
and are therefore widely used. They may all be relat- ables such as the distribution of ages and relatedness
ed to sociodemographic characteristics of the nodes. within the group (but see Whitehead 2009 for limi-
For example, high-ranking individuals generally have tations). For example, groups that are highly struc-
a high centrality in primates (Sueur et al. 2011b), and tured by kin relationships show elevated modularity
old and experienced animals present a strong degree indices, as observed in elephants and in several mul-
and betweenness in elephants (McComb et al. 2011). tilevel societies (e.g. hamadryas baboons, Wittemyer
et al. 2005; Matsuda et al. 2012, 2015). Network cen-
tralisation measures how much a network is centred
3.4.3 Intermediary levels
on some key individuals, for example because of their
Homophily (or heterophily) measures the probability experience or their position in the social hierarchy
that individuals sharing similar characteristics, such (Figure 3.4c). This index is particularly useful when
as rank, age, or nutrient requirements, have stronger describing social structures and for studying diffu-
(or weaker) social relationships or spatial associations sion processes, which typically rely heavily on these
(Ruckstuhl and Neuhaus 2000; Croft et al. 2005). When key individuals (Lusseau and Newman 2004; Carne
studying continuous traits, homophily is typically cal- et al. 2013). Finally, network efficiency measures the
culated using the E/I ratio of Krackhardt (Krackhardt probability of diffusion of an entity (a knowledge or
and Stern 1988). This ratio represents a proportion a pathogen) to the entire group and depends on the
between the relationships or associations observed network topology. Overall, group-level indices thus
(a)
a
c
d
e
f
g
h
i j
k l
m o
n
p
q
r
(b)
1. 2. 3. 4.
(c)
1. Equal network 2. Star network 3. Centralised network
Table 3.2 Individual network measurements of nodes illustrated in Figure 3.4a.
Node Degree Strength Betweenness Eigenvector Clustering
a 1 1 0 0 –
b 3 3 17.5 0.02 0
c 3 3 19.667 0.04 0.33
d 2 3 10.333 0.04 0.33
e 3 4 12.833 0.06 0.5
f 5 5 41.667 0.07 0.4
g 4 4 81 0.11 0.33
h 3 3 17 0.05 0.33
i 1 1 0 0.01 –
j 4 4 81.833 0.27 0.33
k 4 3 16.5 0.29 0.67
l 2 1 0 0.1 –
m 3 2 8.333 0.19 1
n 7 7 46.333 0.51 0.33
o 5 5 6.667 0.39 0.4
p 2 3 0 0.32 1
q 4 5 8 0.4 0.5
r 4 3 9.333 0.26 0.67
s 2 2 0 0.16 1
are highly useful for comparing group or popula- network changes over time according to, for example,
tion structures and properties. However, these indices seasons, changes in group composition, or across gen-
highly depend on factors such as group size or time of erations (Henzi et al. 2009; Pinter-Wollman et al. 2013;
observations. Controlling for these factors is therefore Fisher et al. 2016; Wolf et al. 2018). Temporality is
mandatory in group-level SNA. connected to concepts such as stability, resilience and
network robustness (Naug 2009; Fushing et al. 2013;
Goldenberg et al. 2016; Puga-Gonzalez et al. 2019).
3.4.5 Network dynamics While the study of network efficiency provides a proxy
While SNAs were initially applied to characterise indi- for how an entity (i.e. a disease or an information)
vidual social relationships, they are increasingly used is transmitted, modelling dynamic diffusion process-
with a dynamic perspective to study, for example, the es allows a better understanding of which and when a
stability and resilience of networks over time and gen- group member is infected or informed (Hoppitt et al.
erations or diffusion processes (Allen et al. 2013; Cantor 2010; Rushmore et al. 2014; Pasquaretta et al. 2016;
and Whitehead 2013; Fisher and McAdam 2017; Sueur Romano et al. 2016). Modelling is often necessary to
et al. 2017). Network dynamics can be captured in dif- study social network dynamics (Skyrms and Peman-
ferent ways. Temporality indicates, for example, how a tle 2009; Sueur and Maire 2014; Romano et al. 2018)
Figure 3.4 Graphical representations of some theoretical network properties. (a) A node represents a group member. Nodes with the same
colour code belong to the same subgroup. An edge (line) represents a relationship between two individuals. This social network corresponds to the
social structure of a group of 19 individuals (labelled “a” to “s”). Individual network measures are presented in Table 3.2. (b) Four different types
of networks that illustrate modularity: (1) equal network, all individuals are all equally connected (maximum modularity Q = 0); (2) two-subset
network with 14 relationships between the two subsets (maximum modularity Q = 0.36); (3) two-subset network with only two relationships
between subsets (maximum modularity Q = 0.48); and (4) four-subset network with five relationships between the four subsets (maximum
modularity Q = 0.59). The size of the nodes represents the degree (number of relationships) and the colour represents the subset (one colour per
subset). Thicker lines between nodes represent stronger relationships between individuals. (c) Three networks that illustrate the concept of
centralisation: (1) equal network: all individuals are all equally connected; (2) star network: all group members are only connected to one
individual; (3) centralised network: individuals only have relationships with a few (e.g. dominant) group members and the strength of relationships
(line thickness) decreases with e.g. dominance rank. The larger the size of a node (individual) and the darker the colour, the higher the individual
dominance rank.
because empirical data may be difficult to get or be Aiello C. M., Esque T. C., Nussear K. E., Emblidge P. G., and
ethically questionable (e.g. experimental removals of Hudson P. J. (2018). Associating sex-biased and seasonal
individuals or experimental infections, but see Flack behaviour with contact patterns and transmission risk in
et al. 2005; Flack et al. 2006; Aplin et al. 2015; Firth and Gopherus agassizii. Behaviour 155, 585–619.
Allen J., Weinrich M., Hoppitt W., and Rendell L. (2013).
Sheldon 2015; Firth et al. 2017b).
Network-based diffusion analysis reveals cultural trans-
Finally, studying the concurrent changes and inter-
mission of lobtail feeding in humpback whales. Science
dependencies between diffusion processes and net-
340, 485–488.
work topology is a highly topical area of research Altmann J. (1974). Observational study of behavior-
(Farine et al. 2015; Lopes et al. 2016; Sueur et al. 2017; sampling methods. Behaviour 49, 227–267.
Fisher and Pruitt 2019) which includes both concep- Altmann J., Hausfater G., and Altmann S. A. (1985). Demog-
tual and modelling studies (Banks and Carley 1996; raphy of Amboseli baboons, 1963–1983. Am. J. Primatol.
Puga-Gonzalez et al. 2018). Although empirical tests 8, 113–125.
of these concurrent changes are rare, a recent study Aplin L., Farine D., Morand-Ferron J. et al. (2015). Exper-
on ants demonstrated that workers change their self- imentally induced innovations lead to persistent culture
organised behaviour following an infection to decrease via conformity in wild birds. Nature 518, 538–541.
Archie E. A., Tung J., Clark M., Altmann J., and Alberts S. C.
network efficiency, with an influence on pathogen
(2014). Social affiliation matters: both same-sex and
spread (Stroeymeyt et al. 2018). Hence, SNA can inves-
opposite-sex relationships predict survival in wild female
tigate how group properties influence individual prop- baboons. Proc. R. Soc. B Biol. Sci. 281, 20141261.
erties, for example, when the number of clusters deter- Banks D. L. and Carley K. M. (1996). Models for network
mines the diffusion of disease and/or information and evolution. J. Math. Sociol. 21, 173–196.
thus the health and fitness of individuals. Such effects Banks S. C., Knight E. J., McBurney L., Blair D., and
can influence in return social relationships and thus the Lindenmayer D. B. (2011). The effects of wildfire on mor-
network topology. tality and resources for an arboreal marsupial: resilience to
fire events but susceptibility to fire regime change. PLOS
One 6, e22952.
3.5 Conclusions Barash D. P. (1973). Social variety in yellow bellied marmot
New technologies and analytical methods are revo- (Marmota flaviventris). Anim. Behav. 21, 579–584.
lutionising the sampling of animal behaviours, social Barocas A., Golden H. N., Harrington M. W., McDonald D.
interactions, and association patterns and are thereby B., and Ben-David M. (2016). Coastal latrine sites as social
information hubs and drivers of river otter fission–fusion
opening novel research avenues (Chapter 21). The high
dynamics. Anim. Behav. 120, 103–114.
spatiotemporal resolution of data needed when study-
Bettaney E. M., James R., St Clair J. J. H., and Rutz C. (2015).
ing social processes such as disease or information Processing and visualising association data from animal-
diffusion (often involving rare and brief interindivi- borne proximity loggers. Anim. Biotelemetry 3, 27.
dual contacts) will probably be better attained using Block B. A., Teo S. L. H., Walli A. et al. (2005). Electronic
proximity logging technology rather than classic meth- tagging and population structure of Atlantic bluefin tuna.
ods of observations, particularly when attempting to Nature 434, 1121–1127.
simultaneously track large numbers of individuals. Böhm M., Hutchings M. R., and White P. C. L. (2009). Con-
Conveniently, new analytical tools have been devel- tact networks in a wildlife–livestock host community:
oped to manage, store, and analyse large data sets. Identifying high-risk individuals in the transmission of
bovine TB among badgers and cattle. PLOS One 4, e5016.
Nonetheless, despite all these benefits, biologging will
Bonfil R., Meÿer M., Scholl M. C. et al. (2005). Transoceanic
probably never replace the fine-grain approach and the
migration, spatial dynamics, and population linkages of
experience of the human observer. Whenever possi- white sharks. Science 310, 100–103.
ble, we suggest that combining different techniques Bolt S. L., Boyland N. K., Mlynski D. T., James R., and
probably constitutes the best approach to catch the Croft D. P. (2017). Pair housing of dairy calves and
complexity of animal societies and the functioning of age at pairing: effects on weaning stress, health, pro-
individuals that live and interact together in a myriad duction and social networks. PLOS One 12(1), e0166926.
of ways. https://doi.org/10.1371/journal.pone.0166926.
Borgatti S., Everett M., and Freeman L. (2002). UCINET 6 for
Windows: software for social network analysis. http://
References www.analytictech.com/.
Ackerman J. T., Adams J., Takekawa J. Y. et al. (2004). Effects Boucherie P. H., Loretto M. C., Massen J. J., and Bugnyar
of radiotransmitters on the reproductive performance T. (2019). What constitutes ‘social complexity’ and ‘social
of Cassin’s auklets. Wildl. Soc. Bull. 32(1973–2006), intelligence’ in birds? Lessons from ravens. Behav. Ecol.
1229–1241. Sociobiol. 73, 12.
Boyd J. M., Mackay J. Doney R. G. G., and Jewell P. A. Corbet N .J., Patison K. P., Menzies, D. J., and Swain
(1964). The Soay sheep of the island of Hirta, St. Kilda. D. L. (2018). Using temporal associations to determine
A study of a feral population. Proc. Zool. Soc. Lond. 142, postpartum oestrus in tropical beef cows. Anim. Prod. Sci.
129–163. 58, 1465–1469.
Boyland N. K., James R., Mlynski D. T., Madden J. R., and Creech T. G., Cross P. C., Scurlock B. M., Maichak E. J.,
Croft D. P. (2013). Spatial proximity loggers for record- Rogerson J. D., Henningsen J. C., and Creel S. (2012).
ing animal social networks: consequences of inter-logger Effects of low-density feeding on elk–fetus contact
variation in performance. Behav. Ecol. Sociobiol. 67, rates on Wyoming feedgrounds. J. Wildl. Manag. 76,
1877–1890. 877–886.
Broihanne M.-H., Romain A., Call J. et al. (2018). Mon- Croft D. P., James R., Ward A. J. W., Botham M. S., Mawdsley
keys (Sapajus apella and Macaca tonkeana) and great D., and Krause J. (2005). Assortative interactions and
apes (Gorilla gorilla, Pongo abelii, Pan paniscus, and Pan social networks in fish. Oecologia 143, 211–219.
troglodytes) play for the highest bid. J. Comp. Psychol. 133, Crook J. H. (1964). Adaptive significance of avian social
301–312. organisations. Anim. Behav. 12, 393.
Brown D. D., Kays R., Wikelski M., Wilson R., and Klimley A. Crook J. H. (1970). Social organization and environment—
P. (2013). Observing the unwatchable through acceleration aspects of contemporary social ethology. Anim. Behav. 18,
logging of animal behavior. Anim. Biotelemetry 1, 20. 197–209.
Bueno-Guerra N. and Amici F. (2018). Field and Laborato- Crook J. H. and Gartlan J. S. (1966). Evolution of primate
ry Methods in Animal Cognition: A Comparative Guide. societies. Nature 210, 1200–1203.
Cambridge University Press. Cross P. C., Creech T. G., Ebinger M. R., Heisey D. M.,
Busson M., Authier M., Barbraud C., Tixier P., Reisinger R. Irvine K. M., and Creel S. (2012). Wildlife contact analy-
R., Janc A., and Guinet C. (2019). Role of sociality in the sis: emerging methods, questions, and challenges. Behav.
response of killer whales to an additive mortality event. Ecol. Sociobiol. 66, 1437–1447.
Proc. Natl. Acad. Sci. 116, 11812–11817. Cross P. C., Creech T. G., Ebinger M. R. et al. (2013). Female
Cantor M. and Whitehead H. (2013). The interplay between elk contacts are neither frequency nor density dependent.
social networks and culture: theoretically and among Ecology 94, 2076–2086.
whales and dolphins. Phil. Trans. R. Soc. B 368, 20120340. Csardi G. and Nepusz T. (2006). The Igraph software pack-
Carne C., Semple S., Morrogh-Bernard H., Zuberbühler K., age for complex network research. InterJournal Complex
and Lehmann J. (2013). Predicting the vulnerability of Syst. 1695, 1–9.
great apes to disease: the role of superspreaders and their Daniel T. A., Chiaradia A., Logan M., Quinn G. P., and
potential vaccination. PLOS One 8, e84642. Reina R. D. (2007). Synchronized group association in little
Carpenter, C. R. (1942a). Characteristics of social behavior penguins, Eudyptula minor. Anim. Behav. 74, 1241–1248.
in non-human primates. Trans. New York Acad. Sci. 4, Davis M. J., Thokala S., Xing X., Hobbs N. T., Miller M. W.,
248–258. Han R., and Mishra S. (2013). Testing the functionality and
Carpenter C. R. (1942b). Societies of monkeys and apes. Biol. contact error of a GPS-based wildlife tracking network.
Symp. 8, 177–204. Wildl. Soc. Bull. 37, 855–861.
Castles M., Heinsohn R., Marshall H. H., Lee A. E., Deb D., Wiper S., Russo A., Gong S., Shi Y., Tymosze, C.,
Cowlishaw G., and Carter A. J. (2014). Social networks cre- and Jain A. (2018). Face recognition: primates in the wild.
ated with different techniques are not comparable. Anim. arXiv preprint arXiv:1804.08790.
Behav. 96, 59–67. Dechmann D. K. N., Heucke S. L., Giuggioli L., Safi K., Voigt
Charpentier M. J. E., Harté M., Poirotte C., Meric de Bellefon C. C., and Wikelski M. (2009). Experimental evidence for
J., Laubi B., Kappeler P. M., and Renoult J. P. (2020). Same group hunting via eavesdropping in echolocating bats.
father, same face: deep-learning reveals natural selection Proc. R. Soc. Lond. B 276, 2721–2728.
for paternally-derived signalling of kinship in a wild pri- Douglas M. E., Ji W., and Clout M. N. (2006). MateID:
mate. Sciences Advances 6: eaba3274. design and testing of a novel device for recording con-
Cheney D. L., Silk J. B., and Seyfarth R. M. (2016). Net- tacts between free-ranging animals. Wildl. Soc. Bull. 34,
work connections, dyadic bonds and fitness in wild female 203–207.
baboons. R. Soc. Open Sci. 3, 160255. Doyle R. E., Broster J. C., Barnes K., and Browne W. J. (2016).
Chepko-Sade B. D. and Sade D. S. (1979). Patterns of group Temperament, age and weather predict social interaction
splitting within matrilineal kinship groups. Behav. Ecol. in the sheep flock. Behav. Proc. 131, 53–58.
Sociobiol. 5, 67–86. Drewe J. A., Madden J. R., and Pearce G. P. (2009). The
Chepko-Sade B. D., Reitz K. P., and Sade D. S. (1989). Socio- social network structure of a wild meerkat population:
metrics of Macaca mulatta IV: network analysis of social 1. Inter-group interactions. Behav. Ecol. Sociobiol. 63,
structure of a pre-fission group. Soc. Netw. 11, 293–314. 1295–1306.
Clutton-Brock T. (1974). Primate social organization and Drewe J. A., Weber N., Carter S. P. et al. (2012). Performance
ecology. Nature 250, 539–542. of proximity loggers in recording intra- and inter-species
Clutton-Brock T. (1982). The red deer of Rhum. Natural interactions: a laboratory and field-based validation study.
History 91, 42. PLOS One 7, e39068.
Drewe J. A., O’Connor H. M., Weber N., McDonald R. A., Fisher D. N., Rodríguez-Muñoz R., and Tregenza T. (2016).
and Delahay R. J. (2013). Patterns of direct and indirect Wild cricket social networks show stability across genera-
contact between cattle and badgers naturally infected with tions. BMC Evol. Biol. 16, 151.
tuberculosis. Epidemiol. Infect. 141, 1467–1475. Fisher J. and Hinde R. A. (1949). The opening of milk bottles
Dufour V., Pelé M., Sterck E. H. M., and Thierry B. (2007). by birds. Br. Birds 42, 347–357.
Chimpanzee (Pan troglodytes) anticipation of food return: Flack J. C., Krakauer D. C., and De Waal F. B. M. (2005).
coping with waiting time in an exchange task. J. Comp. Robustness mechanisms in primate societies: a perturba-
Psychol. 121, 145–155. tion study. Proc. R. Soc. Lond. B 272, 1091–1099.
Duncan A. J., Gunn G. J., Lewis F. I., Umstatter C., and Flack J. C., Girvan M., de Waal F. B., and Krakauer D. C.
Humphry R. W. (2012). The influence of empirical con- (2006). Policing stabilizes construction of social niches in
tact networks on modelling diseases in cattle. Epidemics primates. Nature 439, 426.
4, 117–123. Forin-Wiart M.-A., Enstipp M. R., Le Maho Y., and
Ekman, J. (1979). Coherence, composition and territories of Handrich Y. (2019). Why implantation of bio-loggers
winter social groups of the willow tit Parus montanus and may improve our understanding of how animals cope
the crested tit P. cristatus. Ornis. Scand. 10, 56–68. within their natural environment. Integr. Zool. 14,
Ellis, W., FitzGibbon, S., Pye, G. et al. (2015). The role 48–64.
of bioacoustic signals in koala sexual selection: insights Formica V. A., Wood C. W., Larsen W. B., Butterfield R. E.,
from seasonal patterns of associations revealed with GPS- Augat M. E., Hougen H. Y., and Brodie Iii E. D. (2012).
proximity units. PLOS One 10, e0130657. Fitness consequences of social network position in a wild
Emery N. (2016). Bird Brain: An Exploration of Avian Intel- population of forked fungus beetles (Bolitotherus cornu-
ligence. Princeton University Press. tus). J. Evol. Biol. 25, 130–137.
Emlen S. T. and Oring L. W. (1977). Ecology, sexual selec- Foster E. A., Franks D. W., Morrell L. J., Balcom, K. C., Par-
tion, and evolution of mating systems. Science 197, sons K. M., Van Ginneken A., and Croft D. P. (2012). Social
215–223. network correlates of food availability in an endangered
Espinas A. (1877). Des Sociétés Animales. Etude de Psy- population of killer whales, Orcinus orca. Anim. Behav. 83,
chologie Comparée. Germer Baillière. 731–736.
Farine D. R. (2013). Animal social network inference and Fushing H., Wang H., VanderWaal K., McCowan B., and
permutations for ecologists in R using Asnipe. Methods Koehl P. (2013). Multi-scale clustering by building a robust
Ecol. Evol. 4, 1187–1194. and self correcting ultrametric topology on data points.
Farine D. R., Montiglio P.-O., and Spiegel O. (2015). From PLOS One 8, e56259.
individuals to groups and back: the evolutionary implica- Garstang M. (2015). Elephant Sense and Sensibility. Aca-
tions of group phenotypic composition. Trends Ecol. Evol. demic Press.
30, 609–621. Gimenez O., Mansilla L., Klaich M. J., Coscarella M. A.,
Ferreira A. C., Silva L. R., Renna F. et al. (2019). Deep Pedraza S. N., and Crespo E. A. (2019). Inferring animal
learning-based methods for individual recognition in social networks with imperfect detection. Ecol. Model.
small birds. BioRxiv, 862557. 401, 69–74.
Filer J. (2008). Fish cognition and behavior. Fish and Godfrey-Smith P. (2016). Other Minds: The Octopus, the Sea,
aquatic resources series 11. Fish and Fisheries 9(2), and the Deep Origins of Consciousness. Farrar, Straus and
214–215. Giroux.
Firth J. A. and Sheldon B. C. (2015). Experimental manip- Goldenberg S. Z., Douglas-Hamilton I., and Wittemyer
ulation of avian social structure reveals segregation is G. (2016). Vertical transmission of social roles drives
carried over across contexts. Proc. R. Soc. Lond. B 282, resilience to poaching in elephant networks. Curr. Biol.
20142350. 26, 75–79.
Firth J. A., Sheldon B. C., and Brent L. J. (2017a). Indirectly Goodall J. (1986). Social rejection, exclusion, and shun-
connected: simple social differences can explain the caus- ning among the Gombe chimpanzees. Ethol. Sociobiol. 7,
es and apparent consequences of complex social network 227–236.
positions. Proc. R. Soc. Lond. B 284, 20171939. Guttridge T. L., Gruber S. H., Krause J., and Sims D. W.
Firth J. A., Voelkl B., Crates R. A., Aplin L. M., Biro D., Croft (2010). Novel acoustic technology for studying free-
D. P., and Sheldon B. C. (2017b). Wild birds respond to ranging shark social behaviour by recording individuals’
flockmate loss by increasing their social network associa- interactions. PLOS One 5, e9324.
tions to others. Proc. R. Soc. Lond. B 284, 20170299. Haddadi H., King A. J., Wills A. P. et al. (2011). Deter-
Fisher D. N. and McAdam A. G. (2017). Social traits, social mining association networks in social animals: choosing
networks and evolutionary biology. J. Evol. Biol. 30, spatial–temporal criteria and sampling rates. Behav. Ecol.
2088–2103. Sociobiol. 65, 1659–1668.
Fisher D. N. and Pruitt J. N. (2019). Insights from the study of Hamede R. K., Bashford J., McCallum H., and Jones M.
complex systems for the ecology and evolution of animal (2009). Contact networks in a wild Tasmanian devil (Sar-
populations. Current Zoology 66(1), 1–14. cophilus harrisii) population: using social network analysis
to reveal seasonal variability in social behaviour and its Kooyman G. L. (1965). Techniques used in measuring
implications for transmission of devil facial tumour dis- diving capacities of Weddell seals. Polar Rec. 12,
ease. Ecol. Lett. 12, 1147–1157. 391–394.
Hansen M. F., Smith M. L., Smith L. N., Salter M. G., Baxter Körschens M., Barz B., and Denzler J. (2018). Towards
E. M., Farish M., and Grieve B. (2018). Towards on-farm automatic identification of elephants in the wild. arXiv
pig face recognition using convolutional neural networks. preprint arXiv:1812.04418.
Comp. Ind. 98, 145–152. Krackhardt D. and Stern R. N. (1988). Informal networks and
Hazen E. L., Maxwell S. M., Bailey H. et al. (2012). Ontogeny organizational crises: an experimental simulation. Soc.
in marine tagging and tracking science: technologies and Psychol. Q. 51, 123.
data gaps. Mar. Ecol. Prog. Ser. 457, 221–240. Krause J., Krause S., Arlinghaus R., Psorakis I., Roberts S.,
Heaney M., Gray R. D., and Taylor A. H. (2017). Keas per- and Rutz C. (2013). Reality mining of animal social sys-
form similarly to chimpanzees. PLoS One 12, e0169799. tems. Trends Ecol. Evol. 28, 541–551.
Heathcote R .J. P., Darden S. K., Franks D. W., Ramnarine Krause J., Jame, R., Franks D. W., and Croft D. P. (2015).
I. W., and Croft D. P. (2017). Fear of predation drives sta- Animal Social Networks. Oxford University Press.
ble and differentiated social relationships in guppies. Sci. Krause J., Wilson A. D. M., and Croft D. P. (2011). New
Rep. 7, 41679. technology facilitates the study of social networks. Trends
Henzi S., Lusseau D., Weingrill T., Van Schaik C., and Barrett Ecol. Evol. 26, 5–6.
L. (2009). Cyclicity in the structure of female baboon social Krupenye C., Kano F., Hirata S., Call J., and Tomasello M.
networks. Behav. Ecol. Sociobiol. 63, 1015–1021. (2016). Great apes anticipate that other individuals will act
Hinde R. A. and Fisher J. (1951). Further observations according to false beliefs. Science 354, 110–114.
on the opening of milk bottles by birds. Br. Birds 44, Kruuk H. (1972). The Spotted Hyena: A Study of Predation
392–396. and Social Behaviour. University of Chicago Press.
Hirsch B. T., Prange S., Hauver S. A., and Gehrt S. D. (2014). Lathuilière S., Mesejo P., Alameda-Pineda X., and Horaud
Patterns of latrine use by raccoons (Procyon lotor) and R. (2019). A comprehensive analysis of deep regression.
implication for Baylisascaris procyonis transmission. J. IEEE Transactions on Pattern Analysis and Machine Intel-
Wildl. Dis. 50, 243–249. ligence 42(9), 2065–2081.
Holt-Lunstad J., Smith T. B., and Layton J. B. (2010). Social Lavelle M. J., Fischer J. W., Phillips G. E. et al. (2014). Assess-
relationships and mortality risk: a meta-analytic review. ing risk of disease transmission: direct implications for an
PLOS Medicine 7, e1000316. indirect science. BioScience 64, 524–530.
Hoppitt W., Boogert N. J., and Laland K. N. (2010). Detect- Lavelle M. J., Kay S. L., Pepin K. M., Grear, D. A., Campa H.,
ing social transmission in networks. J. Theor. Biol. 263, and VerCauteren K. C. (2016). Evaluating wildlife–cattle
544–555. contact rates to improve the understanding of dynamics of
House J. S., Landis K. R., and Umberson D. (1988). Social bovine tuberculosis transmission in Michigan, USA. Prev.
relationships and health. Science 241, 540. Vet. Med. 135, 28–36.
Hussey N. E., Kessel S. T., Aarestrup K. et al. (2015). Aquatic LeCun Y., Bengio Y., and Hinton G. (2015). Deep learning.
animal telemetry: a panoramic window into the underwa- Nature 521(7553), 436.
ter world. Science 348, 1255642. Lefebvre D., Ménard N., and Pierre, J. S. (2003). Modelling
Imanishi K. (1952). Evolution of the humanity. In: K. Iman- the influence of demographic parameters on group struc-
ishi (ed.) Man, 36–94. Mainichi-shinbunsha. ture in social species with dispersal asymmetry and group
Jacoby D. M. P., Papastamatiou Y. P., and Freeman, R. (2016). fission. Behav. Ecol. Sociobiol. 53, 402–
Inferring animal social networks and leadership: applica- Levin I. I., Zonana D. M., Burt, J. M. and Safran, R .J. (2015).
tions for passive monitoring arrays. J. R. Soc. Interface 13, Performance of Encounternet tags: field tests of miniatur-
20160676. ized proximity loggers for use on small birds. PLOS One
Jarman P. J. (1974). Social organization of antelope in relation 10, e0137242.
to their ecology. Behaviour 48, 215. Lopes P. C., Block P., and König B. (2016). Infection-induced
Ji W., White P. C. L., and Clout M. N. (2005). Contact rates behavioural changes reduce connectivity and the poten-
between possums revealed by proximity data loggers. tial for disease spread in wild mice contact networks. Sci.
J. Appl. Ecol. 42, 595–604. Rep. 6, 31790.
Kappeler P. M. and Watts D. P. (2012). Long-Term Field Lusseau D. and Newman M. E. J. (2004). Identifying the role
Studies of Primates. Springer Science + Business Media. that animals play in their social networks. Proc. R. Soc.
Kappeler P. M., Cremer S., and Nunn C. L. (2015). Sociality Lond. B 271, S477–S481.
and health: impacts of sociality on disease susceptibility Marsh M. K., McLeod S. R., Hutchings M. R., and White
and transmission in animal and human societies. Philos. P. C. L. (2011a). Use of proximity loggers and network
Trans. R. Soc. B Biol. Sci. 370, 20140116. analysis to quantify social interactions in free-ranging
Kays R., Crofoot M. C., Jetz W., and Wikelski M. (2015). wild rabbit populations. Wildl. Res. 38, 1–12.
Terrestrial animal tracking as an eye on life and planet. Marsh M .K., Hutchings M. R., McLeod S. R., and White
Science 348, 2478. P. C. L. (2011b). Spatial and temporal heterogeneities in
the contact behaviour of rabbits. Behav. Ecol. Sociobiol. Pasquaretta C., Klenschi E., Pansanel J., Battesti M., Mery F.,
65, 183–195. and Sueur C. (2016). Understanding dynamics of infor-
Mathis A., Mamidanna P., Cury K. M., Abe T., Murthy V. N., mation transmission in Drosophila melanogaster using a
Mathis M. W., and Bethge M. (2018). DeepLabCut: Mark- statistical modeling framework for longitudinal network
erless pose estimation of user-defined body parts with data (the RSiena Package). Front. Psychol. 7, 20142480.
deep learning. Nature Neurosci. 21, 1281–1289. Paton P. W. C., Zabel C. J., Neal D. L., Steger G. N., Tilgh-
Matsuda I., Zhang P., Swedell L., Mori U., Tuuga A., Bernard man N.G., and Noon B. R. (1991). Effects of radio tags on
H., and Sueur C. (2012). Comparisons of intra-unit rela- spotted owls. J. Wildl. Manag. 55, 617–622.
tionships in nonhuman primates living in multilevel Pelé M., Broihanne M. H., Thierry B., Call J., and Dufour V.
social systems. Int. J. Primatol. 33, 1038–1053. (2014). To bet or not to bet? Decision-making under risk in
Matsuda I., Fukaya K., Pasquaretta C., and Sueur C. (2015). non-human primates. J. Risk Uncertain. 49, 141–166.
Factors influencing grooming social networks: insights Pelé M., Dufour V., Micheletta J., and Thierry B. (2010).
from comparisons of colobines with different dispersal Long-tailed macaques display unexpected waiting abili-
patterns. In: T. Furuichi, J. Yamagiwa, Juichi, and F. Aureli ties in exchange tasks. Anim. Cogn. 13, 263–271.
(eds.), Dispersing Primate Females, 231–254. Springer. Pelé M., Micheletta J., Uhlrich P., Thierry B., and Dufour V.
Matthews A., Ruykys L., Ellis W. et al. (2013). The success of (2011). Delay maintenance in Tonkean macaques (Macaca
GPS collar deployments on mammals in Australia. Aust. tonkeana) and brown Capuchin monkeys (Cebus apella). Int.
Mammal. 35, 65–83. J. Primatol. 32, 149–166.
McBride G. (1964). A general theory of social organization Pinter-Wollman N., Hobson E. A., Smith J. E. et al. (2013).
and behaviour. University of Queensland Papers, Faculty The dynamics of animal social networks: analytical,
of Veterinary Science 1, 75–110. conceptual, and theoretical advances. Behav. Ecol. 25,
McComb K., Shannon G., Durant, S. M., Sayialel K., Slotow 242–255.
R., Poole J., and Moss C. (2011). Leadership in elephants: Poirotte C., Massol F., Herbert A., Willaume E., Bomo P. M.,
the adaptive value of age. Proc. R. Soc. B Biol. Sci. 278, Kappeler P. M., and Charpentier M. J. E. (2017). Mandrills
3270–3276. use olfaction to socially avoid parasitized conspecifics. Sci.
McInnes A. M., McGeorge C., Ginsberg S., Pichegru L., and Adv. 3, e1601721.
Pistorius P. A. (2017). Group foraging increases foraging Prange S., Jordan T., Hunter C., Gehrt, S. D. (2006). New
efficiency in a piscivorous diver, the African penguin. R. radiocollars for the detection of proximity among individ-
Soc. Opensci. 4, 170918. uals. Wildl. Soc. Bull. 34, 1333–1344.
McKinnon E. A. and Love O. P. (2018). Ten years tracking Prange S., Gehrt S. D., and Hauver S. (2011). Frequen-
the migrations of small landbirds: lessons learned in the cy and duration of contacts between free-ranging rac-
golden age of bio-logging. Auk 135, 834–856. coons: uncovering a hidden social system. J. Mammal. 92,
Meise K., Krüger O., Piedrahita P., Müller A., and Trillmich F. 1331–1342.
(2013). Proximity loggers on amphibious mammals: a new Puga-Gonzalez I., Ostner J., Schülke O., Sosa S., Thierry
method to study social relations in their terrestrial habitat. B., and Sueur C. (2018). Mechanisms of reciprocity and
Aquat. Biol. 18, 81–89. diversity in social networks: a modeling and comparative
Miklosi A. 2007. Dog Behaviour, Evolution, and Cognition. approach. Behav. Ecol. 29, 745–760.
Oxford University Press. Puga-Gonzalez I., Sosa S., and Sueur C. (2019). Social style
Nagy M., Ákos Z., Biro D., and Vicsek T. (2010). Hier- and resilience of macaques’ networks, a theoretical inves-
archical group dynamics in pigeon flocks. Nature 464, tigation. Primates 60, 233–246.
890–893. Ralls K., Sanchez J. N., Savage J., Coonan T. J., Hudgens
Naug D. (2009). Structure and resilience of the social net- B. R., and Cypher B. L. (2013). Social relationships and
work in an insect colony as a function of colony size. reproductive behavior of island foxes inferred from prox-
Behav. Ecol. Sociobiol. 63, 1023–1028. imity logger data. J. Mammal. 94, 1185–1196.
Neverova N., Wolf C., Lacey G., Fridman L., Chandra D., Renoult J. P., Guyl B., Mendelson T. C., Percher A., Dorignac
Barbello B., and Taylor G. (2016). Learning human identity J., Geniet F., and Molino F. (2019). Modelling the percep-
from motion patterns. IEEE Access 4, 1810–1820. tion of colour patterns in vertebrates with HMAX. bioRxiv
Newman MEJ. (2010). Networks. An Introduction. Oxford 552307.
University Press. Reynolds, J. J. H., Hirsch, B. T., Gehrt, S. D., and Craft, M. E.
Nguyen N. G., Phan D., Lumbanraja F. R., et al. (2019). (2015). Raccoon contact networks predict seasonal suscep-
Applying deep learning models to mouse behavior recog- tibility to rabies outbreaks and limitations of vaccination.
nition. JBiSE 12, 183–196. J. Anim. Ecol. 84, 1720–1731.
O’Neill C. J., Bishop-Hurley G. J., Williams P. J., Reid Robert K., Garant D., and Pelletier F. (2012). Keep in touch:
D. J., and Swain D. L. (2014). Using UHF proximity does spatial overlap correlate with contact rate frequency?
loggers to quantify male–female interactions: a scoping J. Wildl. Manag. 76, 1670–1675.
study of estrous activity in cattle. Anim. Reprod. Sci. Robert K., Garant D., Wal E. V., and Pelletier F. (2013).
151, 1–8. Context-dependent social behaviour: testing the interplay
between season and kinship with raccoons. J. Zool. 290, pairs during lactation. Mar. Mammal Sci. 19,
199–207. 384–395.
Robertson B., Holland J. D., and Minot E. (2012). Wildlife Scardamaglia R. C., Fiorini V. D., Kacelnik A., and Reboreda
tracking technology options and cost considerations. J. C. (2017). Planning host exploitation through prospect-
Wildl. Res. 38, 653–663. ing visits by parasitic cowbirds. Behav. Ecol. Sociobiol.
Romano V., Duboscq J., Sarabian C., Thomas E., Sueur C., 71, 23.
and MacIntosh A. J. J. (2016). Modeling infection transmis- Schofield D., Nagrani A., Zisserman A., Hayashi M., Mat-
sion in primate networks to predict centrality-based risk. suzawa T., Biro D., and Carvalho S. (2019). Chimpanzee
Am. J. Primatol. 78, 767–779. face recognition from videos in the wild using deep learn-
Romano V., Shen M., Pansanel J., MacIntosh A. J., and Sueur ing. Sci. Adv. 5, eaaw0736.
C. (2018). Social transmission in networks: global efficien- Schreer J. F. and Testa J. W. (1996). Classification of Weddell
cy peaks with intermediate levels of modularity. Behav. seal diving behavior. Mar. Mammal Sci. 12, 227–250.
Ecol. Sociobiol. 72, 154. Schroff F., Kalenichenko D., and Philbin J. (2015). Facenet:
Ropert-Coudert Y. and Wilso, R. P. (2005). Trends and per- a unified embedding for face recognition and clustering.
spectives in animal-attached remote sensing. Front. Ecol. Proceedings of the IEEE Conference on Computer Vision
Environ. 3, 437–444. and Pattern Recognition, 815–823.
Rouco C., Jewell C., Richardson K. S., French N. P., Bud- Seyfarth R. M. (1976). Social relationships among adult
dle B. M., and Tompkins D. M. (2018). Brushtail possum female baboons. Anim. Behav. 24, 917–938.
(Trichosurus vulpecula) social interactions and their impli- Shepard E. (2008). Identification of animal movement pat-
cations for bovine tuberculosis epidemiology. Behaviour terns using tri-axial accelerometry. Endanger. Species Res.
155, 621–637. 10, 47.
Ruckstuhl K. E. and Neuhaus P. (2000). Sexual segre- Siebenaler J. and Caldwell D. K. (1956). Cooperation among
gation in ungulates: a new approach. Behaviour 137, adult dolphins. J. Mammal. 37, 126–128.
361–377. Silk J. B. (2007). The adaptive value of sociality in mam-
Rushmore J., Caillaud D., Hall R. J., Stumpf R. M., Mey- malian groups. Philos. Trans. R. Soc. B-Biol. Sci. 362,
ers L. A., and Altizer S. (2014). Network-based vacci- 539–559.
nation improves prospects for disease control in wild Silk J. B., Alberts S. C., and Altmann J. (2003). Social bonds
chimpanzees. J. R. Soc. Interface 11, 20140349. of female baboons enhance infant survival. Science 302,
Rutz C., Burns Z. T., James R., Ism, S. M. H., Burt, J., Otis 1231–1234.
B., Bowen J., and St Clair J. J. H. (2012). Automated map- Silk J. B., Beehner J. C., Bergman T. J. et al. (2009). The benefits
ping of social networks in wild birds. Curr. Biol. 22, of social capital: close social bonds among female baboons
R669–R671. enhance offspring survival. Proc. R. Soc. B-Biol. Sci. 276,
Ryder T. B., Horton B. M., van den Tillaart M., Morales De 3099–3104.
Dios J., and Moore I. T. (2012). Proximity data-loggers Silk M. J., Weber N., Steward L. C. et al. (2017). Seasonal vari-
increase the quantity and quality of social network data. ation in daily patterns of social contacts in the European
Biol. Lett. 8, 917–920. badger Meles meles. Ecol. Evol. 7, 9006–9015.
Saayman G. and Tayler C. (1973). Social organisation of Skyrms B. and Pemantle R. (2009). A dynamic model of
inshore dolphins (Tursiops aduncus and sousa) in the Indian social network formation. In: T. Gross and H. Sayama
Ocean. J. Mammal. 54, 993–996. (eds.), Adaptive Networks, 231–251. Springer.
Sade D. S. (1972). Sociometrics of Macaca mulatta—linkages Sosa S., Puga-Gonzalez I., Feng H. H., Zhang P., Xiaohua
and cliques in grooming matrices. Folia Primatol. 18, X., and Sueur C. (2018a). A multilevel statistical toolkit
196–223. to study animal social networks: Animal Network Toolkit
Sade D. S. (1989). Sociometrics of Macaca mulatta III: (ANT) R package. bioRxiv 347005.
N-path centrality in grooming networks. Soc. Netw. 11, Sosa S., Pele M., Debergue E. et al. (2018b). Impact of group
273–292. management and transfer on individual sociality in High-
Sade D. S., Altmann M., Loy J., Hausfater G., and Breugge- land cattle (Bos taurus). ArXiv180511553 Q-Bio.
man J. A. (1988). Sociometrics of Macaca mulatta: II. Decou- Sosa S., Sueur C., and Puga-Gonzalez I. (2020). Network
pling centrality and dominance in rhesus monkey social measures in animal social network analysis: their
networks. Am. J. Phys. Anthropol. 77, 409–425. strengths, limits, interpretations and uses. Meth. Ecol.
Saitou T. (1978). Ecological study of social organization in Evol. 12, 10–21.
the great tit, Parus major L. I. Basic structure of the winter St Clair J. J. H., Burns Z. T., Bettaney E. M. et al. (2015). Exper-
flocks. Jap. J. Ecol. 28, 199–214. imental resource pulses influence social-network dynam-
Saraux C., Le Bohec C., Durant J. M. et al. (2011). Relia- ics and the potential for information flow in tool-using
bility of flipper-banded penguins as indicators of climate crows. Nat. Commun. 6, 7197.
change. Nature 469, 203–206. Stacey P. B. and Koenig W. D. (1990). Cooperative Breed-
Sato K., Mitani Y., Naito Y., and Kusagaya H. (2003). ing in Birds: Long Term Studies of Ecology and Behaviour.
Synchronous shallow dives by Weddell seal mother–pup Cambridge University Press.
Stanton M. A. and Mann J. (2012). Early social networks pre- Vander Wal E., Paquet P. C., and Andrés J. A. (2012b). Influ-
dict survival in wild bottlenose dolphins. PLOS One 7, ence of landscape and social interactions on transmission
e47508. of disease in a social cervid. Mol. Ecol. 21, 1271–1282.
Stroeymeyt N., Grasse A.V., Crespi A., Mersch D. P., Cremer Vander Wal E., Paquet P. C., Messier F., and McLoughlin
S., and Keller L. (2018). Social network plasticity decreas- P. D. (2013). Effects of phenology and sex on social prox-
es disease transmission in a eusocial insect. Science 362, imity in a gregarious ungulate. Can. J. Zool. 91, 601–609.
941–945. Walrath R., Deelen T. R. V., and VerCauteren K. C. (2011).
Sueur C. and Maire A. (2014). Modelling animal group fis- Efficacy of proximity loggers for detection of contacts
sion using social network dynamics. PLOS One 9, e97813. between maternal pairs of white-tailed deer. Wildl. Soc.
Sueur C., Jacobs A., Amblard F., Petit O., and King A. J. Bull. 35, 452–460.
(2011a). How can social network analysis improve the Weiss K., Khoshgoftaar T. M., and Wang D. (2016). A survey
study of primate behavior? Am. J. Primatol. 73, 703–719. of transfer learning. J. Big Data 3(1), 9.
Sueur C., Petit O., De Marco A., Jacobs A. T., Watanabe K., Whitehead H. (2008). Analyzing Animal Societies: Quanti-
and Thierry B. (2011b). A comparative network analysis tative Methods for Vertebrate Social Analysis. University
of social style in macaques. Anim. Behav. 82, 845–852. of Chicago Press.
Sueur C., Romano V., Sosa S., and Puga-Gonzalez, I. (2017). Whitehead H. (2009). SOCPROG programs: analysing ani-
Mechanisms of network evolution: a focus on socioeco- mal social structures. Behav. Ecol. Sociobiol. 63, 765–778.
logical factors, intermediary mechanisms, and selection Wilson E. O. (1975). Sociobiology: The New Synthesis. Har-
pressures. Primates 60, 1–15. vard University Press.
Swain D. L., Patison K. P., Heath B. M., Bishop-Hurley Wilson R. P. and McMahon C. R. (2006). Measuring devices
G. J., and Finger A. (2015). Pregnant cattle associations and on wild animals: what constitutes acceptable practice?
links to maternal reciprocity. Appl. Anim. Behav. Sci. 168, Front. Ecol. Environ. 4, 147–154.
10–17. Wilson R. P., Putz K., Peters G., Culik B., Scolaro J. A.,
Tentelier C., Aymes J.-C., Spitz B., and Rives J. (2016). Using Charrassin J. B., and Ropert-Coudert Y. (1997). Long-
proximity loggers to describe the sexual network of a term attachment of transmitting and recording devices to
freshwater fish. Environ. Biol. Fishes 99, 621–631. penguins and other seabirds. Wildl. Soc. Bull. 25, 101–106.
Tosa M. I., Schauber E. M., and Nielsen C. K. (2015). Famil- Wilson R. P., Shepard E. L. C., and Liebsch N. (2008). Prying
iarity breeds contempt: combining proximity loggers and into the intimate details of animal lives: use of a daily diary
GPS reveals female white-tailed deer (Odocoileus virgini- on animals. Endanger. Species Res. 4, 123–137.
anus) avoiding close contact with neighbors. J. Wildl. Dis. Wittemyer G., Douglas-Hamilton I., and Getz W. M. (2005).
51, 79–88. The socioecology of elephants: analysis of the processes
Tremblay Y. and Cherel Y. (1999). Synchronous underwater creating multitiered social structures. Anim. Behav. 69,
foraging behavior in penguins. Condor 101, 179–185. 1357–1371.
Trivers R. L. and Hare, H. (1976). Haploidploidy and the Wolf T., Ngonga Ngomo A., Bennett N., Burroughs R., and
evolution of the social insect. Science 191, 249–263. Ganswindt A. (2018). Seasonal changes in social networks
Uhrich J. (1938). The social hierarchy in albino mice. J. Comp. of giraffes. J. Zool. 305, 82–87.
Psy. 25, 373. Zhang R., Isola P., Efros A. A., Shechtman E., and Wang O.
Vander Wal E., Yip H., and McLoughlin P. D. (2012a). Sex- (2018). The unreasonable effectiveness of deep features as
based differences in density-dependent sociality: an exp- a perceptual metric. Proceedings of the IEEE Conference
eriment with a gregarious ungulate. Ecology 93, 206–212. on Computer Vision and Pattern Recognition, 586–595.
CHAPTER 4
Growth rings across the Tree of Life:

demographic insights from biogenic
time series data
Margaret E. K. Evans, Bryan A. Black, Donald A. Falk, Courtney L. Giebink,
and Emily L. Schultz
4.1 Introduction study of climate system dynamics, even though den-

droecology has a decades-deep literature (Fritts and
The collection of individual-level demographic data
Swetnam 1989). Dendrochronology has a set of prin-
has traditionally involved repeat censuses, either
ciples, practices, and language surrounding these that
through repeat visits to sessile organisms or capture-
give rise to a need for interdisciplinary translation.
recapture of mobile organisms. An appealing and over-
What are the sampling protocols of this discipline?
looked source of demographic information is biogenic
What are ‘cross-dating’ and ‘detrending’? What is a
time series data that can be generated in a single sam-
chronology? When borrowing methods across disci-
pling effort. These records—found on the hard body
plines, it is important to understand the why behind
parts of many plants and animals—offer information
how things are done and think carefully about their
on variation in demographic processes over time, in
application in a different context.
many cases providing information at time scales far
In keeping with the theme of the book, we then high-
exceeding what is typically generated through a repeat
light the diversity of taxa across the Tree of Life from
census approach, while maintaining annual resolution.
which annually resolved demographic time series data
In this chapter, our goals are twofold: first, we seek to
can be generated in a single sampling effort. Further,
make these single-sample time series data more widely
we review other kinds of biogenic time series data,
known to demographers, both from a methodological
beyond growth rings, and other kinds of information to
perspective and in terms of the important knowledge
be gained from growth rings, beyond just their widths.
gaps they can fill. Second, we encourage demographers
We then survey the great variety of demographic pro-
to take on the challenge and opportunity to use these
cesses into which biochronologies have offered insight.
data to improve demographic models and address the
They provide a somatic record of all the individual-
impacts of global change.
level processes that demographers care about. Clearly,
To make biogenic time series more widely appre-
growth increment time series make it possible to inves-
ciated by demographers, we begin by describing the
tigate and parse the many drivers of growth variation,
data (i.e. how they are generated and how they have
such as individual size, climate conditions, competitive
been analysed) and explaining some of the tools and
(or other) interactions, and natural or human-caused
practices of the ‘chronology’ sciences to demographers.
disturbances. Growth rings also act as an internal clock,
Tree-ring science has historically been aligned with
recording the passing of time, so that singular demo-
Earth sciences and geography rather than life sciences;
graphic events in the life cycle, such as birth and death,
the discipline of dendrochronology was founded by
can be dated from these time series, offering insight
an astronomer (Andrew Ellicott Douglass). To this
into their drivers. The experience of an individual with
day, a primary use of tree-ring time series data is the
Margaret E. K. Evans et al., Growth rings across the Tree of Life: demographic insights from biogenic time series data.
In: Demographic Methods Across the Tree of Life. Edited by Roberto Salguero-Gómez and Marlène Gamelon, Oxford University Press.
respect to dispersal or migration, which can be inferred than ring width that can be derived from growth rings,
from ring microchemistry, is also recorded in growth and other biogenic time series data.
ring time series. Questions of evolutionary demogra- Tree rings are perhaps the most well-known exam-
phy (i.e. life history variation among individuals and ple of annually resolved biogenic time series data
across species, individual reaction norms, and local (Douglass 1920; Fritts 1976; Speer 2010). Annual incre-
adaptation) can also be addressed by analysing varia- ments from the ear bones of fish (i.e. otoliths) are also
tion within and among growth time series. Hard body relatively well known, although their study is recent
parts such as bones and teeth are all that remain of most compared to the study of tree rings (Campana and
extinct vertebrates (e.g. dinosaurs and hominins). Thus Neilson 1985; Campana 1990; Black et al. 2019). Both
growth ring samples (and other biogenic time series are curated in open-access international repositories:
data) can help to fill important knowledge gaps about in the case of tree rings, a dedicated repository known
growth, birth, death, movement, and evolution across as the International Tree-Ring Data Bank (ITRDB;
a wide range of organisms. https://www.ncdc.noaa.gov/data-access/paleoclima
While we have structured this chapter to high- tology-data/datasets/tree-ring). Both are distributed
light cross-taxa similarities and differences, we note extensively, covering much of the world’s temperate
that dendrochronology (i.e. the study of tree-ring time forest ecosystems (more than 4,000 sites on six con-
series) is the oldest and most well-developed scientif- tinents), coastal marine ecosystems, and increasing
ic discipline to use growth rings or any other somatic coverage in freshwater systems.
records of demographic processes. The beginnings of In the formation of rings, the material in both trees
dendrochronology date to the early twentieth century and otoliths is not resorbed by the organism and is thus
(Douglass 1914, 1920); like ecology, it is a century-old a stable archive. Tree rings represent secondary tree
science. Hence, many examples throughout the chapter growth (increase in bole diameter) through the forma-
draw from the world of tree-ring science. We hope tion of xylem; the ‘ring’ is simply a radial cross-section
to see the kinds of demographic questions asked of of the annual growth layer. In seasonal climates (those
and methods applied to tree-ring time series applied with a predictable period each year when growth
to other biogenic time series data. ceases), these growth rings are formed yearly. Many
Our second goal is for demographers to come (though not all) conifers reliably form recognisable
away with new ideas about sources of demograph- annual rings, with clear ring boundaries formed by
ic information that can complement more tradition- darkly coloured latewood (Figure 4.1a), which are
al data streams, new ideas about the temporal and xylem cells formed late in the growth season, with
spatial scales at which to think about the demogra- thicker cell walls, smaller lumen area, and a higher
phy of their study organism, and a renewed focus concentration of lignin. Light-coloured earlywood
on the drivers and mechanisms behind demograph- has opposite cell characteristics. Many temperate
ic outcomes. We end by suggesting how information angiosperm trees also reliably make recognisable
from biochronologies may be incorporated into pop- annual growth rings, of varying wood anatomy.
ulation models, bringing to these models a greater Tropical trees are one of the current frontiers of den-
understanding of the drivers of vital rate varia- drochronology, with most species requiring additional
tion and a new capacity to address issues of global study to identify reliable anatomical criteria and the
change. timing of their formation (Silva et al. 2019). In the
otoliths of bony fish, ratios of calcium carbonate and
4.2 Tree and otolith growth rings— protein vary seasonally, forming growth increments.
Yet as in trees, there are environments, such as the
the basics deep ocean, that lack the seasonality to induce annual
Here we lay out two prime examples of growth ring otolith increment formation. Verification of annual
time series data: one from sessile, terrestrial organisms periodicity is often necessary for each new species
in the plant kingdom (trees), and the other from mobile, with such techniques as radiometric dating or mark
aquatic organisms in the animal kingdom (fish). We and recapture (Campana and Neilson 1985; Campana
focus first on the nature of the rings (how they are 1990; Figure 4.1b).
formed) and how they are sampled and developed into An important distinction between these two is that
time series data, along with standard practices for their sampling of otoliths from fish populations is frequently
analysis (i.e. a methodological focus). The subsequent conducted to gather fish age data for stock assess-
two sections of this chapter then briefly survey growth ment (i.e. demographic modelling) and is therefore
rings in other organisms, informative measures other aimed at an unbiased sample of population structure,
G R O W T H R I N G S A C R O S S T H E T R E E O F L I F E : D E M O G R A P H I C I N S I G H T S F R O M B I O G E N I C T I M E S E R I E S D ATA 79
(a) Tree rings (d) Geoduck
Latewood
Earlywood
(b) Split-nose rockfish otolith
(c) Freshwater mussel
3mm
(e) Coral growth rings (f) Lines of

arrested growth
(i) Ibex
horn Outermost
(oldest) increment
(g) Perikymata (h) Cone scar Innermost (youngest)
increment
Figure 4.1 Growth rings across the Tree of Life. (a) Tree rings of the conifers Douglas-fir (Pseudotsuga menziesii) (left; photo by B. A. Black) and
ponderosa pine (Pinus ponderosa) (right; photo by R. J. DeRose), with light-coloured earlywood and dark-coloured latewood. (b) Cross-section of
the otolith (ear bone) of a split-nose rockfish (Sebastes diploproa). (c) Cross-section of a freshwater mussel shell (Margaritifera falcata). (d) Etched
and stained cross-section from the hinge plate region of a Pacific geoduck shell (Panopea abrupta). (Photos in b–d by B. A. Black). (e). Growth
rings of a coral (Porites sp.) (X-ray image by D. Thompson). (f) Growth lines, or lines of arrested growth, in the dorsal rib of Draconyx loureiroi, an
ornithopod of the Late Jurassic (photo by K. Waskow). (g) Perikymata, or growth lines, on the teeth of Plio-Pleistocene hominins (from Le Cabec
et al. 2015). (h) Cone scar on the conifer whitebark pine (Pinus albicaulis) (photo by J. Rapp). (i) Horn growth increments of the ibex (Capra ibex),
an alpine ungulate (from Buntgen et al. 2014).
whereas the sampling of tree rings has often conven- dendrochronologists (Brienen and Zuidema 2006) and
tionally been oriented towards climate reconstruction. the ‘Rosa Lee phenomenon’ by sclerochronologists
As a result, samples in the ITRDB are biased towards (Morrongiello et al. 2012). However, there are relative-
old, climate-sensitive, isolated individuals (Klesse ly unbiased tree-ring collections developed in a forest
et al. 2018; the ‘site and tree selection’ principles inventory context (DeRose et al. 2017; Canham et al.
of dendrochronology). This creates a biased sample 2018), as well as many collections created specifically to
that is not appropriate for population-level inference. answer demographic or ecological questions that have
A sampling bias towards old individuals (desirable the multiple cohort sampling needed for population-
for their long time series) will lead to time-biased level inference (Wykoff and Clark 2002; Youngblood
estimates of growth rates if the risk of mortality is et al. 2004; Guiterman et al. 2018).
greater for faster-growing individuals in a popula- A key step in developing a biogenic time series,
tion, known as ‘slow-grower survivorship bias’ by from tree rings, otoliths, or other somatic records, is
the process of assigning a time of formation to each ring widths that arises from measuring the growth
ring (or other structure). If there were no possibility of of a 3D organism in a linear dimension (evident in
missing or falsely identified growth rings, year assign- Figure 4.1b,d). That is, even if the growth of a tree
ment would simply be a matter of counting backwards in area (stem basal area) is constant, with increasing
from the year of sampling a live organism. Given stem radius, radial increments decline. Depending on
these sources of uncertainty and error, along with sam- the goal of the study, and the sampling that was exe-
ples from dead individuals where the year of forma- cuted towards that goal, detrending may also aim to
tion of the final ring is unknown, dendrochronologists eliminate or preserve the signal of other drivers of
developed the process of ‘cross-dating’ (Douglass 1941; growth variability (e.g. climate, competition, and dis-
Black et al. 2016), which relies upon pattern-matching turbances.). Detrending involves fitting a mathemati-
across samples, analogous to the alignment of DNA cal function to the ring-width time series, after which
sequences, but based on ring-width variation rather observed ring widths are divided by the expected ring
than nucleotide sequence variation. Whereas DNA widths described by the mathematical function (Cook
sequence alignment is based on similarities induced et al. 1990; Speer 2010). This transforms the data from
through ancestor-descendant relationships (related- absolute ring widths to a unitless ring width index with
ness), cross-dating relies upon synchronous patterns (ideally) a constant mean and variance (Figure 4.2).
of increment widths among individuals induced by a Individual-level detrended time series are then aver-
common (environmental) limiting factor. With enough aged to form a chronology, a site-level time series of
replication, it becomes possible to identify missing or growth anomalies (Speer 2010).
false rings (analogous to deletions and insertions) in The current state of the art of detrending are meth-
any one sample. Unlike DNA sequence alignment, ods that disentangle size-related versus other caus-
automated cross-dating of growth rings (i.e. by an algo- es of low-frequency ring-width variation, including
rithm) does not yet exist; a suite of tools and metrics regional curve standardisation (RCS; Briffa et al. 1992)
(Holmes 1983; Wigley et al. 1984; Cook et al. 1990) and ‘signal-free’ detrending (Melvin and Briffa 2008).
are used to verify year assignments made by human Under RCS, all ring-width time series of a given species
judgement and experience. and sampling region are aligned by tree age (rather
Accurate year assignment is essential to answer than calendar year) and detrended with a single nega-
questions about the drivers of ring-width variability. tive exponential curve, or other relatively stiff function
Statistical relationships between ring widths and the (Briffa et al. 1992). Ring-width indices are then aver-
climatic or ecological conditions under which rings aged with respect to calendar year to form a chronol-
were formed can be seriously compromised even at ogy. However, the data requirements for RCS are rel-
modest error rates (Black et al. 2016; i.e. when false atively steep: large sample sizes, individuals of a vari-
rings are identified as an annual growth ring by the ety of ages, and ideally dead-collected material com-
observer, or missing rings are undetected). But there bined with live-collected samples (Helama et al. 2017).
are cases where cross-dating to traditional standards Where these requirements are not met, each measure-
is difficult, for example, because environmental con- ment time series may be fit to a negative exponential
ditions are not strongly limiting to growth and hence (Figure 4.2c) or regression function, but at the expense
interannual variability and synchrony between sam- of discarding low-frequency variability at wavelengths
ples is weak, or when lifespans and hence time series longer than the mean measurement time series (Cook
are short. This highlights the need for formal statistical et al. 1995). If the goal is to remove interdecadal vari-
methods to quantify uncertainty in year assignments, ability in ring widths driven by disturbances (e.g.
which are currently lacking (but see Ricker et al. 2020). gap-phase dynamics) that tend to occur in closed-
The development of such methods would greatly sup- canopy forests, more flexible cubic smoothing splines
port the ‘chronology’ sciences, since the information (Figure 4.2d) may be used. The resulting chronology
content of a time series is inversely proportional to must then be interpreted with the knowledge that the
the uncertainty about year assignments, and interpre- majority of low-frequency variability—regardless of its
tations derived from time series data should be made cause—has been removed.
in light of (quantified) uncertainty. In the sclerochronology literature, in addition to
Following the assignment of a calendar year to each detrending otolith ring-width time series data and
growth ring, it has long been standard practice in den- forming a site-level chronology, it has also been com-
drochronology to detrend ring-width time series, with mon to use mixed effects models to analyse individual-
the primary goal of removing the decline in absolute level variability in absolute growth. These models
(a)
Raw time series
1.2
1.0
0.8
mm
0.6
0.4
0.2
20 40 60 80 100
Age (years)
(b) (c) (d)
3.5 Horizontal line (mean) 2.5 Negative exponential Spline
2.0
Ring width index
3.0 2.0
2.5 1.5
RWI
2.0 1.5
1.5 1.0 1.0
1.0
0.5 0.5 0.5
20 40 60 80 100 20 40 60 80 100 20 40 60 80 100
Age (years) Age (years) Age (years)
Figure 4.2 (a) Raw growth increments (grey line) of the shell of a Pacific geoduck (Panopea abrupta) sampled in the Tree Knob Islands of the
northern British Columbia coast, Canada. Overplotted in (a) are four mathematical functions fitted to the data to ‘detrend’ the raw measurements.
The residuals (ring width index) from three of those detrending functions are shown in panels (b) as a horizontal line, in (c) as a negative
exponential, and (d) as a cubic smoothing spline, with a 50% frequency cut-off set to two-thirds of the time series length. In (a), a fourth, yet more
flexible mathematical function fit to the raw measurements is Friedman’s super smoother (blue line). All panels were created using the
detrend.series function in the R package dplR (Bunn 2008).
are particularly effective in partitioning and attribut- well (Martin-Benito et al. 2011; Fernandez-de-Una et al.
ing variance amongst a suite of predictors (i.e. parse 2016; Foster et al. 2016; Dorado-Linan 2018; Wright
the many factors influencing growth). For example, et al. 2018). Ring-width time series data may also
Morrongiello and Thresher (2015) used MEMs to be fit to nonlinear growth models (see Fernandez-de
attribute variation in growth of the tiger flathead Una et al. 2016; Rollinson et al. 2016; Buechling et al.
(Platycephalus richardsoni) to effects of age, sex, cohort, 2017; Canham et al. 2018). It is worth noting that
year, individual, and geographic area. They found the ecological literature includes arguments against
growth in poleward populations to be driven weakly the analysis of residuals over a single-stage multiple
by ocean temperature, whereas equatorward popula- regression analysis (see Garcia-Berthou 2001; Green
tions were influenced more strongly and negatively 2001; Freckleton 2002; Morrongiello et al. 2012), but
by ocean temperature. The use of hierarchical models also that a MEM approach is inherently data-hungry,
allowed Morrongiello and Thresher (2015) to attribute requiring data on multiple covariates, which are not
growth variation across space, time, and individuals always available.
to causes ranging from individual reaction norms to
environmental gradients covering the geographic dis-
tribution of a species. There are many other examples
4.3 Growth rings in other organisms
of a MEM approach to analysing fish growth from Although trees and fish have been the most extensive-
otolith time series (Weisberg et al. 2010; Helser et al. ly studied to date, other organisms also create a record
2012; Rountrey et al. 2014; Von Biela et al. 2015; Von of growth (or other life events) on hard parts. The tech-
Biela 2016; Smolinski and Mirny 2017; van der Sleen niques of dendrochronology have been applied to the
et al. 2018; Martino et al. 2019). The MEM approach growth rings of other plant growth forms, including
to analysing individual-level absolute growth variabil- shrubs in the Arctic and the Himalayas (Myers-Smith
ity has emerged in the dendroecology literature as et al. 2015a; Weijers et al. 2017; Gamm et al. 2018;
Dolezal et al. 2019; Le Moullec et al. 2019) and peren- humans and their relatives, their study has contribut-
nial herbs with a taproot (‘herb-chronology’; Dee and ed to understanding of the demography of wildebeest
Palmer 2016; Dolezal et al. 2018). Indeed, Buntgen in the Serengeti (Sinclair and Arcese 1995) and marine
(2019) has argued for doing away with plant growth mammals (Hamilton et al. 2017), as well estimates
form categories (e.g. of Raunkiaer) that separate of longevity for diverse organisms (e.g. flying fox-
trees from shrubs from herbs, making the case for es, whale sharks; DivljanPerry-Jones et al. 2006; Ong
fundamental anatomical and physiological similarities et al. 2020).
between them.
Sclerochronology includes, in addition to the otoliths
of fish, the analysis of growth rings of bivalve shells
4.4 Beyond growth rings and ring widths
(Figure 4.1c,d), corals (Figure 4.1e), fish scales, and Other forms of growth or developmental marks or
turtle scutes (Morrongiello et al. 2012). Some of these scars can inform demography, and metrics other than
organisms have surprisingly long lifespans, including the width of a growth ring provide insight as well.
the ocean quahog (Arctica islandica), with a maximum Myers-Smith et al. (2015b) beautifully illustrate that
reported age of 507 years, making it the world’s old- annual axial (stem) increments of certain tundra shrub
est known, reliably aged animal; the Pacific geoduck species can be measured from bud scars or longitu-
(Panopea generosa; ages of 150 years and more); and dinal sectioning (i.e. using winter mark septa). Cone
freshwater mussels (genus Margaritifera, with ages of scars make it possible to quantify and assign to
100 years and more). specific years the reproductive effort of some pines
Bones, teeth, and horns are other examples of hard (see Figure 4.1h; Morgan and Bunting 1992; Redmond
body parts that can yield demographic information. et al. 2016) The spines of cacti and Euphorbia thorns
Skeletochronology, the estimation of age by counting grow in a clearly organised series, recording physio-
growth marks found in the cross-section of bones, has logical variation over time (English et al. 2007, 2010).
been used across a wide variety of vertebrates. Most Further, growth rates are but one kind of information
(but not all) species of dinosaurs formed annual growth to be gained from hard body part time series. Addi-
lines (Figure 4.1f). Extant vertebrates that are known tional characteristics of growth rings and other struc-
to form annual growth lines include Actinopterygia, tures, such as isotopic composition and other forms of
Amphibia, Lepidosauria, and Crocodylia. It was once microchemistry variation, as well as anatomical and
thought that mammals do not form annual growth other forms of microstructure variation, offer insight
marks in bones, but recent work has shown they exist into demographic processes or the drivers and adap-
in large-bodied and slower-growing species, includ- tations behind demographic outcomes. Hypoplasia
ing cervids and primates (Woodward et al. 2013). (i.e. deficiencies in tooth enamel thickness associat-
A cautionary note is warranted: resorption or sec- ed with undernutrition or disease) represent indelible,
ondary reworking of bone tissue can obscure growth chronological records of acute stress. Stressful events
marks. The most reliable bones for ageing are gen- recorded in xylem (wood) include frost damage, fire,
erally long bones, and the largest of them, with defoliation, flooding, and geomorphic events. False
methods to retrocalculate missing years (Woodward and missing rings too are indicators of stress. While we
et al. 2013). have attempted to be comprehensive in our examples
Teeth have been used to study life history varia- of growth rings and other somatic records of demo-
tion and evolution, particularly of hominins and other graphic processes across the Tree of Life, surely many
primates (Smith and Tompkins 1995; Kelly and Smith more opportunities remain to be discovered.
2003). Perikymata are layers of enamel that build up
incrementally on the surface of a tooth as it grows
(Figure 4.1g); each takes about 6–12 days to form in
4.5 Insights into demography
modern humans, and thus they can be used to esti- In this section, we survey the demographic process-
mate how quickly teeth develop. Visualisation tech- es that can be informed by growth ring time series,
niques that bring a new level of detail (synchrotron highlighting the variety of knowledge gaps filled using
microtomography-based virtual histology; Le Cabec examples from the literature. We begin with individ-
et al. 2015) and the combination of skeletochronolo- ual growth, birth, and death, and then we consider
gy and dental development data (Seselj 2017) promise movement and evolution.
to advance our understanding of human evolution. Growth—Across a great diversity of organisms,
While the primary focus has been on the teeth of growth time series provide long records of growth
variability that are used to evaluate specific drivers, using hypoplasia in teeth. Temple (2014) found an
and even parse multiple drivers of growth. Based on increased risk of subsequent mortality following such
horns (see Figure 4.1i) from more than 8,000 indi- events evidenced by hypoplasia. Ogle et al. (2015)
viduals, Buntgen et al. (2014) showed that elevated and Peltier et al. (2018) used a Bayesian stochastic
March to May temperatures, causing earlier snowmelt antecedent modelling framework to quantify ecolog-
and increased food resources, lead to synchronised, ical memory in tree-ring time series data (i.e. lagged
increased growth of ibex (an alpine ungulate) across climatic effects on growth). Indeed, one of the great
a broad area of the eastern Swiss Alps. Growth marks strengths of long growth time series is the ability to
in the bones of loggerhead sea turtles (Caretta caretta) analyse variation across temporal scales, from intran-
have been used to partition age, year, and size as nual to centennial or more.
drivers of growth variation, demonstrating compen- Improved understanding of the drivers of growth
satory growth in response to stochastic environmen- variation, across a great range of time scales, can then
tal variation (Bjorndal et al. 2003). Ong et al. (2015) lead to improved models of the process of growth.
used otolith time series data to show an ontogen- The literature on climate effects on tree growth is
tic shift in the environmental drivers of growth vari- particularly rich (Fritts 1976) because tree rings have
ability of the mangrove jack (Lutjanus argentimacula- often been used as a climate proxy. This accumulat-
tus), from precipitation and other factors influencing ed knowledge has been the foundation for mecha-
river runoff in the estuarine habitat of juveniles to nistic, climate-sensitive models of tree growth such
atmospheric and ocean-basin conditions influencing as the Vaganov-Shashkin model (Vaganov et al. 2006;
the coastal reef habitat of adults. Analysis of annual Tolwinski-Ward et al. 2011) and more recent whole-
growth rings of the coral Montastraea faveoloata put the tree growth process models (Holtta et al. 2010; Hay-
severe coral bleaching event of 1998 into a long-term at et al. 2017; reviewed in Babst et al. 2018). Carbon
context, revealing that water temperatures and solar and oxygen isotopes in tree rings are used to better
irradiance conditions were as high or higher in the past, understand the process of growth at a physiological
pointing at the impact of human-caused stressors (e.g. level (McCarroll and Loader 2004; Gessler et al. 2014).
nutrient pollution, overfishing) on top of climatic con- Ratios of the stable isotopes δ18 O and δ13 C are used
ditions (Carilli et al. 2009). Several recent analyses of to evaluate water use efficiency, the ratio of carbon
tree-ring time series have focused on the joint effects gained to water lost, which is controlled by leaf stom-
of climate and competition on tree growth, includ- atal behaviour. Fractionation of both occurs when leaf
ing their interactions, and how they vary across forest stomata are open for gas exchange; a central question
canopy classes or species (Martin-Benito et al. 2011; is whether increased atmospheric concentration of car-
Fernandez-de-Una et al. 2016; Wright et al. 2018). bon dioxide leads to increased water use efficiency and
Long growth time series are particularly helpful enhanced tree growth, mitigating climate change, or
in answering certain questions about growth that not (Frank et al. 2015; Van der Sleen et al. 2015). Hence
inherently span time, such as ecological resilience these isotope-based studies of tree-ring time series are
or ecological memory. Based on growth time series contributing to a better understanding of whether tree
data from tree rings, Lloret et al. (2011) defined a growth is controlled by the availability of carbon diox-
set of metrics to quantify the response to a stress- ide versus other limiting factors (is a ‘source-limited’
ful event (e.g. drought), including resistance, recov- vs. ‘sink-limited’ process; Fatichi et al. 2014; Zuidema
ery, resilience, and relative resilience. These metrics et al. 2018; Ulrich et al. 2019), which will be used to
have been used, for example, to detect increased vul- improve the representation of the feedback between
nerability to drought after successive drought events forests and climate in coupled atmosphere-biosphere
(Serra-Maluquer et al. 2018), fire (Van Mantgem et al. models.
2018, 2020), or investigate the effect of forest den- Birth and death—In contrast to growth, birth and
sity reduction treatments on resilience to drought death happen only once for each individual in a pop-
stress (Navarro-Cerrillo et al. 2019). Given the multi- ulation, making them ‘rare’ events, particularly in
ple meanings of the term resilience in the ecological long-lived organisms. Tree-ring, otolith, and other time
literature, let alone in the policy world of climate adap- series data can provide unique insight into these key
tation, it is useful to have metrics by which progress demographic processes. For otoliths, the year the first
towards this goal can be measured. Surprisingly, simi- ring was formed can be considered the year of birth,
lar questions about the consequences of stressful events whereas in tree rings, the year the first ring was formed
have been asked about human or hominin populations (i.e. the pith date) represents the year of birth (i.e.
germination) only if samples are collected near the soil not, borrowing strength from repeated measures of
surface (cores collected at other stem heights repre- growth to gain insight into the process of decline (and
sent the year the tree reached that height). Empirically its causes) that can precede death (Ogle et al. 2000;
derived relationships can be used to estimate true year Macalady and Bugmann 2014; Cailleret et al. 2016a,
of birth from increment cores not sampled at ground 2016b). Finally, age distribution data, along with other
level, although most attempts to correlate tree height lines of evidence, can be used to evaluate mortality and
with age find highly variable relationships with much its drivers. Indeed, Zhao et al. (2018) make the argu-
less than annual precision (Wong and Lertzman 2001). ment that, compared to mark-recapture studies, esti-
Once an estimate of the year of birth is made, it mates of mortality rates derived from age distributions
becomes possible to evaluate the conditions under (from otoliths and skeletochronology) are an accurate
which recruitment occurred, or detect strong cohort and useful tool to rapidly assess mortality rates.
structure that provides evidence of episodic recruit- Movement—Otolith microchemistry is a well-
ment. Swetnam and Betancourt (1998) suggested, from established method to determine provenance and
pith dates at 143 sites across the US Southwest, that movement of fishes (Thresher 1999; Kennedy et al.
drought in the late 1500s led to regional-scale mor- 2000). Strontium isotope ratios (87 Sr/86 Sr) and other
tality followed by a pulse of recruitment during a chemical fingerprints are used to infer anadromy,
relatively cool-wet period in the early 1600s. This repre- characterise diadromous migrations, estimate the
sents demographic inference on temporal (centennial) proportion of fish populations made up by wild
and spatial (104 –106 km2 ) scales well beyond those at versus hatchery-raised individuals, and identify habi-
which most demographers operate. Similarly, careful tat use for conservation of wild populations (Hegg
cross-dating of Pacific geoduck growth rings revealed et al. 2015; Willmes et al. 2018; Brennan et al. 2019).
episodic recruitment—two strong recruitment pulses Further, otoliths can be marked chemically en masse,
separated by 60 years (Black et al. 2008). Pith dates have to track movements and abundance (Warren-Myers
further been used to examine the interactive effects of et al. 2018).
fire, climate, and human land uses (e.g. grazing, fire Evolution—Biogenic time series data offer insight
suppression) on tree recruitment in certain forest types into life history variation and evolution, evidence for
across the interior western United States, leading to selection on vital rates and functional traits, as well as
the conclusion that it was lack of fire, more so than a means to quantify the genetic versus plastic basis of
favourable climatic conditions, that drove twentieth- vital rate and trait variability (topics in Chapter 20).
century patterns of recruitment (Brown and Wu 2005; Growth lines in teeth and bones have been particularly
Munier et al. 2014; O’Connor et al. 2014; Taylor et al. important in studying life history variation and evolu-
2016). Most recently, based on pith dates of nearly 3,000 tion of extinct animals such as dinosaurs and hominins
trees at 90 sites, Davis et al. (2019) inferred that climate (Smith and Tompkins 1995; Kelly and Smith 2003). For
change is pushing conditions for post-fire tree recruit- example, perikymata (in teeth) have been used to eval-
ment below critical climate thresholds. Further, cone uate the role of heterochrony and neoteny in human
scars provide unique information on fertility, mak- evolution (Guatelli-Steinberg et al. 2005). Insights into
ing it possible to study masting (in Pinus albicaulis; basic dinosaur life history parameters can be derived
Crone et al. 2011) and how mast years affect subse- from age estimates from the growth lines of bones,
quent recruitment and age structure (in Pinus pumila; including growth rates (from age vs. mass curves), time
Kajimoto et al. 1998). to maturity, and longevity. These have been used to
Similar to birth dates, the date of the last ring formed address some of the most eye-catching questions about
can offer insight into the process of mortality. This par- dinosaurs, such as the evolution of gigantism (Erickson
ticular path towards inference on mortality is limited et al. 2004; Erickson 2005).
to tree rings, since the collection of an increment core Tree-ring time series have been used to assess the
does not harm a tree, whereas an otolith is obtained consequences of individual-level growth heterogene-
through (mortally) destructive sampling. Villalba and ity for population-level growth rate, with the conclu-
Veblen (1998), for example, attributed episodic mortal- sion that fast-growing juveniles contribute twice as
ity of the South American conifer Austrocedrus chilensis much to population growth as slow-growing juve-
to synoptic variation in the El Niño-Southern Oscilla- niles (Zuidema et al. 2009). Tree rings further pro-
tion ocean-atmosphere pattern (ENSO). An alternative vided evidence for the juvenile selection effect, the
approach to investigating mortality is to contrast the preferential survival of fast-growing juveniles to the
growth time series of individuals that died versus did forest canopy (Rozendaal et al. 2010; Zuidema et al
2011). Microstructural variation in tree-ring time series the equatorward edge of its distribution. Other recent
(i.e. variability in xylem cell anatomy; Fonti et al. 2010) examples have used genotype-phenotype association
provides an archive of both adaptive strategy and studies of tree growth ring time series data (Heer
the plastic response of an individual to environmental et al 2018; Housset et al. 2018). Housset et al. (2018)
variation over time. The xylem is the water-transport associated growth metrics on three timescales derived
system of a plant; hence, anatomical characteristics from tree rings collected in a forestry provenance trial
such as the number, length, diameter, wall thickness with single nucleotide polymorphisms at loci that had
and pit characteristics of conduits are under selective demonstrated links to local climate adaptation. Their
pressure with respect to both drought, cold, and other data showed opposing latitudinal clines of cold and
functional roles (i.e. support, storage). The world-wide drought sensitivity, suggesting that assisted migration
phenomenon of drought-induced tree mortality, driv- of southern, drought-tolerant provenances northward
en by global warming through increased evaporative might not be successful because of their sensitivity
demand (Allen et al. 2010), has made hydraulic fail- to cold stress. We expect that genome-wide associa-
ure and resistance to cavitation a focus of a great deal tion studies (i.e. pedigree-free approaches) will soon
of research (Rowland et al. 2015; Anderegg et al. 2016; be applied to tree-ring and other biogenic time series
Greenwood et al. 2017), with quantitative wood anato- data to better quantify the G, E, and G*E compo-
my from tree rings being an emerging line of evidence nents of growth responses to climate, and further, to
used to improve understanding of the process of mor- identify the many loci responsible for variation in the
tality (Powell et al. 2017; von Aryx et al. 2017; Zuidema (polygenic) traits necessary to tolerate environmental
et al. 2018). For example, Roskilly et al. (2019) used extremes anticipated in the twenty-first century.
variation in pit membrane structure to understand Disturbance—Forest ecologists use growth ring
how drought-tolerance, growth rate, and longevity are time series to better understand the influence of
related to one another in Pinus ponderosa: individu- disturbances on tree demography. These range from
als of greatest longevity showed selection towards pit single-tree gap dynamics (Thompson et al. 2007) to
membrane structure associated with drought-tolerance landscape-scale disturbances caused by insect out-
(and slower rates of growth) compared to greater vari- breaks, fire, wind or ice storms, or human activities
ation in pit membrane structure (and growth rates) (Pederson et al. 2008; Axelson et al. 2009; Lynch 2012).
amongst younger trees. Ultimately, investigation of One line of evidence is the detection of a ‘release
xylem anatomical variation unpacks the significance event’, that is, an individual in a population is released
of so-called functional traits like wood density, which from competitive pressure when disturbance kills its
have been predictive of some aspects of tree demogra- neighbours, leading to a change in the (detrended)
phy and life history variation (Visser et al. 2016), but width of growth rings, such as a 25% increase in the
less so of others (mortality; O’Brien et al. 2017; Yang 10-year running mean (Nowacki and Abrams 1997).
et al. 2018), into the underlying components that are Other release criteria and methods (Black and Abrams
truly functional (rather than emergent). 2003; Druckenbrod 2005) address variation among
Further, biogenic time series data offer a golden trees in release response because of forest canopy
opportunity to quantify the response of a single geno- position at the time of disturbance (i.e. individuals in
type (e.g. an individual tree) to varying environmental competitively suppressed vs. dominant positions) or
conditions, and hence the genetic (G), plastic (E), and evolutionary strategy with respect to shade tolerance.
G*E (heritable plastic) basis for adaptation to climate Another approach is to compare growth increment
variation (Evans et al. 2018). In the context of a dial- variability of individuals that are versus are not affect-
lel mating design, Marchal et al. (2019) applied mixed ed by a particular type of disturbance. For example,
effect models of the kind used in the cattle breed- Swetnam and Lynch (1993) reconstructed regional-
ing literature to quantify growth reaction norms in scale insect outbreak cycles by comparing 300-year
response to soil water availability in hybrid larch com- growth time series of host versus nonhost tree species,
pared to its parents. Morrongiello and Thresher (2015) using a spectral analysis to recover pseudo-periodic
used mixed effect modelling of otolith time series to insect population dynamics.
estimate individual-level plasticity, with the conclu- Fire scars have been sampled and developed into
sion that it is currently sufficient across much of the extensive data networks, yielding multicentury, spa-
tiger flathead’s range such that warming waters do tially explicit reconstruction of forest wildfires. Anal-
not negatively impact fitness, while suggesting concern yses across scales of these networks of fire-scarred
about the fitness consequences of future warming at trees have informed our understanding of fire regimes
and their controls, revealing the interplay between abundance (Black et al. 2009; Black et al. 2011, 2014;
bottom-up (topography, vegetation) and top-down Garcia-Reyes et al. 2013). Across the adjacent terres-
(climate) controls of fire (Falk et al. 2007; Falk et al. trial ecosystems of western North America, the NPH
2011). Fire-scar data have further been combined with is further associated with storm track position, and in
birth (pith) dates as described above (Brown and Wu turn winter precipitation, as well as water year riv-
2005; Margolis et al. 2011; O’Connor et al. 2014), to er discharge (Black et al. 2009, 2015). Years of strong
better understand how the fundamental Earth sys- NPH are associated with anomalously high levels of
tem process of fire influences the demography of upwelling, but anomalously low levels of precipitation,
forest ecosystems, including the creation of spatially such that productivity is antiphase across the marine
heterogeneous forest age structure from gap-phase and versus terrestrial systems (Black et al. 2014). Centuries-
cohort responses (Heyerdahl et al. 2014, 2019). Hence long tree-ring time series indicate that the winter NPH
multiple lines of evidence derived from tree rings and associated marine and terrestrial indicators are
(e.g. changes in growth increment time series, records becoming increasingly variable and extreme, and that
of acute stress, and age structure) all have informed for- the spatial footprint of the NPH is increasing, entrain-
est disturbance ecology; in some cases, these approaching synchrony of greater spatial extent. This is expect-
es could be applied to other species and ecosystems ed to reduce bet-hedging effects and resilience across
to understand the role of disturbance in population, species, increasing the risk of extirpation; (Black et al.
range, and ecosystem dynamics. 2017); for example, the extinction of checkerspot butter-
fly populations in the San Francisco region (McLaugh-
lin et al. 2002). Given the importance of the NPH to
4.6 Ecosystem and macrosystem
the key drivers of productivity in marine (upwelling)
demography and terrestrial (precipitation) ecosystems of the region,
Demography has often been conducted at a small spa- its increasing volatility is likely to influence a wide
tial scale (i.e. 1 m2 plots), with a single-species focus, range of demographic processes and species, from the
and relatively short (5–30 year) time spans, a natu- continental slope in the sea to high-elevation forests
ral consequence of the laborious process of building on land.
data sets one year at a time. Biogenic annually resolved A similar story emerges in Western Australia:
time series data, especially when developed into a spa- coupled ocean-atmosphere patterns synchronise dem-
tial network, offer insights at larger spatial and deeper ography across aquatic and terrestrial ecosystems at
temporal scales, and the opportunity to study sever- large spatial scales. Here also, ENSO is an important
al species in an ecosystem, even across ecosystems, driver of conditions in the marine environment, via
to understand the drivers of demography, particular- sea surface temperatures and salinity in the tropical
ly how several organisms are affected by the changing North Coast region and the strength of the Leeuwin
climate system. In this final section, we highlight two current along the subtropical and temperate West
examples of this from the Northeastern Pacific Ocean and South Coasts, as well as in the terrestrial envi-
and Australia. ronment, via precipitation. Ong et al. (2016) showed
Across the ecosystems of the northeast Pacific Ocean increased growth of corals (Porites spp.), two species of
and western North America, the winter (January– marine fishes, and a tree in response to strong La Niña
March) North Pacific High (NPH, defined as mean years, when North Coast sea surface temperatures
sea level pressure at 25◦ N–35◦ N by 125◦ W–145◦ W) and northwest Australian precipitation are higher
exerts strong and synchronous control over produc- than average, and sea surface salinity is lower than
tivity and hence demography of a diverse assemblage average. Ong et al. (2018) further showed synchronous
of species across marine, freshwater, and terrestri- ENSO-driven interannual growth variation among
al ecosystems (Figure 4.3). In the California Current five species of tropical and temperate fishes spanning
Ecosystem, the strength of winter coastal upwelling, 23◦ of latitude. El Niño conditions, associated with
which charges the photic zone with cold, nutrient-rich anomalously warm sea surface temperatures, have
water, is associated with variability in the winter NPH, caused widespread mortality for both fish and corals
which is in turn teleconnected to ENSO (Schwing et al. (Feng et al. 2013); indeed, coral reefs are widely con-
2002). This drives bottom-up trophic effects, evidenced sidered to be the ecosystem most vulnerable to climate
in rockfish otolith growth increments, seabird repro- change–driven demographic collapse.
ductive phenology and success, copepod community, These multispecies, ecosystem- or macrosystem-
and Sacramento and Russian River chinook salmon scale studies illustrate how demographers in the
(a) 3
2 Aleutian Low
Anomaly
1
0 Pacific High
–1
–2 North Pacific High Winter
–3
(b) 3
2
Coastline
1
Anomaly
0
–1 Upwelling
Se
–2
afl
Upwelling
oo
–3
r
Precipitation
–4
(c) 3
Blue oak
2
1
Anomaly
0
–1
–2
Auklet lay date
–3 Splitnose rockfish
–4
1900 1920 1940 1960 1980 2000 2020
Year
Figure 4.3 Cascading effects of (a) the strength of the winter North Pacific High on (b) precipitation in western North America and winter
coastal upwelling in the northeast Pacific Ocean, and on demographic metrics of organisms of these terrestrial and aquatic ecosystems, including
(c) growth increments of blue oak (Quercus douglasii) and the split-nose rockfish (Sebastes diploproa), as well as auklet (Ptychoramphus aleuticus)
lay dates. The precipitation and blue oak time series are inverted to make synchrony easier to identify.
twenty-first century—the age of global change—could same time, demography is sometimes criticised for
profit from a more intimate and mechanistic being phenomenological. Here we set forth a research
understanding of the atmospheric and ocean current agenda for demographers to act upon: it is time to
systems that are changing and profoundly affecting incorporate demographic insights from biogenic time
the organisms they study. Indeed, biogenic time series series data into population models (Figure 4.4), capture
are uniquely positioned to offer insight into both an improved understanding of what drives population
biological and physical systems, since historically dynamics, and better anticipate the impacts of global
their use has been for reconstructing physical system change.
variation, and, we argue, their growing future lays in To clarify, otolith-derived age data have been used
also understanding biological system (demographic) for many years to parameterise stock assessment (age-
variation. length) models for fish (e.g. a von Bertalanffy model)
similar to the age-size models fit to data from dinosaur
bones. However, most of these stock assessment mod-
4.7 A research agenda
els assume constant values for growth parameters,
Across this remarkable range of insights into growth, meaning they only use otoliths for age information
birth, death, movement, evolution, disturbances, (Lee and Punt 2018); they do not use the growth incre-
trophic cascades, and macrosystems gained from ment time series data contained in otoliths to quantify
single-sample time series data, surprisingly little has drivers of variation in growth or incorporate those
been incorporated into population models. At the drivers into population models. The same is true for
Collect Sample Demographic

sample processing analysis
Cross-dating Individual level Population level
Growth
increment Mortality risk and Birth rate and
Pith + covariate drivers (GLMM) drivers (GLMM)
Birth time series
Growth rate and

effects of drivers, Population
either using linear projection matrix,
mixed effects integral projection
modelling (which model, integrated
Earlywood Growth Year may partition into population model,
Latewood ring G, E and G*E or individual-based
effects) or through model
Stress event Fire scar dynamic linear
(state space)
modelling (i.e.
data fusion)
Inverse modelling
Death or of birth and death
age rates from age or
Bark
size structure
Figure 4.4 An overview of how biochronologies can inform population models. We take the example of tree rings, starting with the collection of
a sample, an increment core, using an increment borer. A year of formation is then assigned to each ring using the process of cross-dating.
Additional demographic information derived from the cross-dated growth increment time series may include pith date, stressful events, and age at
the time of sampling or year of death. Individual-level vital rates (and drivers of their variation) that may then be inferred are growth rate (using
linear modelling or dynamic linear modelling) and mortality risk. Birth rate, a population-level parameter, may also be inferred. Forward estimates
of vital rates and their drivers could then be combined in a number of different kinds of demographic models. Alternatively, inverse estimation of
population model parameters may be derived from time series of age or size structure data. Some of these approaches have been used in the
literature (e.g. linear mixed effects modelling of growth increments) and others have not (e.g. Poisson regression of birth dates to estimate birth
rates).
tree rings, which is even more surprising since the model (IPM) (see Chapter 10), or other demographic
increment borer, the tool used to sample a tree for model (Chapters 12 and 14). This would apply to any
its growth rings, was invented by foresters (in the of the above cases of growth modelling—from ibex
late-nineteenth century; Somerville 1891) specifically to loggerhead sea turtles to fish to coral to trees. For
to quantify tree growth. In more than 100 years, example, Mina et al. (2016) used climate sensitivities
tree-ring data have not been used to parameterise derived from a tree-ring-based model of growth (VS-
forestry growth and yield models; instead, like fish- Lite) in an individual-based forest succession model
eries biologists, foresters have primarily used growth (ForClim), which improved representation of the effect
rings to age individuals. of drought. Schultz and Miller (in prep.) used tree-ring
How might estimates from biogenic time series data time series from both live and dead individuals of Pinus
of vital rates and the drivers of their variation be albicaulis to parameterise multiple regression (MEM)
incorporated into population models (Figure 4.4)? One models that form the survival-growth subkernel of an
possibility is to use estimates of baseline growth rate, IPM. That is, ring-width time series from live individu-
along with any effects on growth (of individual size, als were used to estimate growth rates and outer-bark
climate, social status or population density), in a popu- dates from dead individuals were used to estimate size-
lation projection matrix (Chapter 9), integral projection dependent mortality rates. A MEM of ring-width time
series data can similarly be used to parameterise the 2012; Evans et al. 2016; Gonzales et al. 2016; White
growth component of a stand-level forestry growth et al. 2016; Vanderwel et al. 2017; Zipkin et al. 2019).
and yield model (Giebink et al. (in prep.)), which is These approaches should be applied to single-sample
fundamentally similar to an IPM. When estimates of time series data to incorporate them into population
parameters are passed from one model or analysis to models.
another in this sort of sequential manner, it is critical By incorporating biogenic time series data (or
to quantify parameter uncertainty and include it at parameters derived from them), demographic models
the stage of population projection, so that projections could then reflect greater understanding of the caus-
reflect the uncertainty associated with underlying data es of variation in vital rates, and better anticipate the
(‘uncertainty propagation’; Dietze 2017). impacts of global change. Drawing from some of the
Another level of complexity is to combine different insights described in the chapter, demographic models
sources of information on growth (i.e. data fusion) to could then explicitly represent different environmental
estimate the growth parameters in a population model. drivers of juvenile versus adult vital rates, how cli-
One example of this used a hierarchical Bayesian mate stress affects individuals differently across size or
framework to model tree bole diameter increments social status, or the lingering (lagged) effects of climate
(from diameter tape measurements), radial increments stress on performance. Models of migratory fish could
(from increment cores), and their common influences at explicitly represent habitat use, influences in different
once—with two multiple regression models connected basins, or what proportion of the population are wild
to one another via the simple fact that the diame- versus hatchery raised. Biogenic time series data will
ter increment must equal twice the radial increment likely play a key role in estimating and representing in
(Evans et al. 2017). Another example is data fusion models how local adaptation and plasticity constrain
via state space modelling. Clark et al. (2007) treated responses to shifting climate. That is, demographic
tree bole diameter at breast height (DBH) as a latent forecasts should reflect intraspecific variation in vital
state variable; its time evolution is governed by a sim- rates (see Chapter 20); they should follow reactions
ple difference equation, and two data models link norms estimated from data, and biogenic time series
the available observations of growth (DBH remeasure- data are an excellent source of data to do so. The demo-
ments plus tree rings) to the (latent) true state (diameter graphic consequences of changing disturbance regimes
at breast height) of the tree. This model included ran- are an important aspect of global change that also can
dom effects of individual tree and year, but no fixed be modelled with the help of biochronologies. Going
effects; the next step is to include fixed effects that are one step further would be to assimilate time series
known to influence tree growth (e.g. tree size, climate; data from multiple species that are linked via troph-
Heilman et al. (in prep.)). ic relationships, or through common physical drivers,
Information contained in biogenic time series data to trace biotically driven and abiotically driven demo-
can also be incorporated into population models graphic cascades through complex ecological systems.
through inverse modelling. Age structure data derived And it may be at a planetary scale that growth ring
from vertebrate bones (including otoliths) can be used time series offer their most consequential demograph-
to constrain population model parameters that are dif- ic forecasts (i.e. of the impact of climate change on tree
ficult to estimate, such as mortality rate, which oth- growth, the abundance and distribution of forests, and
erwise require mark-recapture data over an extend- hence the expected behaviour of the planetary-scale
ed period (Zhao et al. 2018). Gonzalez and Martorell feedback between forest ecosystems and climate, based
(2013) demonstrated that time series of population on data from tree rings).
structure can be used to accurately estimate individual-
level vital rates for an IPM. Nonidentifiability of pop-
ulation model parameters (i.e. the fact that more than
4.8 Conclusions
one combination of birth and death rates can give rise Time series data generated from natural archives,
to the same age structure) can be greatly reduced by which can be obtained in a single sampling effort,
constraining model structure or parameters with pri- are an underappreciated source of demographic infor-
or knowledge (Gonzalez and Martorell 2013). Indeed, mation. Historically, their use has been for recon-
an emerging trend in the study of demography is the structing physical system variation; we argue that
use of multiple sources of data (e.g. through integrat- their growing future lays in understanding biologi-
ed population modelling; see Chapter 14), data fusion cal system (demographic) variation, and ultimately,
or data assimilation, or inverse modelling (Hartig et al. relationships between the two. They typically extend
the temporal scale of demographic studies profound- Black B. A., Schroeder I. D., Sydeman W. J., Bograd S. J.,
ly, and when developed into spatial networks, provide Wells B. K., and Schwing F. B. (2011). Winter and summer
the opportunity to think more broadly and mecha- upwelling modes and their biological importance in the
nistically about the influence of ecosystem and Earth California Current Ecosystem. Global Change Biology 17,
2536–2545.
system processes across a great range of temporal and
Black B. A., Sydeman W. J., Frank D. C., et al. (2014). Six
spatial scales (e.g. disturbance processes, atmospher-
centuries of variability and extremes in a coupled marine–
ic and ocean circulation patterns) on demography.
terrestrial ecosystem. Science 345, 1498–1502.
Biochronologies further complement other sources of Black B. A., Dunham J. B., Blundon B. W., Brim-Box J., and
demographic information in critical and unique ways, Tepley A. J. (2015) Long-term growth-increment chronolo-
offering traction on vital rates that are otherwise dif- gies reveal diverse influences of climate forcing on fresh-
ficult to estimate—stages of the life history for which water and forest biota in the Pacific Northwest. Global
information is poor because the events are rare (mor- Change Biology 21, 594–604.
tality), organisms are mobile or extinct, or for other Black B. A., Griffin D., Van Der Sleen P., et al. (2016). The
reasons. The challenge now is to combine the infor- value of crossdating to retain high-frequency variabili-
mation contained in biochronologies on vital rates ty, climate signals, and extreme events in environmental
proxies. Global Change Biology 22, 2582–2595.
and the causes of vital rate variation with traditional
Black B. A., Van Der Sleen P., Di Lorenzo E., et al. (2017). Ris-
demographic data streams to better address the conse-
ing synchrony controls western North American ecosys-
quences of global change. tems. Global Change Biology 24(6), 2305–2314.
Black B. A., Andersson C., Butler P. G., et al. (2019).
The revolution of crossdating in marine palaeoecolo-
References gy and palaeoclimatology. Biology Letters 15, 20180665.
Allen C. D., Macalady A. K., Chenchounic H., et al. (2010). http://dx.doi.org/10.1098/rsbl.2018.0665.
A global overview of drought and heat-induced tree mor- Breinen R. J. W. and Zuidema P. A. (2006). Lifetime
tality reveals emerging climate change risks for forests. growth patterns and ages of Bolivian rain forest trees
Forest Ecology and Management 259, 660–684. obtained by tree ring analysis. Journal of Ecology 94,
Anderegg W. R. L., Klein T., Bartlett M., et al. (2016). 481–493.
Meta-analysis reveals that hydraulic traits explain Brennan S. R., Schindler D. E.,Cline T. J. Walsworth T. E.,
cross-species patterns of drought-induced tree mortality Buck G., and Fernandez D. P. (2019). Shifting habitat
across the globe. Proceedings of the National Academy mosaics and fish production across river basins. Science
of Sciences of the United States of America 113(49), 364, 783–786.
5024–5029. Brown P. M. and Wu R. 2005. Climate and disturbance
Axelson J. N., Alfaro, R. I., and Hawkes B. C. (2009). Influ- forcing of episodic tree recruitment in a southwestern
ence of fire and mountain pine beetle on the dynamics of Ponderosa pine landscape. Ecology 86, 3030–3038.
lodgepole pine stands in British Columbia. Forest Ecology Briffa K. R., Jones P. D., Bartholin T. S., et al. 1992.
and Management 257, 1874–1882. Fennoscandian summers from AD 500: temperature
Babst F., Bodesheim P., Charney N., et al. (2018). When tree changes on short and long timescales Climate Dynamics 7,
rings go global: challenges and opportunities for retro- 111–7119.
and prospective insight. Quarternary Science Reviews Buechling A., Matin P. H., and Canham C. D. (2017). Climate
197, 1–20. and competition effects on tree growth in Rocky Mountain
Bjorndal K. A., Bolten A. B., Dellinger T., Delgado C., and forests. Journal of Ecology 105, 1636–1647.
Rost H. (2003). Compensatory growth in oceanic logger- Bunn A. G. 2008. A dendrochronology program library in R
head sea turtles: response to a stochastic environment. (dplR). Dendrochronologia 26, 115–124.
Ecology 84, 1237–1249. Buntgen U. (2019). Re-thinking the boundaries of den-
Black B. A. and Abrams M. D. (2003). Use of boundary-line drochronology. Dendrochronologia 53, 1–4.
growth patterns as a basis for dendroecological release Buntgen U., Jenny H., Liebhold A., et al. (2014). Euro-
criteria. Ecological Applications 13, 1733–1749. pean springtime temperature synchronizes ibex horn
Black B. A., Gillespie D., Maclellan S. E., and Hand C. M. growth across the eastern Swiss Alps. Ecology Letters 17,
(2008). Establishing highly accurate production-age data 303–313.
using the tree-ring technique of crossdating: a case study Cailleret C., Jansen S., Robert E. M. S., et al. (2016a). A syn-
for Pacific geoduck (Panopea abrupta). Canadian Journal of thesis of radial growth patterns preceding tree mortality.
Fisheries and Aquatic Sciences 65, 2572–2578. Global Change Biology 23(4), 1675–1690.
Black B. A., Abrams M. D., Rentch J. S., and Gould J. S. Cailleret M., Bigler, C., Bugmann H., et al. (2016b). Towards
(2009). Properties of boundary-line release criteria in a common methodology for developing tree logistic mor-
North American tree species. Annals of Forest Science 66, tality models based on ring-width data. Ecological Appli-
205. cations 26, 1827–1841.
Campana S. E. (1990). How reliable are growth back- change at Mediterranean read-edge forests. Global
calculations based on otoliths? Canadian Journal of Fish- Change Biology 25, 1296–1314.
eries and Aquatic Science 47, 2219–2227. Douglass A. E. (1914). A method of estimating rainfall by
Campana S. E. and Neilson J. D. (1985). Microstructure of the growth of trees. Carnegie Institute of Washington
fish otoliths. Canadian Journal of Fisheries and Aquatic Publication 192, 101–121.
Science 42, 1014–1032. Douglass A. E. (1920). Evidence of climatic effects in the
Canham C. D., Murphy L., Riemann R., McCullough R., annual rings of trees. Ecology 1, 24–32.
and Burrill E. (2018). Local differentiation in tree growth Douglass A. E. (1941). Crossdating in dendrochronology.
responses to climate. Ecosphere 9, e02368. Journal of Forestry 39, 825–831.
Carilli J. E., Norris R. D., Black B., Walsh S. M., and Druckenbrod D. L. (2005). Dendroecological reconstructions
McField M. (2009). Century-scale records of coral growth of forest disturbance history using time-series analysis
rates indicate that local stressors reduce coral thermal with intervention detection. Canadian Journal of Forest
tolerance threshold. Global Change Biology 16, 1247– Research 35, 868–876.
1257. English, N. B., Dettman D. L., Sandquist D. R., and Williams
Clark J. S., Wolosin M., Dietze M. et al. (2007). Tree growth D. G. (2007). Past climate changes and ecophysiological
inference and prediction from diameter censuses and ring responses recorded in the isotope ratios of saguaro cactus
widths. Ecological Applications 17, 1942–1953. spines. Oecologia 154(2), 247–258.
Cook E., Briffa K., Shiyatov S., and Mazepa V. (1990). English, N. B., Dettman D. L., Sandquist D. R., and
Tree-ring standardization and growth-trend estimation. Williams D. G. (2010). Daily to decadal patterns of
In: E. Cook and L. Kairiukstis (eds.), Methods of Den- precipitation, humidity, and photosynthetic physiolo-
drochronology: Applications in the Environmental Sci- gy recorded in the spines of the columnar cactus,
ences, 104–123. Kluwer. Carnegiea gigantea. Journal of Geophysical Research: Bio-
Cook E. R., Briffa K. R., Meko D. M., Graybill D. A., and geosciences 115. doi:10.1029/2009JG001008. hdl:20.500.
Funkhouser, G. (1995). The ‘segment length curse’ in long 11919/768.
tree-ring chronology development for paleoclimatic stud- Erickson G. M. (2005). Assessing dinosaur growth patterns: a
ies. The Holocene 5, 229–237. microscopic revolution. Trends in Ecology and Evolution
Crone, E. E., McIntire E. J. B., and Brodie J. (2011). What 20, 677–684.
defines mast seeding? Spatiotemporal patterns of cone Erickson G. M. Makovicky P. J., Currie P. J., Norell M. A.,
production by whitebark pine. Journal of Ecology 99, Yerby S. A., and Brochu C. A. (2004). Gigantism and com-
438–444. parative life-history patterns of tyrannosaurid dinosaurs.
Davis K. T., Dobrowski S. Z., Higuera P. E., et al. (2019). Nature 412, 405–412.
Wildfires and climate change push low-elevation forests Evans M. E. K., Merow C., Record S., McMahon S. M., and
across a critical climate threshold for tree regeneration. Enquist B. J. (2016). Towards process-based range model-
Proceedings of the National Academy of Sciences of the ing of many species. Trends in Ecology and Evolution 31,
United States of America 116(13), 6193–6198. 860–871.
Dee J. R. and Palmer M. W. (2016). Application of herbo- Evans M. E. K., Falk D. A., Arizpe A., Swetnam T., Bab-
chronology: annual fertilization and climate reveal annual st F., and Holsinger K. E. (2017). Fusing tree-ring and
ring signatures within the roots of U. S. tallgrass prairie forest inventory data to infer influences on tree growth.
plants. Botany 94, 1–12. Ecosphere 8, e01889.
DeRose R. J., Shaw J. D., and Long J. N. (2017). Building the Evans M. E. K., Gugger P. F., Lynch A. M., et al. (2018).
forest inventory and analysis tree-ring data set. Journal of Dendroecology meets genomics in the common garden:
Forestry 115, 283–291. new insights into climate adaptation. New Phytologist
Dietze M. C. (2017). Ecological Forecasting. Princeton Uni- 218, 401–403.
versity Press. Falk D. A., Miller C., McKenzie D., and Black A. E.(2007).
Divlan A., Perry-Jones K., and Wardle G. A. (2006). Age Cross-scale analysis of fire regimes. Ecosystems 10,
determination in the grey-headed flying fox. Journal of 809–823.
Wildlife Management 70, 607–611 Falk D. A., Heyerdahl E. K., Brown P. M., et al. (2011). Multi-
Dolezal, J., Dvorsky M., Borner A., Wild J., and scale controls of historical forest-fire regimes: new insights
Schweingruber F. H. (2018). Anatomy, Age, and Ecol- from fire-scar networks. Frontiers in Ecology & Evolution
ogy of High Mountain Plants in Ladakh, the Western 9, 446–454.
Himalaya. Springer. Fatichi S., Leuzinger S., and Körner C. (2014). Moving
Dolezal, J., Kopecky M., Dvorsky M., et al. (2019). Sink lim- beyond photosynthesis: from carbon source to sink-driven
itation of plant growth determines tree line in the arid vegetation modeling. New Phytologist 201(13), 1086–
Himalayas. Functional Ecology 33, 553–565. 1095.
Dorado-Linan I., Piovesan G., Martínez-Sancho E., et al. Feng M., McPhaden M. J., Xie S.-P., and Hafner J. (2013). La
(2018). Geographic adaptation prevails over species- Nina forces unprecedented Leeuwin Current warming in
specific determinism in trees’ vulnerability to climate 2011. Scientific Reports 3, 1277.
Fernandez-de-una L., McDowell N. G., Canellas I., and Gea- Greenwood S., Ruiz-Benito P., Martínez-Vilalta J., et al.
Izquierdo G. (2016). Disentangling the effect of compe- (2017). Tree mortality across biomes is promoted by
tition, CO2 , and climate on intrinsic water-use efficiency drought intensity, lower wood density and higher specific
and tree growth. Journal of Ecology 104, 678–690. leaf area. Ecology Letters 20, 539–553.
Fonti P., von Arx G., García-González I., et al. (2010). Study- Guatelli-Steinberg D., Reid D. J., Bishop T. A., and
ing global change through investigation of the plastic Larsen C. S. (2005).Anterior tooth growth periods in
responses of xylem anatomy in tree rings. New Phytolo- Neanderthals were comparable to those of modern
gist 185, 42–53. humans. Proceedings of the National Academy of Sci-
Foster J. R., Finley A. O., D’Amato A. W., Bradford J. B., and ences of the United States of America 102, 14197
Banerjee S. (2016). Predicting tree biomass growth in the e14202.
temperate-boreal ecotone: is tree size, age, competition, or Guiterman C. H., Margolis E. Q., Allen C. D., Falk D. A.,
climate response most important? Global Change Biology and Swetnam T. W. (2018). Long-term persistence and
22(6), 2138–2151. fire resilience of oak shrubfields in dry conifer forests of
Frank D. C., Poulter B., Saurer M., et al. (2015). Water- northern New Mexico. Ecosystems 21, 943–959.
use efficiency and transpiration across European forests Hamilton V., Evans K., and Hindell M. A. (2017). From
during the Anthropocene. Nature Climate Change 5(23), the forests to teeth: visual crossdating to refine age esti-
579–583. mates in marine mammals. Marine Mammal Science 33,
Freckleton R. P. (2002). On the misuse of residuals in ecology: 880–888.
regression of residuals vs. multiple regression. Journal of Hartig F., Dyke J., Hickler T., et al. (2012). Connecting
Animal Ecology 71, 542–545. dynamic vegetation models to data—an inverse perspec-
Fritts H. C. (1976). Tree Rings and Climate. Academic Press. tive. Journal of Biogeography 39, 2240–2252.
Fritts H. C. and Swetnam T. W. (1989). Dendroecology: a Hayat A., Hacket-Pain, A. J., Pretzsch, H., Rademacher,
tool for evaluating variations in past and present for- T. T., and Friend, A. D. (2017). Modeling tree growth
est environments. Advances in Ecological Research 19, taking into account carbon source and sink limita-
111–188. tions. Frontiers in Plant Science 8, http://dx.doi.org/
Gamm C. M., Sullivan P. F., Buchwal A., et al. (2018). Declin- 10.3389/fpls.2017.00182.
ing growth of deciduous shrubs in the warming climate Heer K., Behringer D., Piermattei A., et al. (2018). Link-
of continental western Greenland. Journal of Ecology 106, ing dendroecology and association genetics in natural
640–654. populations: stress responses archived in tree rings asso-
Garcia-Berthou E. (2001). On the misuse of residuals in ciate with SNP genotypes in silver fir (Abies alba Mil.).
ecology: testing regression residuals vs. the analysis of Molecular Ecology 27, 1428–1438.
covariance. Journal of Ecology 70, 708–711. Hegg J. C., Giarrizzo T., and Kennedy B. P. (2015). Diverse
Garcia-Reyes M., Sydeman W. J., Thompson S. A., et al. early life-history strategies in migratory Amazonian cat-
(2013). Integrated assessment of wind effects on Central fish: implications for conservation and management.
California’s pelagic ecosystem. Ecosystems 16(5), 722– PLOS One 10(7), e0129697.
735. Heilman, K. A., M. C. Dietze., A. Arizpe., J. Aragon., A.
Gessler A., Ferrio J. P., Hommel R., Treydte K., Werner R. A., Grey., J. D. Shaw., A. O. Finley., S. Klesse., R. J. DeRose.,
and Monson R. K.(2014). Stable isotopes in tree rings: and M. E. K. Evans. In Review. Fusing tree-ring and forest
towards a mechanistic understanding of isotope fraction- inventory data to forecast the influences of climate, tree
ation and mixing processes from the leaves to the wood. size, and stand density on tree growth.
Tree Physiology 34, 796–818. Helama S., Melvin T. M., and Briffa K. R. (2017). Regional
Giebink C. L., DeRose R. J., Castle M., Shaw J. D., and Evans curve standardization: state of the art. The Holocene 27,
M. E. K. (In prep.). Climatic sensitivities derived from 172–177.
tree rings improve predictions of the Forest Vegetation Helser T. E., Lai H., and Black B. A. (2012). Bayesian
Simulator growth and yield model. hierarchical modeling of Pacific geoduck growth incre-
Gonzalez E. J. and Martorell C. (2013). Reconstructing shifts ment data and climate indices. Ecological Modelling 247,
in vital rates driven by long-term environmental change: 210–220.
a new demographic method based on readily available Heyerdahl E. K., Loehman R. A., and Falk D. A. (2014).
data. Ecology & Evolution 3, 2273–2284 Lodgepole pine-dominated forest in central Oregon’s
Gonzalez E. J., Matorell C., and Bolker B. J. (2016). Inverse Pumice Plateau: historical mixed-severity fires are resis-
estimation of population projection parameters using time tant to future climate change. Canadian Journal of Forest
series of population-level data. Methods in Ecology and Research 44, 593–603.
Evolution 7, 147–156. Heyerdahl E. K., Loehman R. A., and Falk D. A. (2019). A
Green A. J. (2001). Mass/length residuals: measures of body multi-century history of fire regimes along a transect of
condition or generators of spurious results? Ecology 82, mixed-conifer forests in central Oregon, USA. Canadian
1473–1483. Journal of Forest Research 49, 76–86.
Holmes R. L. (1983). Computer-assisted quality control in Margolis E. Q., Swetnam, T. W., and Allen, C. D. (2011). His-
tree-ring dating and measurement. Tree-Ring Bulletin 43, torical stand-replacing fire in upper montane forests of the
69–78. Madrean sky islands and Mogollon plateau, southwestern
Holtta T., Makinen H., Nojd P., Makela A., and Nikinmaa E. USA. Fire Ecology 7, 88–107.
(2010). A physiological model of softwood cambial Martin-Benito D., Kint V., Del Rio M., Muys B., and Canellas
growth. Tree Physiology 30, 1235–1252. I. (2011). Growth responses of West-Mediterranean Pinus
Housset J. M., Nadeau S., Isabel N., et al. (2018). Tree rings nigra to climate change are modulated by competition and
provide a new class of phenotypes for genetic associations productivity: past trends and future perspectives. Forest
that foster insights into adaptation of conifers to climate Ecology and Management 262, 1030–1040.
change. New Phytologist 218, 630–645. Martino J. C., Fowler A. J., Doubleday, Z. A., Grammar G. L.,
Kajimoto T., Onodera H., Ikeda S., Daimaru H., and Seki and Gillanders B. M. (2019). Using otolith chronologies
T. (1998). Seedling establishment of subalpine stone pine to understand long-term trends and extrinsic drivers of
(Pinus pumila) by nutcracker (Nucifraga) seed dispersal on growth in fisheries. Ecosphere 10, e02553.
Mt. Yumori, northern Japan. Arctic and Alpine Research McCarroll D. and Loader N. J. (2004). Stable isotopes in tree
30(4), 408–417. rings. Quaternary Science Reviews 23, 771–801.
Kelley J. and Smith T.M. (2003). Age at first molar emergence McLaughlin J. F., Hellmann J. J., Boggs C. L., and
in early Miocene Afropithecus turkanensis and life-history Ehrlich P. R. (2002). Climate change hastens population
evolution in the Hominoidea. Journal of. Human Evolu- extinctions. Proceedings of the National Academy of
tion 44, 307e329. Sciences of the United States of America 99, 6070–
Kennedy B. P., Blum J. D., Folt C. L., and Nislow K. H. (2000). 6074.
Using natural strontium isotopic signatures as fish mark- Melvin T. M. and Briffa K. R. (2008). A ‘signal-free’ approach
ers: methodology and application. Canadian Journal of to dendroclimatic standardisation. Dendrochronologia 26,
Fisheries and Aquatic Science 57, 2280–2292. 71–86.
Klesse S., DeRose R. J., Guiterman C. H., et al. (2018). Sam- Mina M., Martin-Benito D., Bugmann H., and Cailleret M.
pling bias overestimates climate change impacts on forest (2016). Forward modeling of tree-ring width improves
growth in the Southwestern United States. Nature Com- simulation of forest growth response to drought. Agricul-
munications 9, 5336. tural and Forest Meteorology 221, 13–33.
Le Cabec A., Tang N., and Tafforeau P. (2015). Access- Morgan P. and Bunting S. C. (1992). Using cone scars to esti-
ing developmental information of fossil hominin teeth mate past cone crops of whitebark pine. Western Journal
using new synchotron microtomography-based visualiza- of Applied Forestry (USA) 7(3), 71–71.
tion techniques of dental surfaces and interfaces. PLOS Morrongiello J. R. and Thresher R. E. (2015). A statis-
One 10(4), e0123019. tical framework to explore ontogenetic growth varia-
Le Moullec M., Buchwal A., Van Der Wal R., Sandal L., tion among individuals and populations: a marine fish
and Bremset Hansen B. (2019). Annual ring growth of a example. Ecological Monographs 85, 93–115.
widespread high arctic shrub reflects past fluctuations in Morrongiello J. R., Thresher R. E., and Smith D. C. (2012).
community-level plant biomass. Journal of Ecology 107, Aquatic biochronologies and climate change. Nature Cli-
436–451. mate Change 2, 849–857.
Lee Q. and Punt A. E. (2018). Extracting a time-varying Munier J., Brown P. M., and Romme W. H. (2014). Tree
climate-driven growth index from otoliths for use recruitment in relation to climate and fire in northern
in stock assessment models. Fisheries Research 200, Mexico. Ecology 95, 197–209.
93–103. Myers-Smith I. H., Elmendorf S. C., and Vellend M., et al.
Lloret F., Keeling E. G., and Sala A. (2011). Compo- (2015a). Climate sensitivity of shrub growth across the
nents of tree resilience: effects of successive low-growth tundra biome. Nature Climate Change 5, 887–891.
episodes in old ponderosa pine forests. Oikos 120, Myers-Smith I. H., Hallinger M., Blok D., et al. (2015b).
1909–1920. Methods for measuring arctic and alpine shrub growth:
Lynch A. M. (2012). What tree-ring reconstruction tells us a review. Earth-Science Review 140, 1–13.
about conifer defoliator outbreaks. In: P. Barbosa, D. K. Navarro-Cerrillo R. M., Sanchez-Salguero R., Rodriguez C.,
Letorneau, and A. A. Agrawal (eds.), Insect Outbreaks— et al. (2019). Is thinning an alternative when trees could
Revisited, 125–154. Wiley-Blackwell. die in response to drought? The case of planted Pinus nigra
Macalady A. K. and Bugmann H. (2014). Growth-mortality and P. sylvestris stands in southern Spain. Forest Ecology
relationships in Pinon Pine (Pinus edulis) during severe and Management 433, 313–324.
droughts of the past century: shifting processes in space Nowacki G. J. and Abrams M. D. (1997). Radial-growth
and time. PLOS One 9, e92770. averaging criteria for reconstructing disturbance histories
Marchal A., Schlichting C. D., Gobin R., et al., (2019). Deci- from presettlement-origin oaks. Ecological Monographs
phering hybrid larch reaction norms using random regres- 67, 225–249.
sion. Genes, Genomes, Genetics 9, 21–32.
O’Brien M. J., Engelbrecht B. M. J., Joswig J., et al. (2017). Roskilly B., Keeling E., Hood S., Giuggiola A., and Sala A.
A synthesis of tree functional traits related to drought- (2019). Conflicting functional effects of xylem pit struc-
induced mortality in forests across climatic zones. Journal ture relate to the growth-longevity trade-off in a conifer
of Applied Ecology 54, 1669–1686. species. Proceedings of the National Academy of Sciences
O’Connor C. D., Falk D. A., Lynch A. M., and Swetnam T. W. of the United States of America 116(3), 15282–15287.
(2014). Fire severity, size, and climate associations diverge Rountrey A. N., Coulson P. G., Meeuwig J. J., and Meekan
from historical precedent along an ecological gradient in M. (2014). Water temperature and fish growth: otoliths
the Pinaleno Mountains, Arizona, USA. Forest Ecology predict growth patterns of a marine fish in a changing
and Management 329, 264–278. climate. Global Change Biology 20, 2450–2458.
Ogle K., Whitham T. G., and Cobb N. S. (2000). Tree-ring Rowland L., da Costa A. C. L., Galbraith D. R., et al. (2015).
variation in pinon predicts likelihood of death following Death from drought in tropical forests is triggered by
severe drought. Ecology 81, 3237–3243. hydraulics not carbon starvation. Nature 528, 119–122.
Ogle K., Barber J. J., Barron-Gafford, G. A., et al. (2015). Rozendaal D. M. A., Brienen R. J. W., Soliz-Gamboa C. C.,
Quantifying ecological memory in plant and ecosystem and Zuidema P. A. (2010). Tropical tree rings reveal pref-
processes. Ecology Letters 18, 221–235. erential survival of fast-growing juveniles and increased
Ong J. J. L., Rountrey A. N., Meeuwig J., et al. (2015). juvenile growth rates over time. New Phytologist 185,
Contrasting environmental drivers of adult and juvenile 759–769.
growth in a marine fish: implications for the effects of Schultz E. L. and Miller T. E. X. (in prep.). Incomplete niche-
climate change. Scientific Reports 5, 10859. filling buffers high elevation trees against loss of suitable
Ong J. J. L., Rountrey A. N., Zinke J., et al. (2016). Evi- habitat under climate change.
dence for climate-driven synchrony of marine and terres- Schwing F. B., Murphree T., and Green P. M. (2002). The
trial ecosystems in northwest Australia. Global Change Northern Oscillation Index (NOI): a new climate index
Biology 22, 2776–2786. for the northeast Pacific. Progress in Oceanography 53,
Ong J. J. L., Rountrey A. N., Black B. A., et al. (2018). A 115–139.
boundary current drives synchronous growth of marine Serra-Maluquer X., Mencuccini M., and Martinez-Vilalta J.
fishes across tropical and temperate latitudes. Global (2018). Changes in tree resistance, recovery and resilience
Change Biology 24, 1894–1903. across three successive droughts in the northeast Iberian
Ong J. L., Meekan M. G., Hsu H. H., Fanning L.P., and peninsula. Oecologia 187, 343–354.
Campana S. E. (2020). Annual bands in vertebrae validat- Sesel M. (2017). An analysis of dental development in Pleis-
ed by bomb radiocarbon assays provide estimates of age tocene Homo using skeletal growth and chronological age.
and growth of whale sharks. Frontiers in Marine Science American Journal of Physical Anthropology 163, 531–541.
7, 188. Silva M. S., Funch L. S., and Da Silva L. B. (2019). The
Pederson N., Varner J. M., and Palik B. J. (2008). Canopy growth ring concept: seeking a broader and unambiguous
disturbance and tree recruitment over two centuries in approach covering tropical species. Biological Reviews 94,
a managed longleaf pine landscape. Forest Ecology and 1161–1178.
Management 254, 85–95. Sinclair A. R. E. and Arcese P. (1995). Population con-
Peltier D. M. P.,Barber J. J., and Ogle K. (2018). Quanti- sequences of predation-sensitive foraging: the Serengeti
fying antecedent climatic drivers of tree growth in the wildebeest. Ecology 76, 882–891.
southwestern U.S. Journal of Ecology 106, 613–624. Smith B. H. and Tompkins R. L. (1995). Toward a life histo-
Powell T. L. Wheeler J. K., de Oliveira A. A. R., et al. (2017). ry of the Hominidae. Annual Review of Anthropology 24,
Differences in xylem and leaf hydraulic traits explain 257e279.
differences in drought tolerance among mature Amazon Smolinski S. and Mirny Z. (2017). Otolith biochronology as
rainforest trees. Global Change Biology 23, 4280–4293. an indicator of marine fish responses to hydroclimatic con-
Redmond M. D., Weisberg P. J., Cobb N. S., Gehring C. A., ditions and ecosystem regime shifts. Ecological Indicators
Whipple A. V., and Whitham T. G. (2016). A robust 79, 286–294.
method to determine historical cone production among Somerville W. (1891). An account of Pressler’s growth borer.
slow-growing conifers. Forest Ecology and Management Transactions and Proceedings of the Botanical Society of
368, 1–6. Edinburgh 19, 90–93.
Ricker M., Gutiérrez-García G., Juarez-Guerrero D., and Speer J. H. (2010). The Fundamentals of Tree-Ring Research.
Evans M. E. K. (2020). Statistical age determination of tree University of Arizona Press.
rings. PLOS One 15(9), e0239052. Swetnam T. W. and Betancourt J. L. (1998). Mesoscale dis-
Rollinson C. R., Kaye M. W., and Canham C. D. (2016). turbance and ecological response to decadal climate vari-
Interspecific variation in growth responses to climate ability in the American Southwest. Journal of Climate 11,
and competition of five eastern tree species. Ecology 97, 3128–3147.
1003–1011. Swetnam T. W. and Lynch A. M. (1993). Multicentury,
regional-scale patterns of western spruce budworm out-
breaks. Ecological Monographs 63, 399–424.
Taylor A. H., Trouet V., Skinner C. N., and Stephens S. (2016). Visser M. D., Bruijning M., Wright S. J., et al. (2016). Func-
Socioecological transitions trigger fire regime shifts and tional traits as predictors of vital rates across the life cycle
modulate fire–climate interactions in the Sierra Nevadas, of tropical trees. Functional Ecology 30, 168–180.
USA, 1600–2015 CE. Proceedings of the National Acade- Von Arx G., Arzac A., Fonti P., et al. (2017). Responses of sap-
my of Sciences of the United States of America 113, 13684– wood ray parenchyma and non-structural carbohydrates
13689. of Pinus sylvestris to drought and long-term irrigation.
Temple D. H. (2014). Plasticity and constraint in response Functional Ecology 31, 1371–1382.
to early-life stressors among late/final Jomon period for- Von Biela V. R., Kruse G. H., Mueter F. J., et al. (2015).
agers from Japan: evidence for life history trade-offs from Evidence of bottom-up limitations in nearshore marine
incremental microstructures of enamel. American Journal systems based on otolith proxies of fish growth. Marine
of Physical Anthropology 155, 537–545. Biology 162, 1019–1031.
Thompson R. D., Daniels L. D., and Lewis K. J. (2007). A new Von Biela V. R., Zimmerman C. E., Kruse G. H., et al.
dendroecological method to differentiate growth respons- (2016). Influence of basin- and local-scale environmen-
es to fine-scale disturbances from regional-scale environ- tal conditions on nearshore production in the northeast
mental variation. Canadian Journal of Forest Research 37, Pacific Ocean. Marine and Coastal Fisheries: Dynamics,
1034–1043. Management, and Ecosystem Science 8, 502–521.
Thresher R. E. 1999. Elemental composition of otoliths as a Warren-Myers F. W., Dempster T., and Swearer S. E. (2018).
stock delineator in fishes. Fisheries Research 43, 165–204. Otolith mass marking techniques for aquaculture and
Tolwinski-Ward S. E., Evans M. N., Hughes M. K., and restocking: benefits and limitations. Reviews in Fish Biol-
Anchkaitis K. J. (2011). An efficient forward model of the ogy and Fisheries 28, 485–501.
climate controls on interannual variation in ring width. Weijers S., Buchwal A., Blok D., Loffler J., and Elberling B.
Climate Dynamics 36, 2419–2439. (2017). High Arctic summer warming tracked by
Ulrich D. E. M., Still C., Brooks J. R., Kim Y., and increased Cassiope tetragona growth in the world’s
Meinzer F. C. (2019). Investigating old-growth Ponderosa northernmost polar desert. Global Change Biology 23,
pine physiology using tree-rings, δ13 C, δ18 O, and a 5006–5020.
process-based model. Ecology 100, e02656. Weisberg S., Spangler G., and Richmond L. S. (2010). Mixed
Vaganov E. A., Hughes M. K., and Shashkin A. V. (2006). effects models for fish growth. Canadian Journal of Fish-
Growth Dynamics of Conifer Tree Rings: Images of Past eries and Aquatic Sciences 67, 269–277.
and Future Environments. Springer. White J. W., Nickols K. J., Malone D., et al. Fitting state-
Van Der Sleen P. Groenendijk P., Vlam M., et al. (2015). No space integral projection models to size-structured time
growth stimulation of tropical trees by 150 years of CO2 series data to estimate unknown parameters. Ecological
fertilization but water-use efficiency increased. Nature Applications 26, 2677–2694.
Geoscience 8, 24–28 Wigley T. M. L., Briffa K. R., and Jones P. D. (1984). On the
Van Der Sleen P., Stransky C., Morrongiello J. R., Haslob H., average value of correlated timeseries, with applications
Peharda M., and Black B. A. (2018). Otolith increments in in dendroclimatology and hydrometeorology. Journal of
European plaice (Pleuronectes platessa) reveal temperature Applied Meteorology and Climatology 23, 201–213.
and density-dependent effects on growth (vol 75, pg 1151, Willmes M., Hobbs J. A., Sturrock A. M., et al. (2018). Fishery
2018). ICES Journal of Marine Science 75, 1151. collapse, recovery, and the cryptic decline of wild salmon
Van Mantgem P. J., Falk D. A., Williams E. C., Das A. on a major California river. Canadian Journal of Fisheries
J., and Stephenson N. L. (2018). Pre-fire drought and and Aquatic Science 75, 1836–1848.
competition mediate post-fire conifer mortality in west- Wong C. M. and Lertzman K. P. (2001). Errors in estimat-
ern U.S. National Parks. Ecological Applications 28(7), ing tree age: implications for studies of stand dynamics.
1730–1739. Canadian Journal of Forest Research 31, 1262–1271.
Van Mantgem P.J., Falk D. A., Williams E. C., Das A. J., and Woodward H. N., Padian K., and Lee A. H. (2013). Skele-
Stephenson N. L. (2020). The influence of pre-fire growth tochronology. In: K. Padian and E.-T. Lamm (eds.), Bone
patterns on post-fire tree mortality for common conifers in Histology of Fossil Tetrapods: Advancing Methods, Anal-
western US parks. International Journal of Wildland Fire ysis, and Interpretation, Ch. 7. University of California
29(6), 513–518. Press.
Vanderwel, M. C., Rozendaal D. M. A., and Evans M. E. K. Wright M., Sherriff R. L., Miller A. E., and Wilson T. (2018).
(2017). Predicting the abundance of forest types across the Stand basal area and temperature interact to influence
eastern United States through inverse modelling of tree growth in white spruce in southwest Alaska. Ecosphere
demography. Ecological Applications 27, 2128–2141. 9, e02462
Villalba R. and Veblen T. T. (1998). Influences of large-scale Wykoff P. H. and Clark J. S. (2002). The relationship between
climatic variability on episodic tree mortality in northern growth and mortality for seven co-occurring tree species
Patagonia. Ecology 79, 2624–2640. in the southern Appalachian Mountains. Journal of Ecol-
ogy 90, 604–615.
Yang J., Cao M., and Swenson N. G. (2018). Why function- contribute disproportionately to population growth? A
al traits do not predict tree demographic rates. Trends in new analysis tool for matrix models and its application to
Ecology & Evolution 33, 326–336. rainforest trees. The American Naturalist 174, 709–719.
Youngblood A., Max T., and Coe K. (2004). Stand structure Zuidema P. A., Vlam M., and Chien P. D. (2011). Ages and
in eastside old-growth ponderosa pine forests of Oregon long-term growth patterns of four threatened Vietnamese
and northern California. Forest Ecology & Management tree species. Trees 25, 29–38.
199, 191–217. Zuidema P. A., Poulter B., and Frank D. C. (2018). A wood
Zhao M. Klaassen C. A. J., Lisovski S., and Klaassen M. biology agenda to support global vegetation modeling.
(2018). The adequacy of aging techniques in vertebrates Trends in Plant Science 23(11), 1006–1015.
for rapid estimation of population mortality rates from
age distributions. Ecology and Evolution 9, 1394–1402.
Zipkin E. F., Inouye B. D., and Beissinger S. R. (2019).
Innovations in data integration for modeling populations.
Ecology 100, e02713.
Zuidema P. A., Brienen R. J. W., During H. J., and
Guneralp B. (2009). Do persistently fast-growing juveniles
CHAPTER 5
Longitudinal demographic
data collection
Marlène Gamelon, Josh A. Firth, Mathilde Le Moullec, William K. Petry,
5.1 Introduction: long-term field studies Central America (e.g. Condit et al. 2017), and Europe
(e.g. Hutchings 2010). Longitudinal data have been,
Demographic data can be collected during a single
and continue to, be collected on many taxa across the
visit, or multiple visits, to the population. For instance,
Tree of Life and have provided significant insights in
a single visit to the field might be in principle enough to
ecology, evolution, and demography (Clutton-Brock
collect data on tree rings or fish otoliths, which can then
and Sheldon 2010; Reinke et al. 2019). Importantly,
be used to estimate key demographic parameters retro-
demographic data can be collected at the population
spectively (see Chapter 4). Alternatively, longitudinal
level (e.g. time series of population counts) or at the
studies involve numerous visits to the study popula-
individual level (e.g. monitoring of marked and/or
tion with repeated observations/measurements; this
geo-referenced individuals throughout their life).
kind of approach typically occurs over multiple weeks,
From demographic data collected at the population
months, years, or even decades, depending on the gen-
level (i.e. time series of population counts) and with
eration time of the study species, research aims, and
appropriate methodological tools to analyse them (see
available support for the research programme. The
the next chapters), several questions can be addressed.
collected longitudinal data are then used to estimate
For instance, ‘What is the population size trend?’,
demographic parameters such as annual population
‘What is the population spatial distribution?’, ‘What
abundances (see Chapter 7) and/or survival, growth,
are the effects of changes in climate on population
and reproductive rates (see Chapter 13).
size/distribution?’, and ‘What are the effects of human
Many long-term studies are running worldwide.
activities on population size/distribution?’. Finer-scale
In Antarctica, South America, and Central America,
demographic data can also be recorded. For instance,
the monitoring of avian and mammalian populations
population size time series can be collected with dis-
has already reached over 50 years in some areas (see
aggregation into stage, sex, or morph (e.g. counts of
Taig-Johnston et al. 2017 for a review). In the north-
adult females). This information is not recorded at the
ern hemisphere, some of the longest studies even reach
individual level, but still, it is more useful than total
70 years (Clutton-Brock and Sheldon 2010). Even if
population counts. This better resolution can help in
there has been a strong focus on birds and mam-
tackling some of the questions outlined below with
mals (Festa-Bianchet 2017; Kappeler et al. 2017 for
individual-level data.
some reviews on mammals; Smith et al. 2017; see e.g.
While long-term data collected at the population
Marshall et al. 2018), long-term studies also exist for
level provide significant insights into the popula-
other taxa within the animal kingdom (e.g. corals, see
tion’s dynamics, they usually do not allow a full
Connell et al. 1997; amphibians, see Cayuela et al.
understanding of the underlying demographic mecha-
2020). Long-term field studies are not restricted to
nisms causing changes in population distribution and
animals: long-term plant studies span over 20 years
abundance (Clutton-Brock and Sheldon 2010). Indeed,
in regions of North America (e.g. Ellis et al. 2012),
Marlène Gamelon et al., Longitudinal demographic data collection. In: Demographic Methods Across the Tree of Life.
changes in population abundance over time can result periods of time. Moreover, data collected might
from changes in rates of births (reproduction), deaths be scarce due to low sample sizes and because
(survival), emigration, and/or immigration. That lack researchers are reluctant to individually monitor indi-
of mechanistic resolution of population-level data is viduals in vulnerable populations. In those cases,
precisely the major limitation of longitudinal studies at it might be advisable to take advantage of both
the population rather than individual level. population-level and individual-level data, through,
For instance, climate-induced changes in popula- for example, an integrated modelling approach (see
tion abundance can result from the sensitivity of a Chapter 14).
particular stage class (e.g. a particular age) to climate In this chapter, first we introduce the reader to
conditions that will exhibit lower survival or fecundity. procedures that can be implemented in the field to
In the well-studied Soay sheep (Ovis aries) population, collect specific demographic data on mobile species
Coulson et al. (2001) showed that juvenile survival was (e.g. birds, mammals), at both the population and
particularly sensitive to the North Atlantic Oscillation, individual levels. Second, we present the procedures
a weather phenomenon associated with temperature and the type of demographic data that can be col-
and rainfall. Similarly, within conifer forests, recruit- lected on sessile species (e.g. corals, plants) at both
ment and growth were associated with fire (Tepley levels.
et al. 2017). Interestingly, going beyond population
abundance and exploring the effect of climate and
5.2 Collection of longitudinal
human activities on all vital rates and at all stages can
also allow one to detect reduced survival and/or fecun-
demographic data on mobile species
dity even if population abundance apparently remains 5.2.1 Procedures and type of data collected
stable. For instance, in an eagle owl (Bubo bubo) popula- at the population level
tion, adult survival has dramatically decreased because
of electrocution, despite constant population abun- The problem of imperfect detection while
dance, the decline in survival being balanced through monitoring populations
massive immigration (Schaub et al. 2010), which could Within a mobile species’ range, the population dynam-
only be detected because the data were collected at the ics can be rather complex. A fragmented landscape
individual level. means that individuals from the same population can
From demographic data collected at the individual occupy diverse habitat types in space and time. Deter-
level, it is possible to accurately identify the proximate mining population range size or abundance therein
causes of changes in population size/distribution comes with challenges related to two key aspects:
and the underlying demographic mechanisms. Thus, (1) spatial variability within the studied area and (2)
questions such as ‘What are the effects of changes in cli- individual detectability (Yoccoz et al. 2001). Important
mate on survival/fecundity/immigration/emigration errors in quantification of population size and emer-
rates?’, ‘What are the effects of human activities on gent dynamics can be made if individuals are missed
survival/fecundity/immigration/emigrationrates?’, when they are present at the site, or misidentified, or
‘What are the stage classes the most influenced by when the same individual is counted multiple times
changes in climate?’, and ‘Which stage classes are (Miller et al. 2011). Imperfect detection depends on
most influenced by human activities?’ can be tack- species characteristics (e.g. camouflage), spatial and
led. Identifying the stages and their vital rates (e.g. temporal variation (e.g. migration), and survey char-
survival, fecundity) the most affected by climate acteristics (Cressie et al. 2009; Guillera-Arroita 2017)
changes or human activities is crucial in conser- (Figure 5.1). Uncertainties in the survey character-
vation and management to develop appropriate istics depend on the type of monitoring, the study
targeted strategies (see Frederiksen et al. 2014 for a design, and the analysis (Figure 5.1). If the aim of
review). the monitoring is to estimate population abundance,
In that respect, longitudinal studies at the indi- the higher the number of repeated measurements and
vidual level are powerful. But as with population- sample size, the higher the precision will be. Accu-
level studies, they suffer from limitations. Indeed, rate estimates (i.e. precise and unbiased; Williams et al.
implementing individual-based monitoring is gener- 2002), repeated over time, enable the inference of the
ally costly in terms of money and human resources, drivers of population dynamics, such as climate and
oftentimes requiring expensive materials (e.g. GPS harvest regimes (Thompson et al. 1998; Miller et al.
collars) and experienced fieldworkers during long 2011; Guillera-Arroita 2017) (Figure 5.1). Under- or
L O N G I T U D I N A L D E M O G R A P H I C D ATA C O L L E C T I O N 99
Hierarchical processes
Ecological processes
e.g. climate effects and harvest
S S S
t+1 t+2
ObservationaI processes
e.g. false negative or positive
O O O
Species Spatial Temporal Survey

Reproduction
Survival Abundance
Immigration
Emigration Monitoring Study design Analysis
Sections
N̂ e.g. Chapters
5.2.2-5.3.2 Individual
12 and 13
e.g. CMR
Levels
Sections
5.2.1-5.3.1
Population e.g. Chapter 7
e.g. Distance
sampling
Figure 5.1 Hierarchical organisation of imperfect detection in wildlife systems, inspired from Royle and Dorazio (2008), Zuur et al. (2009), Kéry
and Schaub (2012), and Guillera-Arroita (2017). The true abundance is a latent state (S) changing at each time-step, for example year to year
(from year t+1 to t+2), due to changes in vital rates (survival, reproduction, immigration, emigration) affected by ecological processes. At each
time-step, monitoring of vital rates or population abundance results in observations (O), such as the observed abundance (N̂). Each of these
observations is subject to observational processes, which can be decomposed into distinct components. Population survey is one of the
components and consists in applying the required monitoring in the field, using the appropriate study design (at the population or individual level)
and then using appropriate analyses to make inferences (modified from Le Moullec (2019)).
over-estimation of abundances or range sizes can have species, as well as the spatial coverage vary according
important consequences; for instance, they can lead to the species characteristics, including its rarity and
to ineffective management and incorrect conservation detectability in the environment (Figure 5.2). There-
decisions (Sinclair et al. 2006). after, data can be collected with multiple monitoring
methods and sampling designs (see Figure 5.2): for
A large diversity of methods for instance, camera traps located at random or stratified
monitoring populations designs can be used with capture–recapture or distance
Monitoring populations over years to determine range sampling. Several of the listed monitoring approach-
size and/or abundance has been central in the field es here are also applicable to sessile species (see
of ecology and wildlife management since the 1930s section 5.3.1).
(Krebs 1998; Williams et al. 2002; Sutherland et al. Estimating population abundance can be performed
2013), and a large diversity of methods has emerged on an absolute or relative scale with direct or indirect
since (Seber 1992; Thompson et al. 1998; Sutherland population counts. Absolute abundance methodolo-
2006; Morellet et al. 2011; ENETWILD Consortium gies aim to estimate the true population size, that is, the
et al. 2020). The monitoring method chosen by a man- state variable, while relative abundance methodologies
ager/scientist, the ‘tools’ (e.g. camera traps, perma- (e.g. capture rate, hunting records, sign detection rate,
nent plot) used to increase the overall detectability of activity indices) estimate the population size relative
Abundant species Rare species
High detectability Low detectability
Direct monitoring: Direct monitoring: Indirect monitoring:
Methods Tools Species example Methods Tools Species example Methods Tools Species example
Ground based Ptarmigan [Pedersen et al. 2012] Ground based Saw-wort plant [Schorr 2013] Pellet counts Sika deer Marques et al. 2001]
Distance
Shipboard Porpoise [Hammond et al. 2012] Camera trap Duikers [Howe et al. 2017] Distance Bilby, Rabbit [Lollback et al.
Sampling Distance
Aircraft Polar bear [Aars et al. 2009] Spotlight Red deer [La Morgia et al. 2015] sampling 2017]
sampling
Thermal Fallow deer [Forcadi et al. Acoustic Cetaceas [Marques et al. 2013]
Ground based Chamois [Loison et al. 2006]
imaging 2013]
(entire pop.) Reindeer [Le Moullec et al. 2017] Strip Pellet counts Red deer [Alves et at. 2013]
(permanent Cryptantha flava [Salguero- Strip Spotlight Red fox [Field et al. 2005] transect Acoustic Sperm whale [Ward et al. 2012]
plot-design) Gómez et al. 2014]. Sea transect Red deer [Garel et al. 2010]
eDNA Aquatic parasite [Carraro et al.
anemones [Beuchel et al. 2008] Sign-
Camera trap Wallaby, Water deer, Mara, 2018]
Census Aerial imaging Sea otters [Williams et al. 2017] Capture- detection
(unmarked) Muntjac [Rawcliffe et al. 2008] Scent-station Swift fox [Schauster et al. 2002]
Remote sensing Penguin [Fretwell et al. 2012] recapture rate
Red fox [Ramsey et al. 2014] Track count Lynx [Linnell et al. 2007]
UAVs/AUVs Geese [Chabot & Bird 2012]
Weeds [Lambert et al. 2018] Snap trap Voles [Ims et al. 2011] Capture- Acoustic Minke whale [Martin et al.
Capture
Coral [Parsons et al. 2018] (permanent recapture (SECR) 2013]
rate
plot-design)
Strip Ground based Cowslip [Buckland et al. 2007] Hunting Bag record Red grouse [Cattadori et al.
transect Aircraft Reindeer [Pole et al. 2013] records 2003]
Ungulates [lmperio et al. 2010]
Capture- Point counts European jay [Royle et al. 2007]
recapture (unmarked)
Figure 5.2 Examples of monitoring methods for the estimation of population abundance in animals and plants. A same monitoring method
can be conducted with the help of different tools to increase species detectability and spatial coverage. UAVs = unmanned aerial vehicles;
AUVs = autonomous underwater vehicles; SECR = spatially explicit capture–recapture (unmarked individuals); eDNA = environmental DNA (credit A. Tholfsen / M. Le Moullec).
to a previous survey (Pollock et al. 2002; Hopkins Methods for spatially referenced observations
and Kennedy 2004; O’Brien 2011; Amos et al. 2014). Spatial referencing of individual observations (i.e.
Importantly, the latter approach assumes a constant coordinates, spatial unit) within a population enables
detection probability and thus requires constant survey researchers to relate the frequency of detections of indi-
design, personnel training, species behaviour, habitat viduals to the surrounding conditions, that is spatially
use, and range size between survey events. Under such referenced environmental covariates (Aarts et al. 2012;
conditions, the trends in abundance can inform man- Guillera-Arroita 2017). Thereafter, within the modelled
agers’ decisions. However, relative estimates of popu- spatial distribution of the species, population density
lation size cannot be compared across systems (Amos can be predicted in areas or periods of time not sur-
et al. 2014). veyed, as long as extrapolations are done within the
Abundance estimates from direct individual counts range of data monitored (Sillett et al. 2012). Random
within a population require visual detection of the sampling of individuals across the study region is fun-
organism (i.e. total counts, camera surveys). These damental to unbiased design-based studies. However,
methodologies are often preferred (Morellet et al. model-based studies grant access to a large diversity
2011) and are adapted for abundant, tame, and easi- of study designs and opportunistic count data collec-
ly detectable species, since they are ‘data hungry’ for tion. Aarts et al. (2012) demonstrated the similarities in
robust modelling and often logistically costly (Yoccoz spatial density estimates between count data collected
et al. 2001; Guillera-Arroita 2017) (Figure 5.2). Conse- in discrete space (i.e. number of observations per spa-
quently, these approaches are often restricted to small- tial unit) or in continuous space (e.g. use-availability,
scale studies (but see Yuan et al. 2017). Monitoring where each individual is treated as one observation)
methods for rare or hard-to-detect species, such as and presence–absence data (if the spatial unit corre-
cryptic, nocturnal, forest, or aquatic species, require sponds to single observations). Hence, data to model
the use of specific tools to increase detections and spa- the spatial density can be diverse (Baddeley et al. 2006;
tial coverage (e.g. spotlight, thermal imaging, acous- Zuur et al. 2009; Royle et al. 2013; Ramsey et al. 2015;
tic recorders) (Marques et al. 2013). For instance, to Williams et al. 2017; Antún et al. 2018; Le Moullec et al.
monitor marine species, unmanned underwater vehi- 2019). Spatiotemporal variations in detection probabili-
cles (UUVs) can be coupled with postproduction com- ty are often accounted for, prior to analysing the spatial
puting algorithms to obtain the relevant demograph- density function (i.e. two-stage approach), but in some
ic information (Smale et al. 2012). Notably, the large cases the detection and density function are estimated
diversity of existing methods to monitor populations in simultaneously (i.e. one-stage approach) (Royle et al.
terrestrial systems is still valid in marine environments 2013).
(see Katsanevakis et al. 2012 for a review). In some cas-
es, indirect population counts based on signs detection 5.2.2 Individual-level long-term field studies:
(e.g. fecal pellet counts, track counts, or environmen- concepts, methods, and data
tal DNA) can be better suited to the task (Pollock et al.
2002; Thompson 2004; MacKenzie et al. 2005; Jones Methods that allow individuals within a population
2011). to be uniquely recognised have given rise to a pletho-
The methodology chosen to study a population ra of long-term studies of wild animal systems that
should preferably be associated with a measure of have been formative in our understanding of their ecol-
detection probability that is inherent to certain method- ogy, evolution, and conservation biology. The ability
ologies, such as distance sampling (Buckland et al. to individually recognise animals has not only provid-
2007a), or capture–mark–recapture (Yoccoz et al. 2001). ed mechanistic insights into the drivers of the patterns
Capture–mark–recapture is further developed in the often considered by long-term population-level stud-
next section on individual-based methods, yet infer- ies (such as the movements between, and age struc-
ences on population abundance can be drawn from tures within, populations, and the causes of changes
this methodology too. More generally, information on in fitness or selection), but also proven foundational
organism resightings (marked or unmarked) (Royle in understanding social structure, individual-level
et al. 2013) is often combined with monitoring method- fitness, and the links between different life history
ologies to access information on detection probability, stages and generations (Clutton-Brock and Sheldon
that is total counts (Le Moullec et al. 2017), and camera 2010). This section explores current and developing
traps (Karanth and Nichols 1998). procedures used to collect data at the individual level
in these long-term studies, the types of data these beyond these restrictions, either using mist-netting
approaches provide, and the new implications that to capture, mark, or recapture these birds outside of
recent advances are allowing. the breeding period (such as during winter foraging;
Perrins et al. 1991) or using colour bands (display-
Individual-level monitoring using ing unique combinations of rings of colours) to allow
physical captures non-physical ‘recapture’ via resightings (Ekman 1989).
The fundamental premise of long-term individual- The aforementioned initial developments allowed
based field studies in animal populations of mobile for individual monitoring to be less spatiotemporal-
species has long been, and continues to be, the ly restricted than previous protocols. However, the
individual-level ‘capture, mark, recapture’ (CMR) pro- advances with the largest potential for these particu-
cedure (see Figure 5.1); an individual is ‘captured’ lar systems (and most of the individual-based long-
(physically or simply recorded), ‘marked’ in a unique- term study populations generally) have come from
ly identifiable way (through active intervention or recent technological developments in automated ani-
through documenting individually recognisable char- mal tracking systems (Bridge et al. 2011; Kays et al.
acteristics, such as marks on the tails of whales; Pomilla 2015; Jønsson et al. 2016). For instance, various long-
et al. 2014), and ‘recaptured’ (or ‘resighted’) at a later term study populations of tits now tag captured indi-
time (i.e. re-recording the unique identity of that individuals with passive integrated transponders (or PIT
vidual). The particular protocols employed for each tags) that are either contained within plastic leg rings
of these stages of the individual-level CMR proce- or injected subcutaneously. These PIT tags contain
dure are specific to the species under consideration a microchip with a unique identification code that
and the practicalities of fieldwork. However, follow- can be read, and automatically recorded (i.e. allowing
ing this standard’s underlying logic, these procedures ‘resightings’), by radio-frequency identification (RFID)
are constantly refined as new techniques and technolo- stations. The stations can be placed at nest boxes during
gies become available and applicable to these valuable the breeding period (Firth and Sheldon 2015; Schlicht
long-term study systems. et al. 2015; Firth et al. 2018) or at feeding stations to
One of the best examples of the progress of specif- allow large-scale resighting during the nonbreeding
ic protocols comes from the earliest of the long-term season (Firth and Sheldon 2016).
individual-based field studies: wild birds, particular- More generally, RFID technology is emerging as
ly tits (great tits, Parus major; and blue tits, Cyanistes a particularly good example of applying tracking
caeruleus), which started as long-term study popula- technologies within long-term individual-based study
tions across Europe in the 1940–1950s (Kluijver 1951; populations. This is because RFID is well suited to
Lack 1966). These species readily breed in nest boxes these systems due to the size of the tags (often < 1 g),
during the spring, so individual CMR protocols can their relative affordability, and the lifelong readabili-
quickly be developed. They are based on physically ty of the microchips which do not require an internal
capturing individuals as breeding adults or nestling battery (as they are passive). As such, marking indi-
chicks, marking them with a unique identifier (a met- viduals with unique PIT microchips is now a preva-
al leg ring with a unique code), and recapturing and lent and staple method of individual CMR systems,
identifying them in following breeding seasons. This not just for long-term bird studies (Bonter and Bridge
standard procedure was widely and rapidly adopted 2011) but also across various populations, ranging from
across a range of systems (not limited to animals—see insects to fish to mammals (Gibbons and Andrews
below) and remains a common method across various 2004; Rehmeier et al. 2006; Silcox et al. 2011). Remark-
long-term population bird studies (Grant 1986; Nisbet ably, these new technologies tend to be less invasive.
1989; O’Connor 1991; Perrins et al. 1991). While toe clipping for small mammals has been the rule
for a long time in CMR protocols, these new technolo-
Advancing longitudinal methods: progress gies (e.g. camera trapping and noninvasive DNA; see
in individual-level monitoring also Chapter 1), now favour noninvasive censusing.
Although effective, the nest box ringing procedure The development of a diverse array of exemplary
meant that these early individual CMR studies within methods for individual-based monitoring of animals
these systems were limited to monitoring individuals comes from long-term studies of mammalian systems
at specific locations (e.g. nest boxes) and restricted to (Clutton-Brock and Sheldon 2010; Hayes and Schradin
particular time-frames (i.e. during breeding). As such, 2017; Schradin and Hayes 2017). Indeed, some of
further techniques were soon developed to expand the longest and most substantial individual-based
study populations are mammals, partly due to the in photos gathered from opportunistic sightings or
ease of ‘resighting’ these animals, especially for transect surveys (Eguchi 2014; Urian et al. 2015),
island populations of large mammals where immigra- which makes individual monitoring often very diffi-
tion/emigration do not exist, and habituation is rapid- cult (Mann and Karniski 2017). Advances in a range
ly achievable due to the lack of predation. For example, of tracking technologies, from drone surveys to newly
long-term studies such as those on red-deer (Cervus developed animal-borne tags (Hussey et al. 2015; Kays
elaphus), which began in 1971 on the Isle of Rum, Scot- et al. 2015; Nowacek et al. 2016; Hays et al. 2019), will
land (Clutton-Brock et al. 1982), and Soay sheep, which provide ripe opportunity to rapidly advance the mon-
began in 1985 on St Kilda, Scotland (Clutton-Brock itoring of these systems, by allowing individuals to be
and Pemberton 2004), have been successful in con- monitored over their entire lifetimes.
sistently employing standardised protocols based on The new technological monitoring methods also
capturing and uniquely marking individuals shortly provide the potential for continuously tracking move-
after birth, and then resighting these individuals via ments at fine scales (e.g. within resident territorial
regular censuses of the study populations thereafter. species) as well as over larger distances for migratory
Mammalian systems in less-convenient settings have species, and even nomadic species moving in irregular
benefitted substantially from recent advances in mon- manners (Teitelbaum and Mueller 2019). Further, it is
itoring procedures (Noonan et al. 2015; Hays et al. becoming possible to integrate technology providing
2016; Nowacek et al. 2016). For example, bat species detailed physiological monitoring of individual states,
hold many of the same conveniences that long-term activity, and metabolism into tracking devices (see
bird systems allow (i.e. occupying researcher-made Chapter 2). Such information is particularly valuable
boxes, as well as ease of capture in mist nets). As such, within long-term study systems for which the basic
these systems were also originally based on monitoring underlying ecology is already well researched, and
using individually coded rings but are now converg- this now provides the potential to allow vast advances
ing on the same technological approaches found to be in knowledge surrounding the causes and conse-
useful for monitoring bird systems, for example using quences of individual states (Hayes and Schradin
PIT tags and RFID technology (Fleischmann et al. 2013; 2017; Schradin and Hayes 2017). Yet, one of the most
Law 2018) that allow automated resightings instead important additional advantages of large-scale auto-
of relying on physical captures. Similarly, many mam- mated methods is that these new approaches allow for
mal systems have benefitted from the miniaturisation simultaneous monitoring of all individuals over large
of GPS technologies providing high precision and con- spatiotemporal scales, thus providing unprecedented
stant monitoring (Tomkiewicz et al. 2010; McMahon insight into the social structure of natural populations
et al. 2017). (Krause et al. 2013; Firth and Sheldon 2016) (see
Long-term studies based on species that spend Chapter 3).
large proportions of their time underground (there-
fore restricting the use of GPS) have had to con- 5.3 Collection of longitudinal
sider other approaches. For example, the UK Euro-
demographic data on sessile species
pean badger Meles meles project, which started over
30 years ago (Macdonald and Newman 2002), was tra- Sessile species are those that lack the ability of
ditionally restricted to overground monitoring or cap- (self-)locomotion. From a life cycle perspective, sessile
ture/resighting procedures. Now, magneto-inductive species typically correspond to species where estab-
tracking techniques are available which allow automat- lished life cycle stages are anchored onto a substrate.
ed, continuous, fine-scale monitoring of individuals Naturally, plants adhere to this definition, though
whilst underground (Noonan et al. 2015). These tech- some remarkable exceptions in the Planta kingdom do
niques hold much potential for other long-term stud- exist, such as the polytomic group of tumbleweeds (e.g.
ies for other ground-dwelling species (Schradin and genus Kali, Amarnathus albus, Salsola spp.) and the Rose
Hayes 2017). of Jericho (Anastatica hierochuntica). However, sessili-
One of the most important potential applications of ty is not unique to plants. Indeed, entire taxonomic
technology for longitudinal studies of individual ani- groups of the animal kingdom, such as corals and
mals in relatively inconvenient settings may well be sponges, as well as other species of important economic
for studies of marine mammals (Hazen et al. 2012; value (e.g. barnacles), are sessile. In addition, fungi and
Mann and Karniski 2017). Here, long-term monitoring a significant percentage of bacteria (e.g. Geesey et al.
has long been based on identification of individuals 1978) have sessile modes of life.
The monitoring of sessile species has some advan- Novel light detection and ranging (LIDAR) tech-
tages over that of mobile species, in that mortality in nologies can complement data acquisition through
sessile species cannot be confounded with the individ- their ability to operate in high-density stands (Malhi
ual not being at its previous location due to mobility. et al. 2018). In this regard, the application of the
As such, the classical equation modelling the changes approaches discussed regarding the detectability of
in population size (N) over time as a function of the different life cycle stages in mobile species is just
rates of births, deaths, emigration, and immigration as equally useful here (Figure 5.1). In addition, the
can be reduced to just two terms, births and deaths, development of low-elevation aerial technology
assuming that dispersal (as well as mobility) does not holds great promise, though the challenge remains
occur. However, modelling the demography of sessile in how to navigate this technology in landscapes full
species has its own specific challenges. Indeed, sessile of handicaps, such as rocks, wind/water currents,
species, because they are ‘stuck’ in place for most of tree trunks, lianas/algae, and small stems/debris
their life cycles, have evolved strategies to cope with that may not be adequately identified by navi-
the local environmental conditions (Huey et al. 2002; gating devices equipped with smart technology
Svensson and Marshall 2015; Žádníková et al. 2015). (but see https://www.ox.ac.uk/research/research-
It is precisely the strategies of some of those sessile impact/poetry-motion). Some key applications of
species that causes new challenges for the monitoring drone (unmanned aerial vehicle, UAV) technologies
of natural populations in sessile species. These strate- have been targeted specifically for agricultural use
gies include clonality, dispersal, propagule dormancy, (Saha et al. 2018; Jiménez López and Mulero-Pázmány
vegetative dormancy, creeping, and mimicry, among 2019). These approaches, coupled with artificial
others, and are discussed below. intelligence to automatically distinguish and track
individuals (Ampatzidis et al. 2019), represent a
unique opportunity to quantify population number
5.3.1 Procedures and type of data collected
and even structure in sessile organisms (Figure 5.3).
at the population level It is worth mentioning here that the power of these
Many of the field approaches described above to quan- novel technologies cannot be harnessed efficiently
tify and estimate the population size, distribution, and without taking into account the pertinent advances
structure of mobile species are also relevant to sessile in image analysis (e.g. Maillard et al. 2010; Bruijning
species populations (Figure 5.2). However, field tech- et al. 2018). The monitoring of sessile populations
niques to monitor sessile populations tend to capitalise below water is faced with important challenges due
more on their immobility. Consequently, permanent to obvious logistical considerations, such as lack of
plots, quadrats, and transects are the standard meth- oxygen, strong currents, and high viscosity. But novel
ods used in sessile longitudinal demographic studies. developments from physics now also allow for the
Nonetheless, technological advances are also reshap- treatment of seawater waves as magnifying glasses to
ing how natural populations of sessile species (e.g. evaluate coral reef properties at high resolution using
plants and otherwise) are monitored. UAVs (Chirayath and Earle 2016).
The increase in satellite data resolution now means
that counts of individuals are achievable in ecosys-
tems where density is low, individuals are distributed
5.3.2 Procedures and type of data collected
at random or quasi-random, and where there is little
three-dimensional layering (Bai et al. 2011). Desert and
at the individual level
Mediterranean ecosystems, for instance, are ideal can- While age estimation methods do exist for ses-
didate systems where these technologies show more sile species (Chapter 4), age estimation in species
promise (Peters and Eve 1995). In contrast, monitoring such as plants, corals, or sponges in a way that is
the population dynamics of sessile species in tropi- not destructive—and thus compatible with long-term
cal and marine ecosystems remains more challenging. censuses—is challenging. For that reason, among oth-
This is because, in these cases, populations of sessile ers related to convenience (Ebert 1998), the tracking of
species tend to have complex elevational structures, individuals in sessile species’ populations, as well as
where only adults who have broken through the tight their modelling (e.g. Chapters 9 and 10), tend to include
canopy (or dense marine reef community) are readily information on not only survival and reproduction
observable with satellites, and thus for which little but also changes in individual size (rather than age)
information is directly available for other stages. through time. In fact, size is the most widely used
predictor of fitness components in the demography of inform on the vitality of the individual (e.g. height may
sessile species (Caswell 2001; Caswell and Salguero- be an important trait in light-limiting terrestrial habi-
Gómez 2013). tats, while area/volume ratio is an important one when
Size in the field can be measured according to dif- considering exposure to the environment). The state
ferent morphometric variables, including but not lim- predictor can be based on ontogeny, rather than differ-
ited to the diameter at breast height (e.g. in trees), ent dimensions of size. Classical categories of ontogeny
basal diameter (e.g. in shrubs), area (e.g. in corals have been developed for plants (Gatsuk et al. 1980),
and sponges), degree of modularity/architecture, and though these ignore non-progressive life cycles. It is
number of stems, and so on (Salguero-Gómez et al. recommended that, if a priori knowledge regarding the
2015). When deciding which morphometric variable to life cycle of the species does not exist, researchers col-
collect data for, it is key to consider various factors, lect multiple axes of information regarding the struc-
such as (1) a priori knowledge on the biology of the ture and ontogeny of the individuals in the first few
species and its life cycle and (2) the traits that most field seasons and quickly construct and compare vital
(a) (b)
(c) (d)
Figure 5.3 Examples of monitoring methods for the estimation of population abundance, density, structure, and vital rates in sessile organisms.
(a) Permanent plots can be used to quantify and track population dynamics of organisms fixed on a substrate (credit R. Salguero-Gómez).
(b) A pantograph allows for the spatially explicit depiction of population (and community) structures in a compact, reliable, and fast way (credit
P. Adler). (c) Quadrats and transects to track populations/individuals can be used on both terrestrial and aquatic systems (credit M. Beger).
(d) UAVs, coupled with artificial intelligence algorithms, are starting to allow for the accurate, fast, and convenient measurement of population
dynamics of sessile organisms. Photos show the visible range (left) vs. hyperspectral range (right) of a photo containing a parasite (blue square),
a mistletoe (credit R. Salguero-Gómez).
rate models to evaluate the best predictors. This infor- locations of individuals in a subset of permanent plots
mation can then help guide a more efficient, fine-tuned and trying to mark them again a week later using dif-
(and less-laborious) field monitoring protocol. ferent methods, and then comparing the accuracy and
The kind of data that need to be collected to exam- feasibility of each. Certain GPS systems that offer reso-
ine the population dynamics of sessile species often lution within centimetres are of particular interest here
requires marking or mapping individuals in a way too (Lee and Ge 2006).
that allows researchers to relocate them in the next Certain sessile organisms with cryptic life stages can
site visit. In terrestrial systems, marking/mapping pose interesting challenges to monitor the dynamics of
can be done with sticks, coffee stirrers, pin flags, their populations. A feasible approach here involves
nails, or tags attached to each individual. Each mark- the careful exploration of such stages on the perma-
er needs an unequivocally distinct ID—this can be nent plot in a nondestructive way. For instance, some
done through marking each device with pen or by species undergo dormancy: the aboveground biomass
producing tags that have a unique ID system. Anoth- is completely lacking, but organisms are alive below-
er key attribute of these IDs is that they need to ground. Dormancy comes in the shape of propagule
‘survive’ themselves to the next census. Often, and (e.g. seed) dormancy, vegetative (e.g. established indi-
depending on the climate at the study location and vidual) dormancy, and roots. In this realm, nondestruc-
the kind of material (and presence of herbivores and tive approaches involve carefully excavating roots or
people!), tags can go missing, and the researcher must bulbs to determine if the individual is alive (Bierzy-
update missing IDs with new tags based on a ‘best chudek 1982). An alternative approach here is to make
guess’ system supported by the GPS coordinates of the assumption that if a number of years have elapsed
the surrounding individuals whenever possible. It is since the last time that the individual was observed
recommended that, when analysing the field data aboveground, this individual is declared dead. In this
(e.g. Chapters 9 and 10), the models should be updat- context, prior knowledge about age-based and envi-
ed with perfect and retagged individuals to evaluate ronmentally driven mortality schedules can be cru-
potential sensitivities of outputs to field assumptions cial. However, recent Bayesian statistical frameworks
on ID assignment. Nonetheless, since markers can go inspired in capture–recapture methods can be applied
missing frequently, photography of the study plots for to cope with this uncertainty (Colchero et al. 2012;
every census is strongly recommended. This photo- Paniw et al. 2017).
graphic evidence can become invaluable when trying Clonality is an important challenge in demograph-
to figure out the location of not only individuals but ic studies. Indeed, the incorporation (or not) of explicit
also entire permanent plots located in dynamic ecosys- clonal architecture in demographic studies has been a
tems, such as dunes and marine environments. point of contention since the very inception of mod-
There are pros and cons of using tags of differ- ern plant population ecology (Harper 1981). The con-
ent materials: plastic and wooden ones can rapidly siderations here are important because, although dif-
degrade, whereas aluminium tags tend to attract the ferent segments (i.e. ramets) of a clonal individual
attention of herbivores and vandalism. An additional (i.e. genet) can have a great deal of independence
approach consists of marking the corners of quadrats (Hutchings and Price 1993; Vuorisalo and Hutchings
with metal bars that can be pounded in the soil/rock 1996), and can even compete against each other (Price
and whose positions can be relocated every visit with et al. 1996), considering them all as separate indi-
the help of metal detectors. Once the positions of the viduals or parts of the same one will affect how the
quadrat (or transect) have been located, the Cartesian individual is tagged, how the models are defined and
coordinates of the individuals should allow researchers constructed, and ultimately the outputs from such
to relocate established individuals and identify new demographic analyses (Janovský et al. 2017; Salguero-
recruits. However, it must be noted that sessile indi- Gómez 2018). For instance, data from the same perma-
viduals, contrary to common belief, do move, if only a nent plots where the functional unit individual has been
little bit. Another important consideration is that met- established ignoring clonal links (i.e. the researcher
al markers can leak materials into the soil and also tags seemingly independent units that are in fact
affect the temperature of its microclimate (Nassar et al. part of the same genet as separate individuals) will
2018). Individuals of species’ populations found in inevitably result in the estimation of higher genera-
high-density, overlapping statures and/or small sizes tion times and more variable population growth rates
can prove particularly challenging to mark and relo- than if done on the basis of genetics (Janovský et al.
cate. In these cases, we suggest marking/mapping the 2017).
5.4 The future of long-term studies: of successfully applying new monitoring technologies
new aspects, current biases, and for short substudies within long-term systems (rather
than integrating within the whole standard protocol) to
arising challenges
address specific hypotheses in new ways, for example
Long-term studies clearly hold a large range of ben- using GPS to examine foraging in the long-term
efits compared to a single visit to the field or com- Kalahari meerkat (Suricata suricatta) project (Gall and
pared to short-term field studies. They play a crucial Manser 2018). Further, the development of technolo-
role in our understanding of the drivers of population gies that allow automated manipulation (rather than
dynamics and in the assessment of their demograph- just monitoring) can provide a platform to carry out
ic status (e.g. decline, remain stable, increase in size individual-level experiments within long-term study
over time). As such, they are important to inform pol- systems, for instance applying automated experimen-
icy (see Chapter 19) and to answer societally relevant tal treatments to individuals based on their unique
questions (Hughes et al. 2017). It is also needed to con- RFID codes in the Wytham tit project (Firth and Shel-
sider the potential biases, the aspects still requiring don 2015). Clearly, updating long-running systems
advances, and the potential arising challenges of these with the ever-developing new approaches provides
systems. For instance, in regard to biases in the systems many possibilities to acquire continued research fund-
currently under long-term individual-based monitoring for these crucially important study systems.
ing, it is notable that species that are relatively easy to Finally, a more contemporary challenge arising in
study using traditional CMR techniques are still heav- the new age of monitoring long-term studies is how
ily favoured over ones that it may be difficult to apply to most efficiently deal with the vast amounts of data
these techniques to. Nevertheless, in the dawn of new arising from these systems. Automatically record-
technologies providing novel avenues of monitoring, it ing information for many individuals within these
could now well be the case that other species are par- populations over large periods of time at fine-scale
ticularly well suited to new methods despite been pre- resolutions is the dream of any population ecologist,
viously difficult to monitor with standard approaches. resulting in the production of millions (Firth and Shel-
As such, diversifying long-term individual-based stud- don 2016) or even billions (Noonan et al. 2015) of data
ies across the Tree of Life may shed new light on points. The analytical techniques developed to handle
ecology from the viewpoint of currently understudied this scale of data, ranging from simply dealing with
species and allow previously unrealised potential to be the raw records (e.g. using artificial intelligence and
recognised. machine-learning algorithms) to drawing biologically
One of the most lasting and widely acknowledged relevant conclusions from the patterns within them
challenges of long-term study systems is their ‘mis- (e.g. through comparisons to simulated null models),
match’ with the modern scientific funding structures are arguably just as important advances as those in
that often work in short-term research grants desig- the monitoring technologies themselves (Krause et al.
nated to proposals aimed at pursuing set hypotheses 2013; Kays et al. 2015).
over relatively short periods of time (often 2–5 years) Vast amounts of data are also produced by citizen
(Clutton-Brock and Sheldon 2010). This fact, combined science. Thanks to the development of the Internet,
with the growing ‘publish-or-perish’ climate across computational techniques and statistical tools, com-
biology and most other scientific disciplines, means puters, and accessible interfaces now allow people
that short-term output is often largely favoured over passionate about wildlife to contribute to long-term
longer-term goals—which can, nonetheless, be better field studies. Thus, beyond biologists, research projects
representative of the ecology and evolution of the stud- are engaging millions of individuals worldwide in
ied population. As such, the success of individual- collecting data (see e.g. https://www.ukbms.org for
based long-term study systems often relies on the the UK Butterfly Monitoring Scheme). The number of
continuation of short-term studies taking place with- citizen science projects has exploded since 2000, and
in their larger framework (Schradin and Hayes 2017). volunteers gathering data have already contributed to
Thus, it is now important that technological advances deliver significant insights into the ecological effects
be employed to address this challenge, rather than of climate change (e.g. Bonney et al. 2014; Isaac et al.
magnify it. 2014). Even more recent, modern computational
Continuously applying new methods to long-term science techniques are opening up the opportunity
study systems may allow for their value to be constant- for ‘iEcology’ which aims to monitor populations
ly recognised. There are already numerous examples through passively generated internet data, such as
using millions of Google search users’ activity to Ampatzidis, Y., V. Partel, B. Meyering, and U. Albrecht.
infer bird species’ occurrences or analysing Wikipedia 2019. Citrus rootstock evaluation utilizing UAV-based
users’ locations to assess fish migration patterns (Jaric remote sensing and artificial intelligence. Comput. Elec-
et al. 2020). tron. Agric. 164: 104900.
Antún, M., R. Baldi, L. M. Bandieri, and R. L. D. Agostino.
Thus, it is essential that biologists/ecologists contin-
2018. Analysis of the spatial variation in the abundance of
ue to ‘keep pace’ with these data analytical techniques
lesser rheas using density surface models. Wildl. Res. 45:
at the same rate that they adopt new data collec-
47–54.
tion methods and new approaches to provide usable Baddeley, A., P. Gregori, J. Mateu, R. Stoica, and D. Stoyan
data storage on large scales, which have been devel- (eds). 2006. Case Studies in Spatial Point Process Model-
oped for various long-term studies (Marshall et al. ing. Springer, New York.
2018). There is important incentive for ecologists to Bai, J., J. Li, and S. Li. 2011. Monitoring the plant density of
work more closely with engineers, computer scientists, cotton with remotely sensed data. In: D. Li, Y. Liu, and
and data scientists. Cross- and interdisciplinary work Y. Chen (eds), Computer and Computing Technologies in
offers promising avenues for future developments in Agriculture IV (pp. 90–101). Springer, Berlin.
algorithms to process big data. Once the data have Beuchel, F. and B. Gulliksen. 2008. Temporal patterns
of benthic community development in an Arctic fjord
been gathered, processed, and stored, the question of
(Kongsfjorden, Svalbard): results of a 24-year manipula-
access to these data is a final and important challenge
tion study. Polar Biol. 31(8): 913–924.
to consider (Mills et al. 2015; Whitlock et al. 2016). Bierzychudek, P. 1982. Life histories and demography of
Whether such databases should be open-access and shade-tolerant temperate forest herbs: a review. New Phy-
available for usage by all or whether certain restrictions tol. 90: 757–776.
should be applied in an attempt to favour the continu- Bonney, R., J. L. Shirk, T. B. Phillips, et al. 2014. Next steps
ation of these systems is currently under much debate. for citizen science. Science 343: 1436–1437.
Although such consideration is obviously very useful, Bonter, D. N. and E. S. Bridge. 2011. Applications of radio
the rapidly growing trend towards unrestricted and frequency identification (RFID) in ornithological research:
immediate access to data across biology (Farnham et al. a review. J. Field Ornithol. 82: 1–10.
Bridge, E. S., K. Thorup, M. S. Bowlin, et al. 2011. Technol-
2017; Culina et al. 2018; Sarabipour et al. 2019) sug-
ogy on the move: recent and forthcoming innovations for
gests that open access to such data is not only valuable
tracking migratory birds. BioScience 61: 689–698.
but inevitable, and emphasis should now be placed on Bruijning, M., M. D. Visser, C. A. Hallmann, and
establishing approaches to recognise and maximise the E. Jongejans. 2018. trackdem: automated particle tracking
benefits of this for individual-based long-term studies. to obtain population counts and size distributions from
videos in r. Methods Ecol. Evol. 9: 965–973.
Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake,
Acknowledgements D. L. Borchers, and L. Thomas (eds). 2007a. Advanced
Distance Sampling: Estimating Abundance of Biological
We acknowledge financial support from the Research
Populations. Oxford University Press, Oxford.
Council of Norway through its Centres of Excellence
Buckland, S. T., D. L. Borchers, A. Johnston, P. A. Henrys,
funding scheme (project number 223257). and T. A. Marques. 2007b. Line transect methods for plant
surveys. Biometrics 63(4): 989–998.
Carraro, L., H. Hartikainen, J. Jokela, E. Bertuzzo, and
References A. Rinaldo. 2018. Estimating species distribution and
Aars, J., T. A. Marques, S. T. Buckland, et al. 2009. Estimat- abundance in river networks using environmental DNA.
ing the Barents Sea polar bear subpopulation size. Mar. PNAS 115: 11724–11729.
Mammal. Sci. 25(1): 35–52. Caswell, H. 2001. Matrix Population Models: Construction,
Aarts, G., J. Fieberg, and J. Matthiopoulos. 2012. Compara- Analysis, and Interpretation. Sinauer Associates, Sunder-
tive interpretation of count, presence–absence and point land, MA.
methods for species distribution models. Methods Ecol. Caswell, H. and R. Salguero-Gómez. 2013. Age, stage and
Evol. 3: 177–187. senescence in plants. J. Ecol. 101: 585–595.
Alves J., A. Alves Da Silva, A. M. V. M. Soares, and Cattadori, I. M., D. T. Haydon, S. J. Thirgood, and
C. Fonseca. 2013. Pellet group count methods to esti- P. J. Hudson. 2003. Are indirect measures of abundance a
mate red deer densities: precision, potential accuracy and useful index of population density? The case of red grouse
efficiency. Mamm. Biol. 78: 134–141. harvesting. Oikos 100: 439–446.
Amos, M., G. Baxter, N. Finch, A. Lisle, and P. Murray. 2014. I Cayuela, H., R. A. Griffiths, N. Zakaria, et al. 2020.
just want to count them! Considerations when choosing a Drivers of amphibian population dynamics and
deer population monitoring method. Wildl. Biol. 20: 362–
370.
asynchrony at local and regional scales. J. Anim. Ecol. 89: Festa-Bianchet, M. 2017. When does selective hunting select,
1350–1364. how can we tell, and what should we do about it? Mam-
Chabot D. and D. M. Bird. 2012. Evaluation of an off-the- mal Rev. 47: 76–81.
shelf unmanned aircraft system for surveying flocks of Field, S. A., A. J. Tyre, K. H., Thorn, P. J. O’Connor,
geese. Waterbirds 35: 170–174. and H. P. Possingham. 2005. Improving the efficiency of
Chirayath, V. and S. A. Earle. 2016. Drones that see through wildlife monitoring by estimating detectability: a case
waves—preliminary results from airborne fluid lensing study of foxes (Vulpes vulpes) on the Eyre Peninsula, South
for centimetre-scale aquatic conservation. Aquat. Con- Australia. Wild. Res. 32: 253–258.
serv. Mar. Freshw. Ecosyst. 26: 237–250. Firth, J. A. and B. C. Sheldon. 2015. Experimental manip-
Clutton-Brock, T. H. and J. M. Pemberton (eds). 2004. Soay ulation of avian social structure reveals segregation is
Sheep: Dynamics and Selection in an Island Population. carried over across contexts. Proc. R. Soc. B Biol. Sci. 282:
Cambridge University Press, Cambridge. 20142350.
Clutton-Brock, T. and B. C. Sheldon. 2010. Individuals and Firth, J. A. and B. C. Sheldon. 2016. Social carry-over effects
populations: the role of long-term, individual-based stud- underpin trans-seasonally linked structure in a wild bird
ies of animals in ecology and evolutionary biology. Trends population. Ecol. Lett. 19: 1324–1332.
Ecol. Evol. 25: 562–573. Firth, J. A., B. L. Verhelst, R. A. Crates, C. J. Garroway,
Clutton-Brock, T. H., F. E. Guinness, and S. D. Albon. 1982. and B. C. Sheldon. 2018. Spatial, temporal and individual-
Red Deer: Behavior and Ecology of Two Sexes. University based differences in nest-site visits and subsequent repro-
of Chicago Press, Chicago. ductive success in wild great tits. J. Avian Biol. 49:
Colchero, F., O. R. Jones, and M. Rebke. 2012. BaSTA: e01740.
an R package for Bayesian estimation of age-specific Fleischmann, D., I. O. Baumgartner, M. Erasmy, et al. 2013.
survival from incomplete mark–recapture/recovery data Female Bechstein’s bats adjust their group decisions about
with covariates. Methods Ecol. Evol. 3: 466–470. communal roosts to the level of conflict of interests. Curr.
Condit, R., R. Pérez, S. Lao, S. Aguilar, and S. P. Hubbell. Biol. 23: 1658–1662.
2017. Demographic trends and climate over 35 years in the Focardi S., B. Franzetti, and F. Ronchi. 2013. Nocturnal dis-
Barro Colorado 50 ha plot. For. Ecosyst. 4: 17. tance sampling of a Mediterranean population of fallow
Connell, J. H., T. P. Hughes, and C. C. Wallace. 1997. A deer is consistent with population projections. Wild. Res.
30-year study of coral abundance, recruitment, and dis- 40: 437–446.
turbance at several scales in space and time. Ecol. Monogr. Frederiksen, M., J.-D. Lebreton, R. Pradel, R. Choquet, and
67: 461–488. O. Gimenez. 2014. REVIEW: identifying links between
Coulson, T., E. A. Catchpole, S. D. Albon, et al. 2001. Age, vital rates and environment: a toolbox for the applied
sex, density, winter weather, and population crashes in ecologist. J. Appl. Ecol. 51: 71–81.
Soay sheep. Science 292: 1528–1531. Fretwell, P. T., M. A. LaRue, P. Morin, et al. 2012. An
Cressie, N., C. A. Calder, J. S. Clark, J. M. V. Hoef, emperor penguin population estimate: the first global,
and C. K. Wikle. 2009. Accounting for uncertainty synoptic survey of a species from space. PLOS One 7:
in ecological analysis: the strengths and limitations e33751.
of hierarchical statistical modeling. Ecol. Appl. 19: Gall, G. E. C. and M. B. Manser. 2018. Spatial structure of
553–570. foraging meerkat groups is affected by both social and
Culina, A., T. W. Crowther, J. J. C. Ramakers, P. Gienapp, ecological factors. Behav. Ecol. Sociobiol. 72: 77.
and M. E. Visser. 2018. How to do meta-analysis of open Garel, M., C. Bonenfant, J. -L. Hamann, F. Klein, and
datasets. Nat. Ecol. Evol. 2: 1053–1056. J. -M. Gaillard. 2010. Are abundance indices derived from
Ebert, T. A. 1998. Plant and Animal Populations: Methods in spotlight counts reliable to monitor red deer Cervus elaphus
Demography. Academic Press, San Diego. populations? Wild. Biol. 16: 77–84.
Eguchi, T. 2014. Estimating the proportion of identifiable Gatsuk, L. E., O. V. Smirnova, L. I. Vorontzova,
individuals and group sizes in photographic identification L. B. Zaugolnova, and L. A. Zhukova. 1980. Age states
studies. Mar. Mammal Sci. 30: 1122–1139. of plants of various growth forms: a review. J. Ecol. 68:
Ekman, J. 1989. Ecology of non-breeding social systems of 675–696.
Parus. Wilson Bull. 101: 263–288. Geesey, G. G., R. Mutch, J. W. Costerton, and R. B. Green.
Ellis, M. M., J. L. Williams, P. Lesica, et al. 2012. Matrix 1978. Sessile bacteria: an important component of the
population models from 20 studies of perennial plant microbial population in small mountain streams 1. Lim-
populations. Ecology 93: 951–951. nol. Oceanogr. 23: 1214–1223.
ENETWILD Consortium, S. Grignolio, M. Apollonio, et al. Gibbons, W. J. and K. M. Andrews. 2004. PIT tagging: simple
2020. Guidance on estimation of abundance and density technology at its best. BioScience 54: 447–454.
data of wild ruminant population: methods, challenges, Grant, P. R. 1986. Ecology and evolution of Darwin´s finches.
possibilities. EFSA Support. Publ. 17: 1876E. Princeton University Press, Princeton, NJ.
Farnham, A., C. Kurz, M. A. Öztürk, et al. 2017. Early career
researchers want Open Science. Genome Biol. 18: 221.
Guillera-Arroita, G. 2017. Modelling of species distributions, Janovský, Z., T. Herben, and J. Klimešová. 2017. Accounting
range dynamics and communities under imperfect detec- for clonality in comparative plant demography—growth
tion: advances, challenges and opportunities. Ecography or reproduction? Folia Geobot. 52: 433–442.
40: 281–295. Jarić, I., R. A. Correia, B. W. Brook, et al. 2020. iEcology:
Hammond, P. S., P. Berggren, H. Benke, et al. 2002. Abun- harnessing large online resources to generate ecological
dance of the harbor porpoise and other cetaceans in insights. Trends Ecol. Evol. 35: 630–639.
the North Sea and adjacent waters. J. Appl. Ecol. 39: Jiménez López, J. and M. Mulero-Pázmány. 2019. Drones
361–376. for conservation in protected areas: present and future.
Harper, J. L. 1981. Population Biology of Plants. Academic Drones 3: 10.
Press, London. Jønsson, K. A., A. P. Tøttrup, M. K. Borregaard, S. A. Keith,
Hayes, L. D. and C. Schradin. 2017. Long-term field stud- C. Rahbek, and K. Thorup. 2016. Tracking animal dis-
ies of mammals: what the short-term study cannot tell us. persal: from individual movement to community assem-
J. Mammal. 98: 600–602. bly and global range dynamics. Trends Ecol. Evol. 31:
Hays, G. C., L. C. Ferreira, A. M. M. Sequeira, et al. 2016. 204–214.
Key questions in marine megafauna movement ecology. Jones, J. P. G. 2011. Monitoring species abundance and dis-
Trends Ecol. Evol. 31: 463–475. tribution at the landscape scale. J. Appl. Ecol. 48: 9–13.
Hays, G. C., H. Bailey, S. J. Bograd, et al. 2019. Translating Kappeler, P. M., F. P. Cuozzo, C. Fichtel, et al. 2017. Long-
marine animal tracking data into conservation policy and term field studies of lemurs, lorises, and tarsiers. J. Mam-
management. Trends Ecol. Evol. 34: 459–473. mal. 98: 661–669.
Hazen, E. L., S. M. Maxwell, H. Bailey, et al. 2012. Ontogeny Karanth, K. U. and J. D. Nichols. 1998. Estimation of tiger
in marine tagging and tracking science: technologies and densities in India using photographic captures and recap-
data gaps. Mar. Ecol. Prog. Ser. 457: 221–240. tures. Ecology 79: 2852–2862.
Hopkins, H. L. and M. L. Kennedy. 2004. An assessment Katsanevakis, S., A. Weber, C. Pipitone, et al. 2012. Mon-
of indices of relative and absolute abundance for moni- itoring marine populations and communities: methods
toring populations of small mammals. Wildl. Soc. Bull. dealing with imperfect detectability. Aquat. Biol. 16:
1973–2006 32: 1289–1296. 31–52.
Howe, E. J., S. T. Buckland, M. L. Després-Einspenner, and Kays, R., M. C. Crofoot, W. Jetz, and M. Wikelski. 2015.
H. S. Kühl. 2017. Distance sampling with camera traps. Terrestrial animal tracking as an eye on life and planet.
Methods Ecol. Evol. 8: 1558–1565. Science 348: aaa2478.
Huey, R. B., M. Carlson, L. Crozier, M. Frazier, H. Hamilton, Kéry, M. and M. Schaub. 2012. Bayesian Population Analysis
C. Harley, A. Hoang, and J. G. Kingsolver. 2002. Plants Using WinBUGS: A Hierarchical Perspective. Academic
versus animals: do they deal with stress in different ways? Press, Boston.
Integr. Comp. Biol. 42: 415–423. Kluijver, H. N. 1951. The population ecology of the great tit,
Hughes, B. B., R. Beas-Luna, A. K. Barner, et al. 2017. Long- Parus m. major L. Ardea 39: 1–135.
term studies contribute disproportionately to ecology and Krause, J., S. Krause, R. Arlinghaus, I. Psorakis, S. Roberts,
policy. BioScience 67: 271–281. and C. Rutz. 2013. Reality mining of animal social systems.
Hussey, N. E., S. T. Kessel, K. Aarestrup, et al. 2015. Aquatic Trends Ecol. Evol. 28: 541–551.
animal telemetry: a panoramic window into the underwa- Krebs, C. J. 1998. Ecological Methodology, 2nd edition. Ben-
ter world. Science 348: 1255642. jamin Cummings, Menlo Park, CA.
Hutchings, M. J. 2010. The population biology of the early La Morgia, V., R. Calmanti, A. Calabrese, and S. Focar-
spider orchid Ophrys sphegodes Mill. III. Demography di. (2015). Cost-effective nocturnal distance sampling
over three decades. J. Ecol. 98: 867–878. for landscape monitoring of ungulate populations. Eur J
Hutchings, M. J. and E. A. C. Price. 1993. Does physio- Wildl. Res, 61: 285–298.
logical integration enable clonal herbs to integrate the Lack, D. 1966. Population studies of birds. Clarendon Press,
effects of environmental heterogeneity? Plant Species Biol. Oxford.
8: 95–105. Lambert, J. P. T., H. L. Hicks, D. Z. Childs, and R. P.
Imperio, S., M. Ferrante, A. Grignetti, G. Santini, and Freckleton. (2018). Evaluating the potential of unmanned
S. Focardi. 2010. Investigating population dynamics in aerial systems for mapping weeds at field scales: a case
ungulates: do hunting statistics make up a good index of study with Alopecurus myosuroides. Weed Res. 58(1): 35–
population abundance? Wild. Biol. 16: 205–214. 45.
Ims, R. A., N. G. Yoccoz, and S. T. Killengreen. (2011). Law, B. S. 2018. Long-term research on forest bats: we have
Determinants of lemming outbreaks. PNAS 108(5): the technology. Aust. Zool. 39: 658–668.
1970–1974. Le Moullec, M. (2019) Spatiotemporal variation in abun-
Isaac, N. J. B., A. J. Van Strien, T. A. August, M. P. De Zeeuw, dance of key tundra species: from local heterogeneity to
and D. B. Roy. 2014. Statistics for citizen science: extract- large-scale synchrony. Trondheim: NTNU (ISBN 978-82-
ing signals of change from noisy ecological data. Methods 326-3653-2) Doctoral theses at NTNU, 2019:20.
Ecol. Evol. 5: 1052–1060.
Le Moullec, M., Å. Ø. Pedersen, N. G. Yoccoz, R. Aanes, Martin, S. W., T. A. Marques, L. Thomas, et al. 2013. Estimat-
J. Tufto, and B. B. Hansen. 2017. Ungulate population ing minke whale (Balaenoptera acutorostrata) boing sound
monitoring in an open tundra landscape: distance sam- density using passive acoustic sensors. Mar. Mammal Sci.
pling versus total counts. Wild. Biol. 2017. 29: 142–158.
Le Moullec, M., Å. Ø. Pedersen, A. Stien, J. Rosvold, and McMahon, L. A., J. L. Rachlow, L. A. Shipley, J. S. Forbey,
B. B. Hansen. 2019. A century of conservation: the ongo- T. R. Johnson, and P. J. Olsoy. 2017. Evaluation of micro-
ing recovery of Svalbard reindeer. J. Wildl. Manag. 83: GPS receivers for tracking small-bodied mammals. PLOS
1676–1686. One 12: e0173185.
Lee, I.-S. and L. Ge. 2006. The performance of RTK-GPS for Miller, D. A., J. D. Nichols, B. T. McClintock, E. H. C. Grant,
surveying under challenging environmental conditions. L. L. Bailey, and L. A. Weir. 2011. Improving occupancy
Earth Planets Space 58: 515–522. estimation when two types of observational error occur:
Linnell, J. D. C., P. Fiske, J. Odden, H. Brseth, I. Herfindal, non-detection and species misidentification. Ecology 92:
and R. Andersen. 2007. An evaluation of structured 1422–1428.
snow-track surveys to monitor Eurasian lynx Lynx lynx Mills, J. A., C. Teplitsky, B. Arroyo, et al. 2015. Archiving pri-
populations. Wild. Biol. 13: 456–466. mary data: solutions for long-term studies. Trends Ecol.
Loison, A., J. Appolinaire, J. -M. Jullien, and D. Dubray. Evol. 30: 581–589.
(2006). How reliable are total counts to detect trends Morellet, N., F. Klein, and R. Andersen. 2011. The census and
in population size of chamois Rupicapra rupicapra and management of populations of ungulates in Europe. In:
R. pyrenaica? Wild. Biol., 12: 77–88. R. Putman, M. Apollonio, and R. Andersen (eds), Ungu-
Lollback,G. W., J. E. Dunwoody, R. Mebberson, et al. late Management in Europe (pp. 106–143). Cambridge
2018. Spatial modelling of bilby (Macrotis lagotis) University Press, Cambridge.
and rabbit (Oryctolagus cuniculus) pellets with- Nassar, J. M., S. M. Khan, D. R. Villalva, M. M. Nour,
in a predator-proof enclosure. Austr. Mamm. 40: A. S. Almuslem, and M. M. Hussain. 2018. Compliant
93–102. plant wearables for localized microclimate and plant
Macdonald, D. W. and C. Newman. 2002. Population growth monitoring. Npj. Flex. Electron. 2: 1–12. Nature
dynamics of badgers (Meles meles) in Oxfordshire, U.K.: Publishing Group.
numbers, density and cohort life histories, and a possible Nisbet, I. C. T. 1989. Long-term ecological studies of
role of climate change in population growth. J. Zool. 256: seabirds. Colon. Waterbirds 12: 143–147.
121–138. Noonan, M. J., A. Markham, C. Newman, et al. 2015. A
MacKenzie, D. I., J. D. Nichols, N. Sutton, K. Kawanishi, new Magneto-Inductive tracking technique to uncover
and L. L. Bailey. 2005. Improving inferences in popula- subterranean activity: what do animals do underground?
tion studies of rare species that are detected imperfectly. Methods Ecol Evol. 6: 510–520.
Ecology 86: 1101–1113. Nowacek, D. P., F. Christiansen, L. Bejder, J. A. Goldbo-
Maillard, D., J.-M. Gaillard, M. Hewison, et al. 2010. Ungu- gen, and A. S. Friedlaender. 2016. Studying cetacean
lates and their management in France. In: M. Apollonio, behaviour: new technological approaches and conserva-
R. Andersen, and R. Putman (eds), European Ungulates tion applications. Anim. Behav. 120: 235–244.
and their Management in the 21st Century (pp. 441–474). O’Brien, T. G. 2011. Abundance, density and relative
Cambridge University Press, Cambridge. abundance: a conceptual framework. In: A. F. O’Connell,
Malhi, Y., T. Jackson, L. Patrick Bentley, et al. 2018. New J. D. Nichols, and K. Ullas Karanth (eds), Camera
perspectives on the ecology of tree structure and tree Traps in Animal Ecology (pp. 71–96). Springer,
communities through terrestrial laser scanning. Interface New York.
Focus 8: 20170052. O’Connor, R. J. 1991. Long-term bird population studies in
Mann, J. and C. Karniski. 2017. Diving beneath the surface: the United States. Ibis 133: 36–48.
long-term studies of dolphins and whales. J. Mammal. 98: Paniw, M., P. F. Quintana-Ascencio, F. Ojeda, and
621–630. R. Salguero-Gómez. 2017. Accounting for uncertain-
Marques, F. F. C., S. T. Buckland, D. Goffin, et al. (2001). ty in dormant life stages in stochastic demographic
Estimating deer abundance from line transect surveys of models. Oikos 126: 900–909.
dung: sika deer in southern Scotland. J. Appl. Ecol. 38 (2): Parsons, M., D. Bratanov, K. J. Gaston, F. Gonzalez.
349–363. 2018. UAVs, hyperspectral remote sensing, and machine
Marques, T. A., L. Thomas, S. W. Martin, et al. 2013. Estimat- learning revolutionizing reef monitoring. Sensors 2018
ing animal population density using passive acoustics. (18): 2026.
Biol. Rev. Camb. Philos. Soc. 88: 287–309. Pedersen, Å. Ø., B.-J. Bårdsen, N. G. Yoccoz, N. Lecomte, and
Marshall, H. H., D. J. Griffiths, F. Mwanguhya, et al. 2018. E. Fuglei. 2012. Monitoring Svalbard rock ptarmigan: dis-
Data collection and storage in long-term ecological and tance sampling and occupancy modeling. J. Wild. Manag.
evolutionary studies: the Mongoose 2000 system. PLOS 76(2): 308–316.
One 13: e0190740.
Perrins, C. M., J. D. Lebreton, and G. J. M. Hirons. 1991. Salguero-Gómez, R., O. R. Jones, C. R. Archer, et al.
Bird Population Studies: Relevance to Conservation and 2015. The compadre Plant Matrix Database: an open
Management. Oxford University Press, Oxford. online repository for plant demography. J. Ecol. 103:
Peters, A. J. and M. D. Eve. 1995. Satellite monitoring of 202–218.
desert plant community response to moisture availability. Sarabipour, S., H. J. Debat, E. Emmott, S. J. Burgess,
Environ. Monit. Assess. 37: 273–287. B. Schwessinger, and Z. Hensel. 2019. On the value of
Pollock, K. H., J. D. Nichols, T. R. Simons, G. L. Farnsworth, preprints: an early career researcher perspective. PLOS
L. L. Bailey, and J. R. Sauer. 2002. Large scale wildlife mon- Biol. 17: e3000151.
itoring studies: statistical methods for design and analysis. Schaub, M., A. Aebischer, O. Gimenez, S. Berger, and
Environmetrics 13: 105–119. R. Arlettaz. 2010. Massive immigration balances high
Pomilla, C., A. R. Amaral, T. Collins, et al. 2014. The world’s anthropogenic mortality in a stable eagle owl popu-
most isolated and distinct whale population? Humpback lation: lessons for conservation. Biol. Conserv. 143:
whales of the Arabian Sea. PLOS One 9: e114162. Public 1911–1918.
Library of Science. Schauster, E. R., Gese, E. M., and A. M. Kitchen. 2002. An
Poole, K. G., C. Cuyler, and J. Nymand. (2013). Eval- evaluation of survey methods for monitoring swift fox
uation of caribou Rangifer tarandus groenlandicus sur- abundance. Wildl. Soc. Bull. 30(2): 464–477.
vey methodology in West Greenland. Wild. Biol. 19(3): Schlicht, L., M. Valcu, and B. Kempenaers. 2015. Male
225–239. extraterritorial behavior predicts extrapair paternity pat-
Price, E. A. C., M. J. Hutchings, and C. Marshall. 1996. tern in blue tits, Cyanistes caeruleus. Behav. Ecol. 26:1404–
Causes and consequences of sectoriality in the clonal herb 1413.
Glechoma hederacea. Vegetatio 127: 41–54. Schorr, R. A. (2013). Using distance sampling to esti-
Ramsey, D. S. L., P. A. Caley, and A. Robley. 2015. Esti- mate density and abundance of Saussurea weberi
mating population density from presence–absence data hultén (Weber’s saw-wort). Southwest. Naturalist 58(3):
using a spatially explicit model. J. Wildl. Manag. 79: 378–383.
491–499. Schradin, C. and L. D. Hayes. 2017. A synopsis of long-term
Rehmeier, R. L., G. A. Kaufman, and D. W. Kaufman. 2006. field studies of mammals: achievements, future directions,
An automatic activity-monitoring system for small mam- and some advice. J. Mammal. 98: 670–677.
mals under natural conditions. J. Mammal. 87: 628–634. Seber, G. A. F. 1992. A review of estimating animal abun-
Reinke, B. A., D. A. Miller, and F. J. Janzen. 2019. What dance II. Int. Stat. Rev. Rev. Int. Stat. 60:129–166.
have long-term field studies taught us about population Silcox, D. E., J. P. Doskocil, C. E. Sorenson, and
dynamics? Annu. Rev. Ecol. Evol. Syst. 50: 303–333. R. L. Brandenburg. 2011. Radio Frequency Identifica-
Rowcliffe, J. M., J. Field, S. T. Turvey, and C. Carbone. 2008. tion Tagging: a Novel Approach to Monitoring Surface
Estimating animal density using camera traps without and Subterranean Insects. Am. Entomol. 57:86–93.
the need for individual recognition. J. Appl. Ecol. 45: Sillett, T. S., R. B. Chandler, J. A. Royle, M. Kéry,
1228–1236. and S. A. Morrison. 2012. Hierarchical distance-sampling
Royle, J. A. and R. M. Dorazio. 2008. Hierarchical Modeling models to estimate population size and habitat-specific
and Inference in Ecology: the Analysis of Data from Pop- abundance of an island endemic. Ecol. Appl. 22:
ulations, Metapopulations and Communities. Academic 1997–2006.
Press, Amsterdam. Sinclair, A. R. E., J. M. Fryxell, and G. Caughley. 2006.
Royle, J. A., M. Kery, R. Gautier, and H. Schmid. (2007). Hier- Wildlife Ecology, Conservation and Management, 2nd
archical spatial models of abundance and occurrence from edition. Wiley-Blackwell, Malden, MA.
imperfect survey data. Ecol. Monogr. 77(3): 465–481. Smale, D. A., G. A. Kendrick, E. S. Harvey, et al. 2012.
Royle, J. A., R. B. Chandler, R. Sollmann, and B. Gard- Regional-scale benthic monitoring for ecosystem-based
ner. 2013. Spatial Capture–Recapture. Academic Press, fisheries management (EBFM) using an autonomous
Amsterdam. underwater vehicle (AUV). ICES J. Mar. Sci. 69:
Saha, A. K., J. Saha, R. Ray, S. Sircar, S. Dutta, 1108–1118.
S. P. Chattopadhyay, and H. N. Saha. 2018. IOT-based Smith, J. E., K. D. S. Lehmann, T. M. Montgomery,
drone for improvement of crop quality in agricultural E. D. Strauss, and K. E. Holekamp. 2017. Insights from
field. In: Proceedings of the 2018 IEEE 8th Annual Com- long-term field studies of mammalian carnivores. J. Mam-
puting and Communication Workshop and Conference mal. 98: 631–641.
(CCWC) (pp. 612–615). IEEE, New York. Sutherland, W. J. 2006. Ecological Census Techniques Hand-
Salguero-Gómez, R. 2018. Implications of clonality for age- book, 2nd edition. Cambridge University Press, Cam-
ing research. Evol. Ecol. 32: 9–28. bridge.
Salguero-Gómez, R., H. Kempenich, I. N. Forseth, and Sutherland, W. J., R. P. Freckleton, H. C. J. Godfray, et al.
B. B. Casper. (2014). Long-term individual-level popula- 2013. Identification of 100 fundamental ecological ques-
tion dynamics of a native desert chamaephyte. Ecology tions. J. Ecol. 101: 58–67.
95(2): 577.
Svensson, J. R. and D. J. Marshall. 2015. Limiting resources Vuorisalo, T. and M. J. Hutchings. 1996. On plant sectori-
in sessile systems: food enhances diversity and growth of ality, or how to combine the benefits of autonomy and
suspension feeders despite available space. Ecology 96: integration. Vegetatio 127: 3–8.
819–827. Ward, J. A., L. Thomas, S. Jarvis, et al. 2012. Passive acoustic
Taig-Johnston, M., M. K. Strom, K. Calhoun, K. Nowak, L. A. density estimation of sperm whales in the Tongue of the
Ebensperger, and L. Hayes. 2017. The ecological value of Ocean, Bahamas. Mar. Mammal Sci. 28: E444–E455.
long-term studies of birds and mammals in Central Amer- Whitlock, M. C., J. L. Bronstein, E. M. Bruna, et al. 2016. A
ica, South America and Antarctica. Rev. Chil. Hist. Nat. Balanced Data Archiving Policy for Long-Term Studies.
90: 7. Trends Ecol. Evol. 31: 84–85.
Teitelbaum, C. S. and T. Mueller. 2019. Beyond migration: Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Anal-
causes and consequences of nomadic animal movements. ysis and Management of Animal Populations. Academic
Trends Ecol. Evol. 34: 569–581. Press, San Diego.
Tepley, A. J., J. R. Thompson, H. E. Epstein, and Williams, P. J., M. B. Hooten, J. N. Womble, and M. R. Bower.
K. J. Anderson-Teixeira. 2017. Vulnerability to forest 2017. Estimating occupancy and abundance using aerial
loss through altered postfire recovery dynamics in a images with imperfect detection. Methods Ecol. Evol. 8:
warming climate in the Klamath Mountains. Glob. 1679–1689.
Change Biol. 23: 4117–4132. Yoccoz, N. G., J. D. Nichols, and T. Boulinier. 2001. Monitor-
Thompson, W. L. 2004. Sampling Rare or Elusive Species: ing of biological diversity in space and time. Trends Ecol.
Concepts, Designs, and Techniques for Estimating Evol. 16: 446–453.
Population Parameters, 2nd edition. Island Press, Yuan, Y., F. E. Bachl, F. Lindgren, et al. 2017. Point process
Washington. models for spatio-temporal distance sampling data from
Thompson, W. L., G. C. White, and C. Gowan. 1998. a large-scale survey of blue whales. Ann. Appl. Stat. 11:
Monitoring Vertebrate Populations. Academic Press, San 2270–2297.
Diego. Žádníková, P., D. Smet, Q. Zhu, D. V. D. Straeten, and
Tomkiewicz, S. M., M. R. Fuller, J. G. Kie, and K. K. Bates. E. Benková. 2015. Strategies of seedlings to overcome
2010. Global positioning system and associated technolo- their sessile nature: auxin in mobility control. Front. Plant
gies in animal behaviour and ecological research. Philos. Sci. 6: 218.
Trans. R. Soc. Lond. B Biol. Sci. 365: 2163–2176. Zuur, A. F., E. N. Ieno, N. Walker, A. A. Saveliev, and
Urian, K., A. Gorgone, A. Read, et al. 2015. Recommenda- G. M. Smith. 2009. Mixed Effects Models and Extensions
tions for photo-identification methods used in capture- in Ecology with R. Springer, New York.
recapture models with cetaceans. Mar. Mammal Sci. 31:
298–321.
CHAPTER 6
Drivers of demography: past

challenges and a promise
for a changed future
Pedro F. Quintana-Ascencio, Eric S. Menges, Geoffrey S. Cook,
Johan Ehrlén, and Michelle E. Afkhami
6.1 Population dynamics and biotic critically summarise examples of the main approach-
and abiotic drivers es and identify major accomplishments, challenges,
and limitations. We point to promising approach-
Contemporary demographic research is a daunting es and possible future developments. In the initial
but significant and stimulating endeavour. Global sections, we consider models in closed systems with-
change due to unsustainable human resource acqui- out migration among populations. Later we focus on
sition and consumption is responsible for increasing metapopulation models, emphasising the importance
habitat degradation and fragmentation, alteration of of understanding drivers affecting migration and dif-
disturbance regimes, and shifting climatic conditions. ferential extinction among populations. Finally, we
Even abundant species suffer population declines, and conclude with a discussion of some important and
many others are at risk of extinction (Ceballos et al. general problems associated with assessing relation-
2017). There is a pressing need to understand how ships between environmental drivers and population
populations and metapopulations respond to changes dynamics.
in the environment to mitigate the consequences of
these trends.
Although many demographic models assume deter- 6.2 Challenges with incorporating
ministic or stationary stochastic variation (as inde- drivers into structured population
pendent and identically distributed variables) in vital
rates, these rates and thus population dynamics are
models
nearly always driven by environmental factors. Iden- Demographic studies often require long-term and
tifying the drivers of stochastic population dynamics intensive data collection, but, for most species, data
will often be very helpful in projections and predic- collection has been limited to few sampling years and
tions. If environmental conditions are changing direc- few populations (Menges 2000; Crone et al. 2011, 2013;
tionally or cyclically (i.e. are nonstationary), then we Salguero-Gómez et al. 2015; Gurevitch et al. 2016)
need to identify drivers and changes in drivers to mod- and often collected without proper sampling (e.g.
el population dynamics. In this chapter, we emphasise not representative of the focal systems, subjectively
structured population models, but many of these con- and/or inconsistently sampled). Demographic studies
siderations are relevant for other types of population require the collection of enough information allowing
models. We present a nonexhaustive examination on identifying patterns and associations among variables
how environmental drivers are incorporated in demo- to contribute to the understanding of population
graphic modelling across many types of organisms. We dynamics. Sampling from only a few populations
Pedro F. Quintana-Ascencio et al., Drivers of demography: past challenges and a promise for a changed future.
In: Demographic Methods Across the Tree of Life. Edited by Roberto Salguero-Gómez and Marlène Gamelon,
or years may not capture critical but uncommon populations. In fact, most populations never reach
environmental conditions (e.g. less-optimal sites, equilibria (Crone et al. 2013), and short-term transitory
recently disturbed sites, smaller populations) that (nonequilibrium) dynamics in these species may have
produce different trajectories of population change. a strong influence on population dynamics and persis-
Insufficient spatial and temporal sampling limits the tence (Kendall and Fox 2002). This has many practical
inferences that can be drawn for the effects of drivers ramifications, including challenging our ability to
on populations. For example, species with sporadic understand and evaluate future population changes in
recruitment, source–sink dynamics, or populations rapidly changing environments.
with disproportionate reproductive contributions can Incorporating drivers into population modelling is
result in biased population growth estimates. challenging. It requires identifying the most relevant
Concomitantly to the evaluation of the effect of drivers, estimating their relative effects on vital rates
drivers, there are basic statistical and modelling con- and population dynamics, and recognising their most
siderations. Estimates of critical vital rates should be informative modelling frameworks (e.g. data likeli-
assembled with sufficiently large sample sizes. Some hood distributions, variable interactions, number of
life history stages may be infrequent but important bins in matrices, number of stages). Each driver may
to demography (e.g. large individuals may contribute affect multiple demographic transitions in different
disproportionately to fecundity). Some life history ways. Drivers can also interact with each other in their
stages, such as seed banks and dormant stages in plants effects on vital rates and population dynamics (e.g. Tye
and relatively ephemeral planktonic stages in marine et al. 2016). Although in some systems key drivers of
organisms, may be very hard to study without com- population dynamics are easily identified, the number
plex experiments or sophisticated detection methods. of environmental factors that potentially drive the pop-
Nonetheless, these stages may be key to population ulation dynamics of a focal species in many systems
dynamics (Shefferson et al. 2018; Che-Castaldo et al. is large. For logistical reasons, it is often impossible to
2020). For example, persistent seed banks and dormant examine the effects of all potential drivers. Fitting a
stages in vascular plants may make major contribu- large number of environmental predictors in statistical
tions to population growth and may be particularly models of a limited number of replicates poses oth-
important in buffering populations against extinction er problems. As a consequence, how we select which
during unfavourable conditions (Doak et al. 2002). drivers to study will, of course, affect our conclusions
Small sample sizes may provide poor estimates of vital about the study system. Attentive consideration to
rate means and variances. Covariances of different life major regional disturbances, limiting factors, and the
stage transitions will require much larger sample sizes main human impacts should aid in the identification
than are generally collected, but these covariances may of relevant drivers. Understanding the demograph-
buffer variation within the life cycle (if covariances ic effects of major drivers, particularly those that can
are negative) or create more extreme dynamics than be manipulated or mitigated, can lead to promising
would be predicted from one vital rate alone (if pos- conservation and management strategies.
itive). Careful attention to the correct specification of One challenge associated with assessing the effects
the focal species’ life history during model construc- of climatic variables is that they are multidimension-
tion is fundamental for reliable inference and useful al. For example, effects of temperature might differ
projections (Kendall et al. 2019). among different periods of the year, and the minimum,
Until recently, a significant portion of population mean, maximum, or range might be most relevant for
modelling summarised vital rates from data collected different species. Another difficulty is that all these
without regard to temporal and spatial environmen- aspects of temperature might be correlated to each oth-
tal variation (Ehrlén and Morris 2015). This means er and to other climatic variables, such as precipitation
changes in vital rates and population sizes cannot and snow cover, making strong inferences about exact
be linked to environmental changes or management causality difficult. One approach suggested to avoid
regimes, thus limiting our ability to understand their arbitrary selection of variables when there are many
influence on populations. Population projections of potential drivers of population dynamics is to use func-
deterministic modelling using single matrices and the tional smoothing methods (Teller et al. 2016). These
asymptotic parameters derived from these models can- methods can, for example, explore how climate during
not summarise population change in realistic varying different time windows before an event influence the
environments. Few environments will have such lit- response. Lastly, a major challenge with assessing the
tle variation that they produce consistently changing effects of climatic variables is the temporal scale; years
D R I V E R S O F D E M O G R A P H Y: PA S T C H A L L E N G E S A N D A P R O M I S E F O R A C H A N G E D F U T U R E 117
are usually the replicates, and sufficiently long time is challenging in practice, and most studies include
series are required to disentangle the different effects. only one or a few drivers. Of 136 examined plant demo-
Dagleish et al. (2011) used data from 19 years to exam- graphic studies examining the effects of environmental
ine the effects of climate on three species in a sage- drivers published from 1995 to 2015, the mean number
brush steppe community. They found that the effects of examined drivers was 1.3, more than 70% examined
of precipitation, late-winter snowfall, and summer only one driver, and only 14% of the studies included
temperatures were important to different species. In effects of intraspecific density (Ehrlén et al. 2016).
all species, the responses of population growth rates to Given the problems associated with observational
climate lagged one year or more, suggesting that mod- demography studies, one important complementary
els examining the effects of climatic drivers, as well as way to identify and quantify drivers is by carrying
of other types of drivers, need to consider time-lagged out experiments. Experimental approaches allow the
effects. detection of causal relationships that are hidden in
Assessments of the effects of drivers are usually observational data, corroborate causal relationships
based on correlations between levels of drivers and suggested by correlations, and break up correlations
vital rates and population growth rates across sites or between different environmental drivers. There are
years. Collecting information about variation in differ- two main types of experimental approaches to exam-
ent types of environmental drivers is often relatively ine the effects of environmental drivers. First, we can
easy, and in most cases it represents considerably less manipulate the level of the focal driver at a given site.
work than collecting the demographic data. From that Second, we may carry out transplantation experiments
perspective, it is surprising that many demographic and move species into new environments, representing
studies collect data on only a single driver or no drivers gradients of environmental drivers.
at all. The main limitations of assessing the effects Experimental manipulations of drivers of plant
of environmental drivers using correlations between species in the field have been done relatively frequent-
driver levels and population growth rates lie in iden- ly. Examples of experimental approaches include the
tifying causality and in replicating labour-intensive effects of land use (e.g. Sletvold et al. 2013), herbivory
demographic studies across a sufficient number of sites (e.g. Eckberg et al. 2014), mutualists (e.g. Yule et al.
or over a sufficient number of years. 2013), pathogens (e.g. Roy et al. 2011), competition (e.g.
Prevéy et al. 2010), climatic factors (e.g. Williams et al.
2007), and other abiotic drivers (e.g. Scanga 2014). To
6.3 Quantifying drivers by observation
be able to quantitatively assess the sensitivity of vital
and by experiments rates and population growth rate to changes in a driver
There are two principal ways to quantify the effects from the results of an experimental manipulation, it is
of environmental drivers on vital rates: either through necessary to quantify the level of the driver in the dif-
observations or with experiments. Most studies inves- ferent treatments. However, only 17% of studies that
tigating the effects of environmental drivers on plant used experimental manipulations quantified the lev-
and animal population dynamics use only observation- el of the drivers (Ehrlén et al. 2016). One study (da
al data (Heppell et al. 2000; Ehrlén et al. 2016; Cayuela Silveira Pontes et al. 2012) quantified browsing treat-
et al. 2018). Observational data have the advantage of ments in terms of total edible stem biomass removed,
being relevant in the sense that they examine perfor- finding that only heavy browsing was able to reverse
mance over the actual range of environmental condi- positive population growth rate in the invasive shrub
tions as they are experienced by the focal organisms. species Cytisus scoparius. Still, most studies examin-
However, causal relationships between environmental ing the effects of grazing have compared performance
drivers and vital rates and population growth rates are in plots with ambient levels of grazing against plots
difficult to detect and establish with purely observa- where grazers are excluded, without quantifying graz-
tional studies because different drivers are correlated ing intensity in the treated plot. The lack of informa-
with each other and to intraspecific population den- tion about the levels of the driver means that most
sity. This is because we expect population density to experimental studies carried out only have provided
be positively correlated with carrying capacity, which qualitative information about whether the driver has
in turn is correlated with the level of the environ- an effect or not. Experimental manipulation of popula-
mental drivers. Although in principle it is possible tion drivers of animal species has been less frequent,
to address these problems by including all relevant and examples include effects of population density
drivers and intraspecific densities in the analyses, this (Aars and Ims 2002), food supplementation (Dobson
and Oli 2001), disturbance (Perez-Jorge et al. 2016), and the same factor is not. It is also true that differences in
population bycatch by fisheries (Genovart et al. 2017). environmental conditions might have different effects
A limitation in experimental approaches is that depending on whether they appear every year (as for
parameter combinations used are unlikely ever to spatial variation) or in single years (as for temporal
occur naturally, meaning that the relevance of the variation). Another method is to use information about
results to natural conditions is questionable. However, the effects of the driver on individual performance and
the potential strength of experimental manipulations then estimate the effect of the driver at the population
of drivers in the field is that they enable us to exam- level by simulating the proportion of individuals expe-
ine the effects of environmental conditions that do not riencing different levels of the driver (e.g. Garcia et al.
currently exist but are likely to occur in the future (e.g. 2010).
due to climate change) or at other places. For example, Levels of drivers have been altered in situ, but trans-
Scanga (2014) examined the importance of light regime plantation experiments moving organisms into new
for population growth rates of the gap-dependent environments also allow the exposure of organisms
plant Trollius laxus and extended the gradient of to contrasting conditions. Most transplantation exper-
light availability present in natural subpopulations by iments have focused on one or a few stages of the
including experimentally created canopy gaps. Along life cycle (Hargreaves et al. 2014). However, if these
this extended gradient, she found a unimodal rela- transplantations involve all life cycle stages and are
tionship, with the highest population growth rates at designed to cover gradients of putative environmental
intermediate levels of diffuse light. drivers, then they can provide us with useful infor-
In field experiments, the level of one or several focal mation about the effects of drivers. Transplantation
drivers is manipulated while other factors are allowed experiments also have the advantage that they, if prop-
to vary naturally. We might grow plants under con- erly randomised, break up correlations between the
trolled conditions, for example in a common garden history of an individual and effects of exposure to
and keep the level of all nonfocal drivers at similar lev- current environmental conditions. Finally, transplan-
els. Researchers have conducted common garden/lab- tation experiments allow us to break up correlations
based experiments with corals to assess resilience of between environmental conditions and density that
different genotypes of symbiodinium to thermal stress might bias estimates of driver effects in observational
and outplanting from coral ‘nurseries’ to various habi- studies. One limitation with transplantation experi-
tats to try to restore/reintroduce coral populations. ments is that they do not necessarily break up cor-
However, controlling the level of nonfocal drivers relations among different environmental drivers, thus
often means the effects of many interactions that are suffering partially from the same limitations as obser-
relevant in the field are excluded and that conditions vational studies. In spite of their potential usefulness,
thus might be somewhat artificial. few demographic studies have transplanted organ-
The effects of some drivers, such as soil nutrients, isms along environmental gradients. One example is
soil moisture, and light availability, are often explored Stevens and Latimer (2015), who used a field transplant
using spatial variation over different scales (e.g. in experiment to 14 sites to examine the importance of
terms of variation among populations or patches with- snowpack and disturbance in limiting montane inva-
in populations). The effects of climatic factors, such as sions of two nonnative lowland shrubs, Cytisus scopar-
temperature and precipitation, are mostly examined ius and Spartium junceum. Based on demographic data
using variation among years. Sometimes an integrative analysed by matrix population models, they concluded
approach where both spatial and temporal variation that future predictions of reductions in winter snow-
are combined and evaluated with mixed-model regres- pack and increases in forest disturbance are likely to
sions may be used to assess the effects of drivers such increase the risk of invasion.
as climatic factors and herbivory on population growth Given that it is difficult to know how relevant
rates (e.g. Doak and Morris 2010; Eckhart et al. 2011). the parameter combinations used in experiments are
A critical assumption in such studies is that spatial and for natural conditions, experimental manipulations
temporal variations in drivers have the same effects should ideally be used in conjunction with observa-
on demography. However, this might not always be tional approaches. However, only a small proportion of
true. For example, spatial variation might be corre- studies examining the effects of environmental drivers
lated both with other environmental factors and with on plant population dynamics have used a combi-
intraspecific density in ways that temporal variation in nation of natural variation and experiments (Ehrlén
et al. 2016). In the study by Scanga (2014) mentioned et al. (2014) compared the effects of five different
above, effects of diffuse light were examined using cutting treatments on the performance of the toxic
a combination of natural variation in light availabili- grassland plant Colchicum autumnale. They found that
ty and experimentally created gaps. In another study, population growth rate was significantly lower in plots
Dahlgren and Ehrlén (2011) examined the effect of soil cut earlier in the season.
potassium concentration on the performance of the for- Given that competition is assumed to be one
est herb Actaea spicata, using both natural variation of the most important factors structuring ecological
and experimental potassium additions. Both observa- communities, surprisingly few demographic stud-
tional data and experimental treatments indicated that ies have examined the effects of interspecific and
higher concentrations of potassium increased individ- intraspecific competition. Studies having examined
ual growth and population growth rates. the role of interspecific competition differ consider-
ably regarding how competition has been quantified;
measures used include vegetation height, neighbour
6.4 Continuous drivers (‘stress’)
density, presence of a specific neighbour species, or
Environmental drivers constitute a gradient from those more detailed measures of the neighbouring commu-
that change in a continuous fashion to those involv- nity. It is worth noting that studies examining the
ing relatively discrete events that disrupt vital rates. effects of canopy cover or shading also might be
This difference makes them amenable to distinct ana- regarded as having examined the effects of compe-
lytical and modelling approaches. In this section, we tition, albeit effects mediated by a particular abiotic
discuss drivers that vary continuously, and in section factor. One example of a study with a detailed mea-
6.5 we discuss those involving discrete events. We give sure of interspecific density is Adler et al. (2012),
an overview of the continuous drivers that have been which used 22 years of demographic data to explicit-
investigated in demographic studies, how they have ly quantify both direct effects of climate and indirect
been studied, and some potential problems associated effects mediated by interspecific competition with oth-
with specific drivers. er species for four species in a sagebrush steppe com-
The effects of a range of climatic factors, includ- munity. They found strong effects of climate acting via
ing temperature, precipitation, rainfall anomaly, and the abundance of neighbouring species, showing that
snowfall, on population growth rate have been exam- the relative importance of interspecific competition
ined in many demographic studies. Ehrlén et al. (2016) decreased with increased differences in niches between
found that climatic factors constituted 15.5% of all species.
drivers examined for plants. Given the many abiot- Other types of antagonistic species interactions have
ic factors other than climate that are expected to be also been examined, including herbivory, predation,
drivers of population dynamics, it is somewhat sur- and pathogens. For studies of plant demography, her-
prising that only ~10% of the studies examining the bivory is one of the most-often investigated drivers
effects of drivers on plant population growth rate (e.g. Maron and Crone 2006). The way in which intensi-
included such effects (Ehrlén et al. 2016). Still, a wide ty of herbivory has been quantified varies considerably
range of nonclimatic abiotic factors have been inves- among studies. Many experimental studies have com-
tigated. These include soil water content, dissolved pared ambient levels of herbivory with some type of
oxygen level, soil nutrients, soil type, pH, CO2 lev- herbivore exclusion (Eckberg et al. 2014), while corre-
els, altitude, flooding, inclination, aspect, rock type, lational studies usually estimate intensity as the pro-
and light availability. Like climatic variables, other abi- portion of individuals attacked or average level of
otic variables might often be strongly intercorrelated, damage, for example, in terms of proportion of biomass
and including all relevant variables might be diffi- removed. For studies of predispersal seed predation,
cult. Dahlgren and Ehrlén (2009) examined the effect of the proportion of seeds damaged at the population lev-
eight abiotic drivers on Actaea spicata, a perennial herb, el is often used as an estimate. There are few studies of
finding that only soil potassium concentrations had a the effects of pathogens, but Davelos and Jarosz (2004)
clear effect on population growth rates. used matrix models to examine how chestnut blight
Effects of land use have frequently been examined infection affects population growth rates of American
in terms of intensity of grazing and different mow- chestnuts (Castanea dentata).
ing regimes. Typically, these studies compare areas Incorporation of positive interactions (e.g. mutual-
exposed to different grazing or cutting regimes. Winter ism and facilitation) can have important consequences
for demography of the participating species, although that live in the aboveground tissue of grasses, find-
relatively few studies have examined these effects in ing that these resource provisioning and defensive
population models. Ehrlén et al. (2016) found that mutualists generally increase population growth rates
studies considering mutualism made up only 5.5% of (Yule et al. 2013; Chung et al. 2015).
all drivers examined in environmentally explicit pop-
ulation models of plants. These studies of mutualistic
effects on population dynamics were focused on inter-
6.5 Episodic drivers (‘disturbance’)
actions between plants and animals (pollinators, ants, Drivers that have been characterised by alternative
and dispersal partners). For example, Parker (1997) environmental states include ecological disturbances
simulated the consequences of increased pollination on and site differences. In this section, we explore different
the growth of new populations of an invasive plant, approaches to evaluating the effects of episodic drivers
Cytisus scoparius, by iterating transition matrices over on demography, using fire, hurricanes, harvesting,
short time scales. Her work predicted minimal effects drought, species interactions, and multiple distur-
of pollen limitation for the slow-growing urban pop- bances as instructive examples.
ulations of this invasive plant in contrast to the large, Fire is a dominant ecological disturbance through-
positive effects of increased pollinator visitation pre- out the world (Bond and Keeley 2005) and has strong
dicted for the rapidly growing prairie populations, effects on ecosystems, communities, and populations
thereby considering effects of species interactions and (Bond et al. 2005; Koltz et al. 2018). It is also a strong
land use on population demography. In a paper inves- evolutionary force (Bond and Keeley 2005) and is fre-
tigating a defensive mutualism, Palmer et al. (2010) quently manipulated by humans, through either fire
incorporated the effects of a whole community of ant suppression or prescribed fire. Fire severity and fre-
species on population growth simulations, finding that quency are changing in many parts of the world due to
Acacia drepanolobium, interacting with all four members climate change (Bowman et al. 2009). Given all this, it
of the ant community, had a higher 50-year popula- is not surprising that fire effects (and fire management
tion growth rate compared with individuals associated effects) as drivers of demography have received a fair
with the putative best mutualist. Surprisingly, this was amount of study.
despite two of the other species of ants being putative The simplest approach considers the effects of
parasites. individual fires, often comparing populations affect-
More recently there has been an interest in the role ed by fire with those remaining unburned using
of beneficial interactions with microbes in population projection matrices of fire and no-fire. A study of
demography. All organisms interact with a complex siamang (Symphalangus syndactylus), a frugivorous,
microbiome that can include beneficial symbionts, and Southeast Asian rainforest primate, showed impacts of
much work has established the importance of these El Niño–Southern Oscillation (ENSO)-related fires on
associations for individual level effects in both ani- the demography and persistence by comparing burned
mals (e.g. critical for proper digestion in animals rang- and unburned populations (O’Brien et al. 2003). Alter-
ing from termites to cows to humans; Backhed et al. natively, if prefire data are also available, then a before–
2005) and plants (e.g. rhizobia, mycorrhizal fungi, and after/control–impact design can be used to analyse fire
fungal endophytes confer stress amelioration through as a driver, as in the case of bird assemblages showing
resource provisioning and even defence against her- increases in some species’ abundances after severe fire
bivores; Rodriguez et al. 2009; Larimer et al. 2010). in Montana (Smucker et al. 2005). Finally, other aspects
However, the consequences of these interactions for of the fire regime can be contrasted for their effects
population dynamics have not been deeply investi- on demography. A comparison of burn season and
gated in most cases. For example, to our knowledge, fire frequency on populations of the invasive spotted
only one study has examined the effects of soil micro- knapweed (Centaurea maculosa) showed that summer
biomes in the context of a plant matrix or integral burning was the only treatment that reduced popula-
projection model, despite the substantial literature on tion growth rates, and a life table response experiment
the importance of these interactions. This recent study showed that this effect happened because of reduced
demonstrated that the microbiome increased popula- reproduction (Emery and Gross 2005).
tion growth rates of an endangered plant, especially A more involved approach uses Markovian seq-
in stressful habitats (David et al. 2019). A few studies uences of postfire matrices in simulations to char-
have examined the consequences of beneficial sym- acterise postfire demographic changes. These simu-
bionts with vertically transmitted fungal endophytes lations can characterise observed fire sequences, but
they can also be used in stochastic analyses to pre- (1998) assembled demographic data for the understory
dict optimal fire return intervals and estimate time shrub Ardisia escallonioides into a megamatrix (a set of
postfire to quasi-extinction. For example, Menges matrices nested together in a larger matrix), where each
and Quintana-Ascencio (2004) used matrix selec- demographic matrix was part of a patch-transition
tion to draw from 54 matrices representing different matrix summarising openness with hurricanes and
times-since-fire in Florida scrub. Simulations of Eryn- vegetation recovery. The use of the megamatrix
gium cuneifolium population dynamics showed that revealed some patterns not seen in analyses of individ-
populations were growing only in the first decade ual matrices, such as the importance of small juveniles
postfire and that populations would disappear 30 or to elasticity. Hurricane frequency also interacted with
more years after fire—patterns that were reflected in seed predation, with seed predator effects being great-
observed dynamics (Menges and Quintana-Ascencio est at intermediate hurricane frequencies (Horvitz et al.
2004). Prescribed fire was therefore recommended at 2005).
intervals of 15 years or less. Whether fires are single Similar approaches have been used in characterising
or repeated can have important consequences for the the effect of human harvesting on population dynam-
dynamics of species that recover slowly after fire. Most ics. Matrices representing harvested and unharvested
functional groups of arthropods were more resilient populations and varying the extent of harvest have
to single fire events than to multiple fires, although been usefully applied to various organisms (e.g. Gaoue
resilience varied depending on the functional group and Ticktin 2010). Crowder et al. (1994) used matrix
(Moretti et al. 2006). population models to assess the effect of requiring
Demographic modelling can also incorporate distur- turtle exclusion devices (TEDs) on shrimp harvesting
bances and postdisturbance patterns into projections trawls in the Gulf of Mexico. In this simulation they
of population dynamics. Demographic models can be showed that TEDs would have a benefit on the logger-
used to predict species distributions (e.g. Diez et al. head sea turtle (Caretta caretta) population, by reducing
2014). Merow et al. (2014) modelled Proteaceae shrubs unintentional bycatch and drowning of turtles. These
in South Africa in relation to fire and weather and modelling exercises can define sustainable harvest
their interaction. These approaches can be combined levels.
with spatially explicit models. For example, the land- Another abiotic factor that can strongly influ-
scape dynamics of the Florida scrub plant Hypericum ence demography of populations is episodic drought
cumulicola were affected by patch size and patch iso- events. For example, stochastic matrix models predict-
lation, in addition to fire and demographic factors in ed that higher drought frequency negatively influences
integral projection models (Quintana-Ascencio et al. population growth rate of the endangered yellow-
2018). Population growth was highest under frequent bellied toad (Bombina variegata), potentially threaten-
fire return intervals, at higher elevations, and in larg- ing population persistence and requiring management
er and less-isolated populations, and the models did a practices that reduce exposure to prolonged desicca-
fair job of predicting occupancies and densities across tion stress (Cayuela et al. 2016). The importance of
the landscape. Fire patchiness is another key spa- drought as an episodic driver of population dynam-
tial factor that can drive populations. For example, ics may be growing if climate change increases the
low-intensity fires will tend to be patchy, which may frequency of severe droughts. For instance, a recent
reduce mortality of obligately seeding shrubs and IPM examined how the increased frequency of drought
promote their persistence in a pyrogenic landscape events predicted by Intergovernmental Panel on Cli-
(Ooi et al. 2006). mate Change (IPCC) models for Patagonia would
While less well studied than fire, disturbances from impact the demography of a dominant evergreen tree
hurricanes and other wind/storm events can be strong (Molowny-Horas et al. 2017). The simulations varying
drivers of demography in the tropics and subtrop- the frequency of drought events projected a reduction
ics, particularly for plants and benthic coral com- in density of these tree populations in the mid-term
munities (Pascarella and Horvitz 1998; Vardi et al. under an increased drought scenario.
2012; Edmunds 2015). Vardi et al (2012) used Acrop- While species interactions can be ongoing through-
ora palmata (Elkhorn coral) demographic data to sim- out an organism’s life (see section 6.4), both positive
ulate the effect of hurricane recurrence on the pop- and negative interactions may instead be discrete.
ulation viability of this endangered species, suggest- For example, rare events like mass emergence of
ing that hurricane frequency less than 20 years would periodical cicadas and pathogen outbreaks gener-
result in population extirpation. Pascarella and Horvitz ate discrete species interactions that may strongly
impact demography by enhancing predator growth or framework to apply spatial population dynamics to
reducing host survival (respectively). Further, many the control of agricultural pests. The balance between
interactions are discrete on shorter time scales. In fac- extinction rates and colonisation rates governed the
ultative mutualisms, organisms may or may not asso- rise and fall of spatially separated but interacting pop-
ciate with mutualists, and this can vary throughout ulations, termed metapopulations. Since Levins’ sem-
the organism’s lifetime. For example, the cactus Fero- inal work, the field of metapopulation ecology has
cactus wislizeni produces extrafloral nectaries intended flourished, and the spatially implicit Levins’ metapop-
to attract an ant defender mutualist that reduces her- ulation model has been refined to increase realism.
bivory and increases fruit production; however, over For example, variability in environmental conditions
three years of monitoring, roughly half the plants were can impact the vital rates of local populations. Some
left unguarded (Ford et al. 2015). Because fruit pro- subpopulations can ‘rescue’ declining subpopulations
duction had only weak effects on population growth, in habitat patches through dispersal. Disturbances can
there were no differences in population growth rates synchronise these dynamics at relatively large spatial
between ant and ant-free cacti. and temporal scales, influencing the long-term persis-
tence of the broader metapopulation (e.g. Brown and
Kodric 1977; Ranta et al. 1995; Hanski and Ovaskainen
6.6 Population dynamics in the 2000; Benton et al. 2001; Brook et al. 2009; Desharnais
et al. 2018).
landscape: influence of spatial habitat
A growing number of studies have started to shed
structure and heterogeneity light on the frequency and magnitude with which spa-
The challenges associated with incorporating contin- tially separated populations interact with one another
uous and episodic drivers into structured population (e.g. Paradis et al. 1998; Ricketts 2001; Hanski and
models (see sections 6.4 and 6.5) must be considered Ovaskainen 2003; Smith and Green 2005; Stevens et al.
when assessing how spatial habitat structure and land- 2010; Duncan et al. 2015). As more empirical dispersal
scape heterogeneity influence metapopulation dynam- data are generated, it has become clear that metapopu-
ics. By necessity this extension to the metapopulation lations may function differently than the original patch
requires an understanding of not only the shifting occupancy model conceptualised by Levins (1969).
mosaic of environmental stressors and disturbances Thus, the definition of a metapopulation has evolved
influencing dynamics at the population or local scale to include regular (rather than sporadic) dispersal
but also the critical importance of factors influenc- among populations, relatively infrequent extinction,
ing dispersal of individuals among the local habitat and both local-scale drivers that independently impact
patches within the broader metapopulation. single subpopulations and regional-scale drivers that
The importance of dispersal to metapopulation influence all subpopulations comprising the broader
dynamics has been acknowledged explicitly for over metapopulation.
a century (e.g. Reid 1899; Tansley 1923; Elton 1925; At the local scale, habitat heterogeneity is ubiquitous
Nicholson 1933). This long-held interest of ecologists in nature, and the influence of continuous and episod-
in dispersal, the immigration and emigration of indi- ic drivers on vital rates of populations inhabiting this
viduals, and its influence on the dynamics of spatially heterogeneous environment is somewhat dependent
structured populations should not be a surprise, how- on population size and density. At small population
ever, because, as Elton (1927) eloquently stated, ‘When sizes and low densities, the vagaries of sexual selection
we are studying any particular animal or community (e.g. finding a mate) and environmental disturbances
of animals, we are brought up, sooner or later, against (e.g. a hurricane, tsunami, or wildfire) can result in
questions connected with dispersal’. stochastic extinction of small populations (Courchamp
While dispersal was studied by ecologists in the et al. 1999) (see section 6.5). In addition to these poten-
early 1900s, it was not, with few exceptions, for- tial stochastic effects, vital rates can be impacted by
malised analytically until almost half a century later habitat geometry (patch area and shape). Hokit and
(Skellam 1951, 1952). Research on dispersal contin- Branch (2003) combined a stage-based matrix mod-
ued through the twentieth century, with numerous el with stochastic numerical simulation to explore the
theoretical advances. MacArthur and Wilson (1963, effect of habitat patch area on Sceloperus woodi vital
1967) invoked the role of dispersal capacity as an rates. They found fecundity and survivorship were
implicit mechanism of colonisation in their theory of positively correlated with habitat patch size. In spa-
island biogeography. Levins (1969, 1970) adapted this tially heterogeneous landscapes, populations can be
categorised as sources if they have positive growth breeding dispersal distance, and habitat type. Trenham
rates and sinks if they have negative growth rates et al. (2001) assessed spatial synchrony of demographic
(Pulliam 1988). Gundersen et al. (2001) experimental- patterns in tiger salamanders (Ambystoma californiense)
ly tested this source–sink framework using root voles due to dispersal among a network of breeding
(Microtus oeconomus) to assess the effect of demography ponds. Relatively high rates of salamander disper-
in sink populations on source populations connect- sal resulted in positively correlated breeding popu-
ed via vole dispersal. Dispersal rates tended to vary lations among nearby ponds. However, as distance
with population density; density-dependent disper- between ponds increased, the correlation among mass
sal of voles from source populations to sink habitat and age of individuals decreased. Quintana-Ascencio
patches inhibited the growth rates within the larger et al. (2019) assessed the relative importance of seed
source populations (Gunderson et al. 2001). In plants, bank and seed dispersal on the metapopulation per-
regional source–sink dynamics are closely related to sistence of an endangered and rare plant species.
life history strategy and environmental stability. Clon- They found metapopulation viability of this species
al plants or those with extensive seed banks tend to depended on a strong seed bank coupled with limited
support numerous local populations that can persist dispersal.
through relatively unfavourable environmental condi- At the metapopulation scale, disturbances can result
tions, while plants with relatively high dispersal rates in greater extinction rates, but these effects are land-
(e.g. Jacobaea vulgaris) tend to generate metapopulation scape dependent and tend to be aggregated in cer-
patterns in the landscape, with occupied and unoc- tain regions. If dispersal is nonrandom and biased
cupied patches intermittently connected via dispersal toward regions with greater probability of disturbance,
(Eriksson 1996). However, environmental and climatic it will result in higher extinction rates (Elkin and
factors vary temporally and spatially. If these broad- Possingham 2008). Conversely, if dispersal is asym-
scale environmental gradients generate variability in metric and skewed toward regions of relatively low
habitat quality, the loss of small populations in lower- disturbance, the metapopulation persistence will be
quality habitat patches may result in decreased dis- greater over time. Gouhier et al. (2010) assessed spa-
persal among patches and lower metapopulation-level tial synchrony for Mytilus californianus along the Pacific
fitness (Grear and Burns 2007). Coast of the United States, assessing the relative impor-
For metapopulations, an assessment of the relative tance of demographic processes and dispersal with
importance of environmental variation and dispersal environmental conditions. In a nearshore rocky inter-
on population dynamics is warranted. Density- tidal system with a relatively persistent gradient in
independent spatial autocorrelation of population environmental conditions, there was a weak and incon-
fluctuations controlled by environmental stochasticity sistent spatial correlation among environmental condi-
has been called the Moran effect (cf. Moran 1953). tions and population size at the local scale (Gouhier et
Experimental and empirical support for the role of al 2010). Yet the correlation in abundance of M. californi-
broad-scale environmental conditions synchronising anus switched from synchrony to asynchrony with dis-
dynamics in spatially separated populations is known tance from a given location, suggesting dispersal may
from multiple taxa (e.g. Ranta et al. 1995 [fish], Grenfell control regional-scale patterns of spatial synchrony.
et al. 1998 [sheep], Benton et al. 2001 [mites], Gouhier Carson et al. (2011) found two species of congeneric
et al. 2010 [mussels], Massie et al. 2015 [phytoplank- mussels (Mytilus californianus and M. galloprovincialis)
ton], Kahilainen et al. 2018 [butterflies]). However, spawn and release larvae in different seasons. This
in weakly regulated populations, theoretical models results in divergent dispersal patterns and demograph-
suggest dispersal can modify the magnitude and ic structure due to seasonal shifts in the broad-scale
time-scale of population synchrony (Lande et al. 1999; oceanographic currents. Fletcher et al. (2018) explored
Duncan et al. 2015; Desharnais et al 2018). the effect of random vs. aggregated habitat loss on
Empirical studies generally suggest that greater dis- the dynamics of an insect herbivore, Chelinidea vittiger,
persal increases synchrony. Paradis et al. (1998) were comparing the demographic consequences of local-
one of the first to empirically investigate the effects scale effects of habitat fragmentation (e.g. edge effects)
of dispersal and the feedbacks between environmental with the landscape-scale effects related to the scale of
and demographic factors on population synchrony dispersal. They found the scale of herbivore dispersal
in 53 species of birds. While evidence for popula- and the insect’s ability to move among habitat patch-
tion synchrony was evident across species, the scale es were the primary determinants of whether habi-
of synchrony varied with population size, natal and tat fragmentation had a negative or positive effect on
population growth. If the scale of fragmentation was By combining information about temporal changes
large relative to the dispersal capacity of the herbivore, in drivers with data on how drivers affect vital rates,
movement and reproduction decreased, resulting in a we can model not only stationary environments but
negative impact on population size. also how population dynamics change over time in
To characterise the variability inherent in the envi- environments that undergo directional change. This
ronment, dispersal, and ensuing connectivity pat- is particularly relevant for examining the effects of
terns, useful tools have been developed, particularly changes in climate and environment. Gotelli and
in the fields of landscape and conservation ecology. Ellison (2006) used such an approach and combined
For example, graph and network approaches have data from whole-plot fertilisation experiments on the
been adopted for their ease of comprehension and effect of nitrogen availability on the performance of
ability to simplify large amounts of connectivity data the pitcher plant, Sarracenia purpurea, with predicted
(Urban et al. 2009). From a conservation standpoint, long-term trends in nitrogen deposition. They found
these analytical frameworks have proven useful in that an increase in nitrogen deposition rate of only 1%
examining actual and hypothetical systems of protect- substantially increased the local extinction risk. Sim-
ed areas (Rayfield et al. 2011). In part, the growing ilar models have been used to explore the effects of
acceptance of the use of graphs and networks can be secondary succession mediated by changes in differ-
attributed to the ease with which patches of habitat ent abiotic and biotic factors (e.g. Dahlgren and Ehrlén
separated by an uninhabitable matrix can (at least visu- 2011; Lehtilä et al. 2016).
ally) be viewed as a graph (i.e. a set of nodes connected Some more general aspects of assessing the effects
by edges) or as a network when empirical data are of drivers on population growth rates are important to
available describing the magnitude and directionality consider. One problem is that effects on vital rates and
of individuals dispersing among habitat patches (Dale population growth rate might be nonlinear. Dahlgren
and Fortin 2010; Rayfield et al. 2011). As connectiv- et al. (2011) compared models that accounted vs. not
ity matrices continue to be generated for additional accounted for nonlinear relationships between vital
taxa, along with complementary large-scale environ- rates and state variables, finding that parameterisa-
mental data sets, additional metrics (e.g. modularity; tion method had a large effect on predicted population
Fletcher et al. 2018) can be used to assess how dispersal growth rates. They suggested that because even weak
connects demography among habitat patches. This can nonlinearity in relationships can have large effects on
increase our understanding of how these are impacted model predictions, restricted cubic regression splines
by episodic and continuous environmental drivers. should be considered for parameterising models when-
ever linearity cannot be assumed.
Another complication is that the effect of a focal
6.7 New and promising approaches driver might depend on the level of another driver—
To develop a predictive understanding of how cur- that is, that effects of the two drivers are interactive.
rent and future environmental conditions may influ- However, such interaction effects have rarely been
ence demography at relatively large spatial and tem- incorporated in environmentally explicit demographic
poral scales will require novel approaches. Lee (2017) models. One obvious reason for this is that detecting
and Iles et al. (2019) developed modelling approach- interaction effects requires considerably larger sam-
es allowing for the assessment of increasing and ple sizes. Nicolè et al. (2011) investigated how local
variable environmental drivers in ‘traditional’ elas- habitat quality and climatic variation influenced pop-
ticity analyses. These analytical approaches make it ulation dynamics of the endangered alpine plant Dra-
possible to separate the influence of environmen- cocephalum austriacum. They found that populations
tal driver intensity as well as variability on popu- on steeper slopes had lower population growth rates
lation growth. While these approaches were devel- than populations on more gentle slopes and that pop-
oped for single populations, they lay the ground- ulation growth rates were lower in years with high
work for extending these methods to metapopula- summer temperatures. Interestingly, there was also
tions. Similarly, these emerging large-scale data sets an effect of the interaction between these two fac-
and dynamic approaches, coupled strategically with tors, the negative effects of high summer temperatures
regional-scale vital-rate data, can generate fundamen- being more pronounced in populations on steeper
tal knowledge regarding the role of large-scale environ- slopes. In another example, the mountain date palm
mental variability on population and metapopulation (Phoenix loureirin) in south India is affected by fire,
dynamics. grazing, and human leaf harvest (Mandle and Ticktin
2012). Vital rates were influenced by interactions to project population changes due to novel conditions
among these drivers; harvesting was better tolerated caused by human actions.
in areas that had been recently burned. Other studies
have considered how additional drivers might affect
outcomes by combining demographic models with References
simulations.
Aars, J. and Ims, R. A. (2002). Intrinsic and climatic determi-
To achieve a good understanding of the overall effect
nants of population demography: the winter dynamics of
of a given environmental driver, it is important to tundra voles. Ecology, 83, 3449–3456.
include possible indirect effects. For example, effects of Adler, P.B., Dalgleish, H.J., and Ellner, S.P. (2012). Forecast-
climatic variables may act not only directly on perfor- ing plant community impacts of climate variability and
mance but also via changes in biotic interactions, such change: when do competitive interactions matter? Journal
as interspecific competition or facilitation. The study of Ecology, 100, 478–487.
by Adler et al. (2012), mentioned above, did indeed Backhed, F., Ley, R. E., Sonnenburg, J. L., Peterson, D. A.,
find not only direct effects of several climate variables and Gordon, J. I. (2005). Host–bacterial mutualism in the
but also strong indirect effects via the abundance of human intestine. Science, 307, 1915–1920.
Benton, T. G., Lapsley, C. T., and Beckman, A. P. (2001).
neighbouring species.
Population synchrony and environmental variation: an
Neglecting intraspecific density might result in
experimental demonstration. Ecology Letters, 4, 236–243.
biased estimates of environmental drivers. As a conse- Bond, W. J. and Keeley, J. E. (2005). Fire as a global ‘herbi-
quence, estimates of environmental factors when not vore’: the ecology and evolution of flammable ecosystems.
considering density effects may be incorrect if these Trends in Ecology and Evolution, 20, 387–394.
correlations are strong. Understanding density effects Bond, W. J., Woodward, F. I., and Midgley, G. F. (2005). The
is necessary to predict the long-term impact of environ- global distribution of ecosystems in a world without fire.
mental drivers and equilibrium abundances. Although New Phytologist, 156, 525–538.
effects of intraspecific density are likely to be important Bowman, D. M., Balch, J. K., Artaxo, P., et al. (2009). Fire in
in many systems, few demographic studies with plants the earth system. Science, 324, 481–484.
Brook, B. W., Akcakaya, H. R., Keith, D. A., Mace, G. M.,
have explored the simultaneous effects of environmen-
Pearson, R. G., and Araujo, M. B. (2009). Integrating bio-
tal drivers and density, but there are several with ani-
climate with population models to improve forecasts of
mals (Aars and Ims 2002). It is important to note that species extinctions under climate change. Biology Letters,
negative feedback effects of population density on pop- 5, 723–725.
ulation growth may occur even at low densities (e.g. Brown, J. H. and Kodric-Brown, A. (1977). Turnover rates in
Rodenhouse et al. 1997). Effects of intraspecific density insular biogeography: effect of immigration on extinction.
can explicitly be incorporated into IPMs by modelling Ecology, 58, 445–449.
variation in vital rates as functions of both environmen- Carson, H. S., Cook, G. S., Lopez-Duarte, P.C., and Levin,
tal variation and intraspecific density. Dahlgren et al. L. A. (2011). Evaluating the importance of demograph-
(2016) examined the simultaneous impact of climate ic connectivity in a marine metapopulation. Ecology, 92,
1972–1984.
and intraspecific density on vital rates of the dwarf
Cayuela, H., Arsovski, D., Thirion, J. -M., et al. (2016).
shrub Fumana procumbens over 20 years, using gener-
Demographic responses to weather fluctuations are con-
alised additive mixed models and IPMs. The analyses
text dependent in a long-lived amphibian. Global Change
suggested that the population was regulated by den- Biology, 22, 2676–2687.
sity but with annual fluctuations in response to varia- Cayuela, H., Rougemont, Q., Prunier, J. G., et al. (2018).
tion in weather. Importantly, omission of intraspecific Demographic and genetic approaches to study disper-
density in these models resulted in overestimations sal in wild animal populations: a methodological review.
of the effects of climatic variation and of extinction Molecular Ecology, 27, 3976–4010.
risk. Ceballos, G., Ehrlich, P.R., and Dirzo, R. (2017). Biological
Understanding demographic change as a function annihilation via the ongoing sixth mass extinction sig-
of drivers’ variation will be critical for better sustain- naled by vertebrate population losses and declines. Pro-
ceedings of the National Academy of Sciences, 114, 6089–6096.
able resource use, management, and conservation. This
Che-Castaldo, J., Jones, O. R., Kendall, B. E., et al. (2020).
provides a more mechanistic and potentially predic-
Comments to ‘Persistent problems in the construction
tive approach to understanding population dynamics. of matrix population models’. Ecological Modelling 416,
It allows scientists to consider a range of scenarios that 108913
can reasonably bound likely outcomes. In this way, Chung, Y. Y., Miller, T. E. X., and Rudgers, J. A. (2015).
studying the demographic effect of drivers can help Fungal symbionts maintain a rare host plant population
but demographic advantage drives the dominance of a Doak, D. F. and Morris, W. F. (2010). Demographic compen-
common host. Journal of Ecology, 103, 967–977. sation and tipping points in climate-induced range shifts.
Courchamp, F., Clutton-Brock, T., and Grenfell, B. (1999). Nature, 467, 959–962.
Inverse density dependence and the Allee effect. Trends in Doak, D. F., Thomson, D., and Jules, E. S. (2002). Population
Ecology and Evolution, 14, 405–410. viability analysis for plants: understanding the demo-
Crone, E. E., Menges, E. S., Ellis, M. M., et al. (2011). How graphic consequences of seed banks for population health.
do plant ecologist use matrix population models? Ecology In: Beissinger, S. R., McCullough, D. R. (eds), Popula-
Letters, 14, 1–8. tion Viability Analysis, Chapter 15. University of Chicago
Crone, E. E., Ellis, M. M., Morris, W. F., et al. (2013). Abil- Press, Chicago, IL.
ity of matrix models to explain the past and predict Dobson, F. S. and Oli, M. K. (2001). The demographic basis
the future of plant populations. Conservation Biology, 27, of population regulation in Columbian ground squirrels.
968–978. The American Naturalist, 158, 236–247.
Crowder, L. B., Crouse, D. T., Heppell, S. S., and Martin T. H. Duncan, A. B., Gonzalez, A., and Kaltz, O. (2015). Dis-
(1994). Predicting the impact of turtle excluder devices on persal, environmental forcing, and parasites combine to
loggerhead sea turtle populations. Ecological Applications affect metapopulation synchrony and stability. Ecology,
4, 437–445. 96, 284–290.
Da Silveira Pontes, L., Magda, D., Jarry, M., Gleizes, B., Eckberg, J.O., Tenhumberg, B., and Louda, S. M. (2014).
and L Agreil, C. (2012). Shrub encroachment control by Native insect herbivory limits population growth rate of
browsing: targeting the right demographic process. Acta a non-native thistle. Oecologia, 175, 129–138.
Oecologica, 45, 25–30. Eckhart, V. M., Geber, M. A., Morris, W. F., Fabio, E. S.,
Dahlgren, J. P. and Ehrlén, J. (2009) Linking environmental Tiffin, P., and Moeller, D. A. (2011). The geography of
variation to population dynamics of a forest herb. Journal demography: long-term demographic studies and species
of Ecology, 97, 666–674. distribution models reveal a species border limited by
Dahlgren, J. P. and Ehrlén, J. (2011). Incorporating environ- adaptation. The American Naturalist, 178, S26–S43.
mental change over succession in an integral projection Edmunds, P. J. (2015). A quarter-century demographic anal-
model of population dynamics of a forest herb. Oikos, 120, ysis of the Caribben coral, Orbicella annularis, and projec-
1183–1190. tions of population size over the next century. Limnology
Dahlgren, J. P., García, M. B., and Ehrlén, J. (2011). Nonlin- and Oceanography, 60, 840–855.
ear relationships between vital rates and state variables in Ehrlén, J. and Morris, W. F. (2015). Prediction changes in the
demographic models. Ecology, 92, 1181–1187. distribution and abundance of species under environmen-
Dahlgren, J. P., Bengtsson, K., and Ehrlén, J. (2016). The tal change. Ecology Letters, 18, 303–314.
demography of climate-driven and density-regulated Ehrlén, J., Morris, W. F., Von Euler, T., and Dahlgren,
population dynamics in a perennial plant. Ecology, 97, J. P. (2016). Advancing environmentally explicit struc-
899–907. tured population models of plants. Journal of Ecology, 104,
Dale, M. R. T. and Fortin, M. -J. (2010). From graphs to 292–305.
spatial graphs. Annual Review of Ecology, Evolution, and Elkin, C. M. and Possingham, H. (2008). The role of
Systematics, 41, 21–38. landscape-dependent disturbance and dispersal in
Dalgleish, H. J., Koons, D. N., Hooten, M. B., Moffet, C. A., metapopulaiton persistence. The American Naturalist, 172,
and Adler, P. B. (2011). Climate influences the demogra- 563–575.
phy of three dominant sagebrush steppe plants. Ecology, Elton, C. S. (1925). The dispersal of insects to Spitsber-
92, 75–85. gen. Transactions of the Entomological Society of London, 73,
Davelos, A. L. and Jarosz, A. M. (2004). Demography of 289–299.
American chestnut populations: effects of a pathogen and Elton, C. S. (1927). Animal Ecology. Sidwick & Jackson,
a hyperparasite. Journal of Ecology, 92, 675–685. London.
David, A. S., Quintana-Ascencio, P. F., Menges, E., Thapa- Emery, S. M. and Gross, K. L. (2005). Effects of timing of
Magar, E., Afkhami, M. E. A., and Searcy, C. A. S. (2019). prescribed fire on the demography of an invasive plant,
Living in a microbial world: endangered plant requires spotted knapweed Centaurea maculosa. Journal of Applied
soil microbiome for population persistence. The American Ecology, 42, 60–69.
Naturalist, 194, 488–494. Eriksson, O. (1996). Regional dynamics of plants: a review of
Desharnais, R. A., Reuman, D. C., Costantino, R. F., and evidence for remnant, source–sink, and metapopulations.
Cohen. J. E. (2018). Temporal scale of environmental Oikos, 77, 248–258.
correlations affects ecological synchrony. Ecology Letters, Fletcher, R. J., Reichert, B. E., and Holmes, K. 2018. The neg-
21, 1800–1811. ative effects of habitat fragmentation operate at the scale
Diez, J. M., Giladi, I., Warren, R., and Pulliam, H. R. (2014). of dispersal. Ecology, 99, 2176–2186.
Probabilistic and spatially variable niches inferred from Ford, K. R., Ness, J. H., Bronstein, J. L., and Morris, W. F.
demography. Journal of Ecology 102, 544–554. (2015). The demographic consequences of mutualism:
ants increase host-plant fruit production but not popula- Iles, D. T., Rockwell, R. F., and Koons. D. N. (2019). Shifting
tion growth. Oecologia, 179, 435–446. vital rate correlations alter predicted population respons-
Gaoue, O. G. and Ticktin, T. (2010). Effects of harvest of non- es to increasingly variable environments. The American
timber forest products and ecological differences between Naturalist, 193, E57–E64.
sites on the demography of African mahogany. Conserva- Kahilainen, A., Van Nouhuys, S., Schulz, T., and Saasta-
tion Biology, 24, 605–614. moinen, M. (2018). Metapopulation dynamics in a chang-
García, M. B., Goni, D., and Guzman, D. (2010). Living at the ing climate: increasing spatial synchrony in weather con-
edge: local versus positional factors in the long-term pop- ditions drives metapopulation synchrony of a butterfly
ulation dynamics of an endangered orchid. Conservation inhabiting a fragmented landscape. Global Change Biology,
Biology, 24, 1219–1229. 24, 4316–4329.
Genovart, M., Doak, D. F., Igual, J. M., Sponza, S., Kralj, Kendall, B. E. and Fox, G. A. (2002). Variation among indi-
J., and Oro, D. (2017). Varying demographic impacts of viduals and reduced demographic stochasticity. Conserva-
different fisheries on three Mediterranean seabird species. tion Biology, 16, 109–116.
Global Change Biology, 23, 3012–3029. Kendall, B. E., Fujiwara, M., Diaz-Lopez, J., Schneider, S.,
Gotelli, N. J. and Ellison, A. M. (2006). Forecasting extinc- Voigt, J., and Wiesner, S. (2019). Persistent problems in
tion risk with non-stationary matrix models. Ecological the construction of matrix population models. Ecological
Applications, 16, 51–61. Modeling 406, 33–43.
Gouhier, T., Guichard, F., and Menge, B. A. (2010). Ecological Koltz, A. M., Burkle, L. A., Pressler, Y., et al. (2018). Glob-
processes can synchronize marine population dynamics al change and the importance of fire for the ecology and
over continental scales. Proceedings of the National Academy evolution of insects. Current Opinion in Insect Science, 29,
of Sciences, 107, 8281–8286. 110–116.
Grear, J. S., and Burns, C. E. (2007). Evaluating effects of Lande, R., Engen, S., and Saether, B. -E. (1999). Spatial
low-quality habitats on regional population growth in Per- scale of population synchrony: environmental correla-
omyscus leucopus: insights from field parameterized spatial tion versus dispersal and density regulation. The American
matric model. Landscape Ecology, 22, 45–60. Naturalist, 154, 271–281.
Grenfell, B. T., Wilson, K., Finkenstadt, B. F., et al. (1998). Larimer, A. L., Bever, J. D., and Clay, K. (2010). The inter-
Noise and determinism in synchronized sheep dynamics. active effects of plant microbial symbionts: a review and
Nature, 394, 674–677. meta-analysis. Symbiosis, 51, 139–148.
Gunderson, G., Johannesen, E., Andreassen, H. P., and Ims, Lee, C. T. (2017). Elasticity of population growth with respect
R. A. (2001). Source–sink dynamics: how sinks affect to the intensity of biotic or abiotic driving factors. Ecology,
demography of sources. Ecology Letters, 4, 14–21. 98, 1016–1025.
Gurevitch, J., Fox, G. A., Fowler, N. L., and Graham, C. H. Lehtilä, K., Dahlgren, J. P., Garcia, M. B., Leimu, R., Syrjänen,
(2016). Landscape demography: population change and K., and Ehrlén, J. (2016) Forest succession and population
its drivers across spatial scales. Quarterly Review of Biology, viability of grassland plants: long repayment of extinction
91, 459–485. debt in Primula veris. Oecologia, 181, 125–135.
Hanski I. and Ovaskainen, O. (2000). The metapopulaiton Levins, R. (1969). Some demographic and genetic conse-
capacity of a fragmented landscape. Nature, 404, 755–758. quences of environmental heterogeneity for biological
Hanski I. and Ovaskainen. O. (2003). Metapopulation theory control. Bulletin of the Entomological Society of America, 15,
for fragmented landscapes. Theoretical Population Biology, 237–240.
64, 119–127. Levins, R. (1970). Extinction. In: M. Gesternhaber (ed.), Some
Hargreaves, A. L., Samis, K. E., and Eckert, C. G. (2014) Mathematical Problems in Biology (pp. 77–107). American
Are species’ range limits simply niche limits writ large? A Mathematical Society, Providence, RI.
review of transplant experiments beyond the range. The MacArthur, R. H. and Wilson, E. O. (1963). An equi-
American Naturalist, 183, 157–173. librium theory of insular zoogeography. Evolution, 17,
Heppell, S. S., Caswell, H., and Crowder, L. B. (2000). 373–387.
Life histories and elasticity patterns: perturbation analysis MacArthur, R. H. and Wilson, E. O. (1967). The Theory of
for species with minimal demographic data. Ecology, 81, Island Biogeography. Princeton University Press, Prince-
654–665. ton, NJ.
Hokit, D. G. and Branch, L. C. (2003). Associations Mandle, L. and Ticktin, T. (2012). Interactions among
between patch area and vital rates: consequences for fire, grazing, harvest and abiotic conditions shape palm
local and regional populations. Ecological Applications, 13, demographic responses to disturbance. Journal of Ecology,
1060–1068. 100, 997–1008.
Horvitz, C. C., Tuljapurkar, S., and Pascarella, J. B. (2005). Maron, J. L. and Crone, E. (2006). Herbivory: effects on
Plant–animal interactions in random environments: plant abundance, distribution and population growth.
habitat-stage elasticity, seed predators, and hurricanes. Proceedings of the Royal Society B: Biological Sciences, 273,
Ecology, 86, 3312–3322. 2575–2584.
Massie, T. M., Weithoff, G., Kucklander, N., Gaedke, U., Pérez-Jorge, S., Gomes, I., Hayes, K., et al. (2016). Effects
and Blasius, B. (2015). Enhanced Moran effect by spa- of nature-based tourism and environmental drivers on
tial variation in environmental autocorrelation. Nature the demography of a small dolphin population. Biological
Communications, 6, 5993. Conservation, 197, 200–208.
Menges, E. S. (2000). Population viability analyses in plants: Prevéy, J. S., Germino, M. J., and Huntly, N. J. (2010). Loss of
challenges and opportunities. Trends in Ecology and Evolu- foundation species increases population growth of exot-
tion, 15, 51–56. ic forbs in sagebrush steppe. Ecological Applications, 20,
Menges, E. S. and Quintana-Ascencio, P. F. (2004). Popula- 1890–1902.
tion viability with fire in Eryngium cuneifolium: decipher- Pulliam, H. R. (1988). Source, sinks, and population regula-
ing a decade of demographic data. Ecological Monographs, tion. The American Naturalist, 132, 652–661.
74, 79–99. Quintana-Ascencio, P. F., Koontz, S. M., Smith, S. A.,
Merow, C., Latimer, A. M., Wilson, A. M., McMahon, S. M., Sclater, V. L., David, A. S., and Menges, E. S. (2018).
Rebelo, A. G.,and Silander, J. A. Jr. (2014). On using Predicting landscape level distribution and abun-
integral projection models to generate demographically dance: integrating demography, fire, elevation and
driven predictions of species’ distributions: development landscape habitat configuration. Journal of Ecology, 106,
and validation using sparse data. Ecography, 37, 2395–2408.
1167– 1183. Quintana-Ascencio, P. F., Koontz, S. M., Ochocki, B. M.,
Molowny-Horas, R., Suarez, M. L., and Lloret, F. (2017). et al. (2019). Assessing the roles of seed bank, seed
Changes in the natural dynamics of Nothofagus dombeyi dispersal and historical disturbances for metapopulation
forests: population modeling with increasing drought fre- persistence of a pyrogenic herb. Journal of Ecology, 107,
quencies. Ecosphere, 8, e01708. 2760–2771.
Moran, P. A. P. (1953). The statistical analysis of the Canadian Ranta, E., Kaitala, V., Lindstrom, J., and Linden, H. (1995).
lynx cycle. II Synchronization and meteorology. Australian Synchrony in population dynamics. Proceedings of the Roy-
Journal of Zoology, 1, 291–298. al Society B: Biological Sciences, 1364, 113–118.
Moretti, M., Duelli, P., and Obrist, M. K. (2006). Biodiver- Rayfield, B., Fortin, M. -J., and Fall, A. (2011). Connec-
sity and resilience of arthropod communities after fire tivity for conservation: a framework to classify network
disturbance in temperate forests. Oecologia, 149, 312–327. measures. Ecology, 92, 847–858.
Nicholson, A. J. (1933). Supplement: the balance of animal Reid, C. (1899). The Origin of the British Flora. Dulau & Co.,
populations. Journal of Animal Ecology, 2, 131–178. London.
Nicolè, F., Dahlgren, J. P., Vivat, A., Till-Bottraud, I., and Ricketts, T. H. (2001). The matrix matters: effective isolation
Ehrlén, J. (2011). Interdependent effects of habitat quality in fragmented landscapes. The American Naturalist, 158,
and climate on population growth of an endangered plant. 87–99.
Journal of Ecology, 99, 1211–1218. Rodenhouse, N. L., Sherry, T. W. and Holmes, R. T. (1997).
O’Brien, T. G., Kinnaird, M. F., Nurcahyo, A., Prasetyan- Site-dependent regulation of population size: a new syn-
ingrun, M., and Iqbal, M. (2003). Fire, demography and the thesis. Ecology, 78, 2025–2042.
persistence of siamang (Symphalangus syndactylus: hylo- Rodriguez, R. J., White Jr, J. F., Arnold, A. E., and Redman,
batidae) in a Sumatran rainforest. Animal Conservation, 6, R. S. (2009). Fungal endophytes: diversity and functional
115–121. roles. New Phytologist, 182, 314–330.
Ooi, M. K. J., Whelan, R. J., and Auld, T. D. (2006). Persistence Roy, B. A., Coulson, T., Blaser, W., et al. (2011). Popu-
of obligate-seedling species at the population scale: effects lation regulation by enemies of the grass Brachypodium
of fire intensity, fire patchiness and long fire-free intervals. sylvaticum: demography in native and invaded ranges.
International Journal of Wildland Fire, 15, 261–269. Ecology, 92, 665–675.
Palmer, T. M., Doak, D. F., Stanton, M. L., et al. (2010). Syn- Salguero-Gómez, R., Jones, O. R., Archer, C. R., et al. (2015).
ergy of multiple partners, including freeloaders, increases The COMPADRE plant matrix database: an open online
host fitness in a multispecies mutualism. Proceedings of the repository for plant demography. Journal of Ecology, 103,
National Academy of Sciences, 107, 17234–17239. 202–218.
Paradis, E., Baillie, S. R., Sutherland, W. J., and Gregory, R. Scanga, S. E. (2014). Population dynamics in canopy gaps:
D. (1998). Patterns of natal and breeding dispersal in birds. nonlinear response to variable light regimes by an under-
Journal of Animal Ecology, 67, 518–536. story plant. Plant Ecology, 215, 927–935.
Parker, I. M. (1997.) Pollinator limitation of Cytisus scopar- Shefferson, R. P, Hutchings, M. J., Selosse, M. -A., et al.
ius (Scotch broom), an invasive exotic shrub. Ecology, 78, (2018). Drivers of vegetative dormancy across herbaceous
1457–1470. perennial plant species. Ecology Letters, 21, 724–733.
Pascarella, J. B. and Horvitz, C. C. (1998). Hurricane distur- Skellam, J. G. (1951). Random dispersal in theoretical popu-
bance and the population dynamics of a tropical under- lations. Biometrika, 38, 196–218.
story shrub: megamatrix elasticity analysis. Ecology, 79, Skellam, J. G. (1952). Studies in statistical ecology: i. Spatial
547–563. pattern. Biometrika, 39, 346–362.
Sletvold, N., Dahlgren, J. P., Øien, D. I., Moen, A., and climate and competition. Methods in Ecology and Evolution,
Ehrlén, J. (2013). Climate warming alters effects 7, 171–183.
of management on population viability of threat- Trenham, P. C., Koenig, W. D., and Shaffer, H. B. (2001).
ened species: results from a 30-year experimental Spatially autocorrelated demography and interpond dis-
study on a rare orchid. Global Change Biology, 19, persal in the salamander, Ambystoma californiense. Ecology,
2729–2738. 82, 3519–3530.
Smith, M. A. and Green, D. M. (2005). Dispersal and Tye, M. R., Menges, E. S., Weekley, C., Quintana-Ascencio,
metapopulation paradigm in amphibian ecology and con- P. F., and Salguero-Gómez, R. (2016). A demographic
servation: are all amphibian populations metapopula- ménage à trois: interactions between disturbances both
tions? Ecography, 28, 110–128. amplify and dampen population dynamics of an endemic
Smucker, K. M., Hutto, R. L., and Steele, B. M. (2005). plant. Journal of Ecology, 104, 1778–1788.
Changes in bird abundance after wildfire: importance of Urban, D. L., Minor, E. S., Treml, E. A., and Schick, R. S.
fire severity and time since fire. Ecological Applications, 15, (2009). Graph models of habitat mosaics. Ecology Letters
1535–1549. 12, 260–273.
Stevens, J. T. and Latimer, A. M. (2015). Snowpack, fire, Vardi, T., Williams, D., and Sandin, S. A. (2012). Popula-
and forest disturbance: interactions affect montane inva- tion dynamics of threatned elkhorn oral in the north-
sions by non-native shrubs. Global Change Biology, 21, ern Florida Keys, USA. Endangered Species Research, 19,
2379–2393. 157–1969.
Stevens, V. M., Turlure, C., and Baguette, M. (2010). A meta- Williams, A. L., Wills, K. E., Janes, J. K., Schoor, J. K. V.,
analysis of dispersal in butterflies. Biological Reviews, 85, Newton, P. C. D. and Hovenden, M. J. (2007). Warm-
625–642. ing and free-air CO2 enrichment alter demographics in
Szczys, P., Oswald, S. A., and Arnold, J. M. (2017). Con- four co-occurring grassland species. New Phytologist, 176,
servation implications of long-distance migration routes: 365–374.
regional metapopulation structure, asymmetrical disper- Winter, S., Jung, L. S., Eckstein, R. L., Otte, A., Donath, T. W.,
sal, and population declines. Biological Conservation, 209, and Kriechbaum, M. (2014). Control of the toxic plant
263–272. Colchicum autumnale in semi-natural grasslands: effects of
Tansley, A. G. (1923). Practical Plant Ecology: A Guide for cutting treatments on demography and diversity. Journal
Beginners in the Field Study of Plant Communities. Allen of Applied Ecology, 51, 524–533.
and Unwin, London. Yule, K. M., Miller, T. E. X., and Rudgers, J. A. (2013). Costs,
Teller, B. J., Adler, P. B., Edwards, C. B, Hooker, G., and benefits, and loss of vertically transmitted symbionts
Ellner, S. P. (2016). Linking demography with drivers: affect host population dynamics. Oikos, 122, 1512–1520.
PART 2
Data- and Research Question-Driven

Methods
CHAPTER 7
Abundance-based approaches
Jonas Knape and Andreas Lindén
7.1 Aims of abundance analyses Beyond focus on population levels and pat-
terns, abundance may be used to learn about basic
Data describing the abundance of populations across
population-level parameters to better understand why
time or space, but lacking information on demograph-
populations change. The most fundamental popula-
ic structure, are widely available. Abundance data of
tion parameter available from abundance data, apart
this type are generally available at greater resolution,
from abundance itself, is the population growth rate.
and at larger scales in time and space, than are data
In ecology, the growth rate is studied to understand
on demographic rates. In many cases abundance data
patterns of density dependence and regulation, or
are the only data available about populations at nation-
how populations respond to environmental factors,
al, regional or landscape scales, while we may, at best,
including abundances of interacting species, while in
have estimates of vital rates from a handful of local and
evolutionary studies the growth rate may be used as a
more detailed studies, often of short duration. Abun-
measure of fitness.
dance data therefore form a backbone for the study of
For populations growing continuously in time the
a large number of species and populations in the wild.
per capita rate of increase may be defined as the deriva-
Abundance, N, is a natural aggregate property of
tive of population size with respect to time divided by
populations that plays a fundamental role in applied
current population size. That is, the instantaneous rate
and theoretically oriented ecological studies. Often
of additions of individuals per individual present in
abundance itself is of major interest, and aims may
the population:
be directed to how abundance has changed historical-
dN
ly, how it varies in space, and what the current levels r(t) = /N
of abundance are or are predicted to be in the future. dt
For instance, estimating historical abundance is the The population will then grow when r(t) is positive,
focus of analyses of population trends from monitor- decrease when it’s negative, and be stable when it is
ing programmes, which may be used for red-listing of zero. Alternatively, in discrete time, we may instead
species or as indicators of the state of the environment, define the multiplicative growth rate via
and estimates of current abundance are used to set N
λt = t+1
fisheries quotas or hunting caps, sometimes in combi- Nt
nation with short-term predictions on how harvesting where λt is the per capita increase in population size
might affect population abundance. from time t to t + 1. Commonly discrete growth is stud-
Abundance is also used to study patterns in ied through the logarithm of λt , rt = log(λt ) instead
dynamics. Classical questions addressed in animal of directly. On the logarithmic scale multiplicative
ecology concern the level of synchrony of populations growth processes become additive, making it better
separated in space (Ranta et al., 1995; Liebhold et al., suited for many population analytical approaches. The
2004), the existence of cycles in population dynamics similarity in notation between r(t) and rt is no coinci-
(Barraquand et al., 2017), and how abundance varies dence since if time steps are small, rt is approximately
with environmental characteristics such as across an equal to r(t), and the discrete time population will sim-
altitudinal gradient or with weather (Ives, 1995) or ilarly increase from time t to t + 1 if rt is positive and
climate (Ådahl et al., 2006). decrease when it is negative. From this perspective, the
Jonas Knape and Andreas Lindén, Abundance-based approaches. In: Demographic Methods Across the Tree of Life.
aim of many studies of abundance is to understand of changes in abundance. Similarly, some organisms
how and why the function r(t) or rt varies over time, in including many plants and insects have strongly sea-
space, and as a function of population size or external sonal abundance patterns which makes the timing of
factors. sampling critical. Butterfly monitoring programmes,
for example, may sample abundance on 20 or more
occasions each year to cover the full seasonal cycle.
7.2 Measures of abundance For studying long-term changes in abundance these
What do we mean by abundance? In section 7.1 seasonal samples may be aggregated in some way to
we were deliberately vague about how abundance is annual indexes of abundance, for example by estimat-
defined, and in practice the meaning of abundance ing the area under or the maximum of the seasonal
varies depending on the context. The most intuitive curve (Dennis et al., 2013). Similar measures may be
descriptors of abundance are the total number of indi- applied to organisms with very short generation time,
viduals in the population, or the number of individuals such as bacteria or phytoplankton. Considering sea-
per unit area, known as population density. These sonal patterns has become increasingly important for
measures are commonly applied for several types of many organisms due to accelerating climate change
organisms, including most terrestrial vertebrates and and rapidly changing phenology in a wide range of
arthropods. Other measures of abundance are in use taxa. In some cases, failing to account for changing phe-
and may be preferred for biological or practical rea- nology in abundance at the time of sampling may lead
sons. For example, since tree growth depends on tree to biased estimates of population abundance.
size it can be easier to predict the growth of a pop- It is further important to define the area across which
ulation of trees in a forest using its biomass rather the population is to be studied. Considering the type of
than the number of trees. Fish also grow largely inde- organism and the spatial scale of the questions asked
terminately and with size dependence, which partly are fundamental to this definition. Delimiting the spa-
explains the common use of biomass for studying the tial extent of a population is typically easier for sessile
dynamics of fish populations. There can also be prac- organisms than it is for mobile ones. For the latter a
tical reasons for using fish biomass over the number complicating factor is that it is typically impossible to
of individuals. When populations are assessed using estimate emigration and immigration rates from abun-
trawl surveys or by fishery vessels, it may require dance data. Therefore it is sometimes convenient to
less effort to weigh the catch than it does to count it. define populations over a large enough area so that
For plants with clonal growth it may be difficult to migration does not play a major role in its dynamics.
separate individuals in the field, and counting ram-
ets instead of individuals may be preferred and better
7.2.1 Sampling abundance
represent abundance (Bullock, 1999). Another example
of alternative measures of abundance is areal cover, Sampling abundance in the wild is in general a difficult
which is used to describe the size of some popula- and arduous task. A wide range of sampling meth-
tions of plants. Ultimately, the aims of the study in ods exists, and the methods differ greatly among taxa.
combination with practical considerations should be Sampling may involve counting individuals, counting
used as a guide when deciding on the definition of animal tracks along a predefined transect, mapping
abundance for a population. If the aims are not prop- territories, marking and recapturing individuals, sam-
erly considered, the data to be collected may not be pling areal cover using pin-quadrats or photographs,
able to address the questions that are asked of them camera traps, acoustic monitoring, electro fishing in
(Reynolds et al., 2016). streams, or trawl surveys in the sea. For an overview
Apart from determining what to measure, an impor- of these and other basic methods for obtaining abun-
tant consideration when defining abundance for a pop- dance data we refer the reader to section (5.2.1) and
ulation is to decide when and how often abundance Sutherland (2006).
should be measured. For some species and purposes, a The sampling of abundance almost invariably
single abundance measure per year may be sufficient, involves some degree of observation error, imply-
and this temporal scale of sampling is common for ver- ing that we obtain an estimate of abundance rather
tebrates in particular but also for other organisms. For than abundance itself. Observation error has repercus-
organisms with fast life histories, perhaps with several sions for several types of analyses, and it is generally
generations in the course of a single year, more fre- important to think of the consequences that it may
quent sampling will be necessary for detailed studies have on analyses of abundance data. One obvious
ABUNDANCE-BASED APPROACHES 135
consequence of observation error is that estimated 7.3 Analytic approaches

abundance will tend to be more variable than abun-
dance itself. Failing to consider this could, for example, 7.3.1 Regression-type models
lead to overestimation of extinction risks. Observation One approach to analysing abundance data is to rely on
error also weakens signals in population dynamics, standard regression techniques from statistics, such as
making inference about population processes harder linear models, generalised linear models, mixed mod-
and more uncertain. els, smoothing methods such as additive models, or
Samples of abundance may further be biased. A sparse regression methods. These approaches are used
commonly encountered problem is that individuals in several different contexts of abundance analyses.
may be hard to find, so that measured abundance An example is the use of regression for estimating
tends to be smaller than actual abundance, some- temporal changes in abundance from monitoring data.
times by a considerable margin (see section 5.2.1). Several countries have implemented national monitor-
Some sampling schemes are designed to provide infor- ing programmes to track the status of species over time,
mation of such bias due to detection errors. These primarily for birds, other terrestrial vertebrates, and
include mark–recapture methods (McCrea and Mor- butterflies. Their results are used in conservation set-
gan, 2015) and removal and distance sampling (Buck- tings for red-listing of species that are in decline, as gen-
land et al., 2015). Other sampling schemes make no eral environmental indicators, or for applied research
such attempts and instead provide an index of abun- purposes to study relationships between environmen-
dance. For example, an index of current abundance tal changes and population trends.
may be defined as the proportional abundance rela- The basic premise for trend estimation from abun-
tive to a reference year, or relative indices of abundance dance data is in the form of a regression of an abun-
may be computed for comparisons among sites. dance measure xt against time:
Under certain conditions, an estimate of the number
xt = a + bt + ϵt (7.1)
of individuals corrected for detection can be obtained
from repeated counts at many sites, without any oth- where a is an intercept, b is the trend or change in
er dedicated sampling design (Royle, 2004; Kéry et al., population size, and ϵt is a term denoting sampling
2005). This method is sensitive to model assumptions error. Typically xt is an estimate of the logarithm of
and can give misleading estimates of abundance if the population size rather than the population size at the
assumptions, some of which cannot be checked with- arithmetic scale, so that xt = log(Nt ). If this is the case,
out other types of data, are not met. It has therefore the exponent of the slope coefficient, eb , is interpreted
sparked controversy (Barker et al., 2018). It tends to as the proportional change in the population per unit
work best when individuals are distributed uniform- time, and could be reported in the form of a percentage.
ly in space and are not too mobile, detection rates are This parameter is one of the most typically reported
high, and abundances are small (Knape et al., 2018). statistics from monitoring studies and often applied in
Even with sampling designs intended to quantify evaluation of red-listing criteria.
detection and other observation errors, estimates of The above regression for estimating population
total abundance will often be uncertain and biased. trends applies to data on a single population but can
Obtaining useful relative estimates of abundance can be extended in many ways. For example, if surveys
be substantially easier. Relative estimates contain are conducted over multiple sites the model may be
no information about total abundance, for example extended to study how population trends vary in space
on how many individuals there are in a popula- or with environmental variables in space and time
tion, but they may be sufficient for inference on (Oedekoven et al., 2017). Vanhatalo et al. (2017) used
population growth rates (rt ). Time series of relative one such extension including spatiotemporal effects to
abundance can be applied, for instance, to estimate study outbreak patterns in starfish at the Great Barrier
how fast a population is decreasing. How appro- Reef. Regression models may also be used for impact
priate relative measures of abundance are in this assessments to estimate the effects of infrastructure
case hinges on whether and how observation errors development (Pépino et al., 2012) or extreme natural
vary in time. events (Rochester et al. 2010; Chevalier et al. 2019) on
The large majority of methods discussed in this abundance.
chapter are general enough that they may, perhaps In regression-based approaches a statistical model
with some minor adaptations, be applied to most types is used to analyse the abundance data without nec-
of abundance measures. essarily explicitly specifying a model of population
dynamics. In general, the focus of inference in growing season g. Figure 7.1 shows the results from fit-
regression-based approaches is more towards identi- ting this model. They indicate that population change
fying patterns in and correlatives of abundance and varied in space with fairly stable population size in the
growth rates, rather than to learn about the dynamic south, a strong increase followed by a decline in central
mechanisms behind the population processes. Sweden, and a recent decline from a stable level in the
north.
Example: trends in the common cuckoo
The common cuckoo (Cuculus canorus) is a migrant
7.3.2 Dynamic abundance models
bird with a widespread summer breeding range in
Europe and Asia. It is a brood parasite, laying its eggs An alternative to regression-based approaches is to
in the nests of mainly small-sized passerines (Cramp use population ecological theory as the starting point
1985). The Swedish breeding bird survey consists of for models and apply flexible statistical techniques for
standardised annual transect counts of breeding birds, their assessment. For example, fishery biologist often
including the cuckoo, at around 700 sites laid out on rely on model-based approaches to guide sustainable
a grid across Sweden (Tayleur et al., 2015). The aim is management and harvest of fish stocks (Quinn and
to follow the change in the national population sizes of Deriso, 1999).
common breeding birds. Mathematical models of population dynamics have
Counts at each site are conducted once a year during a long history in ecology. They have been used in
the breeding season, yielding data in the form of the purely theoretical capacities to better understand the
number of seen or heard cuckoos, yit , for site i in year t. foundations of population growth, regulation, or out-
To estimate a smooth trend in the total population size break risks. A classic example is May’s examination of
we may use an extension of eqn. (7.1) chaotic dynamics in simple mathematical population
models when growth rates are high (May 1976). Pop-
yit ∼ NegBin(exp(ai + s(t) + ϵt ), θ)
ulation models are also commonly used and fitted to
where s(t) is now a smooth and possibly nonlinear data in empirical studies. An early example is Gause’s
trend over time, ai is a site-specific random effect, εt is study of population limitation in yeast via a logistic
a random year effect, and the counts are assumed to model (Gause, 1932).
follow a negative binomial distribution with overdis-
persion parameter θ. The random effects are assumed Observation error models
to be normally distributed and independent among, Perhaps the easiest way of linking data to a population
respectively, sites and years. The site effects account model is to assume that the dynamics are determin-
for variation in abundance level and detection rates istic but that abundance is observed with error. This
among sites, while the year effects account for common leads to a regression problem that can be relatively easy
annual variation in the data and hence modelling spa- to solve numerically. While the assumption of deter-
tial synchrony in the unexplained variation (Knape, ministic dynamics is often a gross simplification, it can
2016). make it possible to fit mathematically complex models
The estimated national trend is shown in Figure 7.1. that could otherwise pose problems.
It suggests a slight increase in the cuckoo population We have in the previous section already encoun-
in early 2000 followed by a slight decrease. We may tered an example of an observation error population
ask whether this national trend is consistent across dynamics model, although it was not cast as such. The
Sweden. To address this, we use the average length of regression model of log-population size against time
the growing season as a proxy for gross spatial varia- may be reformulated as a model of this type. The model
tion in habitat and climate over Sweden. The growing can be expressed as
season is short in the northern alpine regions of Swe-
Nt = N0 exp(rt)
den, increases a bit in the northern and mid-latitude (7.2)
yt = Nt exp(ϵt )
forested areas, and is longest in the southern, large-
ly agricultural, regions. We then fit a model similar to Here we have changed the parametrisation of the mod-
the previous one but with a two-dimensional smooth el in eqn. (7.1) by using the initial population size N0 at
term: time 0 in place of a, linked via N0 = ea , and the slope
b has been renamed to r to clarify its relation to the
yit ∼ NegBin(exp(ai + s(t, g) + ϵt ), θ).
growth rate. The first equation is that of a determin-
The smooth interaction term s(t, g) allows trends in istic exponential growth model. This model describes
time to vary in space as a function of the length of the the dynamics of a population with constant per-capita
(a) (b) Short growing season

2.5
Abundance index
2.0
1.5
1.0
0.5
0.0
1995 2000 2005 2010 2015
65 Year
Intermediate growing season
2.5
Abundance index
2.0
1.5
Lat
Vegetative period 1.0

250 0.5
60 200
0.0
150 1995 2000 2005 2010 2015
100 Year
Long growing season
2.5
Abundance index
2.0
1.5
55 1.0
0.5
15 20
Lon 0.0
1995 2000 2005 2010 2015
Year
(c) Overall
2.5
Abundance index
2.0
1.5
1.0
0.5
0.0
1995 2000 2005 2010 2015
Year
Figure 7.1 (a) The length of the growing season in days at each of the survey routes covered by the Swedish breeding bird monitoring
programme. (b) The long-term population trends of common cuckoo are nonlinear in shape and vary across Sweden, here approximated as a
function of the length of the growing season. The trends in abundance over years are illustrated as abundance indices, scaled relative to the
reference year 2000. This is shown for areas with short, intermediate, and long growing season. (c) Abundance indices when the trend is assumed
to be the same across all of Sweden. Shaded regions show approximate 90% confidence intervals for the population change relative to the
reference year.
growth rate and is the solution to the differential When the population is small the ratio N/K is negli-
equation gible and we recover the exponential growth model.
Thus, a population following a logistic model initial-
dN 1 ly grows exponentially at a rate r, the intrinsic growth
= r.
dt N rate, if it starts from a small abundance. As the popu-
The second component of the model in eqn. (7.2) relates lation approaches the carrying capacity K the growth
the observed or measured abundance, yt , to the abun- rate decreases towards zero.
dance Nt via a model for observation error. The ran- This model, which is nonlinear, can be fit to data
dom error term εt is often assumed to be normally assuming that deviations between model predictions
distributed. The relation then describes the log-normal and data are due to observation error only. To do so, it
distribution, which can often be an acceptable choice helps to solve the logistic differential equation to get a
for abundance data in the form of population num- mathematical expression for how abundance changes
bers or biomass of large populations. The equivalence with time under this model. This can be done using
between this model and the linear regression model in basic techniques of calculus and the solution is
eqn. (7.1) means that standard linear regression soft- KN(0)ert
ware can be used to fit it to data by putting log(yt) as N(t) =
K − N(0) + N(0)ert
the response variable and t as the explanatory variable.
If we assume that the logistic model is observed with
The model in eqn. (7.2) highlights the distinguish-
a log-normally distributed observation error at times
ing feature of observation models. They can be parti-
t1 , t2 , …, tm , that is log yi ∼ Normal(N(ti ), σ 2 ), the
tioned into a deterministic model of abundance, here
log-likelihood for the data can be constructed from the
the deterministic exponential growth model, and a
density of the normal distribution as
stochastic observation error model of how observed
abundance arises from the ‘true’ abundance, here a log- m
l(r, K, N(0), σ; y) = − log(2πσ 2 )
normal error distribution. In place of the log-normal 2
1 X
m
distribution, other models for observation errors may
− (log yi − N(ti ))2
sometimes be more appropriate. Binomial, Poisson, or 2σ 2
i=1
negative binomial distributions could be used for dis-
crete count data (Lindén and Mäntyniemi, 2011), a beta The model can then be fit to data by maximising the
distribution for continuous cover data, or a Tweedie likelihood over the parameters r, K, N(0), and σ. Gener-
distribution for biomass data (Dunstan et al., 2013). For ic optimisation routines are available to solve this task
the exponential growth model these observation error (Bolker, 2008). Due to the normally distributed error
distributions lead to a generalised regression problem, (on the log scale), it is sufficient to numerically min-
and generic GLM software can be used to fit such mod- imise the sum of squares over r, K, and N(0), which is
els to data. Most population dynamics models cannot, a nonlinear least squares problem.
however, be framed as linear, and fitting them to data In general, the most important requirement for mak-
in an observation error framework then requires non- ing it technically possible to fit an observation error
linear regression techniques. model to data is that we can predict the population tra-
A population cannot grow indefinitely according to jectory, analytically or numerically, over time (and/or
an exponential growth model because it would, soon- space) for all relevant values of the parameters. Since
er or later, either go extinct or consume all its available the population model is deterministic, population size
resources. To prevent this some form of negative den- is a function of time, t, and the parameter values ϕ (e.g.
sity dependence, a decrease in per-capita growth rate ϕ = (r, K, N(0)) in the logistic model). Technically, the
as populations become large, must be at play (Hixon requirement therefore means that we can compute this
et al., 2002). The importance of regulation was realised function, N(t, ϕ), for various values of t and ϕ.
early in the study of population ecology, and the clas- The population trajectory can then be coupled to
sic logistic population model is built around this idea data using a probability density for the observations
by simply extending the exponential growth model in that is appropriate for the data. For example if the
eqn. (7.2) such that the per-capita growth rate decreases abundance data are counts, a discrete observation error
linearly with abundance: model such as a negative binomial distribution may fit
the data better than the log-normal distribution used
dN 1 above. Formally, the link between data and population
= r(1 − N/K)
dt N trajectory may be described via the probability density
or probability mass function of the observation mod- at a site in England in 1962 (Moore, 1991). We con-
el, p(yt |N(t, ϕ), θ), where θ contains the parameters of sider data for the azure damselfly (Coenagrion puella)
the observation model. From this we may construct a between 1963 and 1989. Counts analysed here are the
general version of the log-likelihood, annual maximum number of individuals recorded over
X multiple visits each year. The counts of this species ini-
l(θ, ϕ|y) = log p(yi ; N(ti , ϕ), θ), tially increased slowly after the creation of the ponds
before they possibly stabilised around some average
which can then be maximised numerically or alterna- level.
tively could be used in a Bayesian framework. In management and conservation applications one
This recipe can make it possible to fit very advanced may need indicators of habitat quality or suitabili-
deterministic models to data. For instance, N(t, ϕ) may ty, which could be measured in terms of population
be implicitly defined as the solution to a differen- dynamical characteristics. For example, the growth
tial equation or a system of differential equations that potential, or how many damselflies ponds could host
describe population dynamics (Nisbet and Gurney on average in a saturated population, could indicate
1982). Even if we cannot solve the differential equation pond quality and suitability. We therefore estimate
analytically, as we could for the logistic and exponen- the intrinsic growth rate (r) and the carrying capac-
tial models, we may be able to solve it numerically and ity (K) by fitting the logistic model for population
compute a close approximation to N(t, ϕ) that can be growth. We assume that the annual counts are nega-
plugged into the likelihood equation above. Nonlinear tively binomially distributed due to observation errors,
discrete time models may also be used. Even deter- with an expected value equal to the population trajec-
ministic individual-based models can in principle be tory (N(t)). Model uncertainty is assessed using para-
fitted this way. That is, one could simulate a determin- metric bootstrap with 5,000 resamples, and also more
istic model of individuals in a population over time, specifically for parameter K, by plotting its profile like-
compute population size from the total number of indi- lihood function.
viduals in the simulation, and then plug the simulated The model fit and residual diagnostics are
deterministic population size into the likelihood. In shown in Figure 7.2. The parameter estimates
practice, care needs to be taken to avoid having more (and 90% confidence intervals) are: initial popu-
parameters than can be reasonably informed by the lation size N(1) = 2.7 (1.1, 4.7), r = 0.23 (0.16, 0.35),
abundance data, and careful consideration of param- K = 210 (105, 108 ), and overdispersion parameter
eter identifiability would be necessary. θ = 0.55 (0.27, 0.76). The huge uncertainty of K, par-
While the principle of estimating observation error ticularly its flat likelihood for values larger than
models this way is straightforward, the likelihood sur- the estimate (Figure 7.2b), reflects that it is unclear
face can for some models be very complex, making whether the population has yet approached an asymp-
estimation difficult in practice. Models with complex tote. Apart from the problems with uncertainty,
dynamics, for instance, can have erratic likelihood sur- analyses of the Dunn–Smyth residuals suggest a
faces that are difficult to deal with using standard reasonable model fit (Figure 7.2c,d).
frequentist or Bayesian methods (Wood, 2010). The results suggest that an exponential growth mod-
Intuitively, fitting observation error models can be el fit the data approximately equally well and that
viewed as fitting a population trajectory curve, speci- these data cannot be used to obtain a reliable esti-
fied by a population model, through the data. Before mate of K. Even a 27-year time series, with sever-
the widespread availability of computing power, the al data points intuitively suggesting that population
relative simplicity in adapting them to data some- growth may have ceased, is not necessarily enough for
times made them the only viable option for estimating this task. Proper uncertainty assessment is crucial in
parameters of a population model. Even today, this this kind of analysis, while a realistic point estimate
simplicity is the main advantage of observation error (K = 210) and reasonable model fit are no guarantee
models, and in cases where fitting a deterministic mod- for a reliable result. At the same time, we have good
el is sufficient to address the question at hand, they are information on other parameters. Regarding popula-
convenient to work with. tion characteristics that could indicate habitat quality,
our only option is to concentrate on the population
Example: damselflies in newly established ponds growth potential (via the estimate of r). The situation
Dragonflies were counted annually immediately fol- could be improved even for K if there would be less
lowing the establishment of a number of small ponds variability in the counts and possibly if new data for
(a) (b)
400 10.0
300 7.5
Dragonfly count
95% CI
Deviance
90% CI
200 5.0
100 2.5
0 0.0
1970 1980 100 1000 10000

Year Carrying capacity (K)
(c) (d)
2
2
1 1
Dunn−Smyth residuals
Sample quantiles
0 0
−1 −1
−2
−2
0 50 100 150 −2 −1 0 1 2
Model fit Theoretical quantiles
Figure 7.2 (a) A deterministic logistic model curve (thick black line) fitted to annual maximum counts of damselflies in newly established ponds
(black points) in England. Green lines show 100 bootstrap samples, illustrating the uncertainty of the model fit. (b) The model deviance (i.e.
difference in –2 log-likelihood) is a shallow function of carrying capacity (K) on the right side of the minimum (K = 210), barely reaching the critical
value for an approximate 90% confidence interval. The upper limit of a corresponding 95% confidence interval goes to infinity, illustrating that we
cannot reliably estimate K. (c) The Dunn–Smyth residuals show no obvious pattern when plotted against the model expectation, and (d) a QQ-plot
of the Dunn–Smyth residuals indicate no apparent deviations from a normal distribution, suggesting a reasonable model fit.
a few years more would become available. Less vari- it (Brännström and Sumpter, 2006). For instance, we
ability in count data may be achieved by improving may take the deterministic discrete time exponential
the monitoring practices or by somehow combining the growth model as the starting point and multiply the
information available from the multiple visits. multiplicative growth rate λ by an error term exp(η t ),
where η t is normally distributed with mean zero and
Stochastic population models variance σ 2 . The stochastic exponential growth model
Variation plays a central role in the dynamics of popu- then is
lations. When using deterministic models, properties
of population dynamics that are inherently stochas- Nt = Nt−1 exp(r + ηt ) (7.3)
tic cannot be studied adequately. For instance, if we
This model is often used in population viability anal-
want to estimate the risk of extinction for some species,
yses to assess extinction risks (Dennis et al., 1991;
a stochastic model is a natural starting point. The
Holmes, 2001; Jacquemyn et al., 2007; Ramula et al.,
majority of analyses of abundance data therefore use
2008). Because the model is stochastic, extinction risk
population models that incorporate variability in the
is nonzero even if the average logarithmic growth rate
dynamics.
r is positive.
In the study of stochastic population dynamics one
The same approach to adding environmental vari-
usually distinguishes between two main types of varia-
ation is being used for other deterministic population
tion denoted environmental and demographic stochas-
models. The deterministic Ricker model, a discrete time
ticity (Engen et al., 1998). Demographic stochasticity
variant of logistic growth, is
operates at the level of individuals and is due to ran-

dom variation in outcomes of life history events such as Nt
survival and reproduction. Environmental stochastici- Nt+1 = Nt exp r 1 −
K
ty, on the other hand, operates at the level of the pop-
ulation and is due to fluctuations in the environment Environmental variation can be added to this model in
affecting all individuals but varying through time. the same way as for exponential growth by adding a
Demographic stochasticity therefore concerns varia- multiplicative error term, resulting in the model
tion in outcomes in life history events, while environ-
Nt
mental stochasticity concerns variability in the average Nt+1 = Nt exp r 1 − + ηt (7.4)
K
rates that govern those outcomes. For example, demo-
graphic stochasticity in survival is the variation caused While adding a normally distributed error term to the
by some individuals surviving and others not, while exponent of the growth rate is common, it does not
environmental stochasticity in survival is the among- always result in a realistic description of growth. Many
year variation in annual survival probabilities. populations have an upper bound to how much they
Demographic stochasticity has the largest relative can increase over a single time period, and such a
effects in small populations where the fate of single bound is not strictly honoured by the additive nor-
individuals can be crucial for population persistence. mally distributed error term. Upper bounds could be
In large populations demographic stochasticity to a enforced by instead adding a multiplicative error term
large extent averages out and becomes less important. to the carrying capacity (Ruokolainen et al., 2009),
In contrast, environmental stochasticity affects popu- resulting in little variation when the population is
lations of all sizes. With abundance data, it is typically far from the equilibrium, or by replacing the expo-
difficult to quantify the contribution of the demograph- nential function for a logistic one (Sköld and Knape,
ic and environmental variation to the total variation in 2018).
the dynamics. An approach often taken is to assume
that demographic variation is negligible and to instead Autoregressive models
focus on environmental variation. An alternative, but related, approach to the analysis
In applications to data, it is common to study of stochastic population dynamics is to rely on statis-
stochastic population dynamics at a discrete time- tical methods for time series analysis. In population
scale. A simple and widely used approach to build a dynamics, this was popularised by Royama (1992),
stochastic population dynamics model with environ- Turchin (1995), and others who used linear autoregres-
mental variation is to start with a deterministic discrete sive (AR) models of the annual log growth rate. The
time model and add a multiplicative error term to simplest model of this type is the AR(1) model, an
autoregressive model with one time lag: mammal dynamics (Cornulier et al., 2013), often in
combination with effects of environmental variables on
Xt+1 = a0 + a1 Xt + ηt (7.5)
population growth rates (Stenseth et al., 2002). Anoth-
where a0 is an intercept, a1 is the first order autoregres- er application is studying insect outbreak dynamics
sive coefficient and η t is a normally distributed error (Peltonen et al., 2002).
term with zero mean. In studying population dynam- Autoregressive models of the logarithm of abun-
ics, Xt is usually taken to be the logarithm of abundance dance may be interpreted directly as population
at time t (i.e. Nt = exp(Xt )). On the arithmetic scale, the models as above. Alternatively, they may be seen as
model then can be written linearisations of nonlinear models around an equilib-
rium point (Månsson et al., 2007). Linearisation can
Nt+1 = Nt exp(a0 + (a1 − 1) log(Nt ) + ηt )
be useful for an improved understanding of non-
Similarly to the stochastic Ricker model above, this linear models or as an approximation to nonlinear
model has an annual growth rate that depends on dynamics for small deviations from an equilibrium
abundance and a random error term representing envi- (Ripa and Heino, 1999).
ronmental stochasticity. It is sometimes referred to as
the Gompertz model and has been frequently used Fitting stochastic population models to data under
to study density dependence (Dennis and Taper 1994) no sampling error
via the autoregressive parameter a1 . When a1 is 1 the In some simple cases, parameter estimates of stochas-
dependence of the growth rate on abundance disap- tic populations models can be obtained fairly easily.
pears, implying density independence. Values of a1 less For example, when the exponential growth model with
than 1 imply a negative effect of increased density on environmental stochasticity is assumed to be observed
the growth rate. If a1 is also larger than −1, Nt will reach without error, the logarithms of the annual growth
an equilibrium value around which the population will rates are independent and normally distributed
fluctuate. The equilibrium value is exp(a0 /(1 − a1 )) and
log(Nt+1 /Nt ) ∼ Normal(r, σ 2 ).
may be interpreted as a stochastic carrying capacity in
a manner similar to K in the stochastic Ricker model in Estimates of r and σ 2 can therefore be obtained from,
eqn. (7.4). respectively, the sample mean and the sample vari-
By adding further autoregressive terms to the model, ance of the logarithm of the observed annual growth
delayed density dependence can be studied. Delayed rates (Dennis et al., 1991; Morris and Doak, 2002). Most
density dependence occurs when population size at stochastic models, however, cannot be transformed
previous time steps affects current growth rates. This to a set of independently observed quantities in this
can happen through age structure in the population way. One then has to be aware of the temporal sta-
(Fromentin et al., 2001), through interactions with oth- tistical dependence among abundances, induced by
er species (Turchin, 2003), or due to effects of unknown the stochastic population model. Handling this depen-
autocorrelated environmental variables (Ripa and Ives, dence when estimating stochastic population models,
2007). In the first two cases it may be argued that and dynamic statistical models in general, requires
fundamental biological processes are missing from the some care. Software specifically adapted for time series
model, but adding delayed density dependence can be is available for certain models and greatly simplifies
seen as a phenomenological short-cut for modelling model estimation. In particular, there are several soft-
the desired dynamical properties without making the ware options available for fitting AR models. This is
resulting model overly complex. The simplest model also the case for ARMA models that are an extension of
with delayed density dependence is the AR(2) model AR models that include moving average (MA) terms.
The MA terms in ARMA models can be useful for
Xt+1 = a0 + a1 Xt + a2 Xt−1 + ηt
implicitly accounting for measurement error (in con-
The second order autoregressive coefficient a2 repre- trast to the more explicit approach described in the
sents effects of the abundance two time steps previous- section 7.3.3), or for accounting for interactions with
ly, not captured by the effect of abundance in the previ- other, unobserved, species (Ives et al., 2010). For non-
ous time step. This model can, in contrast to the AR(1) standard models for which dedicated software is not
model, produce quasi-cycles in dynamics (stochastic readily available, common approaches are to fit models
cycles which get out of phase over time). It has been using maximum likelihood (ML) or restricted maxi-
used to study cycles and spatial synchrony (Buonaccor- mum likelihood (REML) methods in a frequentistic
si et al., 2001) in birds (Watson et al., 2000) and small framework, or Markov chain Monte Carlo (MCMC)
in a Bayesian framework. Likelihoods for stochastic for the unobserved population ‘states’ Nt and an obser-
population models observed without error can often vation error model for which observations yt at differ-
be computed without too much trouble. For example, ent points in time are independent given the popula-
a likelihood for the stochastic discrete Ricker model tion process. Thus, the temporal statistical dependence
is similar to the likelihood of a linear regression of in the data is assumed to arise from the population
log(Nt+1 /Nt ) against Nt . Therefore, ML estimates for process only.
this and similar, models can be obtained using stan- This basic model can be extended in several ways.
dard regression software. These estimates, however, One example is the Gompertz state-space model with
do not necessarily behave in the same way as they log-normal observation error (Dennis et al., 2006). The
do in an ordinary regression problem. In contrast to model is obtained by replacing the exponential growth
standard regression, the temporal dependence in time model by a Gompertz model in eqn. (7.6). Taking the
series can cause estimates to be biased in small sam- logarithm of abundance and observations, this model
ples, which can be problematic in practice since most may be written
ecological time series are short, and uncertainty esti-
Xt+1 = a0 + a1 Xt + ηt
mates obtained from regression are typically not valid (7.7)
log(yt ) = Xt + ϵt
in the time series setting. One way to explore these
issues is to use simulations (Bence 1995). The Gompertz model, which includes the exponential
model as a special case (a1 = 1), is in the form of a lin-
ear Gaussian state-space model. Essentially, this means
7.3.3 State-space models
that both the observation and process equations are
The two approaches to estimation of dynamic mod- linear with Gaussian errors. State-space models that
els from abundance data we have seen so far assumed are linear and Gaussian are relatively easy to work
either that the population process is deterministic or with because efficient computational tools are available
that abundance was observed without error. Both of for them, in particular the Kalman filter. Two of the
these assumptions are usually unsatisfactory, as in most influential statistical books on state-space mod-
reality we expect considerable unpredictability in the els are almost exclusively devoted to models of this
population process, as well as difficulties in accu- kind (Harvey, 1990; Durbin and Koopman, 2012). Lin-
rately measuring abundance. Moreover, observation ear state-space models can have multivariate observa-
error has been shown to have the capacity to severe- tions as well as multivariate states, which allows joint
ly bias estimates of population parameters and pro- modelling of spatially separated populations, inclusion
cesses in dynamic models when data are assumed of age structure, species interactions, incorporating
to be observed without error. Biased estimates under lagged density dependence, seasonal effects, dealing
these circumstances have been reported for density with missing covariates and covariates observed with
dependence (Shenk et al., 1998; Lebreton and Gimenez, error (Almaraz et al., 2012), or dynamic factor analysis
2013), extinction risk, sustainable harvesting strategies to look for common trends among multiple popula-
(Walters and Ludwig, 1981), and effects of autocorrelat- tion time series (Zuur et al., 2003). In short, linear
ed environmental covariates (Lindén and Knape, 2009). state-space models offer many possibilities and a lot of
State-space models deal simultaneously with observa- flexibility.
tion and population process errors and may be used to For some types of data a log-normal or Gaus-
try to overcome these problems and to provide more sian observation distribution may not be appropriate,
realistic descriptions of data. and it is necessary to go beyond the linear state-
A simple state-space model is the exponential space model framework. The observation equation
growth state-space model (Lindley 2003; Holmes et al. may then be changed, for example to a negative bino-
2007) mial distribution for counts or to a beta-binomial
distribution for pin-point cover data (Damgaard, 2012).
Nt+1 = Nt exp(r + ηt )
(7.6) Similarly, nonlinear dynamics such as the Ricker
yt = Nt exp(ϵt )
model can be accommodated by modifying the pop-
The model combines the stochastic exponential growth ulation process model. State-space models may also
model from eqn. (7.3) with the observation model be used to combine a population dynamics mod-
from the exponential observation error model from el with an observation model derived from more
eqn. (7.2). It illustrates the fundamental structure of complex sampling designs within a single model. For
population state-space models, with a dynamic model instance, a model of population dynamics may be
combined with a model for distance sampling data other, each plot receiving a different fertilisation
and be jointly estimated. It is also possible to plug regime.
in existing estimates of variances in population abun- One may ask to what extent there is synchrony in
dances, along with the abundance estimates them- the dynamics among the plots. Are the populations in
selves, into state-space models in a two-stage approach the different plots behaving independently, in strong
(Newman et al., 2014). synchrony, for example due to shared external (e.g.
Fitting state-space models to data can be done by weather) or internal (e.g. seed dispersal) drivers, or
specialised algorithms, by numerical ML estimation somewhere in between where both local features of the
in cases where the likelihood can be readily comput- plots and shared drivers contribute to the dynamics.
ed (notably the linear state-space model), in Bayesian To address these questions we use a state-space mod-
settings using MCMC, or by various kinds of approxi- el to deal with observation error and to analyse the
mation methods. structure of the temporal variation. Handling of obser-
The joint treatment of observation and process errors vation error is important in this context since it will
makes state-space models compelling, and, for many tend to dilute any synchrony in dynamics among the
reasons, they are often to be preferred over meth- plots.
ods assuming only process error, or only observation For each of the 18 plots the data consist of pres-
error. However, they also bring substantial computa- ence of the species in 25 random 0.1 m2 quadrats. The
tional and complexity costs. Estimating a state-space quadrats are small relative to the size of the plots. It
model often takes orders of magnitude more time is therefore reasonable to assume that the observed
than a simpler alternative. For small or moderate- species presences in the quadrats are independent sam-
ly sized data sets, model fitting can be reasonably ples of the occurrence in each plot. We therefore model
quick despite longer computation times than alter- observation error, which in this case is due to sampling
native methods, but it can be more problematic for variation, using a binomial model:
larger data sets. Furthermore, it is easy to overparam-
eterise state-space models by including parameters for yit ∼ Bin(25, pit )
which there is not enough information in the data to
estimate them well (Knape, 2008). Such identification where yit is the number of observed occurrences in
problems can be particularly acute in the state-space plot i in year t and pit is the probability of the species
model setting because there is usually considerable occurring in a random quadrat. This probability may
uncertainty about the observation and process error be viewed as an index of abundance, observed with
variances, unless one includes explicit information on error, whose dynamics we are interested in. Since the
either of them. For example, explicit information about index pit is a probability it is restricted to the interval
observation error may be included from replicate mea- from 0 to 1, and we model it at the logit scale to satisfy
surements of the same population within short time this restriction. Hence, we define the dynamic process
periods and can aid the estimation of the observa- model in terms of xit = logit(pit ). For this we use the
tion error variance. In general, problems with identi- AR(1) process in eqn. (7.5) as a starting point, but add
fication are not always intuitive or easy to diagnose, some components to it.
but tools aimed at checking estimability can be useful First, potential synchrony could be caused by expo-
(Lele et al., 2007). sure to common fluctuating environmental variables.
We include the average amount of rainfall in June, July,
and August as one such potential variable. Addition-
Example: thistles in wheat fields ally, in 1994, 2001, 2008, and 2015–2016 all plots were
At Rothamsted in the United Kingdom, long-term fallowed and no data were collected. To account for
experiments with wheat fields have been run- effects of the fallows we include a separate param-
ning since the late 1850s (Moss et al., 2004). As eter for each of the four fallow periods that allow
part of these experiments, data on weed species for instant changes in abundance. Second, the error
are collected at wheat fields that are not treat- terms η in the AR(1) model in eqn. (7.5) represent pro-
ed with herbicides. Here we consider annual cess error, in other words unexplained variation in
data on the perennial creeping thistle (Cirsium the growth rates among years. To study synchrony in
arvense) (Tiley, 2010) from 1991 to 2018 at 18 plots addition to that potentially caused by the measured
that consist of small fields lined up next to each covariates, rain and fallow, we allow the process errors
to be correlated among plots. Specifically, we assume 7.4 Conclusions

that the variance of the process errors consists of two
When analysing population dynamics, it is common
components, σn2 and σc2 , with a total variance σn2 + σc2 .
that the only available data are estimates of popula-
The first component is the variance due to variation
tion sizes or indices of relative abundance. This chapter
in local dynamics and corresponds to errors that are
focused on such situations where demographically
independent among plots. The second component is
explicit data are not available or not used.
variance that is shared among plots in such a way that
Multiple demographic mechanisms can give rise to
the covariance is allowed to decline exponentially as
very similar patterns of abundance. The lack of direct
a function of distance with cov(ηi , ηj ) = σc2 exp(−ϕ ·
information about demographic processes in abun-
distance(i, j)) for plot i and j. In the special case that
dance data implies that we cannot make strong infer-
ϕ is equal or close to zero the dependence of the
ence about the demographic drivers behind abundance
correlation on distance disappears and the correla-
patterns. Instead, the main aim of most abundance
tion is the same regardless of distance. This is often
analyses is to reveal the patterns themselves. Typical
referred to as compound symmetry and would sug-
analyses done for abundance data are trend estimation,
gest that factors that are shared among all plots are
indicating the state and direction of population status,
driving synchrony. On the other hand, the correla-
assessment of population characteristics such as the
tion dropping with distance would suggest that fac-
growth potential, linking population change to envi-
tors operating at a smaller spatial scale are driving
ronmental variables, estimating density dependence
synchrony.
or cycles, or assessing spatial synchrony and other
The resulting process model at the logit scale is
spatiotemporal patterns in population fluctuations.
described by the equation
Consistent with the information content in the data,
xit+1 = a0i + a1 xit + c · raint + dft 1(fallow in year t) + ηit analytic models for abundance data usually contain lit-
tle detail about demographic mechanisms. Rather, joint
where c is the slope for the effect of rain and dft
effects of demographic processes are combined into a
are effects for the fallow years, with 1(.) an indica-
fairly simple process model. Stochastic components of
tor function being 1 if there was a fallow in year t
the analytic models can be divided into observation
and zero otherwise. We assume that the autoregres-
errors and process errors. Observation errors are due to
sive parameter, the effect of rainfall, and the response
randomness in the counting or estimation uncertainty,
to the fallowing are identical among plots at the log-
that is, variability in the data which does not exist in the
it scale. Intercepts a0i are allowed to vary among plots
actual population. Process error, in turn, is true varia-
to allow for different rates of growth due to the differ-
tion in the population dynamics, which cannot be pre-
ent fertilisation regimes. The initial logit scale indices
dicted by the model and is therefore treated as random
at year 1, xi1 , are treated as unknown parameters to be
unexplained variation. Process errors may be due to
estimated.
unmeasured or missed environmental variables, varia-
We fit a Bayesian version of the model to data using
tion in population structure, or so-called demographic
a custom implementation in the statistical program-
stochasticity. Simple statistical models of abundance
ming language Stan (Carpenter et al., 2017). Figure 7.3
data presume that all randomness is either observation
shows the estimated dynamics and parameters, and
errors or process error, but in the real world both are
how the correlation in the dynamics among plots
inevitably present to some extent. State-space models
changes as a function of the distance between the
are a group of models that simultaneously include both
plots. The results suggest that a large part of the syn-
sources of stochasticity and may often provide better
chrony in dynamics is mainly driven by local fea-
inference, but they are technically more challenging to
tures that extend only a few plots away, with only
fit and they require more from the data.
some potential synchrony remaining between plots
Analyses of population abundance are, arguably,
at opposite ends. We further see that there is some
most informative and useful when the data consist
evidence for a slight positive effect of rainfall and
of long time series and have spatial replication. This
that populations were reduced after the fallows in
often allows drawing conclusions about abundance
2001, 2008, and 2015–2016 but not after the fallow
patterns at scales in time and space that are otherwise
in 1994.
not possible, or practical, to study.
(a)
Plot: 1 Plot: 2
0.6
0.4
0.2
0.0
p
Plot: 3 Plot: 4
0.6
0.4
0.2
0.0
1990 2000 2010 1990 2000 2010
Year
(b)
3
−3
x
−6
−9
1990 2000 2010
Year
(c)
1.00 2
0.75 0
Correlation
Coefficient
0.50 −2
0.25 −4
0.00 −6
0 5 10 15 Rain Fallow 1 Fallow 2 Fallow 3 Fallow 4
Distance Coefficient
Figure 7.3 (a) One step ahead predictions of the abundance index at the first four plots. Black points show the estimated abundance index p in
each year. Red lines connect these estimated indices to the abundance index predicted by the model in the following year. (b) Same as (a) but for
all 18 plots together and on the logit scale. (c) Estimated correlation in process errors among the plots as a function of distance between them.
(d) Estimates of coefficients (at the logit scale) for rainfall and each of the four fallows. In (c) and (d) green lines show 90% credible intervals.
References Dennis, B. & Taper, M.L. (1994). Density dependence in time

series observations of natural populations: estimation and
Almaraz, P., Green, A.J., Aguilera, E., Rendón, M.A. & Bus-
testing. Ecological Monographs, 64, 205–224.
tamante, J. (2012). Estimating partial observability and
Dunstan, P.K., Foster, S.D., Hui, F.K.C. & Warton, D.I.
nonlinear climate effects on stochastic community dynam-
(2013). Finite mixture of regression modeling for high-
ics of migratory waterfowl. Journal of Animal Ecology, 81,
dimensional count and biomass data in ecology. Journal
1113–1125.
of Agricultural, Biological, and Environmental Statistics, 18,
Ådahl, E., Lundberg, P. & Jonzén, N. (2006). From climate
357–375.
change to population change: The need to consider annual
Durbin, J. & Koopman, S.J. (2012). Time Series Aanalysis by
life cycles. Global Change Biology, 12, 1627–1633.
State Space Methods. Oxford University Press.
Barker, R.J., Schofield, M.R., Link, W.A. & Sauer, J.R. (2018).
Engen, S., Bakke, Ø. & Islam, A. (1998). Demographic and
On the reliability of N-mixture models for count data.
environmental stochasticity-concepts and definitions. Bio-
Biometrics, 74, 369–377.
metrics, 54, 840–846.
Barraquand, F., Louca, S., Abbott, K.C., et al. (2017). Moving
Fromentin, J.M., Myers, R.A., Bjornstad, O.N., et al. (2001).
forward in circles: Challenges and opportunities in mod-
Effects of density-dependent and stochastic processes on
elling population cycles. Ecology Letters, 20, 1074–1092.
the regulation of cod populations. Ecology, 82, 567–579.
Bence, J.R. (1995). Analysis of short time series: correcting
Gause, G.F. (1932). Experimental studies on the struggle for
for autocorrelation. Ecology, 76, 628–639.
existence: I. Mixed population of two species of yeast.
Bolker, B.M. (2008). Ecological Models and Data in R. Princeton
Journal of Experimental Biology, 9, 389–402.
University Press.
Harvey, A.C. (1990). Forecasting, Structural Time Series Models
Brännström, Å. & Sumpter, D.J.T. (2006). Stochastic ana-
and the Kalman Filter. Cambridge University Press.
logues of deterministic single-species population models.
Hixon, M.A., Pacala, S.W. & Sandin, S.A. (2002). Popu-
Theoretical Population Biology, 69, 442–451.
lation regulation: Historical context and contemporary
Buckland, S.T., Rexstad, E.A., Marques, T.A. & Oedekoven,
challenges of open vs. closed systems. Ecology, 83,
C.S. (2015). Distance Sampling: Methods and Applications.
1490–1508.
Springer.
Holmes, E.E. (2001). Estimating risks in declining popula-
Bullock, J. (1999). Plants. Ecological Census Techniques: A
tions with poor data. Proceedings of the National Academy of
Handbook. Cambridge University Press Cambridge.
Sciences of the United States of America, 98, 5072–5077.
Buonaccorsi, J.P., Elkinton, J.S., Evans, S.R. & Liebhold, A.M.,
Holmes, E.E., Sabo, J.L., Viscido, S.V. & Fagan, W.F. (2007). A
et al. (2001). Measuring and testing for spatial synchrony.
statistical approach to quasi-extinction forecasting. Ecolo-
Ecology, 82, 1668–1679.
gy Letters, 10, 1182–1198.
Carpenter, B., Gelman, A., Hoffman, M.D., et al. (2017). Ives, A.R. (1995). Predicting the response of populations to
Stan: A probabilistic programming language. Journal of environmental change. Ecology, 76, 926–941.
Statistical Software, 76, 1–32. Ives, A.R., Abbott, K.C. & Ziebarth, N.L. (2010). Analysis of
Chevalier, M., Russell, J.C. & Knape, J., et al. (2019). ecological time series with ARMA(p,q) models. Ecology,
New measures for evaluation of environmental perturba- 91, 858–871.
tions using Before-After-Control-Impact analyses. Ecolog- Jacquemyn, H., Brys, R., Hermy, M. & Willems, J.H. (2007).
ical Applications, 29, e01838. Long-term dynamics and population viability in one of
Cornulier, T., Yoccoz, N.G., Bretagnolle, V., et al. (2013). the last populations of the endangered Spiranthes spiralis
Europe-wide dampening of population cycles in keystone (Orchidaceae) in the Netherlands. Biological Conservation,
herbivores. Science, 340, 63–66. 134, 14–21.
Cramp, S. (1985). Handbook of the Birds of Europe, the Mid- Kéry, M., Royle, J. & Schmid, H. (2005). Modeling avian
dle East, and North Africa: Volume IV Terns to Woodpeckers. abundance from replicated counts using binomial mixture
Oxford University Press, Oxford. models. Ecological Applications, 15, 1450–1461.
Damgaard, C. (2012). Trend analyses of hierarchical pin- Knape, J. (2016). Decomposing trends in Swedish bird popu-
point cover data. Ecology, 93, 1269–1274. lations using generalized additive mixed models. Journal
Dennis, E.B., Freeman, S.N., Brereton, T. & Roy, D.B. (2013). of Applied Ecology, 53, 1852–1861.
Indexing butterfly abundance whilst accounting for miss- Knape, J. (2008). Estimability of density dependence in mod-
ing counts and variability in seasonal pattern. Methods in els of time series data. Ecology, 89, 2994–3000.
Ecology and Evolution, 4, 637–645. Knape, J., Arlt, D., Barraquand, F., Berg, Å., Chevalier, M.,
Dennis, B., Munholland, P.L. & Scott, J.M. (1991). Estima- Pärt, T., Ruete, A. & Żmihorski, M. (2018). Sensitivi-
tion of growth and extinction parameters for endangered ty of binomial N-mixture models to overdispersion: The
species. Ecological Monographs, 61, 115–143. importance of assessing model fit. Methods in Ecology and
Dennis, B., Ponciano, J.M., Lele, S.R., Taper, M.L. & Sta- Evolution, 9, 2102–2114.
ples, D.F. (2006). Estimating density dependence, process Lebreton, J.-D. & Gimenez, O. (2013). Detecting and estimat-
noise, and observation error. Ecological Monographs, 76, ing density dependence in wildlife populations. Journal of
323–341. Wildlife Management, 77, 12–23.
Lele, S.R., Dennis, B. & Lutscher, F. (2007). Data cloning: Ramula, S., Puhakainen, L., Suhonen, J. & Vallius, E. (2008).
Easy maximum likelihood estimation for complex ecolog- Management actions are required to improve the viabili-
ical models using Bayesian Markov chain Monte Carlo ty of the rare grassland herb Carlina biebersteinii. Nordic
methods. Ecology Letters, 10, 551–563. Journal of Botany, 26, 83–90.
Liebhold, A., Koenig, W.D. & Bjørnstad, O.N. (2004). Spa- Ranta, E., Kaitala, V., Lindstrom, J. & Linden, H., et al. (1995).
tial synchrony in population dynamics. Annual Review of Synchrony in population dynamics. Proceedings: Biological
Ecology, Evolution, and Systematics, 35, 467–490. Sciences, 262, 113–118.
Lindén, A. & Knape, J. (2009). Estimating environmental Reynolds, J.H., Knutson, M.G., Newman, K.B., Silverman,
effects on population dynamics: Consequences of obser- E.D. & Thompson, W.L. (2016). A road map for design-
vation error. Oikos, 118, 675–680. ing and implementing a biological monitoring program.
Lindén, A. & Mäntyniemi, S. (2011). Using the negative bino- Environmental Monitoring and Assessment, 188, 399.
mial distribution to model overdispersion in ecological Ripa, J. & Heino, M. (1999). Linear analysis solves two puz-
count data. Ecology, 92, 1414–1421. zles in population dynamics: The route to extinction and
Lindley, S.T. (2003). Estimation of population growth and extinction in coloured environments. Ecology Letters, 2,
extinction parameters from noisy data. Ecological Applica- 219–222.
tions, 13, 806–813. Ripa, J. & Ives, A.R. (2007). Interaction assessments in
May, R.M. (1976). Simple mathematical models with very correlated and autocorrelated environments. The Impact
complicated dynamics. Nature, 261, 459–467. of Environmental Variability on Ecological Systems (eds
Månsson, L., Ripa, J. & Lundberg, P. (2007). Time series mod- D.A. Vasseur & K.S. McCann), pp. 111–131. Springer
elling and trophic interactions: Rainfall, vegetation and Netherlands.
ungulate dynamics. Population Ecology, 49, 287–296. Rochester, C.J., Brehme, C.S., Clark, D.R., et al. (2010).
McCrea, R. & Morgan, B. (2015). Analysis of Capture-Recapture Reptile and amphibian responses to large-scale wild-
Data. Taylor & Francis, Boca Raton. fires in southern California. Journal of Herpetology, 44,
Moore, N.W. (1991). The development of dragonfly com- 333–351.
munities and the consequences of territorial behaviour: A Royama, T. (1992). Analytical Population Dynamics. Springer
27 year study on small ponds at Woodwalton Fen, Cam- Netherlands.
bridgeshire, United Kingdom. Odonatologica, 20, 203–231. Royle, J.A. (2004). N-mixture models for estimating popu-
Morris, W.F. & Doak, D.F., et al. (2002). Quantitative Conser- lation size from spatially replicated counts. Biometrics, 60,
vation Biology. Sinauer. 108–115.
Moss, S.R., Storkey, J., Cussans, J.W., Perryman, S.A.M. & Ruokolainen, L., Linden, A., Kaitala, V. & Fowler, M.S.
Hewitt, M.V. (2004). The Broadbalk long-term experiment (2009). Ecological and evolutionary dynamics under
at Rothamsted: What has it told us about weeds? Weed coloured environmental variation. Trends in Ecology &
Science, 52, 864–873. Evolution, 24, 555–563.
Newman, K.B., Buckland, S.T., Morgan, B.J.T., et al. (2014). Shenk, T.M., White, G.C. & Burnham, K.P. (1998). Sampling-
Modelling population dynamics using closed-population variance effects on detecting density dependence from
abundance estimates. Modelling Population Dynamics: temporal trends in natural populations. Ecological Mono-
Model Formulation, Fitting and Assessment Using State-Space graphs, 68, 445–463.
Methods (eds K.B. Newman, S.T. Buckland, B.J.T. Morgan, Sköld, M. & Knape, J. (2018). Bounding reproductive rates
R. King, D.L. Borchers, D.J. Cole, P. Besbeas, O. Gimenez in state-space models for animal population dynamics.
& L. Thomas), pp. 123–145. Methods in Statistical Ecology. Ecosphere, 9, e02215.
Springer. Stenseth, N.C., Mysterud, A., Ottersen, G., et al. (2002).
Nisbet, R.M. & Gurney, W. (1982). Modelling Fluctuating Ecological effects of climate fluctuations. Science, 297,
Populations. John Wiley & Sons. 1292–1296.
Oedekoven, C.S., Elston, D.A., Harrison, P.J., et al. (2017). Sutherland, W.J. (Ed.). (2006). Ecological Census Techniques:
Attributing changes in the distribution of species abun- A Handbook, 2nd edn. Cambridge University Press.
dance to weather variables using the example of British Tayleur, C., Caplat, P., Massimino, D., et al. (2015). Swedish
breeding birds. Methods in Ecology and Evolution, 8, birds are tracking temperature but not rainfall: Evidence
1690–1702. from a decade of abundance changes. Global Ecology and
Peltonen, M., Liebhold, A.M., Bjørnstad, O.N. & Williams, Biogeography, 24, 859–872.
D.W. (2002). Spatial synchrony in forest insect outbreaks: Tiley, G.E.D. (2010). Biological flora of the British Isles: Cirsi-
roles of regional stochasticity and dispersal. Ecology, 83, um arvense (L.) Scop. Journal of Ecology, 98, 938–983.
3120–3129. Turchin, P. (2003). Complex Population Dynamics: A Theoreti-
Pépino, M., Rodríguez, M.A. & Magnan, P. (2012). Impacts cal/Empirical Synthesis. Princeton University Press.
of highway crossings on density of brook charr in streams. Turchin, P. (1995). Population regulation: old arguments
Journal of Applied Ecology, 49, 395–403. and a new synthesis. Population Dynamics: New Approach-
Quinn, T.J. & Deriso, R.B. (1999). Quantitative Fish Dynamics. es and Synthesis (eds N. Capuccino & P. Price), pp. 19–40.
Oxford University Press. Elsevier.
Vanhatalo, J., Hosack, G.R. & Sweatman, H. (2017). Spa- Watson, A., Moss, R. & Rothery, P. (2000). Weather and syn-
tiotemporal modelling of crown-of-thorns starfish out- chrony in 10-Year population cycles of rock ptarmigan and
breaks on the Great Barrier Reef to inform control red grouse in Scotland. Ecology, 81, 2126–2136.
strategies (M. Bode, Ed.). Journal of Applied Ecology, 54, Wood, S.N. (2010). Statistical inference for noisy nonlinear
188–197. ecological dynamic systems. Nature, 466, 1102–1104.
Walters, C.J. & Ludwig, D. (1981). Effects of measurement Zuur, A.F., Tuck, I.D. & Bailey, N. (2003). Dynamic fac-
errors on the assessment of stock-recruitment relation- tor analysis to estimate common trends in fisheries time
ships. Canadian Journal of Fisheries and Aquatic Sciences, 38, series. Canadian Journal of Fisheries and Aquatic Sciences, 60,
704–710. 542–552.
CHAPTER 8
Life tables: construction

and interpretation
Owen R. Jones
8.1 Introduction Jones and Vaupel 2017), within species between the
sexes (Clutton-Brock and Isvaran 2007; Bonduriansky
There is remarkable diversity in the life history
et al. 2009; Maklakov et al. 2009; Griffinet al. 2017),
strategies employed by different species and among
or depending on environmental conditions, especial-
populations. What unites these diverse strategies is
ly during early development (Descamps et al. 2008;
the fundamental balancing equation for population
Reed et al. 2008; Bouwhuis et al. 2010; Boonekamp
growth: Nt+1 = Nt + Bt – Dt + It – Et , where the pop-
et al. 2014).
ulation’s size at time t+1 (Nt+1 ) is calculated from the
Life tables have been a fundamental tool for explor-
population size at time t (Nt ), and where B repre-
ing this diversity by showing how various demo-
sents the number of offspring produced, D represents
graphic metrics, chief among them the probability of
deaths, and I and E represent immigration and emigra-
death and fertility, change with advancing age. In this
tion respectively between time t and t+1. Where births
chapter, I aim to provide broad coverage of the calcu-
and immigration exceed deaths and emigration, the
lation and use of life tables, with a particular focus on
population will grow. It is convenient to consider only
nonhuman animals.
closed populations so that Nt+1 = Nt + Bt – Dt .
The use of life tables has a long history. They orig-
These equations break down population growth into
inate in John Graunt’s bills of mortality published
processes of birth and death (and migration) that occur
in 1662 (Sutherland 1963) and have been extensively
to individuals. However, the equations assume that
used in all aspects of human demography for centuries
all individuals have identical probabilities of death
(Prestonet al. 2000). Their use has not been restrict-
and of producing offspring, which is of course usu-
ed to humans, but serious use in nonhumans began
ally not true. A major source of demographic hetero-
only in the twentieth century. Perhaps the earliest non-
geneity is age (or developmental stage). For example,
human life tables were those published in a series of
over their life course both birds and mammals tend to
articles led by Raymond Pearl exploring mortality pat-
experience bathtub-shaped mortality trajectories (Siler
terns using fruit fly (Drosophila melanogaster) beginning
1979) with relatively high infant and juvenile mortal-
with Pearl and Parker (1921). A later article in the
ity that declines with age towards the age of matu-
series broadened the taxonomic scope by presenting
rity, and eventually increases with the march of time
life tables for other ‘lower organisms’, including var-
and effects of senescence (Jones et al. 2008). Fertili-
ious other types of Drosophila, the freshwater Hydra
ty also varies through the mammalian life course: In
fusca (now Hydra oligactis), the slug Agriolimax agrestis
humans, fertility is unimodal, increasing to a peak
(now Deroceras agreste), the Oriental cockroach Blat-
and then declining with menopause, while in most
ta orientalis, and mice (Mus musculus), with the aim
other mammals there is no such menopausal fertili-
of comparing survivorship curves to understand how
ty decline, but rather an increase to a plateau after
mortality varies with age and among species (Pearl
maturity (Johnstone and Cant 2019). These patterns
and Miner 1935). Pearl and Miner used these com-
are not universal and demographic trajectories show
parisons to develop a categorisation of survivorship
remarkable diversity among species (Jones et al. 2014;
Owen R. Jones, Life tables: construction and interpretation. In: Demographic Methods Across the Tree of Life.
curves based on the distribution of mortality risk 4

through the life course. Deevey (1947) further used
life tables to explore mortality in species as diverse 3
as barnacles (Semibalanus balanoides), Dall’s sheep (Ovis
Age (years)
dalli), blackbird (Turdus merula), and rotifer (Floscu- 2
laria conifera) and redefined Pearl and Miner’s (1935)
survivorship categorisation as the now-familiar type
1
I, II, and III survivorship, which result from increas-
ing, constant, and decreasing mortality risk respective-
0
ly (Pearl and Miner 1935; Deevey 1947; Jones 2016).
0 1 2 3 4 5 6
Later work by Caughley (1966) used life tables to com-
Time (years)
pare mortality trajectories among mammals, with the
central finding that, despite great variation in typical Figure 8.1 Lexis diagram illustrating the difference between cohort
life span, mammals generally have U- or ‘fish-hook’- and period approaches. The figure shows the fate of three cohorts of
shaped mortality trajectories. four individuals. Time is represented by the horizontal axis and age by
Since then, life tables have been put to a diversity the vertical axis. Each line represents an individual born at the point
of uses including studying sex differences in mortality marked with a filled circle and dying at the time marked with an open
(e.g. Clutton-Brock and Isvaran 2007), testing theo- circle. The height of the open circle on the vertical axis thus represents
ries of life history evolution (Purvis and Harvey 1995) age at death. The pink area highlights the cohort born in the interval
0–1. The cohort can be followed until all four individuals have died.
and human evolution (Burger et al. 2012; Colchero
The green area highlights the period in the interval 2–3. There are six
et al. 2016; e.g. Levitis et al. 2013), and the effects of individuals that enter the period at time=2 at various ages, and three
caloric restriction on mortality (e.g. Carey et al. 2002), of these individuals die. Four new individuals are born, which all
among many others. Although most nonhuman life survive until the end of the period. To calculate mortality risk, the
tables have been used for animal populations, they period life table methods must account for the number of individuals
have also been used to study plant demography, from ‘at risk’ during the period.
herbs (Leverich and Levin 1979; Hegazy 1990) to trees
(Van Valen 1975; Harcombe 1987). Life tables form the
basis for age-structured population modelling, includ- interval (i.e. the population at risk) and the number
ing the application of the Euler–Lotka equation (Lotka that die and the number of offspring produced with-
1907), which is perhaps the most important equation in this interval in order to correctly calculate mortality
in mathematical demography. Life tables have obvious and fertility. The calculation of life tables can be divid-
similarities with age-based (Leslie) matrix population ed into approaches—cohort-based and period-based
models (MPMs) (Leslie 1945; Caswell 2001), and pro- (also known as static or vertical life tables)—where, in
ducing an MPM from a life table that includes both both cases, the rows correspond to age intervals and
survival and fertility information is straightforward the first column indicates the age intervals over which
(and vice versa). It is also possible to approximate an the life table is calculated. The choice of interval large-
age-based life table from a stage-based (Lefkovitch) ly depends on the preferences of the researcher and on
MPM (Cochran and Ellner 1992; Caswell 2001). MPMs the life history of the species. It is often a year, but not
are covered in detail in Chapter 9. necessarily so, and can vary within the life table. For
In this chapter, I focus on single decrement life example, in human life tables the age intervals at the
tables. That is, life tables where individuals have only start of life are commonly 1 year, then 4 years, and then
one possible exit from a state (i.e. death). Multiple broaden to 5 years in later life.
decrement life tables allow more than one exit (e.g. As their names imply, cohort-based methods aim
multiple mutually exclusive causes of death) and are to track the fate of a set of individuals born within a
uncommon in nonhuman studies (see Chapter 4 in particular time-frame (e.g. in a particular year), while
Preston et al. (2000) for excellent coverage of this topic). period life tables focus on the fate of a population of
mixed ages during a particular time period such as a
year (Figure 8.1). These differences can be important,
and we should expect measures derived from cohort
8.2 Cohort and period life tables and period life tables to align only if the environment
The central aim of most life tables is to describe how has been constant and unchanging. There is no agreed
the risk of death, and sometimes fertility, changes with universal standard for the life table layout or nomen-
age. The life table, therefore, makes an accounting of clature, but in this chapter (see Tables 8.1, 8.2, and 8.3) I
the number of individuals that enter a particular age follow the nomenclature of Preston et al. (2000). Here
L I F E TA B L E S : C O N S T R U C T I O N A N D I N T E R P R E TAT I O N 153
subscripts n and x define the length and start of the and follow them until (ideally) all individuals in the
age interval respectively (i.e. n px is the probability of group are dead (Figure 8.1). The raw data used for
survival from age x to x+n). Where the age interval is the construction of these life tables are thus the age
consistently the same (e.g. 1 year), the n subscript is at death of individuals, which may also be record-
normally omitted. ed as the number of individuals entering each age
In the published literature, life table-like data are interval, and the methods assume that all individu-
reported in diverse ways: Sometimes data are present- als in the population are equally likely to be sam-
ed as numbers of individuals entering an age interval pled ( Caughley 1977; Schwartz and Armitage 1998;
and/or dying in the interval, while at other times Preston et al. 2000). I illustrate a cohort life table by
data may be presented as the proportion surviving calculating a life table for tree swallows (Tachycine-
to a particular age (i.e. survivorship standardised to ta bicolor) using cohort-based data obtained from the
begin at 1); or data may be reported as age-specific DATLife database ( Max-Planck Institute for Demo-
probability of survival or (less-commonly) hazard or graphic Research [Germany] 2018 ), which sources the
death rate of individuals in the interval (the slope of original data from Butler (1988) (Table 8.1). The first
the survivorship curve multiplied by –1). The close column (x) represents the exact age at the start of the
mathematical relationships among columns in the life interval, and the length of the interval (n) is the dif-
table mean that usually a complete actuarial life table ference between the values of x in consecutive rows.
can be derived from these initial data, but knowl- Then, lx is the number of individuals entering the inter-
edge of sample size (e.g. initial cohort size or the val at age x, with the first entry, therefore, being the
number of individuals in an age interval) can be use- number of individuals in the entire cohort and subse-
ful to infer confidence for the estimated demograph- quent entries being the number surviving to each age
ic quantities. In sections 8.2.1 and 8.2.2, I describe (x). From this information, I can calculate the number
cohort and period approaches and illustrate them with dying between ages x and x+n (n dx ) as the difference
examples. between lx and l x+1 and the probability of dying (n qx )
or surviving (n px ) the interval as n dx /lx and 1 – (n dx /lx ),
respectively.
8.2.1 Cohort life tables Before proceeding towards the calculation of age-
Cohort life tables focus on a group of individuals born specific death rate, it is first useful to consider what
within a particular time interval, for example a year, is meant by a ‘rate’ and why this differs from a
Table 8.1 Cohort life table for tree swallows (Tachycineta bicolor) calculated for a cohort of 320 individuals from x, lx , and n ax data obtained
from DATLife. Deaths are assumed to be distributed uniformly during the interval, so n ax = 0.5. Values in columns marked by asterisks have been
rounded to 3 decimal places (dp).
Exact Number Number Probability Probability Person- Person- Life Death rate Average
age left dying of dying of years lived years lived expectancy in the number of
alive at between between surviving between above age from age cohort years lived
age x ages ages x and between ages x and x x* between in the time
x and x+n* ages x and x+n ages x and interval by
x+n x+n* x+n* those dying
in the time
interval *
x lx n dx n qx n px n Lx Tx ex n mx n ax
0 320 253 0.791 0.209 193.5 272.0 0.850 1.308 0.500

1 67 40 0.597 0.403 47 78.5 1.172 0.851 0.500
2 27 16 0.593 0.407 19 31.5 1.167 0.842 0.500
3 11 7 0.636 0.364 7.5 12.5 1.136 0.933 0.500
4 4 2 0.500 0.500 3 5.0 1.250 0.667 0.500
5 2 1 0.500 0.500 1.5 2.0 1.000 0.667 0.500
6 1 1 1.000 0.000 0.5 0.5 0.500 2.000 0.500
(n ax · n dx ). Estimation of n ax requires accurate time

10
of death data that is normally not available for non-
9
humans. A simplifying assumption is that deaths are
8
distributed uniformly through the age interval such
7 that n ax = 0.5n. However, if deaths are concentrated ear-
6 ly during the interval, for example, due to high infant
ID
5 mortality in the first interval, a smaller n ax value could

4 be appropriate for that interval. An n ax value may
3 also be informed by strong seasonality in deaths. In
2 the animal demography literature, it is often implicitly
1 assumed that individuals entering the interval survive
0 1 2 3 4 5 6 7 8 9 10 the entire interval (i.e. n ax = 1). Whether assumptions
Age (years) about ax , and consequently Lx , make an appreciable
difference to analytical outcomes depend on the life
Figure 8.2 Understanding life expectancy (e0 ) and the average history of the organism concerned and on the questions
length of time that those that die in the interval survive within the
being addressed.
interval (ax ). The ax is used to estimate Lx , the person-years lived
From n Lx one can calculate the number of person-
during the interval. The horizontal lines represent life lines for ten
years lived beyond age x(Tx ) as the sum of n Lx from
individuals, with age at death marked with an open circle. Life P
expectancy from any particular age is the average remaining length of x onwards (i.e. Tx = ∞ a=x n La ). The age-specific death
the lines. For example, life expectancy from birth (age = 0) is the rate (hazard) (n mx ) is calculated as the number of
average length of the lines (6.503 years) and life expectancy from age deaths in the age interval divided by the number
4 is the average length of the lines conditional on reaching the vertical of person-years lived in the age interval (i.e. mx =
dashed line (2.882 years). In this case, individual number 9 is 0
n dx /n Lx ). Finally, life expectancy from age x (ex is
excluded because the individual dies before reaching 4 years. In most
calculated as Tx /lx (Figure 8.2).
data sets, the precise age at death is not recorded and instead it is
only recorded that the individual did not survive to the end of the age
interval. However, dying individuals usually survive through at least 8.2.2 Period life tables
some of their ‘final’ interval (here marked with a red segment), and
this is accounted for by the ax value which is the average length of As their name suggests, period life tables focus on a
time survived within this final interval, for each age interval. Here, the particular time period, such as a year (Figure 8.1), and
overall ax is the average length of the red segments (0.503 years), but are also known as static or vertical life tables. They aim
this could vary with age: typically, ax is smaller for younger ages. to model what would happen to an imaginary (‘syn-
thetic’) cohort that experienced the mortality condi-
probability. In demography, a rate refers to the occur- tions of that period. They are sometimes favoured over
rence of an event in relation to the amount of exposure cohort life tables because following an entire cohort
to the risk of occurrence of that event. The timing of the until all members are dead can be impractical, espe-
event during the interval is therefore important: Indi- cially for longer-lived organisms. The main assump-
viduals that survive the interval are exposed to risk tions of this approach are that the age distribution is
for the entire duration of the interval, while those that stationary (i.e. the proportion of individuals in each
die are only exposed to risk until they die. I therefore age class is constant, and population growth rate is
calculate the death rate as the number of deaths occur- constant) and that every individual in the population
ring during a given time interval (e.g. a particular year has an equal probability of being sampled (Caughley
of life), divided by the number of ‘person’-years lived 1977; Schwartz and Armitage 1998; Preston et al. 2000).
during that interval (n Lx ). The difference in denomi- These are strong assumptions that are likely rarely
nator between the calculations for probability of death met in wild animals and, although ‘moderate’ devia-
and death rate mean that these two quantities are dis- tions are probably acceptable (Stearns 1992, p. 25), the
tinct. use of period life tables in fluctuating populations has
The number of ‘person’-years lived during that inter- been robustly cautioned against (Anderson et al. 1981;
val (n Lx ) is calculated as the number of individuals Menkens and Boyce 1993).
entering the interval at age x (lx ), multiplied by their Nevertheless, understanding period life tables is
probability of survival (n px ), plus the average length of useful and I will illustrate their construction using
time that those that die in the interval survive with- data for the European starling (Sturnus vulgaris)
in the interval (n ax ) (Figure 8.2): n Lx = (n px · lx ) + from the DATLife database (Max-Planck Institute
Table 8.2 Period life table for European starling (Sturnus vulgaris) calculated using x, n Nx , n Dx , and n ax data from DATLife (original data sourced
from Deevey 1947) and with l0 set to 1.000. See text for explanation of nomenclature. Noninteger values are rounded to 3 dp.
x n Nx n Dx n ax n mx n qx n px lx n dx n Lx Tx ex
0 203 99 0.500 0.488 0.392 0.608 1.000 0.392 0.804 2.051 2.051
1 166 69 0.500 0.416 0.344 0.656 0.608 0.209 0.503 1.247 2.052
2 97 51 0.500 0.526 0.416 0.584 0.399 0.166 0.316 0.744 1.866
3 46 27 0.500 0.587 0.454 0.546 0.233 0.106 0.180 0.428 1.840
4 19 8 0.500 0.421 0.348 0.652 0.127 0.044 0.105 0.248 1.953
5 11 5 0.500 0.455 0.370 0.630 0.083 0.031 0.068 0.143 1.728
6 6 4 0.500 0.667 0.500 0.500 0.052 0.026 0.039 0.076 1.450
7 2 1 0.500 0.500 0.400 0.600 0.026 0.010 0.021 0.037 1.400
8 1 1 0.000 1.000 1.000 0.000 0.016 0.016 0.016 0.016 1.000
for Demographic Research (Germany) 2018) and perfectly with the intuitive result obtained by dividing
originally sourced from Deevey (1947). The starting the number dying in the interval (800) by the number
point for constructing a period life table is the calcu- entering the interval (1,000). The derivation of this n qx
lation of age-specific death rate (hazard) (n mx ) from formula can be seen in Preston (2000, p. 43).
the number of individuals alive in the middle of an age The age-specific probability of survival (n px ) is
interval (n Nx ) and the number of those individuals that 1 – n qx . The final age interval is a special case because
die in the observation period (n Dx ): n mx = n Dx /n Nx . In it extends from x to infinity. It is, therefore, appropriate
this case (Table 8.2) there are 203 individuals within the to assume that all individuals die in this interval, that
age interval 0–1, and 99 of them die; thus, the death rate is, ∞ qx = 1.000 and ∞ px = 0.000.
is 99/203 = 0.488. (Note that I am assuming that these At this point, we can choose a population size to
n Nx values are for the middle of the age interval.) start our synthetic cohort (l0 ), the ‘radix’, and calcu-
It is worth emphasising again here that death late lx+n as lx · n px . In human life tables, the radix is
rates are distinct from probabilities of death and can often set as 1 million or 100,000, though 1.000 is also
exceed 1. This is clear if you recall that the numera- convenient because then the lx values are survivorship,
tor in the relevant equation is the number alive in the or the probability of survival to age x. The values of
middle rather than the start of the interval and that the the subsequent columns will vary in proportion to the
mortality rate is the slope of the survivorship curve value chosen for the radix. The number of individu-
multiplied by –1. The effect of this difference is best als dying in a time interval (n dx ) is lx – lx+1 , and the
illustrated with an example: In an animal population number of person-years lived in the interval (n Lx ) is
with 1,000 individuals alive at the start of the interval n · lx+n + n ax · n dx for all but the final row. For the final
and very high mortality risk, one may see 400 dying row, which has an open interval, ∞ Lx = lx /∞ mx . Person-
and 600 remaining alive by the time they reach the mid- years lived above age x (Tx ) is the sum of n Lx from x
P
dle of the interval and then a further 400 dying during onwards (i.e. Tx = ∞ a=x n La ) and life expectancy from
the second half of the interval (a total of 800 dying), age x (e0x ) is Tx /lx .
producing a mortality rate (n mx ) of 800/600 = 1.333. In addition to the cohort and period life table meth-
But how can we calculate the age-specific probability ods described above, life tables based on ‘snapshots’
of death (n qx ) from n mx ? This relies on knowledge of of the standing age structure of the population or
how deaths are distributed within the age interval: If on the ages of death of individuals across multiple
deaths are concentrated early in the interval, n qx is larg- cohorts have been used extensively in nonhumans
er. Thus, n ax and n mx can be used to calculate n qx as (Udevitz and Ballachey 1998; Udevitz and Gogan
follows: n qx = (n · n mx )/(1 + (n – n ax ) ·n mx ). For example, 2012). As with the period methods described above,
using the example from above, where n mx is 1.333, n ax these approaches normally assume that populations
is 0.5, and n = 1, we can check that we obtain a sensible are stable and stationary (Caughley 1977), though
result: 1.333/(1 + 0.5 × 1.333) = 0.800. This corresponds this assumption can be relaxed if a time series of age
structures are available (Hoenig and Gedamke 2007; age-specific fertility rate (n bx ) for Bali myna (Leu-
Udevitz and Gogan 2012). Other methods exist to copsar rothschildi). This life table is based on data
estimate life tables from nonstable populations if the from the DATLife database (Max-Planck Institute for
population growth rate (Udevitz and Ballachey 1998) Demographic Research [Germany] 2018) which was
or mortality rates (Fryxell 1986) are known. When the originally sourced from Ricklefs et al. (2003). I assume
stationarity assumptions are met, the age structure of that the fertility reported in DATLife, in this case, is
the population will be the same as the standardized based on the number of females entering the interval,
survivorship (lx , where l0 is standardised to 1), and rather than person-years lived during the interval (i.e.
it is straightforward to calculate the other columns DATLife fertility = total number of births/Nx ). I there-
in the life table from this starting point using the fore calculate n bx as a rate by multiplying the DATLife-
mathematical relationships among them. reported fertility with Nx , to calculate the number of
offspring produced by the sampled population, then
dividing by Lx to calculate the fertility rate. This fer-
tility information includes offspring of both sexes, so I
8.2.3 Fertility further divide by 2 to obtain an estimate of females only
Until now I have focused on mortality trajectories, but (i.e. I assume a 50:50 sex ratio at birth). In practice, with
information on how fertility changes with age is nec- wild animal data, it is unlikely that decisions about
essary to address many fundamental topics. For non- using Nx or Lx (and the values used for ax in the Lx cal-
humans, life tables that include fertility are generally culation) will make an appreciable difference for most
based on the female fraction of the population because use cases, but users should be aware of the potential
measuring male reproductive output is hard. Methods impacts.
for estimating fertility vary depending on the nature of The data for this life table end at age 18 when lx
the organism but usually involve carefully monitoring is still 0.116. I have arbitrarily closed the life table by
a sample of females to observe them with confirmed setting n Dx to equal n Nx for this interval. This has con-
offspring (or not). More recently, genetic sampling sequences for the calculation of some of the subsequent
methods are being used to confirm parenthood (see derived columns (e.g. life expectancy will be underesti-
Chapter 1). Like death, fertility can be expressed as a mated) and emphasises the difficulty of collecting data
rate by dividing the number of births in an interval by for cohort life tables. Nevertheless, for some impor-
the number of person-years lived by individuals enter- tant metrics that require fertility information, this is
ing the interval. Nevertheless, in many studies on wild not a problem because fertility (n bx ) has dropped to
animals, fertility is expressed as the average number of zero at these advanced ages. For example, the calcula-
female offspring born within an age interval to females tion of net reproductive rate (R0 ), generation time (T),
that enter the interval (i.e. assuming that ax = 1). and intrinsic population growth rate (r) are unaffect-
Assessment of fertility is complicated by the fact that ed by this truncation. Net reproductive rate (R0 ), or the
the number of births can be hard to estimate in wild average number of females produced by a female dur-
animals and may be ambiguous: Do we refer to live ing her life, is calculated as the sum of the lx bx series
P
births or all births (including, for example, stillbirths (R0 = lx bx ), which in this case is 4.674 female off-
and miscarriages)? Do we include offspring of both spring per female. Generation time (G), the average age
sexes or focus only on the female fraction of the pop- of the parents of all offspring produced by a cohort,
ulation? The fact that very early mortality of offspring is a particularly useful quantity (Gaillard et al. 2005)
makes detection difficult means that part of a fertility that can be calculated by weighting the sum of the
estimate may include some neonatal survival. More- production of offspring by the age at which they are
P P
over, the denominator (n Lx ) can also be ambiguous: Do produced and dividing by R0 (G = lx bx x/ lx bx ).
we include all individuals in our calculation of person- In this example, the generation time is 4.662 years.
years? Just the females? Only sexually active females? Population growth rate can then be estimated by divid-
And so on. These decisions must be made with refer- ing the natural logarithm of R0 by generation time
ence to the specific research question but are impor- (r ≈ ln(R0 )/G, which is 0.331 in this case). Another
tant when making comparisons across populations or important quantity, reproductive value, which is an
species and have consequences for modelling popula- individual’s fitness expressed in terms of expected con-
tion dynamics. tribution to future population growth, is calculated as
Notwithstanding the above caveats, Table 8.3 illus- the number of offspring produced by individuals from
trates the construction of a life table that includes age x onwards divided by the number of individuals
Table 8.3 Period life table for female Bali myna (Leucopsar rothschildi) calculated for a cohort of 644 individuals from x, Nx , n ax , and n bx data
obtained from DATLife. Deaths are assumed to be distributed uniformly during the interval, so n ax = 0.5, except for the final interval where it has
been set to 0 and where qx is set to 1 to arbitrarily complete the life table. Noninteger values have been rounded to 3 dp Nomenclature is the same
as in Table 8.2, with the addition of fertility rate (n bx ). Fertility values reported in DATLife are for offspring of both sexes and I assume the sex ratio
is 50:50.
x Nx n Dx lx n qx n px n Lx Tx ex n mx n ax n bx
0 644 93 1.000 0.144 0.856 597.500 5757.500 8.940 0.156 0.500 0.050
1 551 45 0.856 0.082 0.918 528.500 5160.000 9.365 0.085 0.500 0.530
2 506 20 0.786 0.040 0.960 496.000 4631.500 9.153 0.040 0.500 0.895
3 486 9 0.755 0.019 0.981 481.500 4135.500 8.509 0.019 0.500 0.855
4 477 13 0.741 0.027 0.973 470.500 3654.000 7.660 0.028 0.500 1.000
5 464 54 0.720 0.116 0.884 437.000 3183.500 6.861 0.124 0.500 0.620
6 410 38 0.637 0.093 0.907 391.000 2746.500 6.699 0.097 0.500 0.540
7 372 37 0.578 0.099 0.901 353.500 2355.500 6.332 0.105 0.500 0.465
8 335 35 0.520 0.104 0.896 317.500 2002.000 5.976 0.110 0.500 0.375
9 300 40 0.466 0.133 0.867 280.000 1684.500 5.615 0.143 0.500 0.395
10 260 15 0.404 0.058 0.942 252.500 1404.500 5.402 0.059 0.500 0.300
11 245 11 0.380 0.045 0.955 239.500 1152.000 4.702 0.046 0.500 0.320
12 234 24 0.363 0.103 0.897 222.000 912.500 3.900 0.108 0.500 0.190
13 210 33 0.326 0.157 0.843 193.500 690.500 3.288 0.171 0.500 0.235
14 177 29 0.275 0.164 0.836 162.500 497.000 2.808 0.178 0.500 0.230
15 148 24 0.230 0.162 0.838 136.000 334.500 2.260 0.176 0.500 0.130
16 124 25 0.193 0.202 0.798 111.500 198.500 1.601 0.224 0.500 0.000
17 99 24 0.154 0.242 0.758 87.000 87.000 0.879 0.276 0.500 0.000
18 75 75 0.116 1.000 0.000 0.000 0.000 0.000 0.000 0.000
of age x (assuming a stable age distribution). Further 2015; Colchero et al. 2016). These metrics can be con-
development of these relationships and measures can sidered life history traits of an organism living in a
be found in any good textbook on population biology particular place at a particular time and are thus use-
(e.g. Gotelli 2001; Neal 2018). ful in a comparative evolutionary context. For example,
using life tables derived from MPMs, researchers have
8.3 The richness of life table information explored the evolution of the pace and shape of senes-
cence in plants (Baudisch 2011; Baudisch et al. 2013)
From the life tables described it is possible to derive a and have explored the structure and evolution of life
dizzying array of metrics, some of which I have already history across animals and plants using a large num-
mentioned. Many of these metrics can be considered ber of life-table-derived metrics (Salguero-Gómez et al.
as part of the life table since they vary through time 2016; Salguero-Gómez and Jones 2017; Healy et al.
within the life table and apply to particular age inter- 2019).
vals (e.g. life expectancy, reproductive value, life span
distribution [the empirical density function of the age
at death], indicators of the force of selection [Baudisch 8.4 Complications, correlations,
2005]). Others are single-value summaries such as pop-
ulation growth rate, generation time, or measures of
and limitations
the ‘shape’ of mortality such as entropy or life span The life table construction methods described are
equality (Demetrius 1978; Baudisch 2011; Wrycza et al. designed to use ideal data, where individuals in
the population have perfect detectability and, for 2001), even though age might be more appropriate
period-type life tables, where the population structure under ideal conditions (see Chapter 9). Until recent-
is stationary. They also rely on cohorts being followed ly, researchers were satisfied with a simplified (st)age
until (almost) all individuals are dead or where the structure (e.g. the two age classes of ‘Juvenile’ and
period-data include sufficient sample sizes across all ‘Adult’) that assumes constant mortality and fertili-
ages. Such data are relatively easy to obtain from ty within those classes. However, vital rates can vary
laboratory-based studies of model species like med- throughout adulthood (Jones et al. 2014), and senes-
flies (e.g. Carey et al. 2002) or fruit flies (e.g. Pearl and cence is commonplace, at least in birds and mammals
Parker 1921) or even naked mole rats (Sherman and (Jones et al. 2008; Nussey et al. 2013). Failure to account
Jarvis 2002). However, following the fate of most wild for these age-varying vital rates can have consequences
animals is difficult, and alternative approaches such for understanding population dynamics and in some
as using standing age structure (to produce ‘static life cases can influence estimates of extinction risk (Robert
tables’) are common (Udevitz and Gogan 2012). Other et al. 2015; Colchero et al. 2019).
examples include the use of data from marine mammal Furthermore, many organisms are naturally struc-
strandings (e.g. Stolen and Barlow 2003), bird ringing tured by something other than age, for example,
(banding) recoveries (e.g. Lack 1943), skull recovery distinct developmental stages (e.g. eggs, larvae, and
(Gonzalez and Crampe 2001; e.g. Simmons et al. 1984), adults in insects), and in some groups, notably
fish otolith growth rings (e.g. Carlson et al. 2003), and plants, life histories can be complex, including both
fossil dinosaurs (Erickson et al. 2006; Erickson et al. progression and retrogression (e.g. shrinkage) (e.g.
2009), where age can be estimated. This small selec- Salguero-Gómez and Casper 2010). For these cases,
tion of citations highlights the variety of data that can MPMs are a flexible and powerful modelling tool
be coerced into the life table framework, but it is clear (Caswell 2001; see Chapter 9), and in cases where life
that there may sometimes be nuances in these data that histories are structured by quantitative traits (e.g. body
complicate estimation of demographic trajectories. size, stem diameter, plant height) IPMs are similar-
Developments in the statistical field of survival ly useful alternatives (Ellner and Rees 2006; Coulson
analysis help researchers estimate demographic 2012; see Chapter 10).
trajectories and quantify the effect of covariates, while
accounting for the various nuances in the available
data. These nuances include censoring, where not
8.5 Conclusions
all individuals have experienced the event (death, I have given a broad overview of the construction and
in the present context) by the end of the study, and use of life tables, with a focus on nonhuman studies.
truncation, where the time of birth (and therefore age) The use of life tables has a long history and they are
is unknown. In addition, observational bias leading still ubiquitous in human demographic studies and
to the violation of the assumption that all individuals in controlled laboratory studies. Numerous metrics
have an equal chance of observation can be problem- of importance to population biology can be calculat-
atic. The statistical approaches designed to cope with ed from them, including generation time, population
these issues include long-standing methods such as growth rate, reproductive value, and entropy mea-
bivariate nonparametric analysis (Kaplan–Meier esti- sures. Their use has been somewhat superseded in
mates: Kaplan and Meier 1958), proportional hazard field-based demographic studies on nonhuman ani-
approaches (Cox 1972), and accelerated failure time mals, largely because of the difficulty in meeting the
models (Wei 1992). In addition, and especially for wild required assumptions. Alternatives such as MPMs
animals where individual-based data are available, (Chapter 9), IPMs (Chapter 10), and Bayesian capture–
capture–recapture modelling (Lebreton et al. 1992; recapture methods (Chapter 13) have become popular.
Lebreton and Gaillard 2016) has become increasingly Nevertheless, an understanding of life tables and the
popular. Some of these methods explicitly model mor- estimates derived from them is useful to appreciate and
tality over continuous age with a functional form (e.g. understand much of life history theory.
Gompertz, Weibull, Siler) (Colchero and Clark 2012;
Colchero et al. 2012). These approaches are covered in
detail in Chapter 13. Data and code accessibility
Difficulties in estimating age in wild animals can Data and code to calculate the life tables and associated
also be a problem, and so researchers have often opted quantities and to produce the plots in this chapter
to use stage- rather than age-structured approaches are available at Zenodo: https://doi.org/10.5281
like Lefkovitch MPMs (Lefkovitch 1965; Caswell /zenodo.3860588.
References from incomplete mark—recapture/recovery data with

covariates. Methods in Ecology and Evolution/British Ecolog-
Anderson, D. R., Wywialowski, A. P., & Burnham, K. P.
ical Society, 3(3), 466–470.
(1981). Tests of the assumptions underlying life table
Colchero, F., Rau, R., Jones, O. R., et al. (2016). The emer-
methods for estimating parameters from cohort data.
gence of longevous populations. Proceedings of the National
Ecology, 62(4), 1121–1124.
Academy of Sciences, 113(48), E7681–E7690.
Baudisch, A. (2005). Hamilton’s indicators of the force of
Colchero, F., Jones, O. R., Conde, D. A., et al. (2019).
selection. Proceedings of the National Academy of Sciences,
The diversity of population responses to environmental
102(23), 8263–8268.
change. Ecology Letters, 22(2), 342–353.
Baudisch, A. (2011). The pace and shape of ageing. Methods
Coulson, T. (2012). Integral projections models, their con-
in Ecology and Evolution/British Ecological Society, 2(4), 375–
struction and use in posing hypotheses in ecology. Oikos,
382.
121(9), 1337–1350.
Baudisch, A., Salguero-Gómez, R., Jones, O. R., et al. (2013).
Cox, D. R. (1972). Regression models and life-tables. Journal
The pace and shape of senescence in angiosperms. Journal
of the Royal Statistical Society. Series B, Statistical Methodolo-
of Ecology, 101(3), 596–606.
gy, 34(2), 187–202.
Bonduriansky, R., Maklakov, A., Zajitschek, F., & Brooks, R.
Deevey, E. S. (1947). Life tables for natural populations of
(2009). Sexual selection, sexual conflict and the evolution
animals. Quarterly Review of Biology, 22(4), 283–314.
of ageing and life span. Functional Ecology, 22(3), 443–453.
Demetrius, L. (1978). Adaptive value, entropy and survivor-
Boonekamp, J. J., Salomons, M., Bouwhuis, S., Dijkstra, C., &
ship curves. Nature, 275(5677), 213–214.
Verhulst, S. (2014). Reproductive effort accelerates actuari-
Descamps, S., Boutin, S., Berteaux, D., & Gaillard, J. M.
al senescence in wild birds: an experimental study. Ecology
(2008). Age-specific variation in survival, reproductive
Letters, 17(5), 599–605.
success and offspring quality in red squirrels: evidence of
Bouwhuis, S., Charmantier, A., Verhulst, S., & Sheldon, B. C.
senescence. Oikos, 117(9), 1406–1416.
(2010). Individual variation in rates of senescence: natal
Ellner, S. P. & Rees, M. (2006). Integral projection mod-
origin effects and disposable soma in a wild bird popula-
els for species with complex demography. The American
tion. Journal of Animal Ecology, 79(6), 1251–1261.
Naturalist, 167(3), 410–428.
Burger, O., Baudisch, A., & Vaupel, J. W. (2012). Human mor-
Erickson, G. M., Currie, P. J., Inouye, B. D., & Winn, A. A.
tality improvement in evolutionary context. Proceedings of
(2006). Tyrannosaur life tables: an example of nona-
the National Academy of Sciences, 109(44), 18210–18214.
vian dinosaur population biology. Science, 313(5784),
Butler, R. W. (1988). Population dynamics and migra-
213–217.
tion routes of tree swallows, Tachycineta bicolor, in North
Erickson, G. M., Makovicky, P. J., Inouye, B. D.,
America. Journal of Field Ornithology, 59(4), 395–402.
Zhou, C.-F., & Gao, K.-Q. (2009). A life table for Psit-
Carey, J. R., Liedo, P., Harshman, L., et al. (2002). Life history
tacosaurus lujiatunensis: initial insights into ornithischian
response of Mediterranean fruit flies to dietary restriction.
dinosaur population biology. Anatomical Record, 292(9),
Aging Cell, 1(2), 140–148.
1514–1521.
Carlson, J. K., Cortés, E., & Bethea, D. M. (2003). Life
Fryxell, J. M. (1986). Age-specific mortality: an alternative
history and population dynamics of the finetooth shark
approach. Ecology, 67, 1687–1692.
(Carcharhinus isodon) in the northeastern Gulf of Mexico.
Gaillard, J.-M., Yoccoz, N. G., Lebreton, J.-D., et al. (2005).
Fishery Bulletin, 101(2), 281–292.
Review of Generation time: a reliable metric to measure life-
Caswell, H. (2001). Matrix Population Models. Sinauer
history variation among mammalian populations. The Ameri-
Associates, Sunderland, MA.
can Naturalist, 166(1), 119–123; discussion 124–128.
Caughley, G. (1966). Mortality patterns in mammals.
Gonzalez, G. & Crampe, J.-P. (2001). Mortality patterns in a
Ecology, 47(6), 906–918.
protected population of isards (Rupicapra pyrenaica). Cana-
Caughley, G. (1977). Analysis of Vertebrate Populations.
dian Journal of Zoology, 79(11), 2072–2079.
John Wiley & Sons, London.
Gotelli, N. J. (2001). A Primer of Ecology. Sinauer Associates,
Clutton-Brock, T. H., & Isvaran, K. (2007). Sex differences in
Sunderland, MA.
ageing in natural populations of vertebrates. Proceedings.
Griffin, R. M., Hayward, A. D., Bolund, E., Maklakov, A. A.,
Biological Sciences/The Royal Society, 274(1629), 3097–3104.
& Lummaa, V. (2017). Sex differences in adult mortality
Cochran, M. E. & Ellner, S. (1992). Simple methods
rate mediated by early-life environmental conditions.
for calculating age-based life history parameters for
Ecology Letters, 21(2), 235–242.
stage-structured populations. Ecological Monographs,
Harcombe, P. A. (1987). Tree life tables. Bioscience, 37(8),
62(3), 345–364.
557–568.
Colchero, F. & Clark, J. S. (2012). Bayesian inference on age-
Healy, K., Ezard, T. H. G., Jones, O. R., Salguero-Gómez, R.,
specific survival for censored and truncated data. Journal
& Buckley, Y. M. (2019). Animal life history is shaped by
of Animal Ecology, 81(1), 139–149.
the pace of life and the distribution of age-specific mor-
Colchero, F., Jones, O. R., & Rebke, M. (2012). BaSTA: an R
tality and reproduction. Nature Ecology & Evolution, 3(8),
package for Bayesian estimation of age-specific survival
1217–1224.
Hegazy, A. K. (1990). Population ecology and implica- Tree of Life—database. https://datlife.org/ (accessed 9
tions for conservation of Cleome droserifolia: a threatened July 2019).
xerophyte. Journal of Arid Environments, 19, 269–282. Menkens, G. E. & Boyce, M. S. (1993). Comments on the
Hoenig, J. M. & Gedamke, T. (2007). A simple method for use of time-specific and cohort life tables. Ecology, 74(7),
estimating survival rate from catch rates from multiple 2164–2168.
years. Transactions of the American Fisheries Society, 136(5), Neal, D. (2018). Introduction to Population Biology. Cam-
1245–1251. bridge University Press, Cambridge.
Johnstone, R. A. & Cant, M. A. (2019). Evolution of Nussey, D. H., Froy, H., Lemaitre, J.-F., Gaillard, J.-M.,
menopause. Current Biology, 29(4), R112–R115. & Austad, S. N. (2013). Senescence in natural popula-
Jones, O. R. (2016). Life history patterns. In R. M. Kliman tions of animals: widespread evidence and its implica-
(ed.), Encyclopaedia of Evolutionary Biology (pp. 366–371). tions for bio-gerontology. Ageing Research Reviews, 12(1),
Academic Press, Oxford. 214–225.
Jones, O. R. & Vaupel, J. W. (2017). Senescence is not Pearl, R. & Miner, J. R. (1935). Experimental studies on the
inevitable. Biogerontology, 18(6), 965–971. duration of life. XIV. The comparative mortality of cer-
Jones, O. R., Gaillard, J.-M., Tuljapurkar, S., et al. (2008). tain lower organisms. Quarterly Review of Biology, 10(1),
Senescence rates are determined by ranking on the 60–79.
fast-slow life-history continuum. Ecology Letters, 11(7), Pearl, R. & Parker, S. L. (1921). Experimental studies on the
664–673. duration of life. I. Introductory discussion of the dura-
Jones, O. R., Scheuerlein, A., Salguero-Gómez, R., et al. tion of life in Drosophila. The American Naturalist, 55(641),
(2014). Diversity of ageing across the tree of life. Nature, 481–509.
505(7482), 169–173. Preston, S., Heuveline, P., & Guillot, M. (2000). Demog-
Kaplan, E. L. & Meier, P. (1958). Nonparametric estima- raphy: Measuring and Modeling Population Processes.
tion from incomplete observations. Journal of the American Blackwell, Oxford.
Statistical Association, 53, 457–481. Purvis, A. & Harvey, P. H. (1995). Mammal life-history evo-
Lack, D. (1943). The age of the blackbird. British Birds; an lution: a comparative test of Charnov’s model. Journal of
Illustrated Magazine Devoted to the Birds on the British List, Zoology, 237(2), 259–283.
36(9), 166–175. Reed, T. E., Kruuk, L. E. B., Wanless, S., Frederiksen, M.,
Lebreton, J.-D. & Gaillard, J.-M. (2016). Wildlife demog- Cunningham, E. J. A., & Harris, M. P. (2008). Reproduc-
raphy: population processes, analytical tools and man- tive senescence in a long-lived seabird: rates of decline
agement applications. In R. Mateo, B. Arroyo, & J. T. in late-life performance are associated with varying costs
Garcia (eds), Current Trends in Wildlife Research (pp. 29–54). of early reproduction. The American Naturalist, 171(2),
Springer, Cham. E89–E101.
Lebreton, J.-D., Burnham, K. P., Clobert, J., & Ricklefs, R. E., Scheuerlein, A., & Cohen, A. (2003). Age-
Anderson, D. R. (1992). Modeling survival and testing related patterns of fertility in captive populations of
biological hypotheses using marked animals: a unified birds and mammals. Experimental Gerontology, 38(7),
approach with case studies. Ecological Monographs, 62(1), 741–745.
67–118. Robert, A., Chantepie, S., Pavard, S., Sarrazin, F., &
Lefkovitch, L. P. (1965). The study of population growth in Teplitsky, C. (2015). Actuarial senescence can increase the
organisms grouped by stages. Biometrics, 21(1), 1. risk of extinction of mammal populations. Ecological Appli-
Leslie, P. H. (1945). On the use of matrices in certain popu- cations: A Publication of the Ecological Society of America,
lation mathematics. Biometrika, 33, 183–212. 25(1), 116–124.
Leverich, W. J. & Levin, D. A. (1979). Age-specific survivor- Salguero-Gómez, R. & Casper, B. B. (2010). Keeping plant
ship and reproduction in Phlox drummondii. The American shrinkage in the demographic loop. Journal of Ecology,
Naturalist, 113(6), 881–903. 98(2), 312–323.
Levitis, D. A., Burger, O., & Lackey, L. B. (2013). The human Salguero-Gómez, R. & Jones, O. R. (2017). Life history trade-
post-fertile lifespan in comparative evolutionary context. offs modulate the speed of senescence. In R. P. Shefferson,
Evolutionary Anthropology: Issues, News, and Reviews, 22(2), O. R. Jones, & R. Salguero-Gómez (eds), The Evolution
66–79. of Senescence in the Tree of Life (pp. 403–421). Cambridge
Lotka, A. J. (1907). Relation between birth rates and death University Press, Cambridge.
rates. Science, 26(653), 21–22. Salguero-Gómez, R., Jones, O. R., Jongejans, E., et al. (2016).
Maklakov, A. A., Bonduriansky, R., & Brooks, R. C. (2009). Fast–slow continuum and reproductive strategies struc-
Sex differences, sexual selection, and ageing: an experi- ture plant life-history variation worldwide. Proceedings of
mental evolution approach. Evolution; International Journal the National Academy of Sciences, 113(1), 230–235.
of Organic Evolution, 63(10), 2491–2503. Schwartz, O. A. & Armitage, K. B. (1998). Empirical consid-
Max-Planck Institute for Demographic Research (Ger- erations on the stable age distribution. Oecologia Montana,
many). (2018). DATLife—the demography across the 7(1–2), 1–6.
Sherman, P. W. & Jarvis, J. U. M. (2002). Extraordinary life Sutherland, I. (1963). John Graunt: a tercentenary tribute.
spans of naked mole-rats (Heterocephalus glaber). Journal of Journal of the Royal Statistical Society. Series A, 126(4), 537.
Zoology, 258(3), 307–311. Udevitz, M. S. & Ballachey, B. E. (1998). Estimating survival
Siler, W. (1979). A competing-risk model for animal mortal- rates with age-structure data. Journal of Wildlife Manage-
ity. Ecology, 60(4), 750–757. ment, 62(2), 779–792.
Simmons, N. M., Bayer, M. B., & Sinkey, L. O. (1984). Demog- Udevitz, M. S. & Gogan, P. J. P. (2012). Estimating sur-
raphy of Dall’s sheep in the MacKenzie Mountains, North- vival rates with time series of standing age-structure data.
west Territories. Journal of Wildlife Management, 48(1), 156– Ecology, 93(4), 726–732.
162. Van Valen, L. (1975). Life, death, and energy of a tree.
Stearns, S. C. (1992). The Evolution of Life Histories. Oxford Biotropica, 7(4), 259–269.
University Press, Oxford. Wei, L. J. (1992). The accelerated failure time model: a useful
Stolen, M. K. & Barlow, J. (2003). A model life table for bot- alternative to the cox regression model in survival analy-
tlenose dolphins (Tursiops truncatus) from the Indian River sis. Statistics in Medicine, 11, 1871–1879.
Lagoon system, Florida, U.S.A. Marine Mammal Science, Wrycza, T. F., Missov, T. I., & Baudisch, A. (2015). Quantify-
19(4), 630–649. ing the shape of aging. PLOS One, 10(3), e0119163.
CHAPTER 9
Introduction to matrix
population models
Yngvild Vindenes, Christie Le Coeur, and Hal Caswell
9.1 Introduction out computations, as matrix algebra is much more

efficient.
Population growth, extinction risk, and changes in trait
MPMs are discrete time population models where
distributions over time are examples of population-
individuals are classified into discrete stages (Caswell,
level processes that are studied in ecology and
2001). Individuals within each stage are treated as
evolutionary biology. They are also intricately con-
identical, in the sense that they are assumed to expe-
nected, since population growth ultimately depends
rience the same vital parameters describing demo-
on the demographic processes of individuals (such
graphic processes and stage transitions. Indeed, the
as survival, growth, and reproduction), which in turn
goal of defining stages is to accomplish this as close-
depend on the state of individuals (the individual state,
ly as possible. MPMs were introduced by Leslie (1945,
or i-state). The i-state may reflect properties such as
1948) for age-structured populations and later extend-
age, life cycle stage, or morphological traits. Structured
ed to developmental stages (Lefkovitch, 1965). Today,
population models play a key role in linking such indi-
MPMs are one of the most widely applied structured
vidual variation in state variables to population-level
population models, because of many computational as
processes (Caswell and John 1992). Matrix population
well as analytical advances. A comprehensive presen-
models (MPMs) represent a class of structured popula-
tation of MPMs, their analysis and applications, is pro-
tion model which is widely applied in ecology and evo-
vided by Caswell (2001). The purpose of this chapter
lutionary biology. This chapter will introduce MPMs
is to give an introduction to age structured and stage
and present a few of their applications.
structured MPMs and their key components. We also
The state of a population (or p-state) is a distribu-
show how to calculate common model results using
tion of individuals in the i-state categories. This can be
different empirical examples.
written as a vector n(t), and the transformation that
A key defining property of MPMs and other struc-
maps the population from one time to the next can be
tured population models is the incorporation of indi-
written as a matrix: hence the name ‘matrix population
vidual differences, which provides the link between
models’. Doing so opens a host of powerful mathemat-
individual- and population-level processes (Caswell
ical methods to extract the population consequences
and John, 1992). Individual differences can arise from
of the individual life cycle. There is misapprehension
a range of properties, like age, developmental stage,
that matrix mathematics is difficult. It is not. Because
morphological, and behavioural traits, and external
so much of ecology involves relations between multi-
factors like location (Bolnick et al., 2003; Dall et al.,
dimensional variables, and because matrices exist pre-
2004; Caswell, 2009; Vindenes and Langangen, 2015).
cisely to describe such relations, the basics of matrix
Understanding individual variation has always been
algebra belong in the mathematical knowledge of
a central problem in evolutionary biology, because
every population ecologist. In this chapter, we will
trait variation is one of the prerequisites for evo-
also write out some simple matrix expressions as sum-
lution. MPMs, and their life table ancestors, have
mations and multiplications, to aid in understand-
long been part of population ecology (Deevey, 1947).
ing. This is definitely not the way to actually carry
Yngvild Vindenes, Christie Le Coeur, and Hal Caswell, Introduction to matrix population models. In: Demographic Methods Across the Tree of Life.
There also exists a tradition, dating back to Lotka and analyse an MPM for each case. The R code, together
Volterra, based on models in which all individuals with more detailed information for each example,
are treated as equal. While such models yield impor- are provided in the online supplementary material
tant theoretical insights of population dynamics, they (www.oup.com/companion/SalgueroGamelonDM).
ignore the underlying mechanisms acting on individu- Our examples include a mammal, a bird, and an
als. By providing the crucial link from such individual invertebrate. However, we emphasise that MPMs (and
processes (survival, development, growth, matura- other demographic models) and their analyses depend
tion, reproduction, migration, etc.) to the population on the structure of the life cycle, and the rates applying
dynamics, structured population models like MPMs to individuals within that life cycle, and not at all on
have been central to our understanding of ecology and the taxon to which they are applied. This is part of the
life history evolution (Caswell, 2001). power of demographic analyses.
The number of potential state variables to include This chapter describes basic aspects of demograph-
in an MPM is potentially infinite, and an important ic models formulated in terms of matrices and vectors.
part of constructing an MPM for a given population We assume that the reader is familiar with basic matrix
is to choose the state variables. As is true for oth- operations (addition and multiplication of matrices
er modelling choices, this choice will depend on the and vectors, inverses of matrices, eigenvalues, and
study system, study questions, and available data and eigenvectors). If you are not familiar with these, there
may involve trade-offs with other modelling choic- are many resources available, online and in books (e.g.
es such as analytical tractability. The chosen state Caswell, 2001).
variables will affect the model results, and the con-
sequences of choosing a certain structure may be
9.2 An age-structured MPM
assessed along with other assumptions of the model.
Statistical methods exist to evaluate choices of state We start by describing an age structured MPM,
variables (Caswell 2001; Chapter 12). In general, a state before extending this to a general case of stage struc-
variable can represent any property or combination of ture. We will assume a constant environment and
properties relevant to describe the demographic pro- no density dependence, and for simplicity consider
cesses of individuals. Age is commonly used, going a female-based model with annual time-steps (other
back to life table analyses (Pearl, 1923; Deevey, 1947) time intervals are also possible). Age-structured MPMs
and the early development of MPMs (Leslie, 1945, are widely used in demography and life history theory
1948). Classical life history theory was also developed and are closely linked to life tables (Chapter 8).
for age-structured populations (Fisher, 1930; Roff, 1992;
Charlesworth, 1994). Since several physiological and 9.2.1 Life table for age structured population
developmental processes change with age, age is often
a good predictor of demographic processes. However, As an example, consider a perennial plant that repro-
this is not always the case, and other variables, such as duces at age 2 and 3 and where all individuals die
size, may be more important in many species, in partic- before reaching age 4. The life table for this fictive
ular with indeterminate growth. Some state variables population is given in Table 9.1.
are naturally discrete (e.g. distinct life cycle stages), Here, pj denotes the annual survival probability from
while others are continuous, such as size, and must age j to age j + 1, lj denotes the cumulative survival
be discretised in a matrix model. Integral projection from age 0 to age j (by definition l0 = 1), and mj is the
models (IPM; Chapter 10) are developed to incorporate fecundity, that is, the number of seeds (age 0) produced
continuous state variables, but in the end the underly- at exact age j (we assume a pulse reproduction). The
ing theory and the eventual analyses are very similar last column shows the product of lj mj .
to MPMs (Ellner and Rees, 2006). The life table can be used to calculate various life
In the following sections, we first describe an age history characteristics of the population. For the plant
structured MPM and then extend this to a general
Table 9.1 Life table of plant example.
stage structured MPM, dissecting the main model ele-
ments. This is followed by a brief introduction to some Age j Annual survival Cumulative Fecundity lj m j
common results from analysis of an MPM, includ- pj survival lj mj
ing parameters describing the long-term population
0 0.2 1 0 0
dynamics and mean life history of the individu-
1 0.9 0.2 0 0
als in the population. We then present three case 2 0.6 0.18 3 0.54
studies representing different kinds of demographic 3 0 0.108 6 0.648
structure and demonstrate how to construct and
I N T R O D U C T I O N TO M AT R I X P O P U L AT I O N M O D E L S 165
Figure 9.1 Relationship between

discrete age class j and continuous age x
1 2 3 Age class (upper figure, modified from fig. 2.1 of
0 1 2 3 Age Caswell, 2001), and the life cycle graphs
for a pre- and post-reproductive census.
Pre-reproductive census: Post-reproductive census: Census time is indicated on the
age-timeline using grey circles and black
b₃ = p₀m₃ b₃= p₂m₃
triangles, respectively. Parameters
b₂= p₀m₂ b₂ = p₁m₂ applying to age class (sj and bj ) are
shown in red, and parameters applying to
1 2 3 1 2 3
s₁ = p₁ s₂ = p₂ s₁ = p₀ s₂ = p₁ actual age / birthday j (mj and pj ) are
b₁= p₀m₁ b₁= p₀m₁ shown in black. For simplicity only three
age classes are shown.
example, the expected lifetime reproduction (net repro-

P n2 (t + 1) = s1 n1 (t), (9.2)
ductive rate) is, for instance, given by R0 = kj=0 lj mj =
0+0+0.54+0.648 = 1.188, and the cohort generation time
P n3 (t + 1) = s2 n2 (t). (9.3)
is given by G = R10 kj=0 jlj mj = (2 · 0.54+3 · 0.648)/1.188 ≈
2.5. Finally, the long-term population growth rate λ is
This is a linear system of equations, which can be
the solution of the Euler–Lotka equation,
written in matrix form:
X
k
nt+1 = Ant (9.4)
λ− j l j m j = 1
j=0
    
n1 b1 b2 b3 n1
0.54λ−2 + 0.648λ−3 = 1.
 n2  =  s1 0 0   n2  .
This equation can only be solved numerically, and for n3 t+1 0 s2 0 n3 t
the plant example we find λ ≈ 1.07. In a matrix model,
all of these parameters (R0 , G, λ) and others can also where A is the population projection matrix; this age-
be calculated directly from the projection matrix and classified case is called the Leslie matrix (named after
its main components, as shown in section 9.4. First, early works of Leslie, 1945, 1948). The first row con-
we describe the age structured MPM using the plant tains the fertilities, while the subdiagonal contains the
example. survival probabilities. Next we consider how these are
defined from the underlying age-dependent parame-
ters of the life table.
9.2.2 Projecting growth of age
structured population
9.2.3 Defining Leslie matrix elements for
To describe how an age structured population is grow-
pre- and post-reproductive censuses
ing from one time step to the next, we need to
keep track of the number of individuals in each of The fertility coefficients bj and the survival probabili-
the age classes. Following Caswell (2001), we always ties sj in the Leslie matrix describe transitions among
denote the first age class as age class 1, irrespective the age classes from one time step to the next. How-
of census time (see Figure 9.1; note that some authors ever, because matrix models are defined in discrete
use 0 to denote the first age class in the case of a time, while age is an inherently continuous variable,
post-reproductive census). Using the plant example linking these parameters to the underlying age-specific
described above, the population vector nt = [n1 , n2 , n3 ]′t parameters requires some careful definitions regard-
(the symbol ′ indicates the transpose) describes the ing (1) census time (when the population is counted
number of individuals in each age class at time t. Let during each time interval), and (2) the mode of repro-
sj describe the survival probability of an individual duction (birth flow or birth pulse) of the study popu-
in age class j to next year and bj the number of off- lation. We have, for simplicity, assumed a birth-pulse
spring produced by an individual in age class j that population where individuals reproduce on their birth-
enter next year’s population (fertility coefficient). With day (for birth flow populations see e.g. Caswell, 2001,
these parameters we can describe the growth of each Chapter 2.4.1). Regarding census time we will con-
age class sider the two most common examples, where census
is either right before (pre-reproductive) or right after
n1 (t + 1) = b1 n1 (t) + b2 n2 (t) + b3 n3 (t), (9.1) (post-reproductive) the pulse reproduction (Caswell,
(a) (b)
Pre-reproductive census Post-reproductive census
150 Age class 1 150

Age class 2
Age class size

Age class size
100 Age class 3 100
50 50
Figure 9.2 Projected growth in the age
structured plant example, assuming a
0 0 pre-reproductive census (a) or post-reproductive
0 5 10 15 20 0 5 10 15 20 census (b). In both cases the asymptotic
Years Years population growth rate is λ = 1.07.
2001, Chapter 2.4.2). The Leslie matrix elements will Figure 9.2 shows the projected growth of the three
then depend on the age-specific parameters as follows: age classes in the plant example, using a pre- and post-
reproductive census. In the latter, the offspring age
1. With a pre-reproductive census, offspring (seedlings) class consists of seeds and is therefore larger compared
will be nearly 1 year old when counted, and age to the offspring class with pre-reproductive census.
classes correspond to actual age at counting (e.g. Note that errors in the definition of projection matrix
individuals in age class 1 are 1 year old when count- elements seem to be a frequent issue in the literature
ed), so that sj = pj . The fertility coefficients are where matrix models are used, in particular for mod-
then given by bj = mj p0 , the product of the fecun- els with a post-reproductive census (Kendall et al.,
dity mj and the first-year survival probability p0 2019). With the exception of the asymptotic growth
(Figure 9.1), so that the resulting Leslie matrix for rate λ, all model results from an MPM will general-
the plant example is given by ly depend on the census time. This is not surprising,
since populations are changing also within each time
    interval, so it matters at which time point during the
b1 b2 b3 m 1 p0 m 2 p0 m 3 p0 time interval we make the census. This is also impor-
A =  s1 0 0  =  p1 0 0  tant to keep in mind when doing comparative analyses
0 s2 0 0 p2 0 of MPM results from different studies—such compar-
  isons should be made for models with similar census
0 0.6 1.2
=  0.9 0 0 . time.
0 0.6 0
9.3 Stage-structured MPMs

2. With a post-reproductive census, the first age class
will consist of newborn individuals (seeds), and age Age is a very special kind of population structure:
class does not correspond to actual age (individu- individuals age exactly 1 year for every year that
als in age class j are counted at age j − 1), so in this passes, it is impossible to move backwards in age,
case sj = pj−1 . The fertility in age class j is then and reproduction always produces individuals in the
given by bj = mj pj−1 = mj sj , the product of the first age class. We will now turn to a more general
parental survival probability until the next age, and kind of structure, considering a population classi-
the fecundity at the next age (fig. 9.1). For the plant fied into k discrete stages, with no a priori restric-
example, the Leslie matrix for a post-reproductive tion on what kind of transitions are permitted. The
census is then given by change in population vector from one time step to
the next is still given by eqn. (9.4), but the projection
    matrix A is now a general stage-structured projection
b1 b2 b3 m 1 p0 m 2 p1 m 3 p2
matrix.
A =  s1 0 0  =  p0 0 0 
It is helpful to write this matrix in terms of compo-
0 s2 0 0 p1 0
nents that capture the different demographic processes
 
0 2.7 3.6 of survival, reproduction, and transitions. To reduce
=  0.2 0 0 . the size of expressions, suppose that the life cycle
0 0.9 0 includes k = 3 stages. Then write
now routine using a computer program like R(R Devel-

   opment Core Team, 2020) or Matlab, for which several
g11 g12 g13 s1 0 0
A =  g21 g22 g23   0 s2 0  packages also exist. We show some examples using R
g31 g32 g33 0 0 s3 in the supplementary material.
  
q11 q12 q13 b1 0 0

+ q21 q22 q23   0 b2 0  (9.5) 9.4 Basic analytical results from MPMs
q31 q32 q33 0 0 b3 MPMs can be used to calculate a number of prop-
erties of the dynamics and life history of the pop-
= GDiag(s) + QDiag(b) (9.6) ulation and to investigate consequences of chang-
ing parameters (Caswell, 2001). These analyses go far
=U+F (9.7) beyond the simple numerical projection of the pop-
ulation over time, using eqn. (9.4). It is convenient
This general projection matrix depends on four main to divide analytical results into those describing tran-
demographic processes: (1) Survival, governed by the sient, short-term dynamics (Chapter 11) and those
stage-specific survival probabilities sj ; (2) transitions describing asymptotic, long-term dynamics. The latter
among stages (conditional on survival), governed by include measures of population growth and structure,
the probabilities gij that an individual from stage j reproductive value, invasion fitness, selection gradi-
P
enters stage i next time step ( ki=1 gij = 1); (3) repro- ents, equilibria, resilience, and sensitivity and elasticity
duction, governed by the fertility coefficients bj ; and (4) analysis of these properties. Here we describe some
distribution of new offspring among stages, governed basic analytical results for the general stage structured
by the probabilities qij that an offspring produced in model defined above. Analysis of stochastic MPMs is
P
stage j enters stage i ( ki=1 qij = 1). We refer to the qij described in Chapter 11. An overview of the variables
as offspring transition rates and the gij as individual and parameters defined here is given in Table 9.2.
transition rates.
We have now written A in terms of four component
9.4.1 Asymptotic properties from the
matrices, which capture each of these four demograph-
ic processes, and then as the sum of a matrix U that
projection matrix
describes the survival and transitions among stages Properties of the asymptotic population growth and
of extant individuals, and a matrix F that describes structure are calculated in terms of the eigenvalues
production of offspring from and to the various stages. and eigenvectors of A (Caswell, 2001, Chapter 4.5).
These component matrices appear in any MPM, and These include i) the long-term population growth rate
by adjusting G and Q, different kinds of demograph- λ, which is the dominant eigenvalue of A, ii) the sta-
ic structure can be obtained. The age structured Leslie ble stage structure w = [w1 , ..., wk ]′ , which is the
matrix from equation 9.4 is a special case of the general right eigenvector associated with λ and scaled so that
Pk
projection matrix: With age structure, surviving indi- j=1 wj = 1, and iii) the stage-specific reproductive
viduals will get one year older each time step, so that values v = [v1 , ..., vk ], which is the left eigenvector
gij = 1 for i = j + 1 and zero otherwise. Similarly, all off- associated with λ. This vector can be scaled in vari-
P
spring enter the first age class, so that qij = 1 for i = 1 and ous ways; here we scale it so that kj=1 wj vj = w′ v =
zero otherwise. We show examples with different kinds 1. These parameters can also be found by projecting
of transition matrix structures in section 9.5. the population vector over a long time interval using
Although we have written the projection matrix A A. Regardless of the initial population vector, after
as a constant matrix, depending on the situation it can a period of transient fluctuations the population will
be time dependent (stochastic or deterministic), depen- gradually reach the stable stage structure w where each
dent on environmental factors, or nonlinear (density- stage grows with rate λ. In the context of life history
or frequency-dependent). Depending on the nature of theory, λ is also a measure of fitness of the phenotype
the model, different kinds of analyses can be done described by A (see Metz et al. (1992) for discussion).
based on the projection matrices, analytical or numeri- The reproductive values describe the relative
cal. In any MPM, the most straightforward way to use contribution of individuals in each stage to future
the projection matrix (or matrices) is for numerical pro- population size, compared to other stages. Fisher
jection of the population vector over time by iterating (1930) introduced the concept of reproductive value
eqn. (9.4), to see how the number and distribution of for a continuous-time age structured model and
individuals in each stage is changing. Such projection is demonstrated that the total reproductive value, which
Table 9.2 Overview of variables and parameters for a general stage-structured matrix model. Model
results assume no density dependence and a constant environment.
Parameter Explanation
Pk
Nt = j=1 nj,t Population size at time t.
k Number of discrete stages.
′ Transpose of matrix or vector.
Diag(x) Diagonal matrix with elements of vector x
on the main diagonal, and 0 elsewhere.
s = [s1 , ..., sk ]′ Stage-specific survival probabilities.

b = [b1 , ..., bk ]′ Stage-specific fertility coefficients (next year’s offspring number).
gij Transition probability from stage j to stage i.
qij Probability that offspring produced in stage j is assigned to stage i.
A = GDiag(s) + QDiag(b) Projection matrix, dimension k × k.

G Transition matrix conditional on survival, elements gij .
U = GDiag(s) Survival and transition matrix.
Q Offspring transition matrix conditional on survival, elements qij .
F = QDiag(b) Offspring production matrix.
λ Long-term (asymptotic) growth rate, dominant eigenvalue of A.

w = [w1 , ..., wk ]′ Stable stage structure, right eigenvector associated with λ.
v = [v1 , ..., vk ] Reproductive values, left eigenvector associated with λ.
N=P (I − U)−1 Fundamental matrix, elements nij .

ηj = i nij Life expectancy in stage j.
R = F(I − U)−1 Generation to generation projection matrix.
R0 Net reproductive rate, dominant eigenvalue of R.
λ
TM = vFw Mean age of parent at stable growth.
P
in our notation would be V(t) = ki=1 ni (t)vi = v′ n(t), 2019b). One of the simplest and most commonly used
grows exponentially with rate λ, even when the analyses finds the sensitivities of λ to the elements of A.
population is not currently at the stable structure w. Recent developments using matrix calculus have great-
This result applies also to the discrete time stage- ly extended the range of sensitivity analysis (Caswell,
structured models described here (Engen et al., 2007; 2019b), but we do not address these here.
Vindenes et al., 2012) as well as to integral projection The sensitivity of λ can be calculated directly from
models (Vindenes et al., 2011). The stable structure and the stable structure and reproductive values (using our
reproductive value vectors are key properties of the scaling of v above; Caswell, 2001, Chapter 9),
population dynamics of MPMs and also occur in other
calculations such as sensitivity analysis of population dλ
= vi w j . (9.8)
growth rate. daij
It is important to note that asymptotic properties
such as λ are consequences of the life cycle and the The sensitivity matrix, with sensitivity of λ to each
environment that determines the demographic rates. matrix element, is thus given by vw’.
They show what would happen if the environment In population genetics, these sensitivities are inter-
remains constant, but they do not require any assump- preted as stage-specific selection gradients linking the
tion that the environment will remain constant. life history (as defined by A) to fitness measured by
λ (Lande, 1982). It is also possible to calculate sen-
sitivities of λ (or any other output from an MPM)
9.4.2 Sensitivity analysis of population
with respect to other parameters related to the life his-
growth rate tory (Caswell, 2001, Chapter 9). Sensitivities of λ to
Sensitivity analysis asks, and answers, the question, underlying parameters, such as the vital rates (survival
how would some result change if the parameters from probability, transition probability, etc.), can be found
which it is calculated were to change (Caswell, 2001, by applying the chain rule:
dλ X dλ daij but they make it possible to decompose the variance

= . (9.9) among individuals into each component (Hamel et al.,
dx daij dx
ij
2016; Hartemink et al., 2017). Understanding the con-
sequences of heterogeneity as well as its relative con-
The elasticity of λ is the proportional rate of change in tribution to the variance in life history parameters is
response to a proportional change in x and is given by an important area of research (Vindenes et al., 2008;
d ln λ
d ln x
= λx dλ
dx
. Tuljapurkar et al., 2009; Cam et al., 2016; Caswell et al.,
2018; van Daalen and Caswell, 2020). In Box 9.1 we dis-
cuss how the use of matrix models has contributed to
9.4.3 Life history characteristics this field.
The matrices F and U (elements of A) contain infor-
mation that can be used to calculate different char-
9.5 Case studies with different
acteristics of the underlying life history, such as the
(stage-specific) lifetime reproductive success, the gen- demographic structures
eration time, and stage-specific life expectancies. By adjusting the transition matrices G and Q of the gen-
The lifetime reproductive success, or net repro- eral model, different kinds of demographic structure
ductive rate, R0 describes the mean generation- can be included (Vindenes et al., 2012). In the following
to-generation population growth and the expected we present some of these, with numerical examples.
lifetime reproduction of an individual (Caswell, 2001). R code for calculations is provided as supplementary
The generation-to- generation projection matrix is giv- material.
en by R = F(I − U)−1 , and its dominant eigenvalue
corresponds to R0 (Section 5.3.4 in Caswell, 2001). See
9.5.1 Case study 1: Age structure
Caswell (2011) and van Daalen and Caswell (2017) for
a deeper analysis that provides not only the mean
in Asian elephants
lifetime reproductive success, but also the variance, In an age-structured model, transitions among the
skewness, and other measures of variability among stages (in this case the age classes) are simple: surviv-
individuals. ing individuals always get one time step older, and will
Generation time is a key life history characteristic, enter the next age class with probability 1 (thus gij = 1
for which several measures exist. One common mea- for i = j + 1 and zero otherwise). All offspring enter the
sure is the mean age of parents of offspring in a population in the first age class (thus qij = 1 for i = 1 and
population at the stable stage distribution (Caswell, zero otherwise).
2001). Using a Markov chain definition of the popula- As an example of an age structured MPM we consid-
tion dynamics in a general stage structured population, er the female-based model for Asian elephants (Elephas
Bienvenu and Legendre (2015) demonstrated that this maximus) developed by Goswami et al. (2014) to study
λ
value can be calculated as TM = vFw (here adapted to potential consequences of conflict-induced mortality.
use our scaling of v). Here we consider their baseline model with no such
The life expectancy in each stage can be calculated extra mortality. The main model parameters are the
from the fundamental matrix N = (I − U)−1 (Section survival probabilities s and fertility coefficients b from
5.3.41 in Caswell, 2001). In this matrix, element nij is the age class 1 to 60 (Figure 9.3a; supplementary material).
expected number of time-steps an individual currently Goswami et al. (2014) assumed a post-reproductive
in stage j will spend in stage i before it dies. The remain- census, so that offspring in age class 1 are newborn
ing life expectancy, conditional on current stage, is thus (actual age 0), and the survival from age class i to
given by the column sums of the fundamental matrix, i + 1 then corresponds to the survival probability of
P
ηj = i nij . In models with only one offspring stage, the actual age i − 1, si = pj−1 (for instance, s1 = p0 is the
life expectancy at birth is given by the column sum for survival probability from actual age 0 to age 1). The
this stage. fertility of age class i is given by bi = mi si and defines
In the models discussed so far, all differences among next year’s number of female offspring produced by a
individuals are due to a single demographic variable, female currently in age class i.
that we have called ‘stage’. However, it is possible The long-term growth rate is λ ≈ 1.05 for this model,
to add additional variables that capture heterogene- indicating an annual growth of about 5% in the absence
ity beyond that contributed by stage. The techniques of extra mortality and density-dependent reproduction
for this are beyond this Chapter (Caswell et al., 2018), (Goswami et al., 2014). The stable age structure w
Box 9.1 Effects of individual heterogeneity
Individual heterogeneity refers to variation among individu- example, habitat structure is a source of observed het-
als that results in their experiencing different demographic erogeneity. If we ignore habitat type, the resulting model
rates. Heterogeneity is a topic of interest in different bio- becomes purely age structured (supplementary material).
logical fields, including statistical ecology, behavioural biol- This model has the same value of λ = 1.00 as the model
ogy, and conservation biology (e.g. Vaupel et al., 1979; including habitat structure, but other model results differ,
Dall et al., 2004; Vindenes et al., 2008; Caswell, 2009; such as the mean lifetime reproduction and the age-specific
Tuljapurkar et al., 2009; Bolnick et al., 2011; Cam et al., reproductive values. For instance, the generation time would
2016; Hamel et al., 2016; Jenouvrier et al., 2018). These be 7.67 years rather than 7.69 (supplementary material).
fields have focused on different questions, from identifying In this case, the habitat heterogeneity had rather small
unobserved heterogeneity in demographic data to evaluating effects only on fertility, so ignoring it had small effects
its consequences for model results. on model results. However, if we in addition to habitat
It is seldom realised however, that MPMs, and other also ignored age, which has much larger effects on fer-
demographic models, are essentially an approach to incor- tility, the estimated generation time would be 5.51 years,
porate and account for individual heterogeneity. The choice underestimating by 28%. A similar comparison approach
to structure a model by age, size, stage, habitat, or some- has been used to evaluate effects of ignoring heterogene-
thing else is a judgement of what kind of heterogeneity ity on other parameters, such as parameters characterising
is most important. The entries in the population projection stochastic population dynamics (demographic and environ-
matrix capture the differences in rates experienced by indi- mental variance, e.g. Vindenes et al., 2008; Vindenes and
viduals of different ages, sizes, and so on. In other words, Langangen, 2015).
individual heterogeneity is nothing new. A main concern, Another important question is the relative contribution
however, is estimating and evaluating the consequences of of heterogeneity to the variance among individuals in life
heterogeneity that is not included in the description of the history outcomes, compared to other sources of individu-
life cycle, perhaps because it is observed but neglected, or al differences, including random variation. These methods
perhaps because it is invisible or unobserved. Try as you may are described at length in Caswell et al. (2018) and van
to pick the most important state variables, there will always Daalen and Caswell (2020), and make it possible to quanti-
be residual variation among individuals. fy the fraction of the variance in some outcome that is due
The identification and estimation of unobserved hetero- to heterogeneity. The approach has so far been applied to
geneity in demographic data, so that it can be included in a longevity (Hartemink et al., 2017; Hartemink and Caswell,
model, is difficult. It is no surprise that estimating something 2018; Seaman et al., 2019) and lifetime reproductive out-
without observing it is a difficult statistical problem. Recent put (Jenouvrier et al., 2018; van Daalen and Caswell, 2020).
advances, such as finite mixture models, sometimes make Hartemink et al. (2017) considered effects of hidden hetero-
it possible to identify groups of individuals that share the geneity (‘frailty’) on longevity in humans and found that most
same demographic rates, but where the difference between of the variation could be explained by individual stochas-
the groups is not explained by any apparent trait or prop- ticity (random variation within each heterogeneity group).
erty (Hamel et al., 2016; Hartemink and Caswell, 2018; Jenouvrier et al. (2018) detected three heterogeneity groups
Jenouvrier et al., 2018). Once heterogeneity is included in a in a population of southern fulmars (Fulmarus glacialoides)
model, whether as observed or unobserved state variables, and found that this heterogeneity explained 3.7% of the vari-
the two main areas of research concern are (1) evaluat- ance in age at first reproduction, 5.9% in longevity, and 22%
ing the potential consequences of ignoring heterogeneity in lifetime reproductive success.
for various model results, and (2) evaluating the relative Theory for using demographic models like MPMs to esti-
contribution of heterogeneity to the variance in life history mate and evaluate consequences of heterogeneity is still in
parameters, compared to other sources of variation among development, and there are several other questions where
individuals. this theory and methods will be relevant (Cam et al.,
One approach to studying consequences of ignoring het- 2016), including senescence (McDonald and Fitzpatrick,
erogeneity is to compare results from models including the 1996; Knape et al., 2011), extinction risk (Conner and White,
relevant state variable capturing heterogeneity or not (Vin- 1999), and patterns of selection gradients (Caswell and
denes et al., 2008). For instance, in the black-headed gull Salguero-Gómez, 2013; Hernández et al., 2020).
(a) (b)
1.00 1.00 0.08 2.0
0.75 0.75 0.06 1.5
Wi
0.50 0.50 0.04 1.0
Vi
bi
Si
0.25 0.25 0.02 0.5
0.00 0.00 0.00 0.0

1 10 20 30 40 50 60 1 10 20 30 40 50 60
Age class i Age class i
Annual survival (si) Annual fertility (bi) Stable structure (wi) Reproductive value (vi)
(c) (d)
40 5 0.10
4 0.08
30
3
dλ /dbi
0.06
dλ /dsi
20
ηi
2 0.04
10
1 0.02
1 0 0.00
1 10 20 30 40 50 60 1 10 20 30 40 50 60
Sensitivity to survival (si) Sensitivity to fertility (bi)
Figure 9.3 Vital parameters and results from an age-structured female-based model for Asian elephants. (a) Survival probability si and fertility bi
across age class i. (b) Stable age structure wi and reproductive value vi across age class i. (c) Life expectancy η i in age class i. (d) Sensitivity of λ to
survival probability and fertility.
declines monotonically with age class (Figure 9.3b), These sensitivities are shown in Figure 9.3d, together
which is always true for growing age-structured popu- with the sensitivities to fertility in each age class,
lations. The reproductive value increases with age class which are given by the first row of the sensitivity
at first and peaks around or shortly after the age at matrix. These patterns are also universal for growing
first reproduction and thereafter declines with age class age-structured populations, with the sensitivities being
(Figure 9.3b)—another common pattern for such life highest at young age classes and then declining with
histories. age (Hamilton, 1966).
The remaining life expectancy in each age class
is shown in figure 9.3c. Because the survival proba-
bility increases slightly during the first years of life 9.5.2 Case study 2: Life cycle stages in an isopod
(Figure 9.3a), the life expectancy also increases towards
Many species are characterised by discrete life his-
a peak in early life and then declines with age (Figure
tory stages, reflecting ontogenetic development
9.3c). In contrast to life expectancy, generation time
(Lefkovitch, 1965). A commonly used model for such
depends on both survival and fertility, and for this ele-
stages (the ‘standard size-classified model’ of Caswell
phant model, the mean age of mothers is TM = vFw λ
≈
(2001)) proposes that, at each time step, surviving
28 years.
individuals in stage j will either make a transition
With a post-reproductive census, the sensitivity of λ
to stage j + 1, with probability γ j , or remain in their
to survival sj in age class j, taking into account effects on
current stage with probability (1 − γ j ). The transition
both age class transitions and reproduction, is given by
matrix thus takes the values gj+1,j = γj and gjj = (1 − γj )
∂λ for j = 1, ..., k − 1, gkk = 1 for the final stage and zero else-
= vj+1 wj + v1 wj mj . (9.10)
∂sj where. As in the age structured model, all offspring
enter the first stage 1 (thus qij = 1 for i = 1 and zero both reproductive stages (Kammenga et al., 2001). With
otherwise). a post-reproductive census, the fertility coefficients in
As an example of a species with distinct life cycle stage 4 and 6 are
stages we consider the isopod Porcellio scaber (common
rough woodlouse). In a toxicology study, Kammenga b4 = 0.5m4 [γ3 s3 + (1 − γ4 )s4 ] (9.11)
et al. (2001) defined a matrix model for this species,
with monthly time-steps and seven distinct life b6 = 0.5m6 [γ5 s5 + (1 − γ6 )s6 ]. (9.12)
history stages: 1) oocytes, 2) eggs/offspring (in mar-
supium), 3) juveniles, 4) first reproductive stage, The factor 0.5 represents an assumption of equal sex
5) non-reproductive adult, 6) second reproductive ratio of fertilised eggs. These fertility coefficients con-
stage, 7) senescent non-reproductive adult. Transi- tain one component corresponding to reproduction
tion probabilities γ j were calculated as the reciprocal by females that survive and make a transition from
of the average stage duration (Kammenga et al., the preceding stage and one component from females
2001). that survive and remain in the reproductive stage. The
The average number of oocytes produced by repro- resulting projection matrix for the isopods is given by
ductive individuals over a month is m4 = m6 = 12 for
 
(1 − γ1 )s1 0 0 0.5m4 [γ3 s3 + (1 − γ4 )s4 ] 0 0.5m6 [γ5 s5 + (1 − γ6 )s6 ] 0
 γ1 s1 (1 − γ2 )s2 0 0 0 0 0 
 
 (1 − γ3 )s3 0 
 0 γ2 s2 0 0 0 
 
A= 0 0 γ3 s3 (1 − γ4 )s4 0 0 0 .
 
 0 0 0 γ4 s4 (1 − γ5 )s5 0 0 
 
 0 0 0 0 γ5 s5 (1 − γ6 )s6 0 
0 0 0 0 0 γ6 s6 s7
(9.13)
Vital rate values for this model are shown in Figure 9.6 For this model the long-term growth rate is λ ≈ 1.07,
as well as in supplementary information, and are taken indicating a monthly population growth of around
from the control model of Kammenga et al. (2001) with- 7%. The stable stage structure is shown in figure 9.6c;
out any toxicity effects, and the corresponding life cycle it contains a high proportion of oocytes, eggs, and
graph for this model is given in figure 9.4. Life cycle juveniles compared to adults. The reproductive val-
graphs are useful tools for constructing MPMs, in par- ue is higher for adults, especially in the two repro-
ticular for more complex stage structured models with ductive stages (Figure 9.6d). The last stage consists
various forms of transitions among the stages. Figure of non-reproducing senescent individuals with no
9.5 shows the number of individuals in each stage contribution to future reproduction; thus the repro-
over time, starting from a population consisting of 10 ductive value is zero for this stage. If we were
oocytes. After an initial period of transient fluctuations, only concerned with λ, this stage could be omit-
the population reaches a stable stage structure, and all ted from the model. But if we want to know the
stages grow steadily at rate λ. entire population’s structure, or if the model included
0.5m6[J₅s₅ + (1-J₆)s₆] Figure 9.4 Life cycle graph for the

isopod example of case study 2, with
0.5m4[J₃s₃ + (1-J₄)s₄] monthly transitions corresponding to the
projection matrix of eqn. (9.13). Stages
are 1) Oocytes, 2) Eggs, 3) Juveniles, 4)
J1s1 J2s2 J3s3 J4s4 J5s5 J6s6 First reproductive stage,
1 2 3 4 5 6 7 5) Non-reproductive adults, 6) Second
reproductive stage, 7) Senescent. The final
stage does not contribute to population
(1–J1)s1 (1 –J2)s2 (1 –J3)s3 (1 –J4)s4 (1 –J5)s5 (1 –J6)s6 s7 growth rate, λ.
40 Stage i
1
2
30 3
Stage size ni,t
4
5 Figure 9.5 Projection of stage-specific
20 6
7 abundance in the isopod case study,
starting from a population of 10 oocytes
10 and using the projection matrix in eqn.
(9.13). Stages are 1) Oocytes,
2) Eggs, 3) Juveniles, 4) First reproductive
0 stage, 5) Non-reproductive adults,
0 5 10 15 20 25 30 35 40 6) Second reproductive stage,
Time (months) 7) Senescent.
Table 9.3 Sensitivities in the isopod model, for stages 1 to 6. For the senescent stage 7 the sensitivity to survival is 0.
∂λ ∂λ ∂λ
Stage j To survival, ∂sj
To transition, ∂γj
To fecundity, ∂mj
1 v1 w1 (1 − γ1 ) + v2 w1 γ1 w1 (v2 − v1 )s1
2 v2 w2 (1 − γ2 ) + v3 w2 γ2 w2 (v3 − v2 )s2
3 v3 w3 (1 − γ3 ) + v4 w3 γ3 + 0.5v1 w4 γ3 m4 w3 (v4 − v3 )s3 + 0.5v1 w4 m4 s3
4 v4 w4 (1 − γ4 ) + v5 w4 γ4 + 0.5v1 w4 (1 − γ4 )m4 w4 (v5 − v4 )s4 − 0.5v1 w4 m4 s4 0.5v1 w4 [γ3 s3 + (1 − γ4 )s4 ]
5 v5 w5 (1 − γ5 ) + v6 w5 γ5 + 0.5v1 w6 γ5 m6 w5 (v6 − v5 s5 ) + 0.5v1 w6 m6 s5
6 v6 w6 (1 − γ6 ) + v7 w6 γ6 + 0.5v1 w6 (1 − γ6 )m6 w6 (v7 − v6 )s6 − 0.5v1 w6 m6 s6 0.5v1 w6 [γ5 s5 + (1 − γ6 )s6 ]
density dependence, this stage would have to be Regarding other life history parameters for this
included. case model, the net reproductive rate is R0 ≈ 1.77, life
The expressions for sensitivities of λ to stage-specific expectancy in stage 1 is η 1 = 5.44 months, while the
survival probability sj , transition rate γ j and fecundity generation time is Tm ≈ 7.67 months.
mj are given in table 9.3, and the resulting values are
shown in figure 9.7, together with the corresponding
9.5.3 Case study 3: Age and habitat structure
elasticities. These are found by applying the chain rule,
in black-headed gulls
as in eqn. (9.9). Population growth rate λ is most sen-
sitive and most elastic to the survival and the growth Individuals rarely differ from each other by just one
probabilities of stages 2 and 3, which also contain a state variable. Multistate MPMs describe the dynam-
high proportion of the population (Figure 9.6). Overall, ics of populations structured by more than one state
the sensitivities of λ to survival and transition proba- variable, and age by stage models are an important
bilities are much higher than those to fecundity, while example. Here, the stage can represent any of the famil-
the elasticities are more similar in scale. Note that in iar i-state variables (size, developmental stage, physi-
the reproductive stages 4 and 6, the sensitivity to tran- ological condition, spatial location, habitat quality, or
sition probability is negative. In stage 4, individuals genotype), and the inclusion of age permits the stage-
moving to stage 5 lose the opportunity to reproduce specific transitions, fertilities, and survival probabili-
and have to wait at least one more time step to enter ties to vary with age. To analyse multistate models, the
the next reproductive stage, resulting in a lower repro- population vector must contain all combinations of the
ductive value in stage 5 than in 4 and a corresponding states, and the population projection matrix has a block
negative sensitivity to the transition. Thus, this sensi- structure corresponding to the arrangement of stages.
tivity analysis shows that increasing the probability of We show a simple example here, but the possibilities
remaining in the first reproductive stage will increase are very general (Caswell et al., 2018).
λ, even though there is another reproductive stage Lebreton (1996) developed an age by stage MPM for
later. black-headed gulls (Chroicocephalus ridibundus), where
(a) (b)
6 Monthly survival (si) 6 1.00
Monthly fertility (bi)

0.75
4 4
0.50
bi
Si
gj
2 2
0.25
0 0 0.00
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Stage i Stage i
(c) (d)
0.4 4
0.3 3
Wi
0.2 Vi 2
0.1 1
0.0 0
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Stage i Stage i
Figure 9.6 Vital rates and results from stage-structured MPM for isopods in case study 2. Stages are 1) Oocytes, 2) Eggs, 3) Juveniles, 4) First
reproductive stage, 5) Non-reproductive adults, 6) Second reproductive stage, 7) Senescent. Results representing the two reproductive stages are
shown in dark colour. (a) Stage-specific survival probability sj and fertility bj (fertility depends on other vital rates, see eq. (9.11) and eq. (9.12)).
(b) Stage-specific probability γ j of making a transition to the next stage. (c) Stable stage structure. (d) Stage-specific reproductive values.
(a) (b) (c)

0.4 0.4 0.4 0.4 0.4 0.4
Sensitivity of Oto mj
Elasticity of Oto mj
Sensitivity of Oto γj
Elasticity of Oto γj
Sensitivity of Oto sj
Elasticity of Oto sj
0.2 0.2 0.2 0.2 0.2 0.2
0.0 0.0 0.0 0.0 0.0 0.0
–0.2 –0.2 –0.2 –0.2 –0.2 –0.2

1 2 3 4 5 6 7 1 2 3 4 5 6 7 1 2 3 4 5 6 7
Stage i Stage i Stage i
Sensitivity Elasticity
Figure 9.7 (a) Sensitivities and elasticities of λ to underlying vital rates of stage specific survival probabilities sj (b) transition probabilities γ j
(c) and oocyte production mj (c), for the isopod example in case study 2. Stages are 1) Oocytes, 2) Eggs, 3) Juveniles, 4) First reproductive stage,
5) Non-reproductive adults, 6) Second reproductive stage, 7) Senescent.
the stage represents breeding habitat quality. This habitat (Figure 9.8d), in particular for ages 2–5 after
model was also used by Hunter and Caswell (2005) in dispersal has occurred.
a demonstration of the vec-permutation matrix Hunter and Caswell (2005) calculated the sensitivi-
approach to constructing such models. This goes ties of λ to the dispersal probabilities p and q. Using
beyond the scope of this chapter but is described in our notation and scaling, these are given by
detail in Caswell et al. (2018). The model has two dλ P ∂λ ∂aij
= i,j ∂a = −v2 w1 s1 + v7 w1 s1
habitat types (‘good’ and ‘poor’ quality) and five age dp ij ∂p
(9.15)
classes, where the last class contains individuals of age = w1 (v7 − v2 )s1
5 and above. Only the 1-year-olds disperse among the
and
two habitats.
Including age within habitat, we get 10 stages, dλ
where the first five correspond to ages 1–5 in the good = v2 w6 s6 − v7 w6 s6 = w6 s6 (v2 − v7 ). (9.16)
dq
habitat and the last five correspond to ages 1−5 in the
poor habitat. The resulting 10 × 10 projection matrix With the baseline values of p = 0.25 and q = 0.375 these
for black-headed gulls is given by are dλ
dp
≈ −0.04 and dλ dq
≈ 0.02. Thus, as expected,

increased transition out of the good quality habitat has
b1 b2 b3 b4 b5 0 0 0 0 0 
 (1 − p)s1 0 0 0 0 qs6 0 0 0 0 
a negative effect on λ, while increased transition out of
 
 0 s2 0 0 0 0 0 0 0 0  the poor quality habitat is positive. The sensitivities of
 
 0 0 s3 0 0 0 0 0 0 0 
  λ to age-specific survival and fecundity in the two habi-
 0 
 0 0 0 s4 s5 0 0 0 0 
A= . tats are shown in Figure 9.8e,f, confirming that the vital
 0 0 0 0 0 b6 b7 b8 b9 b10 
 
 ps1 0 0 0 0 (1 − q)s6 0 0 0 0  parameters of the good quality habitat have a higher
 
 0 0 0 0 0 0 s7 0 0 0 
  influence on the population growth.
 0 0 0 0 0 0 0 s8 0 0 
0 0 0 0 0 0 0 0 s9 s10
(9.14)
9.6 Conclusions
Here, the upper left block of the matrix describes age- Computational and analytical advances have opened
specific survival and reproduction in the good habitat, new application of MPMs in a range of theoretical and
while the lower right matrix describes the same pro- applied fields. In conservation biology, for instance,
cesses for the poor habitat. The upper right matrix sensitivity analysis can be used to identify vulnerable
describes transition from the poor to the good habitat stages and predict responses to environmental change
(only possible in age 1), which occurs with probability (e.g., Jenouvrier et al., 2009; Hunter et al., 2010; van
q, while the lower left matrix describes transition from de Pol et al., 2010). For harvested populations or in
the good to the poor habitat, which occurs with prob- pest control, MPMs can be used to optimise the man-
ability p. These probabilities can be varied to study the agement effort and strategies with respect to different
consequences of different dispersal patterns (Lebreton, goals and risks (e.g., Bieber and Ruf, 2005). MPMs are
1996; Hunter and Caswell, 2005). widely used in ecotoxicology to evaluate the popula-
According to Lebreton (1996), the age-specific sur- tion consequences of pollutant effects on individual
vival probabilities are the same in the two habitats, life cycles (e.g., Li et al., 2014; Miller et al., 2020).
while age-specific fecundities differ, leading to higher Because the essence of MPMs is the incorporation of
fertility in the good quality habitat (Figure 9.8b; param- individual differences through stages, they also rep-
eter values in the supplementary material). For the resent a natural framework to study potential conse-
special case without any dispersal among the habitats quences of ignoring such differences (Box 9.1; Vindenes
(p = q = 0), we get a separate age-structured MPM for and Langangen 2015). In combination with modern
each habitat. The long-term growth rate is then λ ≈ 1.01 statistical methods for identifying unobserved hetero-
in the good habitat and λ ≈ 0.97 in the poor habi- geneity, MPMs represent a powerful tool to decom-
tat. When dispersal is included (p = 0.25, q = 0.375 in pose sources of variation in life history characteristics
the baseline model), the long-term growth rate for the (Caswell, 2009; Hamel et al., 2016; Jenouvrier et al.,
entire population, coupling the good and poor habi- 2018).
tats, is λ ≈ 1.00. The stable structure is skewed towards We have introduced matrix models through a time-
more individuals in the good habitat, because of high- invariant stage-structured MPM and shown how dif-
er reproduction combined with a larger probability of ferent kinds of demographic structure can be obtained
remaining in than leaving this habitat (Figure 9.8c). by adjusting the transition probabilities. The ability of
Reproductive values are higher in the good quality matrices to mathematically describe transformations of
(a) (b) 0.4

1.00
0.3
0.75
0.2
0.50
bi
si
0.1
0.25
0.0
0.00
1 2 3 4 5 1 2 3 4 5
(c) (d)
0.4 1.5
0.3
1.0
0.2
wi
vi
0.5
0.1
0.0 0.0
1 2 3 4 5 1 2 3 4 5
(e) (f)
Figure 9.8 Vital parameters and model
0.5 0.25 results from the age by stage structured
Sensitivity of
0.4 0.20 model for black-headed gulls, case

Sensitivity of
O to bj
0.3 0.15 study 3. The last age class includes ages

O to sj
0.2 0.10 ≥ 5. (a) Survival probability si (same

0.1 0.05 in the two habitats). (b) Fertility bi .
0.0 0.00 (c) Stable stage structure.
1 2 3 4 5 1 2 3 4 5
(d) Reproductive values. (e) Sensitivity of λ
Age class i Age class i to survival probability. (f) Sensitivity of λ
Good habitat Poor habitat to fertility.
multiple variables, like the population structure vector, Caswell et al., 2018), and (9) models describing indi-
makes MPMs the most flexible of any of the frame- viduals in terms of their close relatives (Pavard and
works for structured population models. Our three Branger, 2012) and entire kinship networks (Caswell,
case studies only scratch the surface of all the possi- 2019a, 2020). Coupled with these developments has
ble kinds of demographic structure that can be con- been the introduction of a new, powerful, and much
sidered. Developments and applications of MPMs go more general approach to the sensitivity analysis of
far beyond the linear, time-invariant models here. The every aspect of population dynamics (transient and
list is long, but we should draw attention to impor- asymptotic, deterministic and stochastic, linear and
tant recent developments of MPMs in (1) stochas- nonlinear) using the methodology of matrix calculus
tic models (e.g., Haridas and Tuljapurkar, 2005; Tul- (Caswell, 2019b).
japurkar and Haridas, 2006; Steinsaltz et al., 2011), A particularly important development has been
(2) nonlinear models for density dependence (Cush- theory and applications that couple MPMs explicit-
ing et al., 2002; Caswell, 2008), (3) nonlinear two-sex ly to the stochastic movement of individuals through
models for frequency dependence (Shyu and Caswell, their life cycle. This leads to a rich connection with
2018), (4) stage-structured epidemic models (Klepac Markov chain methods, which have been used to anal-
and Caswell, 2011; Metcalf et al., 2012), (5) models yse longevity and occupancy times (e.g., Tuljapurkar
for species invasion dynamics (Neubert and Caswell, and Horvitz, 2006; Horvitz and Tuljapurkar, 2008;
2000; Neubert and Parker, 2004), (6) models including Caswell, 2009; Roth and Caswell, 2018) and have led
explicit population genetics and selection (de Vries and to a complete stochastic theory for variation among
Caswell, 2019a, 2019b), (7) models for second-order individuals in lifetime reproductive output (Caswell,
systems, where individual fate depends not only on 2011; van Daalen and Caswell, 2017), and measures
present stage but also on previous stage (de Vries and of population health (Caswell and Zarulli, 2018). The
Caswell, 2018), (8) models that can incorporate any stochastic fates of individuals within an MPM provide
number of i-state dimensions (Roth and Caswell, 2016; the variability that leads to demographic stochasticity
in population dynamics and to connections between Caswell, H. 2001. Matrix population models. 2nd edition.
MPMs, multitype branching processes, and diffusion Sinauer Associates, Sunderland, Massachusetts.
models (Vindenes et al., 2008; Caswell and Vindenes, Caswell, H. 2008. Perturbation analysis of nonlinear
2018). matrix population models. Demographic Research 18:
59–116.
MPMs are increasingly applied also in compara-
Caswell, H. 2009. Stage, age and individual stochasticity in
tive analyses (Chapter 18). The growing open access
demography. Oikos 118: 1763–1782.
databases COMPADRE (2016) and COMADRE (2017)
Caswell, H. 2011. Beyond R0: demographic models for
contain published projection matrices from the liter- variability of lifetime reproductive output. PLOS One
ature; they cover an increasing number of species of 6:e20809.
plants and animals. The metadata for these studies Caswell, H. 2019a. The formal demography of kin-
provide information about the study system, location, ship: A matrix formulation. Demographic Res. 41:
phylogeny, source publication, and other factors use- 679–712.
ful for comparative analyses. MPMs can now be used Caswell, H. 2019b. Sensitivity Analysis: Matrix Methods in
to study patterns of variation in demographic parame- Demography and Ecology. Demographic Research Mono-
ters across the Tree of Life. For instance, Che Castaldo graphs, Springer Nature Switzerland, Cham.
Caswell, H. 2020. The formal demography of kinship. II.
et al. (2018) used plant MPMs from COMPADRE (2016)
Multistate models, parity, and sibship. Demographic Res.
to evaluate the extent to which demographic results
42: 1097–1144.
from one species can be extrapolated to related or bio- Caswell, H., and A. M. John, 1992. From the individual to
logically similar species, as commonly advocated in the population in demographic models. Pages 36–61 in
conservation when demographic data are lacking. Oth- D. L. DeAngelis and L. J. Gross, editors. Individual-based
er examples of applications include studies aiming to models and approaches in ecology. Chapman & Hall,
identify life history strategies in variable environments New York.
(McDonald et al., 2017), to understand evolution of Caswell, H., and R. Salguero-Gómez. 2013. Age, stage and
senescence in plants (Caswell and Salguero-Gómez, senescence in plants. J. Ecol. 101: 585–595.
2013), or to identify diverging patterns of senescence Caswell, H., and Y. Vindenes. 2018. Demographic vari-
ance in heterogeneous populations: Matrix models and
across the Tree of Life (Jones et al., 2014).
sensitivity analysis. Oikos 127: 648–663.
This chapter has provided only a glimpse of possible
Caswell, H., and V. Zarulli. 2018. Matrix methods in health
applications of MPMs, but even a model assuming demography: a new approach to the stochastic analysis of
time-independent vital rates yields much informa- healthy longevity and DALYs. Population Health Metrics
tion about the underlying life history and popula- 16: 8.
tion dynamics. Through the methods provided in this Caswell, H., C. de Vries, N. Hartemink, et al. 2018. Age x
chapter and in the cited references, MPMs allow a stage-classified demographic analysis: a comprehensive
deeper understanding of the underlying mechanisms approach. Ecol. Monogr. 88: 560–584.
shaping ecological and evolutionary processes. Charlesworth, B. 1994. Evolution in age-structured pop-
ulations. 2nd edition. Cambridge University Press,
Cambridge.
Che Castaldo, J. P., C. C. Che Castaldo, and M. C.
References Neel. 2018. Predictability of demographic rates based
Bieber, C., and T. Ruf. 2005. Population dynamics in wild on phylogeny and biological similarity. Conservation 32:
boar Sus scrofa: ecology, elasticity of growth rate and 1290–1300.
implications for the management of pulsed resource con- COMPADRE, 2016. Plant Matrix Database. Available at
sumers. J. Appl. Ecol. 42: 1203–1213. www.compadre-db.org. Max Planck Institute for Demo-
Bienvenu, F., and S. Legendre. 2015. A new approach to the graphic Research (Germany), version 4.0.1, accessed 5
generation time in matrix population models. Am. Nat. October 2016.
185: 834–843. COMADRE, 2017. Animal Matrix Database. Available at
Bolnick, D. I., R. Svanback, J. A. Fordyce, et al. 2003. The www.comadre-db.org. Max Planck Institute for Demo-
ecology of individuals: Incidence and implications of graphic Research (Germany), version 2.0.1, accessed
individual specialization. Am. Nat. 161: 1–28. 3 September 2017.
Bolnick, D. I., P. Amarasekare, M. Araújo, et al. 2011. Why Conner, M., and G. White. 1999. Effects of individual hetero-
intraspecific trait variation matters in community ecology. geneity in estimating the persistence of small populations.
Trends Ecol. Evol. 26: 183–192. Nat. Resource Model. 12: 109–127.
Cam, E., L. M. Aubry, and M. Authier. 2016. The conundrum Cushing, J. M., R. F. Costantino, B. Dennis, et al. 2002. Chaos
of heterogeneities in life history studies. Trends Ecol. Evol. in ecology: experimental nonlinear dynamics. Elsevier,
31: 872–886. New York.
Dall, S. R., A. I. Houston, and J. M. McNamara. 2004. The Hunter, C. M., H. Caswell, M. C. Runge, et al. 2010. Cli-
behavioural ecology of personality: consistent individual mate change threatens polar bear populations: a stochastic
differences from an adaptive perspective. Ecol. Lett. 7: demographic analysis. Ecology 91: 2883–3897.
734–739. Jenouvrier, S., H. Caswell, C. Barbraud, et al. 2009. Demo-
de Vries, C., and H. Caswell. 2018. Demography when graphic models and IPCC climate projections predict the
history matters: construction and analysis of second- decline of an emperor penguin population. Proc. Natl.
order matrix population models. Theor. Ecol. 11: Acad. Sci. U.S.A. 106: 1844–1847.
129–140. Jenouvrier, S., L. M. Aubry, C. Barbraud, et al. 2018. Inter-
de Vries, C., and H. Caswell. 2019a. Selection in two- acting effects of unobserved heterogeneity and individual
sex stage-structured populations: Genetics, demogra- stochasticity in the life history of the southern fulmar.
phy, and polymorphism. Theor. Popul. Biol. 130: J. Anim. Ecol. 87: 212–222.
160–169. Jones, O. R., A. Scheuerlein, R. Salguero-Gómez, et al. 2014.
de Vries, C., and H. Caswell. 2019b. Stage-structured Diversity of ageing across the tree of life. Nature 505:
evolutionary demography: linking life histories, popula- 169–173.
tion genetics, and ecological dynamics. Am. Nat. 193: Kammenga, J. E., C. A. M. v. Gestel, and E. Hornung. 2001.
545–559. Switching life-history sensitivities to stress in soil inverte-
Deevey, E. S. J. 1947. Life tables for natural populations of brates. Ecol. Appl. 11: 226–238.
animals. Quart. Rev. Biol. 22: 283–314. Kendall, B. E., M. Fujiwara, J. Diaz-Lopez, et al. 2019. Per-
Ellner, S. P., and M. Rees. 2006. Integral projection mod- sistent problems in the construction of matrix population
els for species with complex demography. Am. Nat. 167: models. Ecol. Model. 406: 33–43.
410–428. Klepac, P., and H. Caswell. 2011. The stage-structured
Engen, S., R. Lande, B.-E. Sæther, et al. 2007. Using repro- epidemic: linking disease and demography with a
ductive value to estimate key parameters in density- multi-state matrix approach model. Theor. Ecol. 4:
independent age-structured populations. J. Theor. Biol. 301–319.
244: 308–317. Knape, J., N. Jonzén, M. Sköld, et al. 2011. Individual het-
Fisher, R. A. 1930. The genetical theory of natural selection. erogeneity and senescence in Silvereyes on Heron Island.
Clarendon Press, Oxford. Ecology 92: 813–820.
Goswami, V. R., D. Vasudev, and M. K. Oli. 2014. The Lande, R. 1982. A quantitative genetic theory of life history
importance of conflict-induced mortality for conservation evolution. Ecology 63: 607–615.
planning in areas of human–elephant co-occurrence. Biol. Lebreton, J. D. 1996. Demographic models for subdivid-
Cons. 176: 191–198. ed populations: the renewal equation approach. Theor.
Hamel, S., N. G. Yoccoz, and J.-M. Gaillard. 2016. Assess- Popul. Biol. 49: 291–313.
ing variation in life-history tactics within a population Lefkovitch, L. P. 1965. The study of population growth in
using mixture regression models: a practical guide for organisms grouped by stages. Biometrics 21: 1–18.
evolutionary ecologists. Biol. Rev. 92: 754–775. Leslie, P. H. 1945. On the use of matrices in certain popula-
Hamilton, W. D. 1966. The moulding of senescence by natu- tion mathematics. Biometrika 33: 183–212.
ral selection. J. Theor. Biol. 12: 12–45. Leslie, P. H. 1948. Some further notes on the use of
Haridas, C. V., and S. Tuljapurkar. 2005. Elasticities in vari- matrices in population mathematics. Biometrika 35:
able environments: properties and implications. Am. Nat. 213–245.
166: 481–495. Li, W.-H., Y.-R. Ju, C.-M. Liao, et al. 2014. Assessment of
Hartemink, N., and H. Caswell. 2018. Variance in animal selenium toxicity on the life cycle of Caenorhabditis elegans.
longevity: contributions of heterogeneity and stochastic- Ecotoxicology 23: 1245–1253.
ity. Popul. Ecol. 60: 89–99. McDonald, D. B., and J. W. Fitzpatrick. 1996. Actuarial
Hartemink, N., T. I. Missov, and H. Caswell. 2017. Stochas- senescence and demographic heterogeneity in the Florida
ticity, heterogeneity, and variance in longevity in human scrub jay. Ecology 77: 2373–2381.
populations. Theor. Popul. Biol. 114: 107–116. McDonald, J. L., M. Franco, S. Townley, et al. 2017. Diver-
Hernández, C. M., S. F. van Daalen, H. Caswell, et al. gent demographic strategies of plants in variable environ-
2020. A demographic and evolutionary analysis of mater- ments. Nat. Ecol. Evol. 1: 0029.
nal effect senescence. Proc. Natl. Acad. Sci. U.S.A. 117: Metcalf, C. J. E., J. Lessler, P. Klepac, et al. 2012. Structured
16431–16437. models of infectious disease: inference with discrete data.
Horvitz, C. C., and S. Tuljapurkar. 2008. Stage dynamics, Theor. Popul. Biol. 82: 275–282.
period survival, and mortality plateaus. Am. Nat. 172: Metz, J. A. J., R. M. Nisbet, and S. A. H. Geritz. 1992. How
203–215. should we define ‘fitness’ for general ecological scenarios?
Hunter, C. M., and H. Caswell. 2005. The use of the vec- Trends Ecol. Evol. 7: 198–202.
permutation matrix in spatial matrix population models. Miller, D. H., B. W. Clark, and D. E. Nacci. 2020. A multi-
Ecol. Model. 188: 15–21. dimensional density dependent matrix population model
for assessing risk of stressors to fish populations. Ecotoxi- Tuljapurkar, S., and C. V. Haridas. 2006. Temporal autocor-
col. Environ. Safety 201: 110786. relation and stochastic population growth. Ecol. Lett. 9:
Neubert, M. G., and H. Caswell. 2000. Demography and 324–334.
dispersal: calculation and sensitivity analysis of invasion Tuljapurkar, S., and C. C. Horvitz. 2006. From stage to age in
speed for structured populations. Ecology 81: 1613–1628. variable environments: life expectancy and survivorship.
Neubert, M. G., and I. M. Parker. 2004. Projecting Ecology 87: 1497–1509.
rates of spread for invasive species. Risk Analysis 24: Tuljapurkar, S., U. K. Steiner, and S. H. Orzack. 2009.
817–831. Dynamic heterogeneity in life histories. Ecol. Lett. 12:
Pavard, S., and F. Branger. 2012. Effect of maternal and 93–106.
grandmaternal care on population dynamics and human van Daalen, S. F., and H. Caswell. 2017. Lifetime repro-
life-history evolution: A matrix projection model. Theor. ductive output: individual stochasticity, variance, and
Popul. Biol. 82: 364–376. sensitivity analysis. Theor. Ecol. 10: 355–374.
Pearl, R. 1923. Introduction to medical biometry and statis- van Daalen, S. F., and H. Caswell. 2020. Variance as a
tics. W. B. Saunders, Philadelphia. life history outcome: sensitivity analysis of the contribu-
R Development Core Team, 2020. R: A Language and Envi- tions of stochasticity and heterogeneity. Ecol. Model. 417:
ronment for Statistical Computing. R Foundation for Sta- 108856.
tistical Computing, Vienna. Vaupel, J. W., K. G. Manton, and E. Stallard. 1979. The impact
Roff, D. A. 1992. The evolution of life histories: theory and of heterogeneity in individual frailty on the dynamics of
analysis. Chapman & Hall, New York, U.S.A. mortality. Demography 16: 439–454.
Roth, G., and H. Caswell. 2016. Hyperstate matrix models: Vindenes, Y., S. Engen, and B.-E. Sæther. 2008. Individ-
extending demographic state spaces to higher dimen- ual heterogeneity in vital parameters and demographic
sions. Methods Ecol. Evol. 7: 1438–1450. stochasticity. Am. Nat. 171: 455–467.
Roth, G., and H. Caswell. 2018. Occupancy time in sets of van de Pol, M., Y. Vindenes, B.-E. Sæther, et al. 2010. Effects
states for demographic models. Theor. Popul. Biol. 120: of climate change and variability on population dynamics
62–77. in a long-lived shorebird. Ecology 91: 1192–1204.
Seaman, R., T. Riffe, and H. Caswell. 2019. Changing con- Vindenes, Y., S. Engen, and B.-E. Sæther. 2011. Integral
tribution of area-level deprivation to total variance in age projection models for finite populations in a stochastic
at death: a population-based decomposition analysis. BMJ environment. Ecology 92: 1146–1156.
Open 9:e024952. Vindenes, Y., B.-E. Sæther, and S. Engen. 2012. Effects of
Shyu, E., and H. Caswell. 2018. Mating, births, and tran- demographic structure on key properties of stochastic
sitions: a flexible two-sex matrix model for evolutionary density-independent population dynamics. Theor. Popul.
demography. Popul. Ecol. 60: 21–36. Biol. 82: 253–263.
Steinsaltz, D., S. Tuljapurkar, and C. Horvitz. 2011. Deriva- Vindenes, Y., and Ø. Langangen. 2015. Individual hetero-
tives of the stochastic growth rate. Theor. Popul. Biol. 80: geneity in life histories and eco-evolutionary dynamics.
1–15. Ecol. Lett. 18: 417–432.
CHAPTER 10
Integral projection models

Edgar J. González, Dylan Z. Childs, Pedro F. Quintana-Ascencio,
10.1 Introduction those that include density dependence [Metcalf et al.

2008]; stochastic dynamics [Childs et al. 2004; Ellner
When the dynamics of a population is modelled in
and Rees 2007]; both continuous and discrete state vari-
discrete time, but the variable that structures the popu-
ables [Childs et al. 2003; Ellner and Rees 2006]; or time-
lation is continuous (e.g. size), rather than discrete (e.g.
lagged dynamics Kuss et al. 2008]). In the case studies
developmental stage) as in the case of matrix popula-
presented here, we will consider a simple IPM with a
tion models (MPMs, Chapter 9), an integral projection
continuous state variable and a more complex one with
model (IPM) represents an ideal approach to describe
an individual and two environmental variables.
the behaviour of the population; it also takes advantage
An important difference between IPMs and MPMs is
of such nature to make the model more robust to small-
that the former necessarily require the construction of
er data sets (Ramula et al. 2009) and better incorporate
statistical models to describe the vital rates as a func-
other explanatory variables (e.g. environment [Merow
tion of the state variable, whereas the latter, although
et al. 2014a] and density dependence [Metcalf et al.
largely benefitted by their use, can be constructed with-
2008]). In contrast, when a continuous state variable is
out them (but see Box 9.1 in Chapter 9). Thus, this
discretised, as required by an MPM, a subjective num-
chapter aims at providing an introduction to IPMs,
ber of classes and class boundaries have to be intro-
emphasising the statistical aspects relevant for the con-
duced, a decision that can influence outputs derived
struction of the vital-rate models and exemplifying
from such models (Salguero-Gómez and Plotkin 2010).
them in an animal and a plant system. For a thorough
IPMs closely relate to MPMs, and thus all the prop-
explanation of the logic, analyses, and applications of
erties and analyses that have been developed for them
IPMs, we refer the reader to Ellner et al. (2016).
(see Chapter 9) can also be applied to IPMs. How-
ever, IPMs also inherit the assumptions of MPMs. In
its simplest form, an IPM assumes that the population
10.2 Constructing a basic IPM
is surveyed in discrete time periods (ideally coincid-
ing with or close to the pulse of reproduction), the As with MPMs, state variables, which determine
population is density independent (i.e. the popula- differential demographic behaviours among individ-
tion changes over time in an exponential manner), uals, include characteristics both of the individual
the population has a deterministic behaviour (i.e. the itself and of the environment it inhabits. The for-
dynamics of the population is strictly determined by mer includes genetic, ontogenic, physiological, or
the vital rates, which in turn are strictly determined behavioural attributes (Ellner et al. 2016; Rees and Ell-
by the state variables), the population dynamics is conner 2016), whilst the latter involve biotic or abiotic
stant over time (i.e. the dynamics does not change over conditions, measured at the individual, subpopulation,
time), and the state of a population at time t+1 strict- or population levels (Hegland et al. 2010; Merow et al.
ly depends on its state at time t. IPMs that consider 2014b). In the construction of an IPM, one assumes
more complex scenarios where one or more of these that an individual in a particular state performs
assumptions are eliminated have been developed (e.g. demographically similarly to other individuals in this
Edgar J. González et al., Integral projection models. In: Demographic Methods Across the Tree of Life.
same state. Also, one has to distinguish between individuals (G), and number (b) and size (Q) of new-
dynamic and static state variables, depending on borns.
whether an individual can or cannot change from one Once the relevant vital processes are identified, these
state of the variable to another during its lifetime. For must be integrated in the kernel, the function that
example, individuals may change in size over their life, describes the dynamics of the population of interest
whilst they may not change their genetic configuration between times t and t+1. The dynamics of the pre-
or the edaphic environment they experience. This dis- existing individuals and that of the newborns describe
tinction is important, as in the former case, one has to the dynamics of the whole population. Let us focus
consider in the IPM a function describing the proba- on the former. At time t there is a probability density
bilities among states when describing their life cycles. function dt (z) that describes the probability an individ-
Here, we will construct an IPM using a dynamic state ual in the population has of having a size z. If at this
variable: individual size. moment the population consists of Nt individuals, with
To start, let us consider a closed population (i.e. with their sizes distributed according to dt , we can define
no immigration or emigration) structured by indivi- the function as
dual size and identify the relevant vital processes. In a
census performed at a discrete time t+1, the popula- nt (z) = Nt dt (z) (10.1)
tion consists of the individuals that existed when cen- R R
sus at time t was performed and survived to time t+1 Note that Z nt (z)dz = Nt Z dt (z)dz = Nt , as dt (z) is
(nP in Figure 10.1) and those that, through reproduc- a probability density and integrates to one over the set
tion, entered the population between these two time of possible size values Z. Thus, although nt (z) is not a
points and survived to time t+1 (nB in Figure 10.1). The probability density, it behaves as one, in the sense that
size of the individuals that survived to time t+1 might it is continuous, and, if normalised (i.e. divided by its
have increased, remained, or decreased; the individu- integral), it integrates to one. Therefore, for the remain-
als born between the two time points must have a size der of the chapter we will refer to this type of function
at time t+1. Thus, in our case, the set of vital process- as density functions, restricting the term probability den-
es that we must consider to adequately describe the sity function to those that strictly integrate to one.
life cycle are: survival (s), change in size of existing If s(z) denotes the survival function, which gives us
the probability an individual of size z has of surviving
to time t+1, and if at time t we had a density function
Intercensal period nt (z), at time t+1 we will have a density s(z) . nt (z) of the
Census t Census t+1 surviving individuals of size z. Note that by making s
Reproduction period to depend only on z and not on Nt , we are assuming
nt+1(z’) = nP(z’)+nB(z’) that population size does not impact survival—that is,
that survival is density independent.
Newborn size
n (z’) Next, a surviving individual must have a size at time
Q(z’, z) B
Reproduction t+1; the function that describes the probability an indi-
nt(z) pb(z)⋅b(z)
vidual that at time t had a size z has of having a size z′
Survival at time t+1 is denoted by G(z′ , z) and should have the
R
s(z) property that Z G(z′ , z)dz′ = 1 for all z values. There-
np(z’) fore, the density of the surviving individuals of size z
Change in size G(z’, z) at time t that reached a size z′ at time t+1 is
Figure 10.1 Components of an IPM. The individuals in the nP (z′ , z) = G (z′ , z) · s (z) · nt (z) (10.2)
population at census time t (left rectangle) are structured according to
nt (z). During the intercensal period, a fraction of them survive
according to s(z) and change to a size z′ at time t+1 with probability To estimate the density of the individuals of size z′ at
G(z′ , z). The surviving individuals at time t+1 will distribute according time t+1, irrespective of their original size at time t, we
to nP (z′ ). The reproduction period occurs over part of the intercensal need to integrate this function over all possible original
period, and a fraction pb (z) of the individuals bear b(z) newborns, z sizes. Thus,
which at time t+1 will have a size z′ with probability Q(z′ , z). At time
t+1 there will be newborns that survived to this moment and are Z
distributed as nB (z′ ). The density of individuals at time t+1 results nP (z′ ) = G (z′ , z) · s (z) · nt (z) dz (10.3)
from the addition of nP (z′ ) and nB (z′ ). Z
INTEGRAL PROJECTION MODELS 183
As with eqn. 10.1, if we now integrate over all If we define the kernel as
R
possible z′ sizes, Z nP (z′ )dz′ = NP is the total number
of individuals that existed at time t and survived to K (z′ , z) = G (z′ , z) · s (z) + Q (z′ , z) · b (z) · pb (z) (10.7)
time t+1.
Let us next focus on the newborns. In a closed popu- we have that
Z
lation, the number of newborns will strictly depend on
the number of reproductive individuals. Given a den- nt +1 (z′ ) = K (z′ , z) · nt (z) dz, (10.8)
Z
sity nt (z) of existing individuals in the populations, a
size-dependent fraction of them, pb (z), will reproduce. and, as stated above,
Thus, we will have a density pb (z) . nt (z) of reproduc- Z
tive organisms. Each of these individuals will produce nt +1 (z′ ) dz′ = Nt +1 . (10.9)
Z
a size-dependent number of newborns, b(z), that will
be alive at time t+1; that is, b(z) . pb (z) · nt (z) newborns Eqns. 10.7 and 10.8 contain the structure of a basic IPM
will be alive at time t+1. Again, note that b does not and describe how the density of individuals at time t,
depend on Nt , and thus we are assuming that fecundity nt , turns into that at time t+1, nt+1 , while considering
is density independent. the size-structured dynamics of the population. Evi-
Finally, given that the size of the newborn at time dently, alternative kernel structures can be conceived
t+1 is relevant, we need to know the probability Q(z′ , z) as each species, state variable, and census timing will
that a newborn has of having a size z′ at time t+1, given present their particular modelling considerations; sev-
that its parent had a size z at time t; again, this function eral structures are explored by Ellner et al. (2016). Also,
R
must satisfy that Z Q(z′ , z)dz′ = 1 for all z values. Thus, not all vital processes may be easily quantifiable, in
the density of the newborns of size z′ that originated which case we can resort to inverse estimation (see
from parents of size z is given by Box 10.1).
As stated above, the kernel, K(z′ , z), is the continu-
nB (z′ , z) = Q (z′ , z) · b (z) · pb (z) · nt (z) (10.4) ous equivalent of the transition matrix in the MPMs
(eqn. 9.1). As with MPMs, we can extract the asymp-
Again, we can integrate over all original parents of size totic properties from the kernel, such as the asymptotic
z and get population growth rate (λ), perform sensitivity and
elasticity analyses (see section 10.5), and, by iterating
Z eqn. 10.8, we can project the population structure (and
nB (z′ ) = Q (z′ , z) · b (z) · pb (z) · nt (z) dz, (10.5) size, using eqn. 10.9) over time to determine the trajec-
Z
tory a population will be expected to follow under the
the density of the newborns of size z′ at time t+1, and assumptions of the model.
R ′ ′ Finally, IPMs are mechanistic models, in the sense
Z nB (z )dz = N B is the total number of newborns at
time t+1. In constructing this function, we are relating that eqn. 10.6 describes the way in which the size struc-
the size of the parent with that of its descendant. This ture changes over time as a function of the underlying
relation can be based on evidence from the census data processes responsible for such change. However, func-
(see case study 1, section 10.7.1); however, as we will tions s, G, b, pb , and Q must still be defined. These could
see in case study 2 (section 10.7.2), we sometimes do not in turn be described mechanistically and several efforts
have evidence of such relation or we cannot associate have been made to achieve this, for example, by includ-
parents with newborns, and thus we have to model Q ing energy budget (Smallegange et al. 2017) or genetic
only as a function of z′ . (Coulson et al. 2011) dynamics as drivers of individ-
Given that the population at time t+1 consists of ual performance. In the case when no mechanism
the individuals that existed at time t and survived to was considered in studying the population, one can
time t+1, plus the newborns present at time t+1, fol- resort to phenomenological (i.e. nonmechanistic) rela-
lowing eqns. 10.3 and 10.5, the density of the existing tions between the state variable and the vital processes;
individuals of size z′ at time t+1 is such would be a semimechanistic IPM (Ellner et al.
1998). Here, to provide a general framework that does
Z not depend on the identity of the state variable used,
nt +1 (z′ ) = nP (z′ ) + nB (z′ ) = [G (z′ , z) · s (z) we will present modelling options to establish phe-
Z
nomenological relationships between state variables
+ Q (z′ , z) · b (z) · pb (z)] · nt (z) dz (10.6)
and vital rates.
Box 10.1 Dealing with missing life cycle parts: an inverse modelling approach
The lack of information on a particular stage of the life Let us assume that in the Soay sheep example (see
cycle of the species of interest or the inability to follow section 10.7.1) we do not have information on the weight-
individual dynamics through time are common situations dependent number of newborns each ewe has per year. This
in demographic research. For instance, seed dispersal and would be the case if surveying could not be performed during
germination in many plants, death in long-lived trees, and the breeding season and newborns could not be associat-
migration and hibernation in animals are processes that can ed to their parents. Still, we know that the size structure
be hard to record; our inability to do so will render imper- of the population fluctuates over time and thus the number
fect censuses. Under these scenarios, many researchers can of reproductive females; if reproduction were weight depen-
rely on off-field experiments to estimate transition probabil- dent, we would also expect a correlated fluctuation in new-
ities over these unobserved processes (e.g. Williams et al. borns. However, this correlation is blurred by the survival and
2010; Zambrano and Salguero-Gómez 2014), but these change in weight of ewes, as the population size structure
experiments can hardly reproduce the vagaries of natural reflects all these processes. Thus, we have to use the survival
conditions. An alternative to this challenge is to recognise and weight data, a traditional input of an IPM, along with the
that the output of an IPM, that is, the population size and population size and structure, the IPM output, to estimate
structure (n and N in eqns. 10.8 and 10.9, respectively), the pb and b functions in eqn. 10.7. However, these functions
contains information on these unobserved processes and involve a large number of parameters (associated to both
that we can use it to our advantage. In an inverse modelling fixed and random variables), which cannot be estimated from
approach, we use the output of a model as its input to infer the limited information that population size and structures
the value of the parameters associated to these missing pro- provide.
cesses (González et al. 2016). It is worth noting that IPMs Thus, we must fit a simpler fecundity function: if we
are better suited for this approach than MPMs, because the know that ewes can have up to two newborns per year
former depend usually on fewer parameters than the latter, (Clutton-Brock and Pemberton 2004), a linear logistic model
and the larger the number of parameters, the harder it is to with an upper asymptote of two would involve the esti-
estimate them inversely. mation of only an intercept and a size slope, and would
(a) (b)
2.0
200
1.5
Number of newborns
Population size
150
1.0
100
0.5
50
0.0
10 20 30 1986 1988 1990 1992 1994 1996
Initial weight (kg) Year
Boxed Figure 10.1 Inverse estimation of fecundity in Soay sheep. (a) Number of newborns produced as a function of ewe
weight. Green indicates directly estimated number, corresponding to b(z) . pb (z) in eqn. 10.7, using individual fecundity data
(dots); red indicates inversely estimated fecundity, corresponding to f(z) in eqn. 10.B1, using population data (i.e. population
structures and sizes). (b) Population sizes observed during the study period. Green indicates projected sizes using individual
data; red indicates projected sizes using individual and population data.
approximate the number of newborns produced by a ewe. to www.oup.com/companion/SalgueroGamelonDM), we use

As with b(z) in eqn. 10.32, we multiply this model by differential evolution adaptive Metropolis (implemented in
0.5 to consider only female lambs. This model can be for- the dream package; Vrugt et al. 2009) to estimate the
mulated as parameters. A comparison between the fecundities estimat-
ed through direct and inverse modelling is shown in Boxed
f (z) = 0.5 · 2 · logistic(β0 + β1 · z), (10.B1)
Figure 10.1a.
and approximates the product b(z) . pb (z) in eqn. 10.7. Finally, we have to emphasise that, given that inverse
We want to find the values of these two parameters that modelling aims at estimating vital-rate parameters
maximise the likelihood of both the individual-level (sur- that translate into population sizes and structures that
vival, weight of existing individuals and newborns) and match those observed, this approach translates into
population-level (population size and structure) data. As an IPM that better predicts population-level patterns
shown in the code provided in the Appendix (please go (Boxed Figure 10.1b).
10.3 Modelling vital rates in terms we assume that, on average, they relate linearly, we
of the state variable can use eqn. 10.11 to construct our change-in-size and
size-of-newborn functions, G(z′ , z) and Q(z′ , z), respec-
We have several options on how to relate the census tively. Given that size can only take positive values,
data on the state variable and those on the vital rates. we usually log-transform this variable to make it span
First, we have to recognise that usually the response all real numbers, as required by the normal distribu-
variables (i.e. the vital rates) will be measurements (G tion, or use strictly positive distributions such as the
and Q in eqn. 10.7), probabilities (s and pb ), or counts log-normal or gamma.
(b). To model measurement variables, linear models
(LMs) are commonly used, whilst generalised linear
models (GLMs) are needed to model probabilities and 10.3.2 Generalised linear models
counts. In the following subsections we will introduce
As stated, vital processes may describe probabilities
these two types of models, starting with LMs.
or counts. Since such variables do not (and cannot be
made to) span all real numbers, and normality can-
10.3.1 Linear models not be achieved without such range, we need models
that allow for different ranges of the response variable;
The basic LM associates a response variable with
these are known as GLMs (Zuur et al. 2013).
an explanatory one, assuming independence among
The probabilities we have to model come from bina-
data points, linearity, normality, and homoscedastici-
ry events (0: dead, 1: alive for the s function in eqn.
ty. Thus, if y is a continuous (measurement) variable
10.7; and 0: nonreproductive, 1: reproductive for the pb
that can take any real value and x is an explanatory
function); such events can be described with a binomial
variable, we can resume the above as
distribution. Thus, if we substitute the normal distri-
y ∼ Normal(β0 + β1 · x, σε ). (10.10) bution in eqn. 10.10 for this distribution, we have the
correct value range to model these events. However,
This equation states that y follows a normal distri- changing the distribution is not the only modifica-
bution with mean β 0 + β 1 . x (the mean of y changes tion we have to make; given that a line spans all real
linearly with respect to x) and standard deviation σ ε ; numbers, we have to transform it to a more appro-
β 0 , β 1 , and σ ε are parameters that have to be estimated priately scale; in the case of probabilities, the logistic
from data. If we remember the equation of the normal function is a transformation that restricts a line to the
probability density function, we can express the above [0, 1] interval. Similar to eqn. 10.10, we can describe
equation as mathematically this GLM as
[ y−(β0 +β1 x)]2
1 −
f (y, x) = √ e 2σε2
(10.11) y ∼ Binomial(Nt , logistic(β0 + β1 · z)), (10.12)
2πσε
Note that, stated in this way, we show that this is a where Nt (z) is the number of events involved and
bivariate function. Thus, if y and x are sizes at time logistic(β 0 + β 1 . z) is the size-dependent probability
t+1 and t, respectively (i.e. z′ and z in eqn. 10.6), and of success (i.e. of having a value of 1); β 0 and β 1 are
parameters that have to be estimated from the data. have to reassign the probability in the tails of the nor-
This size-dependent probability can be used to describe mal probability outside of the [L, U] interval to these
the mean behaviour of our binary events s and pb . Thus, extreme values. By doing this we are avoiding the
accidental eviction of individuals from the population
s (z) = logistic(β0 + β1 · z) and
(10.13) due to their too small/large size (see more details in
pb (z) = logistic(β0 + β1 · z). Williams et al. 2012). To achieve this we can redefine the
bivariate function in eqn. 10.10 as a restricted function
For count variables (in our case b in eqn. 10.7), the neg-
(Easterling et al. 2000),
ative binomial distribution is the most commonly used
one, and the exponential function is a transformation
Fr (z′ , z)= F (z′ , z) , if L ≤ z′ ≤ U,
that restricts a line to positive values. This GLM can be
expressed as = F (L, z) , if z′ < L, and (10.16)
= F (U, z) , if z′ > U.
b ∼ Negative binomial(exp(β0 + β1 · z), θ), (10.14)
where θ is a scale parameter. The mean of this distribu- Third, to implement eqn. 10.8, we would require to ana-
tion is lytically integrate K(z′ , z) . nt (z) over Z. However, this is
computationally impractical, and one usually resorts to
b (z) = exp(β0 + β1 · z). (10.15) numerical methods to approximate this integral (East-
erling et al. 2000). A common approach is the midpoint
Therefore, eqns. 10.11, 10.13, and 10.15 fully define the
rule (Figure 10.2; see Zuidema et al. 2010 and Ellner
functions involved in the construction of the kernel in
et al. 2016 for alternatives), where the kernel is evaluat-
eqn. 10.7.
ed on a partition of the [L, U] interval into m classes of
equal width h = (U – L)/m. Thus, if zi = L + (i – 0.5) h are
10.4 Combining the vital functions the midpoints of these classes, the integral at midpoint
in the IPM z′ j is approximated as
Once the vital processes have been modelled as a func-

ZU X
m
tion of the state variables, we would like to use the asso- nt+1 z′j = K z′j , z nt (z) dz ≈ hK z′j , zi nt (zi ) .
ciated functions, s, G, pb , b, and Q, through eqn. 10.6, to i=1
L
combine them in the kernel. Although eqn. 10.6 clear- (10.17)
ly states how the functions combine, its computational
implementation requires a few steps. As the error in this approximation decreases (quadrat-
First, we must establish the (closed) interval of val- ically) with m, we will get better approximations with
ues of the state variable Z = [L, U] we will assume a large m-value (m >100; Ellner et al. 2016).
an individual can take. We set Z as a closed interval As stated above, we can use the theory developed for
because it makes biological sense that individuals can MPMs to analyse an IPM if we note that h . K(z′ j , zi ) is
only display state values within a certain interval. Evi- an m×m matrix. Thus, if we denote K = h . K(z′ j , zi ) and
dently, L and U must be chosen so that no potential define the matrices G = h . G(z′ j , zi ) and Q = h . Q(z′ j , zi ),
individual in the population is excluded. However, as and the vectors s = s(zi ), b = b(zi ) and pb = pb (zi ), we can
we will see next, if these two values make relatively express eqn. 10.7 with the notation of Chapter 9 as
good biological sense, the dynamics will be correctly
modelled. K = GDiag (s) + QDiag (b) Diag (pb ) . (10.18)
Second, by restricting our state variable to take val-
ues only within the [L, U] interval, we have to take The only difference with that chapter is that the F
into account that, if we originally considered a differ- matrix is QDiag(b)Diag(pb ). Therefore, K can be used
ent scale to model a vital process, we have to correct as the transition matrix in the previous chapter.
the scale. For example, if we log-transformed size we Finally, we want to emphasise that resorting to dis-
are assuming that this variable can take any real val- cretisation is not an advocacy for MPMs. The main
ue and that a normal distribution (eqn. 10.11) correctly difference is that the vital-rate models were construct-
describes the variance around the mean in G and Q. ed without any categorisation, and thus, for example,
To correct the scale of the variable, we can assume that the s vector does not contain the average survival of
any individual with size z > U or z < L at time t+1 the observed individuals within each category, but the
will actually behave demographically as one having estimated one following a model of the relationship
a size equal to U or L, respectively. Accordingly, we between the continuous state variable and survival.
Continuous density n (z) Discretised density n (z )

t t i
Product h⋅K(z’ , z )⋅n (z )

j i t i m
Discretised density n (z’ ) = ∑h⋅K(z’ , z )⋅n (z )

i=1
t+1 j j i t i
z h z z’
Discretisation Addition
Multiplication
Continuous kernel K(z’, z) Discretised kernel K(z’ , z )
j i
z z’
z’ z’
z z
Figure 10.2 Midpoint rule integration procedure to iterate an IPM. (a) Starting with the continuous population size density at time t, nt (z)
(black), we discretise it (grey); here the partition has 20 classes. (b) Under the same partition, we discretize the kernel, K(z′ , z), (c) multiply it by the
discrete size structure, and (d) obtain a discrete size structure at time t+1 (black), closely resembling the corresponding continuous size density at
time t+1, nt+1 (z′ ) (grey).
10.5 Perturbation analysis: sensitivity at some point (z′ 0 , z0 )? Ellner et al. (2016) show that
and elasticity this change can be calculated in the same way as
with MPMs (eqn. 9.2). Therefore, we can calculate
Perturbation analyses evaluate the effect that a pertur- the sensitivity (S) of λ to a change of K(z′ , z) at point
bation of the kernel or some other underlying part of (z′ 0 , z0 ) as
it has on the measures derived from the IPM; in the
remainder of the section, we will focus on the popu- ∂λ ′
SK (z′0 , z0 ) = (z , z0 ) = v (z′0 ) · w (z0 ) , (10.19)
lation growth rate (λ), as it is a commonly examined ∂K 0
metric—but we emphasise that perturbations can be where v(z′ ) are the reproductive values and w(z) the
done with regards to any outcome of the model. For stable (size) structure associated to K(z′ , z). Note
some metrics perturbations, analytical formulae exist that, as in Chapter 9, v and w are scaled so that
R ′ ′
R
(see Caswell 2001), and for others, numerical meth- Z v(z )dz = Z w(z)dz = 1.
ods are necessary (‘brute force’ sensu Morris and Doak In turn, the elasticity (E) of λ to a change of K(z′ , z)
2002). at point (z′ 0 , z0 ) is
In evaluating the effect of a perturbation, we might
∂λ
z′ , z0 K z′0 , z0 ∂λ ′
be interested in its absolute effect on λ, or on its effect EK (z′0 , z0 ) = ∂K λ0 = · (z , z0 )
relative to its original state; sensitivity and elasticity ′
λ ∂K 0
K(z0 , z0 )
analyses evaluate these absolute and relative effects,
K z′0 , z0
respectively. An important assumption we make in = · SK (z′0 , z0 ) . (10.20)
performing these analyses is that vital processes are λ
independent from one another, and thus perturbations
do not propagate to other processes; for a discussion on 10.5.2 Perturbation on the pre-existing
this matter in MPMs see van Tienderen (1995) and De and newborn sections of the kernel
Kroon et al. (2000), and in IPMs see Griffith (2017).
As we described in section 10.2 and as is shown in
Figure 10.1 (blue vs. orange areas), the dynamics
described by K(z′ , z) can be decomposed into the pre-
10.5.1 Perturbation on the kernel
existing and newborn sections (U and F matrices in
A perturbation on the kernel addresses the question: Chapter 9)—that is, K(z′ , z) = U(z′ , z) + F(z′ , z), where
How much does λ change when we perturb K(z′ , z) U(z′ , z) = G(z′ , z) . s(z) and F(z′ , z) = Q(z′ , z) . b(z) . pb (z).
We may be interested in evaluating the sensitivity (or is a perturbation of s(z) at a point z0 and all points z′ ;
elasticity) of λ to a change in one of these compo- that is,
nents of the system. Using the chain rule and some ZU
derivative properties, we can show that the sensitiv- ∂λ ∂λ
Ss (z0 ) = (z0 ) = (z0 ) dz′ .
ity of U(z′ , z) at (z′ 0 , z0 ) is equal to the sensitivity of ∂s ∂s
L
K(z′ , z):
Again, using derivative rules and the results of eqn.
∂λ ∂λ ∂K ∂λ ∂ [U + F]
SU = = · = · 10.21 we obtain
∂U ∂K ∂U ∂K ∂U

∂λ ∂U ∂F ∂λ ∂λ ∂λ ∂U ∂G · s ∂G ∂s
= · + = · [1 + 0] = SK . (10.21) = · = SK · = SK · ·s+G·
∂K ∂U ∂U ∂K ∂s ∂U ∂s ∂s ∂s ∂s
Note that, for simplification, we are omitting the point = SK · G. (10.25)
of evaluation (z′ 0 , z0 ) in the equation. The same proce-
dure shows that SF , the sensitivity of F(z′ , z), is equal to Thus,
SK . ZU ZU
∂λ
The elasticities EU and EF , however, are not the same, Ss (z0 ) = (z0 ) dz′ = SK (z′ , z0 ) · G (z′ , z0 ) dz′ .
∂s
as they differ in their denominators: L L
∂λ ∂λ (10.26)
U F
EU = ∂U = · SK and EF = ∂F = · SK .
λ λ λ λ Under the same procedure, we can show that
U F
(10.22) ZU
Sb (z0 ) = SK (z′ , z0 ) · Q (z′ , z0 ) · pb (z0 ) dz′ and
Considering that the elasticity of λ to a change in a
L
demographic component f is always the sensitivity of
λ to a change in f divided by the quotient λ/f (i.e. Ef = ZU
Sf /(λ/f )), for the remainder of the section we will Sp b (z0 ) = SK (z′ , z0 ) · Q (z′ , z0 ) · b (z0 ) dz′ . (10.27)
focus only on the sensitivities, leaving to the reader the L
calculation of this division.
Note that these are the sensitivities of λ to a pertur-
bation of a vital rate at a particular point z0 . If we
10.5.3 Sensitivities on the vital rates were interested in the sensitivity to a perturbation of
the entire vital rate, we only have to integrate over
If we are interested in perturbing a particular vital rate,
all points z. Thus, for example, perturbing the entire
we have to remember that we have bivariate and uni-
survival would correspond to
variate vital rate functions (i.e. G(z′ , z) and Q(z′ , z), and
s(z), b(z), and pb (z), respectively). For the former, and ZU
using G(z′ , z) to exemplify the procedure, the sensitiv- Ss = Ss (z) dz. (10.28)
ity of λ to a perturbation of G(z′ , z) at a point (z′ 0 , z0 ) L
can be derived using again the chain rule, the product
rule, and the results of eqn. 10.21 as 10.5.4 Sensitivities to vital rate parameters
∂λ ∂λ ∂U ∂G · s
SG = = · = SK · Finally, we may be interested in how a perturbation
∂G ∂U ∂G ∂G in a parameter θ of a vital rate function f impacts λ.

∂G ∂s Now, vital rate functions can take many forms differ-
= SK · ·s+G· = SK · s. (10.23)
∂G ∂G ent from those described in eqns. 10.11, 10.13, and 10.15
Similarly, the sensitivity of λ to a perturbation of (see e.g. section 10.6.1). Thus, the sensitivity of λ to a
Q(z′ , z) at point (z′ 0 , z0 ) is perturbation on a particular parameter has no gener-
al expression. However, due to the chain rule, we only
SQ = SK .b.pb . (10.24) have to focus on the derivative of the vital rate function
For the univariate vital rates, and using s(z) to exem- with respect to the parameter to correctly calculate the
plify the procedure, a perturbation of s(z) at a point z0 sensitivity function:
∂λ ∂λ ∂f ∂f 10.6.1 Extensions of the linear model

Sθ = = · = Sf · . (10.29)
∂θ ∂f ∂θ ∂θ
The statistical theory on LMs includes alternative mod-
To exemplify this, we can calculate the sensitivity of λ els when the data violate some or all of the assumptions
to a perturbation in β 1 of s in eqn. 10.13 at point z0 . of a traditional LM: independence among data points,
Thus, linearity in the relation between the explanatory and
h i the response variables, and homoscedasticity in the
1 response variable around its mean. These alternatives
∂s ∂
1+e−(β0 + β1 z)
Sβ 1 = Ss · = Ss · provide a high degree of flexibility in the phenomeno-
∂β1 ∂β1
logical models we can construct.
= Ss (z0 ) · [s (z0 )]2 · e−(β0 + β1 z0 ) · z0 . Nonindependence among data points occurs when
(10.30) these are considered as spatially or temporarily related.
Data points may be spatially associated if, for example,
The interpretation of parameter-level sensitivities is several plots were set within the experimental design,
not always straightforward. For instance, in the above and thus we need to consider all data within a plot
example, the sensitivity to a perturbation of β 1 will as nonindependent; data are temporally related if they
be non-negative for all z values (since all factors are come from an individual that was measured more
non-negative); that is, a change in the survival of indi- than once. In such cases, the LM framework allows for
viduals will affect λ in the same direction irrespective the inclusion of random factors (plot, individual, year,
of their size. However, if survival is described by a etc.), which can make any parameter of the model to
nonlinear function, a perturbation in β 1 can produce vary between values of these factors. The models that
both negative and positive sensitivities on λ. Consider allow including random factors are the so-called linear
the survival equation s in the case study 1 (eqn. 10.32), mixed models (Bolker et al. 2009; Zuur et al. 2013).
which includes a nonlinear function: a spline (i.e. Nonlinearity usually arises when mechanistic rela-
a function of interconnected polynomials); in this tionships between response and explanatory variables
example the sensitivity to a perturbation in β 1 at size are considered. However, we can also construct non-
z0 is linear phenomenological models within the previous
framework. These are the generalised additive models
Sβ1 = Ss (z0 ) · [s (z0 )]2 · e−[β0 +β1 ·spline(z0 )] · spline(z0 ). (Ozgul et al. 2012; Hastie 2017; Wood 2017), which
(10.31) use splines to fit any nonlinear relationship (e.g.
Figure 10.3a).
Since Ss (z0 ), [s(z0 )]2 , and exp{–[β 0 + β 1 . spline(z0 )]}
Finally, heteroscedasticity can be addressed by mod-
are non-negative, the sign of the sensitivity is deter-
elling σ ε in eqn. 10.11, not as a constant but as a function
mined by the sign of the spline at z0 . In our case, the
of z, be it linear or not (Easterling et al. 2000; Rigby et al.
spline takes positive values for small and large individ-
2017).
uals and negative for medium-size ones (Figure 10.3a).
Therefore, the sensitivity to a perturbation in β 1 will
be negative for these latter individuals, and thus an
10.6.2 Model selection
increase in β 1 in medium-size individuals will translate
into a lower λ. If we measured either one or more individual char-
acteristic or environmental factor or a combination
of both, it is likely that we may want to construct a
10.6 Further considerations in the kernel that considers the relevance of these variables
in describing the vital rates. Under such a scenario,
construction of vital-rate models we will first consider multiple models regarding how
Having presented the theory around IPMs, the con- well (or poorly) the variables describe each vital rate; in
struction of the vital-rate models, and their integration addition to linear vs. nonlinear relations between each
in the kernel, we can now consider two important sub- state variable and the vital rate, we have to consider the
jects for the construction of the vital-rate models: (1) the potential interactions that may exist between state vari-
extensions of the LM that allow the use of data that vio- ables in the effect they have on each vital rate. Every
late its assumptions, and (2) the selection procedure for option will translate into a model, and we will end
a set of competing vital-rate models. up with a set of models from which we have to select
(a) (b) (c)

1 1.00 35
30
Survival probability
Final weight (kg)

0 0.75
Fitted spline
25
–1 0.50
20
–2 0.25 15
10
–3 0.00
10 20 30 10 20 30 10 20 30
initial weight (kg) Initial weight (kg) Initial weight (kg)
(d) (e) (f)

1.00 2.00
Reproduction probability
Number of newborns
Offspring weight (kg)

0.75 1.75 15
0.50 1.50
10
0.25 1.25
5
0.00 1.00
10 20 30 10 15 20 25 30 10 15 20 25 30
Initial weight (kg) Initial weight (kg) Ewe weight (kg)
(g) (h)
30 200
Final weight (kg)
Population size
k 150
0.03
20 0.02
0.01
0.00 100
10
50
10 20 30 1986 1988 1990 1992 1994 1996

Initial weight (kg) Year
Figure 10.3 Results on modelling of Soay sheep demographic behaviour. (a) Spline fitted to the survival data (spline in s in eqn. 10.32);
(b) weight-dependent survival; (c) individual change in weight between consecutive time points; (d) weight-dependent probability of reproduction;
(e) weight-dependent fecundity; (f) weight of newborns as a function of parent weight (blue and red lines indicate single and twin pregnancies,
respectively); (g) heat map of the kernel (eqn. 10.7) as a function of individual size at two consecutive time points; (h) population sizes observed
during the study period (green indicates projected sizes using the IPM).
the best one. Information criteria, among which the 2004); thus, for this function we consider the binary
most commonly used is the sample-corrected Akaike explanatory variable b. The Appendix (please go
information criterion (AICc), allow the comparison of to www.oup.com/companion/SalgueroGamelonDM)
models taking into account their goodness-of-fit, their presents the set of models considered and the model
complexity, and the amount of information available selection procedure followed. Modelling was per-
(i.e. sample size). We consider the best-supported mod- formed in R (R Core Team 2020) using packages lme4
el as the one having the smallest value of the criterion. (for generalised linear mixed models; Bates et al. 2015),
For a full description of the theory around model selec- gamm4 (for generalised additive mixed models; Wood
tion see Burnham and Anderson (2002). and Scheipl 2017), and AICcmodavg (for model selec-
tion; Mazerolle 2017).
The (marginal) structures of the selected models
10.7 Case studies were:
Here we exemplify IPMs with two study systems: an
s (z) = logistic(β0 + β1 · spline (z)),
animal and a plant population. The R code associat-
ed to the construction of the IPMs is included in the −
[z′ −(β0 +β1 z)]2
Appendix (please go to www.oup. com/companion/ G (z′ , z) = √1 e
2 πσε
2σε2
,
SalgueroGamelonDM). pb (z) = logistic(β0 + β1 · z), (10.32)
b (z) = 0.5 · (1 + logistic(β0 + β1 · z), and
10.7.1 Case study 1: Soay sheep population {z′ −[β0 +β1 z+β2 b]2 }
−
structured by weight Q (z′ , z, b) = √1 e
2 πσε
2σε2
.
We will start with a relatively simple case study on The graphs associated to these models are presented
an animal system: The database on the Soay sheep in Figure 10.3b–f. Note that survival is described non-
(Ovis aries). The data come from a survey of unique- linearly (through a spline, Figure 10.3a) by the state
ly marked female sheep inhabiting the Village Bay variable, while pb (z) is linear, both in the logistic scale
of Hirta Island, St Kilda archipelago, Scotland, Unit- (Figure 10.3b,d). G(z′ , z) and Q(z′ ) follow a structure
ed Kingdom, between 1986 and 1996 (Clutton-Brock similar to eqn. 10.11; since Q(z′ , z) depends on b, sin-
and Pemberton 2004; Coulson 2012). The survey year gle pregnancies produce heavier newborns than twin
ran from 1 August to 31 July, and the state variable ones (Figure 10.3f; blue and red lines, respectively).
was the weight of the sheep. Females reach a maxi- Note that in b(z) we add 1 to the logistic model, which
mum longevity of 16 years and can conceive at around ranges in the [0, 1] interval, to make it range in the
7 months of age. A maximum of two newborns are [1, 2] newborns interval and then multiply it by 0.5 to
born per female sheep in April of each year, so they are obtain the number of newborn females (Figure 10.3e).
4 months old by the time the population is surveyed. These functions were integrated into the kernel follow-
Because the sex of the newborn was not recorded, we ing eqn. 10.7 (Figure 10.3g; note that the darker line
assume that there is a 50% chance that it is female. reflects the product of the s and G functions and the
To model the vital rates, we first have to recognise lighter one reflects the product of pb , b, and Q). The
that, since an individual can live for more than a year model is then iterated following eqn. 10.8 (Figure 10.2),
and all individuals in a given year experienced the producing estimated population sizes that reflect the
same environmental conditions, we have to take into average (marginal) dynamics of the population over
account the nonindependence between the data associ- the 10-year study period (green line in Figure 10.3h).
ated to the individual and the year of study; we can also Finally, Box 10.1 demostrates how, in a scenario
consider nonlinear relations between weight (z, kg) and where information on the weight-dependent number
each vital rate. Additionally, to model reproduction we of newborns produced by each ewe were not available,
have to modify function b in eqn. 10.15, as the num- an inverse estimation procedure could be used to
ber of newborns is restricted to two; thus, a binary approximate the function (b) describing it.
event (0 = single newborn, 1 = two newborns), as
in eqn. 10.12, is a better modelling approach. Final- 10.7.2 Case study 2: Perennial herb population
ly, we have to modify function Q to account for the structured by height, elevation, and time
fact that it is known that in this species the weight of since fire
the newborn (z′ ) depends both on the weight of the
ewe (z) and on whether it was the product of a single We will now model the dynamics of a more complex
or twin pregnancy (b) (Clutton-Brock and Pemberton study system. Hypericum cumulicola (Hypericaceae) is a
(a) (b) (c)

1.00 1.00
Reproduction probability
75
Final height (cm)

0.75 0.75
50
0.50 0.50
25 0.25
0.25
0.00 0 0.00
0 25 50 75 0 25 50 75 0 20 40 60 80
Initial height (cm) Initial height (cm) Initial height (cm)
(d) (e) (f)

8 1000
log(Number of recruits)
Population size
6 0.02
800
Density
4
0.01 600
2
400
0 0.00
0 20 40 60 80 0 25 50 75 1995 2000 2005 2010
Initial height (cm) Final height (cm) Year
(g)
80
60
Final height (cm)
k
0.03
0.02
0.01
40
0.00
20
0
0 20 40 60 80
Initial height (cm)
Figure 10.4 Results on modelling of Hypericum cumulicola demographic behaviour. (a) Height-dependent survival; (b) individual change in
height between consecutive years; (c) height-dependent probability of reproduction; (d) height-dependent fecundity; (e) density function of the
height of newborns; (f) population sizes observed during the study period (green indicates projected sizes using the IPM); (g) heat map of the
kernel (eqn. 10.7) as a function of individual size at two consecutive years. Graphs a–d and g present the models evaluated over the individual
state variable and at mean values of two environmental variables (relative altitude and time since fire).
small perennial herbaceous species endemic to central had to assume that newborns have the same size dis-
Florida, United States (Estill and Cruzan 2001). Flow- tribution irrespective of the size of their parents. The
ering occurs mostly between mid-summer and mid- graphs associated to these models, evaluated at the
autumn. Every year, starting in spring, established mean values of e and h, are presented in Figure 10.4a–e.
individuals may develop multiflowered reproductive We also include the kernel evaluated at these mean
branches that senesce in the winter. Aboveground values (Figure 10.4g) and the population sizes derived
individuals are killed by fire (Quintana-Ascencio and from the model (green line in Figure 10.4f). Although
Morales-Hernández 1997), but seeds in the soil can sur- seed dormancy has a significant role for H. cumulicola,
vive it (Quintana-Ascencio et al. 1998). H. cumulicola is in this implementation we ignore this state for heuristic
mostly limited to gaps between vegetation patches in purposes.
Florida rosemary scrub (Menges et al. 2017). The patch-
es occur at variable elevation, thus affecting H. cumuli-
cola demography probably due to differential water References
availability and community composition (Quintana-
Bates, D., Maechler, M., Bolker, B. and Walker, S., 2015. Fit-
Ascencio et al. 2018). We estimate the vital rates using
ting linear mixed-effects models using lme4. Journal of
a 21-year-long annual census database across 12 sites, Statistical Software, 67(19), pp. 1–48.
using as state variables an individual characteristic, Bolker, B. M., Brooks, M. E., Clark, C. J. et al., 2009. Gener-
height (z, cm), and two environmental factors—relative alized linear mixed models: a practical guide for ecology
elevation (e, m) and time since fire (h, years), measured and evolution. Trends in Ecology and Evolution, 24(3),
at the site level. We also consider the nonindependence pp. 127–135.
of data points due to the study year and the spatial Burnham, K. P. and Anderson, D. R. 2002. Model Selec-
arrangement of data points into sites, gaps, and the tion and Multimodel Inference: A Practical Information-
0.5×1 m study quadrats placed within each gap. Theoretic Approach, 2nd edition. Springer, New York.
Caswell, H. 2001. Matrix Population Models: Construction,
Given that we are dealing with more than one state
Analysis, and Interpretation, 2nd edition. Sinauer Asso-
variable, we have to construct a large set of models and
ciates, Sunderland, MA.
perform model selection. Due to space limit, we refer
Childs, D. Z., Rees, M., Rose, K. E., Grubb, P. J. and Ellner,
the reader to the Appendix of this chapter (please go S. P., 2003. Evolution of complex flowering strategies: an
to www.oup.com/companion/SalgueroGamelonDM), age–and size–structured integral projection model. Pro-
where, for each vital rate, we construct all pos- ceedings of the Royal Society of London. Series B: Biolog-
sible models and perform model selection. The ical Sciences, 270(1526), pp. 1829–1838.
(marginal) structures of the selected models for each Childs, D. Z., Rees, M., Rose, K. E., Grubb, P. J. and
vital rate were Ellner, S. P., 2004. Evolution of size–dependent flower-
ing in a variable environment: construction and analysis
of a stochastic integral projection model. Proceedings of
s (z, e, h) = logistic(βo + β1 · spline (z, e, h)),
the Royal Society of London. Series B: Biological Sciences,
−
{z′ −[β0 +β1 ·spline(z, e, h)]}2 271(1537), pp. 425–434.
1
G (z′ , z, e, h) = √ e 2σε2
, Clutton-Brock, T. H. and Pemberton, J. M. (eds), 2004. Soay
2 πσε Sheep: Dynamics and Selection in an Island Population. Cam-
pb (z, e, h) = logistic(βo + β1 · spline (z, e, h)), bridge University Press, Cambridge.
Coulson, T., 2012. Integral projections models, their con-
b (z, e, h) = exp(β0 + β1 · spline (z) struction and use in posing hypotheses in ecology. Oikos,
121(9), pp. 1337–1350.
+ β2 · e + β3 · h + β4 · e · h)and
Coulson, T., MacNulty, D. R., Stahler, D. R., Wayne,
′ −β )2
1 −
(z 0 R. K. and Smith, D. W., 2011. Modeling effects of
Q (z′ ) = √ e 2σε2
. (10.33) environmental change on wolf population dynamics,
2 πσε
trait evolution, and life history. Science, 334(6060),
pp. 1275–1278.
s(z, e, h), G(z′ , z, e, h), pb (z, e, h), and b(z, e, h) have more
De Kroon, H., Van Groenendael, J. and Ehrlén, J., 2000.
complex structures compared with eqn. 10.11 as they Elasticities: a review of methods and model limitations.
included a spline that relates nonlinearly the explana- Ecology, 81(3), pp. 607–618.
tory variables with the vital rates. Q(z′ ) in turn has a Easterling, M. R., Ellner, S. P. and Dixon, P. M., 2000. Size-
simpler structure, as we could not relate the size of specific sensitivity: applying a new structured population
the parent (z) with that of its descendant (z′ ) and thus model. Ecology, 81(3), pp. 694–708.
Ellner, S. P. and Rees, M., 2006. Integral projection mod- decisions in a stochastic, density-dependent environment.
els for species with complex demography. The American Proceedings of the National Academy of Sciences, 105(30),
Naturalist, 167(3), pp. 410–428. pp. 10466–10470.
Ellner, S. P. and Rees, M., 2007. Stochastic stable popu- Morris, W. F. and Doak, D. F. 2002. Quantitative Conservation
lation growth in integral projection models: theory and Biology. Sinauer Associates, Sunderland, MA.
application. Journal of Mathematical Biology, 54(2), pp. Ozgul, A., Coulson, T., Reynolds, A., Cameron, T. C. and
227–256. Benton, T. G., 2012. Population responses to perturba-
Ellner, S. P., Bailey, B. A., Bobashev, G. V., Gallant, A. R., tions: the importance of trait-based analysis illustrated
Grenfell, B. T. and Nychka, D. W., 1998. Noise and non- through a microcosm experiment. The American Natural-
linearity in measles epidemics: combining mechanistic ist, 179(5), pp. 582–594.
and statistical approaches to population modeling. The Quintana-Ascencio, P. F. and Morales-Hernández, M. 1997.
American Naturalist, 151(5), pp. 425–440. Fire-mediated effects of shrubs, lichens and herbs on the
Ellner, S. P., Childs, D. Z. and Rees, M., 2016. Data-Driven demography of Hypericum cumulicola in patchy Florida
Modelling of Structured Populations. Springer, Cham. scrub. Oecologia 112(2), pp. 267–271.
Estill, J. C. and Cruzan, M. B. 2001. Phytogeography of rare Quintana-Ascencio, P. F., Dolan, R. W. and Menges, E. S.
plant species endemic to the southeastern United States. 1998. Hypericum cumulicola demography in unoccupied
Castanea, 1, pp. 3–23. and occupied Florida scrub patches with different time-
González, E. J., Martorell, C. and Bolker, B. M., 2016. Inverse since-fire. Journal of Ecology 86(4), pp. 640–651.
estimation of integral projection model parameters using Quintana-Ascencio, P. F., Koontz, S. M., Smith, S. A., Sclater,
time series of population-level data. Methods in Ecology V. L., David, A. S. and Menges, E. S., 2018. Predicting
and Evolution, 7(2), pp. 147–156. landscape-level distribution and abundance: integrating
Griffith, A. B., 2017. Perturbation approaches for integral demography, fire, elevation and landscape habitat config-
projection models. Oikos, 126(12), pp. 1675–1686. uration. Journal of Ecology, 106(6), pp. 2395–2408.
Griffith, A. B., Salguero-Gómez, R., Merow, C. and McMa- R Core Team, 2020. R: A Language and Environment for Statis-
hon, S., 2016. Demography beyond the population. Jour- tical Computing. R Foundation for Statistical Computing,
nal of Ecology, 104(2), pp. 271–280. Vienna.
Hastie, T. J., 2017. Generalized additive models. In: Wood, Ramula, S., Rees, M. and Buckley, Y. M., 2009. Integral pro-
S. N. (ed.), Statistical Models in S (pp. 249–307). Routledge, jection models perform better for small demographic data
Abingdon-on-Thames. sets than matrix population models: a case study of two
Hegland, S. J., Jongejans, E. and Rydgren, K., 2010. Inves- perennial herbs. Journal of Applied Ecology, 46(5), pp.
tigating the interaction between ungulate grazing and 1048–1053.
resource effects on Vaccinium myrtillus populations with Rees, M. and Ellner, S. P., 2016. Evolving integral
integral projection models. Oecologia, 163(3), pp. 695– projection models: evolutionary demography meets
706. eco-evolutionary dynamics. Methods in Ecology and
Kuss, P., Rees, M., Ægisdóttir, H. H., Ellner, S. P. and Evolution, 7(2), pp. 157–170.
Stöcklin, J., 2008. Evolutionary demography of long-lived Rigby, R. A., Stasinopoulos, D. M., Heller, G. Z. and De
monocarpic perennials: a time-lagged integral projection Bastiani, F., 2017. Distributions for modelling location,
model. Journal of Ecology, 96(4), pp. 821–832. scale, and shape: using GAMLSS in R. www.gamlss.org
Mazerolle, M. J., 2017. AICcmodavg: model selection and (accessed 5 March 2018).
multimodel inference based on (Q)AIC(c). R package ver- Salguero-Gomez, R. and Plotkin, J. B. 2010. Matrix dimen-
sion 2, pp. 1–1. sions bias demographic inferences: implications for com-
Menges, E. S., Crate, S. J. H. and Quintana-Ascencio, P. F. parative plant demography. The American Naturalist,
2017. Dynamics of gaps, vegetation, and plant species 176(6), pp. 710–722.
with and without fire. American Journal of Botany 104, Smallegange, I. M., Caswell, H. Toorians, M. E. M. and De
pp. 1825–1836. Roos A. M. 2017. Mechanistic description of population
Merow, C., Dahlgren, J. P., Metcalf, C. J. E. et al., 2014a. dynamics using dynamic energy budget theory incorpo-
Advancing population ecology with integral projection rated into integral projection models. Methods in Ecology
models: a practical guide. Methods in Ecology and Evo- and Evolution, 8(2), pp. 146–154.
lution, 5(2), pp. 99–110. Van Tienderen, P. H. 1995. Life cycle trade-offs in matrix
Merow, C., Latimer, A. M., Wilson, A. M., McMahon, S. population models. Ecology, 76(8), pp. 2482–2489.
M., Rebelo, A. G. and Silander Jr, J. A., 2014b. On using Vrugt, J. A., Ter Braak, C. J. F., Diks, C. G. H., Robinson,
integral projection models to generate demographically B. A., Hyman, J. M. and Higdon, D., 2009. Accelerat-
driven predictions of species’ distributions: development ing Markov chain Monte Carlo simulation by differential
and validation using sparse data. Ecography, 37(12), pp. evolution with self-adaptive randomized subspace sam-
1167–1183. pling. International Journal of Nonlinear Sciences and
Metcalf, C. J. E., Rose, K. E., Childs, D. Z., Sheppard, A. W., Numerical Simulation, 10(3), pp. 273–290.
Grubb, P. J. and Rees, M., 2008. Evolution of flowering
Williams, J. L., Auge, H. and Maron, J. L., 2010. Testing Zambrano, J. and SalgueroGómez, R., 2014. Forest
hypotheses for exotic plant success: parallel experiments fragmentation alters the population dynamics of a
in the native and introduced ranges. Ecology, 91(5), pp. late-successional tropical tree. Biotropica, 46(5), pp.
1355–1366. 556–564.
Williams, J. L., Miller, T. E. and Ellner, S. P., 2012. Avoiding Zuidema, P. A., Jongejans, E., Chien, P. D., During, H.
unintentional eviction from integral projection models. J. and Schieving, F., 2010. Integral projection models
Ecology, 93(9), pp. 2008–2014. for trees: a new parameterization method and a val-
Wood, S. N., 2017. Generalized Additive Models: An Introduc- idation of model output. Journal of Ecology, 98(2),
tion with R. Chapman and Hall/CRC Press, Boca Raton, pp.345–355.
FL. Zuur, A. F., Hilbe, J. M. and Ieno, E. N., 2013. A Beginner’s
Wood, S. and Scheipl, F., 2017. gamm4: generalized addi- Guide to GLM and GLMM with R: A Frequentist and Bayesian
tive mixed models using ‘mgcv’ and ‘lme4’. R package Perspective for Ecologists. Highland Statistics, Newburgh.
version 0. pp. 2–5. https://cran.r-project.org/web/ pack-
ages/gamm4/index.html.
CHAPTER 11
Transient analyses of population

dynamics using matrix
projection models
David N. Koons, David T. Iles, and Iain Stott
11.1 Introduction value that does not change over time. Although they
are undoubtedly abstractions of nature, MPMs for
In this chapter we address the burgeoning topic of
constant environments are often used to characterise
transient population dynamics using matrix projection
the general (or average) properties of particular popu-
models (MPMs), which are introduced in Chapter 9
lations or life histories, and a well-developed suite of
and widely used for examining the dynamics of pop-
analytical methods are available to understand their
ulations structured by discrete life cycle stages (e.g.
dynamics (Caswell 2001; see Chapter 9). Because of
developmental phases, sex) or discretized continuous
their analytical tractability, models for constant envi-
life history traits (e.g. age, size; Caswell 2001). The
ronments have been the focus of most studies that
popularity of the MPM hails in part from its reliance
utilise MPMs (Boyce et al. 2006). In contrast, time-
on algebra and calculus, which are familiar to most
varying models allow vital rates to change over time as
quantitatively inclined ecologists given the early foun-
a result of density dependence, environmental covari-
dations built by pioneers such as Alfred Lotka (1924)
ates, unexplained stochastic variation, or a combina-
and Robert May (1981). Vindenes et al. provide a lucid
tion of variables (Caswell 2001). Time-varying models
overview of how to conduct asymptotic analyses of
introduce additional layers of realism but require more
MPMs in Chapter 9, with valuable explanation of ter-
sophisticated analytical tools, many of which are under
minology and extensions to a wide array of environ-
continued development.
mental conditions.
The vast majority of studies that use MPMs and
Integral projection models (IPMs) (Easterling et al.
IPMs for constant environments focus on the long-
2000), eloquently reviewed in Chapter 10 by González
term, asymptotic dynamics once stage structure has
et al., are continuously structured population models
stabilised. In reviewing these concepts, Vindenes et al.
in discrete time and akin to infinite-dimensional MPMs
in Chapter 9 introduce us to the ‘transient dynamics’
with a stage width of zero. Though transient anal-
or ‘transients’ that occur before asymptotic dynamics
yses are more mature for MPMs, the limit–property
are reached as a result of an initially nonstable stage
connection between discrete classes of an MPM and
structure. This prompts the question of how influential
continuous states of an IPM should allow many of the
a nonstable stage structure can be on transient pop-
analytical tools developed for MPMs to be applied to
ulation dynamics. The near-term dynamics resulting
IPMs, and when not, numerical simulation is often a
from an initial stage structure skewed toward individ-
sufficient replacement (Ellner et al. 2016).
uals with low reproductive value will be very different
It is useful to distinguish models for constant vs.
compared to one skewed toward individuals with high
time-varying environments; they have fundamentally
reproductive value: intuition alone tells us that the for-
different dynamics and require a different set of tools
mer will initially perform more poorly than the latter
to study. MPMs for constant environments assume
(see section 2.2 for verification).
that each vital rate is described by a single numerical
David N. Koons, David T. Iles, and Iain Stott, Transient analyses of population dynamics using matrix projection models.
Given the potential impact of nonstable stage struc- are referred to in the chapter, and we provide a concise
ture on population dynamics, analyses of transients overview of these tools in Table 11.1 to assist the reader.
are becoming popular in an era of global change
because of numerous situations that cause nonstable
stage structures. For example, stage structure can be
11.2 Examples
altered by sudden biotic forces (e.g. disease, novel pre- 11.2.1 Example life cycle models
dation pressure), abiotic forces (e.g. extreme weather
events, fire), and anthropogenic forces (e.g. harvesting, We herein illustrate analyses of transient population
manipulation of habitat) that quickly remove individ- dynamics using two published MPMs, but every-
uals of certain stages from the population (defined thing we demonstrate is readily applicable to pop-
here as a demographic ‘disturbance’). Alternative- ulation models for other types of organisms. We
ly, the vital rates of certain stages can change with downloaded vital-rate data for both examples from
less immediate impact on the number of individuals COM(P)ADRE (Salguero-Gómez et al. 2015, 2016) with
in each stage (defined here as a demographic ‘per- the assistance of the ShinyPop application (https://
turbation’). Demographic disturbances affect the cur- iainmstott.shinyapps.io/shinypop/). Our first exa-
rent stage structure, whilst demographic perturbations mple focuses on a study of the lady orchid (Orchis
change the stable stage structure: both mechanisms purpurea) in eastern Belgium. This study was conduct-
create a mismatch between vital rates and stage struc- ed by Jacquemyn et al. (2010), and we focus specifically
ture (remember the fundamental projection equation on their data for the third plot in a light environ-
from Chapter 9, nt+1 = Ant , and the IPM analogue in ment, which was one of six study plots and monitored
Chapter 10). over 6 years. When averaged over time and rounded,
Disturbances may be common: for example, the the stage-classified MPM for the lady orchid in light
leading edge of a plant or marine invertebrate invasion plot 3 is
or colonisation consists entirely of dispersing propag-  
ules that later progress through the life cycle, whereas 0 0 0 0 0 25.31
 0.02 0 
terrestrial animal and marine vertebrate colonisations  0 0 0 0 
 
originate from the (usually adult or subadult) stages  0 0.09 0 0 0 0 
Aorchid = 
most likely to disperse (Shigesada and Kawasaki 1997),  0 0 0.9 0.24 0 0 
 
and humans directly manipulate stage structure when-  0 0 0 0.76 0.41 0.29 
ever they relocate, reintroduce, or stock individuals 0 0 0 0 0.57 0.70
into a wild habitat. Human interventions often force (11.1)
prolonged perturbations, where populations are man-
aged to boost or suppress particular vital rates. Pro- The six discrete stages were based on developmen-
longed environmental perturbations may also result tal phases: protocorms, tubers, seedlings, juveniles,
from surpassing an environmental ‘tipping point’ or nonflowering plants, and flowering plants (Jacquemyn
shifts to alternative stable states of an ecosystem et al. 2010). After each time step, surviving individuals
(Jenouvrier et al. 2012). These are just a few examples of either advanced to the next stage (subdiagonal of eqn.
when transients might actually be of greater relevance 11.1), remained in a stage (diagonal), or, for some flow-
to ecologists than asymptotic dynamics (Hastings 2004; ering plants, reverted to a nonflowing plant (supradi-
Hastings et al. 2018). agonal). Because the asymptotic growth rate of Aorchid
Our goal in this chapter is to provide a gentle is λ1 = 1.02, we will use this population to illustrate how
introduction to the concepts and analysis of transi- transients might be used to further enhance population
ent dynamics using MPMs such that readers are pre- growth of this beautiful and locally abundant orchid
pared to dive into reviews on these topics (Ezard that is otherwise rare and threatened near the northern
et al. 2010; Stott et al. 2011; Stott 2016) and then extent of the species’ range (Jacquemyn et al. 2010).
apply analyses of transients to relevant questions and Our second example focuses on a study of polar
problems of their own. We begin with the analysis bears (Ursus maritimus) in the southern Beaufort Sea off
of transient dynamics in constant environments and the northern coast of Alaska, United States. This study
then highlight recent extensions to time-varying envi- was conducted by Hunter et al. (2010) and occurred
ronments, providing comparisons to asymptotic ana- over 5 years (though the study is ongoing). When
logues throughout. A large number of demographic averaged over time and rounded, the stage-classified
tools for analysing asymptotic and transient dynamics MPM for polar bears is
T R A N S I E N T A N A LYS E S O F P O P U L AT I O N DY N A M I C S U S I N G M AT R I X P R O J E C T I O N M O D E L S 199
Table 11.1 A concise overview of various ways to analyse MPMs that are discussed further within the text. When applicable, the key references
refer to seminal books or review papers, which contain citations of the original work that developed the theoretical basis for a given demographic
tool. To find applied examples of a tool, we highly recommend that readers use a search engine to find papers that have cited the original works or
our key references here.
Demographic tool Type of environment Interpretation Key references
General properties of model
Asymptotic analysis of Constant(time- Measures long-term population growth Caswell 2001 and references therein
deterministic model invariant) rate, (st)age structure, reproductive value,
generation time, and more when population
structure has stabilised
Transient analysis of Constant Measures properties of transient dynamics Ezard et al. 2010; Stott et al. 2011;
deterministic model (including amplification, attenuation, dura- Stott 2016; and references therein
tion, inertia, transient envelope, and other
transient behaviour of specific disturbances)
before population structure has stabilised
Stochastic analysis Time-varying(changing) Measures long-term stochastic growth rate Tuljapurkar 1990 and references
(λs ) and other demographic outcomes in therein
a time-varying environment once popula-
tion structure has reached a multivariate
stationary distribution
Transient analysis of time- Time-varying Measures realised population growth rates Gotelli and Ellison 2006; Hunter
varying model (λreal,t ), abundances, density dependence, et al. 2010; Schaub et al. 2013
population viability, and other demograph-
ic outcomes in any type of time-varying
environment without assuming stationary
conditions
Prospective perturbation analysis
Asymptotic elasticities for Constant Measure effects of proportional change Caswell 2001 and references therein
deterministic models in vital rates on asymptotic demographic
outcomes, most commonly λ1
Transient elasticities for Constant Measure effects of proportional change Stott 2016 and references therein
deterministic models in vital rates or components of initial
stage structure on transient demographic
outcomes
Stochastic elasticities Time-varying Measure effects of proportional change in Tulajpurkar et al. 2003; Haridas and
mean and variance of vital rates on λs Tuljapurkar 2005; Haridas et al. 2009
Transient elasticities for time- Time-varying Measure effects of proportional change Caswell 2007; Haridas et al. 2009
varying models in vital rates on transient demographic
outcomes in a time-varying environment
Retrospective perturbation analysis
Asymptotic LTRE with random Constant Measures the contribution of vital rates to Caswell 1989; Horvitz et al. 1997
design variation in λ1 among deterministic models
Stochastic LTRE Time-varying Measures the contribution of vital rates and Caswell 2010; Davison et al. 2010
stage structure to differences in λs between
or among different populations
Transient LTRE Time-varying Measures the contribution of vital rates and Koons et al. 2016; Maldonado-
stage structure to temporal variation (or Chaparro et al. 2017
other summaries) of λreal,t within a single
population
  is away from its stable stage structure, we projected

0 0 0 0 0 0.46
  the population dynamics of lady orchids over 10 years
 0.86 0 0 0 0 0 
  with all individuals initially in the first stage (i.e. n0 =
 0 0.86 0 0 0 0 
Abear =  [1, 0, . . . , 0]), representing a hypothetical colonisation
 0 0 0.86 0.50 0.38 0.90 
  wavefront beginning solely with protocorms. In this
 0 0 0 0.40 0.07 0 
0 0 0 0 0.44 0 case, the population experienced a phenomenon called
‘attenuation’ in which unfavourable short-term pop-
(11.2)
ulation growth rates led to lower densities (bottom
The six discrete stages were respectively classified solid line in Figure 11.1a) than a population with
as age 2, 3, and 4 subadults, adult females avail- equivalent vital rates but initiated from the stable
able to breed without dependent young, adult females stage structure (a.k.a. the ‘stable equivalent’; dashed
accompanied by one or more cubs of the year, and line in Figure 11.1a). In particular, this initial stage
adult females accompanied by one or more yearling structure yielded the largest possible transient attenu-
cubs. Equation 11.2 pertains to the female compo- ation (reduction) in population density, which likely
nent of the polar bear population and assumes an occurred because stage 1 individuals had the lowest
equal sex ratio among surviving offspring (see Hunter asymptotic reproductive value, but the formal link
et al. 2010 for details about the underlying lower- between the reproductive values associated with an
level vital rates that comprise each matrix entry). initial stage structure and resulting transient dynamics
Because the asymptotic growth rate of Abear is λ1 = remains to be proven. In contrast, when we initiated
0.94, we will use this population to illustrate how tran- a lady orchid population with 100% of the individ-
sients might be used to inform conservation strategies uals in the last stage (i.e. n0 = [0, . . . , 0, 1]) with
for declining or threatened and endangered species highest reproductive value, which could represent
(Caughley 1994). human introduction of ornamental flowering orchids,
the population experienced an opposite phenomenon:
favourable short-term population growth led to the
11.2.2 Transient analysis in constant
largest possible transient ‘amplification’ (increase) in
environments
population density relative to a stable equivalent (top
To demonstrate analyses of transient population dyna- solid line in Figure 11.1a).
mics in constant environments, we scaled matrices Upon closer inspection, there are interesting dynam-
Aorchid and Abear by their respective asymptotic pop- ics in these transient amplifications and attenuations
ulation growth rates (i.e. divided each matrix by its that several researchers have suggested as key indices
dominant eigenvalue). This results in λ1 = 1 for any for comparison across populations and species. Across
standardised matrix, thereby removing its long-term any of the possible initial stage structures, the maxi-
trend but preserving other asymptotic properties (e.g. mum amplification and attenuation after the first time
stable stage structure, elasticities). This facilitates com- step are respectively called the ‘reactivity’ (ρ̄1 ; Neubert
parison across populations and species and provides and Caswell 1997) and ‘first time-step attenuation’
for clear interpretation of any transient index of popu- (ρ1 ; Townley et al. 2007; Townley and Hodgson 2008).
lation density as a multiple of asymptotic density. With These instantaneous transient responses will always be
a scaled matrix, indices >1 indicate a transient popu- transient in nature but are useful for gauging imme-
lation density that is greater than asymptotic density diate population responses to demographic pertur-
produced by a stable stage structure, an index >0 but bations or disturbances. Limiting one’s evaluation of
<1 indicates a density less than asymptotic density, and transients to the first time interval, however, risks miss-
an index = 1 pertains to a population projected from its ing the full extent of transient dynamics, and there
stable stage structure (Stott et al. 2011). are other indices to consider. For example, the greatest
Using a standard eigen-analysis described in magnitudes of amplification and attenuation that are
Chapter 9, we found that for the standardised matrix possible at any time step, even if they are dissipated at a
Aorchid , the quantities of the stable stage structure later time, are called the ‘maximal amplification’ (ρ̄max )
(0.92, 0.02, 0.00, 0.00, 0.02, 0.04) and reproductive and ‘maximal attenuation’ (ρmin , maximal in attenu-
values (1, 46.24, 522.57, 590.62, 604.07, 638.01) were ated magnitude but minimal in numerical value; see
greatest for stages 1 and 6 respectively. To illustrate the Figure 11.1). Measurement of these maximal transient
transient behaviour that can occur when a population responses is effective for capturing the full extent
(a)
Lady orchid
_
ρ1
2
_ _
ρmax ρ∞
Transient envelope
0
log (Density)
–2
r∞
–4 r1 rmin
–6
0 1 2 3 4 5 6 7 8 9
(b)
0.4
Polar bear
0.2
log (Density)
0.0
–0.2
–0.4
0 1 2 3 4 5 6 7 8 9
Time
Figure 11.1 Standardised transient population dynamics, excluding the influence of asymptotic growth, for (a) the mean lady orchid MPM in a
constant environment. Bold lines indicate extreme amplifications and attenuations of transient densities on the log scale for respective initial stage
structures with 100% of the individuals in last stage (top bold line) or first stage (bottom bold line). Labelled points along the paths of transient
amplification and attenuation are as described in the main text, and these extremities encapsulate the transient envelope (grey areas), which
covers the range of possible transient dynamics for all other case-specific initial stage structures. The dashed line indicates a reference projection of
population dynamics from an initially stable stage structure. (b) Analogous projections of standardised transient population dynamics for the polar
bear example. Note the different behaviours of transient amplifications compared to attenuations over time.
of transient dynamics: Regardless of the initial stage thought of in this way, structured populations are mov-
structure, population density should always be small- ing objects with mass (structured density). Nonstable
er than maximal amplification and larger than max- stage structure exerts an inertia on abundance, causing
imal attenuation. These four near-term indices were it to move in directions other than that exerted by the
all originally defined as transient bounds (Neubert vital rates acting on their own (i.e. acting on a stable
and Caswell 1997; Townley et al. 2007; Townley and stage structure).
Hodgson 2008), but analogues also exist for projec- In an applied context, population inertia could help
tions from case-specific initial stage structures (i.e. any or hinder a manager’s ability to achieve near- and
credible structure where individuals are not stacked long-term population goals (Koons et al. 2006a, 2006b).
into a single stage that yields maximal transients). Consider the maximum bounds for amplified (ρ̄∞ ) and
Collectively, these capture the majority of variation attenuated inertia (ρ∞ ) for lady orchids in Figure 11.1a.
in near-term transient responses that occur across Since this population is predicted to grow slightly in
different initial conditions and are coined the ‘tran- the long run (λ1 = 1.02), a nonstable stage structure that
sient envelope’ (grey areas in Figure 11.1; Stott et al. produces a large amplified inertia like ρ̄∞ (resulting
2011). from an initial preponderance of flowering plants in
An important long-term outcome of an initially the population) would be a hindrance for managers
nonstable stage structure and ensuing transients is attempting to stabilise or reduce population density
that they affect asymptotic population density n∞ but but a benefit to those wishing to more rapidly enhance
not asymptotic population growth, which means that population density. A large attenuated inertia (result-
unlike λ1 , density and abundance are not ergodic, ing from an initial population consisting predominant-
meaning they are not independent of initial conditions ly of protocorms) would result in opposite outcomes
(see Caswell 2001, chapter 4; see also Stott et al. 2010b relevant to management goals.
for explanation of reducible vs. irreducible matrices The outcomes of transient dynamics and population
and implications for transients). A generalisation of inertia can nevertheless be very different across popu-
‘population momentum’ (Keyfitz 1971), population lations and species (e.g. Koons et al. 2005, 2006b; Stott
inertia I measures the long-term ratio of the densi- et al. 2010a; Gamelon et al. 2014; Iles et al. 2016). Take
ty of a population originating from any initial stage for example polar bears, for which entries of the sta-
structure relative to a reference initial stage structure, ble stage structure (0.05, 0.05, 0.04, 0.51, 0.23, 0.11) and
with the reference most often being the stable equiv- reproductive values (1, 1.10, 1.21, 1.33, 1.47, 1.76) asso-
alent (Tuljapurkar and Lee 1997; Koons et al. 2007). ciated with standardised matrix Abear were greatest for
Though population inertia is not a transient, it nicely stages 4 and 6. Though a somewhat different life his-
captures the net effect of transient dynamics on long- tory pattern compared to lady orchids, an initial stage
term population density. When the reference initial structure with 100% of the individuals in the first stage
stage structure equals the stable structure, population (n0 = [1, 0, . . . , 0]) also produced the greatest atten-
inertia is defined as uation in transient dynamics and long-term inertia in
density relative to a stable equivalent (bottom solid line
eT v*1 n0 w1 in Figure 11.1b). When we initiated a polar bear pop-
I= (11.3)
eT n0 ulation with 100% of the individuals in the last stage
(n0 = [0, . . . , 0, 1]), the population experienced maximal
where e is a vector with each entry equal to one, amplification in both transient dynamics and inertia
T denotes the transpose of a vector, * denotes the com- (top solid line in Figure 11.1b), again likely because the
plex conjugate transpose, n0 is as defined above, and final stage has the greatest reproductive value. Reloca-
w1 and v1 are the dominant right and left eigenvectors tions or colonisations by adults would therefore induce
of a projection model (Caswell 2001; Koons et al. 2007). amplified transient dynamics and population inertia.
In introductory physics courses we are taught about But it is nevertheless important to note the vastly dif-
the momentum/inertia of moving objects with mass, ferent magnitudes in standardised transients between
such as a maritime ship attempting to turn east at the lady orchid and polar bear examples that result
a fixed point from a northerly course. By the time from their different life histories (Figure 11.1). R code
the ship achieves a perfect easterly bearing, it will is provided in Appendix 11.1 (please go to www.oup
have drifted north of the intended turning point as a .com/companion/SalgueroGamelonDM) for building
result of the ship’s physical inertia. Although not often graphics like that in Figure 11.1.
11.2.3 Transient perturbation analysis survival of stage 4 adult polar bears without offspring
in constant environments (row 4, column 4 of eqn. 11.5) and their chance of sur-
viving to produce and care for cubs (row 5, column 4):
Section 11.2.2 examined the effects of disturbances
to stage structure, assuming vital rates remain  
0 0 0 0 0 0.04
unchanged. However, analysis of an MPM or IPM  0 
 0.04 0 0 0 0 
would be incomplete without also investigating the  
 0 0.04 0 0 0 0 
consequences of perturbations to the vital rates. Pros- EλA1bear = 
 0 0 0.04 0.26 0.09 0.10 
pective perturbation analysis (see Chapters 9 and 10)  
 0 0 0 0.23 0.02 0 
answers the question: ‘All else being equal, what
0 0 0 0 0.14 0
would happen to the population dynamics if one-
(11.5)
by-one, each life cycle vital rate (or entry of stage
structure) were changed by the same amount?’ These
which if increased, should induce the greatest propor-
are highly valuable tools for managers of natural
tional increase in λ1.
populations whose everyday business is ‘changing
A strict focus on perturbation analysis of the asymp-
demography’ to achieve population objectives. A per-
totic population growth rate could limit one’s eval-
turbation analysis can provide information about the
uation of population dynamics for informing actions
life cycle stages and vital rates that can be targeted in
to achieve near-term population objectives. Perturba-
management interventions to give maximum impact
tions of several transient indices could be considered
on the focal demographic outcome for minimum effort
for informing near- to moderate-term management of
(while remembering that one should also consider the
population density or abundance, but population iner-
logistical and financial realities of actually inducing
tia is a valuable omnibus measure because it essentially
changes in demographic parameters; Nichols and
combines the effects of transient dynamics over time,
Hines 2002; Baxter et al. 2006). Elasticities are prospec-
it correlates well with other transient indices, and it is
tive perturbation tools for comparing the demographic
also relevant to long-term abundance objectives (Stott
outcomes of proportional changes in demographic
et al. 2011). When initiating each of the example popu-
parameters (Caswell 2000, 2001), which can help
lations with 100% of the individuals in stage 1, we had
identify the changes in demographic parameters that
previously shown that this induces a maximal atten-
would be most effective at achieving desired demo-
uation of population inertia for lady orchids (Figure
graphic outcomes (e.g. a targeted change in population
11.1a), which could be problematic since the species
growth rate; see Heppell 1998).
is generally rare and threatened near the northern
A basic elasticity analysis of standardised Aorchid
extent of the species’ range (Jacquemyn et al. 2010). The
indicates that the asymptotic population growth rate
same initial stage structure for polar bears also induced
for lady orchids in a constant environment is, not sur-
a maximal attenuation in population inertia, which
prisingly, most sensitive to proportional changes in
would be problematic since the population is predicted
survival and stasis of individuals in the flowering plant
to decline by 6% per time step in the long run.
stage:
Using perturbation methods developed by Koons
  et al. (2007), we found that in terms of a focus on
0 0 0 0 0 0.02
 0  life-cycle vital rates, the attenuated inertia for lady
 0.02 0 0 0 0 
  orchids was most sensitive to proportional changes
 0 0.02 0 0 0 0 
EλA1orchid =  in survival and transition of nonflowering plants to
 0 0 0.02 0.01 0 0 
  becoming flowering plants and also the survival and
 0 0 0 0.02 0.13 0.16 
0 0 0 0 0.19 0.41 stasis of nonflowering plants:
(11.4)  
0 0 0 0 0 1.18
 0 
All else being equal, maintaining a large pool of flow-  1.28 0 0 0 0 
 
ering plants in the population therefore induces the  0 0.63 0 0 0 0 
EIAorchid = 
greatest proportional increase in the asymptotic pop-  0 0 0.25 −0.18 0 0 
 
ulation growth rate for lady orchids. The asymptotic  0 0 0 −0.55 −1.88 0.38 
population growth rate for standardised Abear , how- 0 0 0 0 −2.55 1.44
ever, was most sensitive to proportional changes in (11.6)
Note that these elasticities sum to zero, not one, and 0

that negative values imply that, quite counterintu-
itively (but we promise it to be true!), increasing the –1
matrix-level vital rate will decrease the numerical
value of population inertia. Contrary to elasticity –2
log (Density)
patterns for the asymptotic population growth rate
(eqn. 11.4), managers wishing to dampen attenuated –3
inertial effects of the initial stage structure on the
–4
eventual density of lady orchids should, all else being
equal, try to ‘decrease’ the survival of nonflowering –5
plants. This management action would reduce the
fraction of nonflowering plants in the population. But –6
we note that perpetually decreasing this vital rate will
also decrease λ1 and long-term population density, 0 2 4 6 8
which highlights the care that must be taken when Time
interpreting the transient and asymptotic impacts
of manipulating a species’ vital rates (Figure 11.2, Figure 11.2 Projections of standardised population dynamics,
light-grey line). Another management option might be excluding the influence of asymptotic growth, for the lady orchid MPM
to instead focus on increasing the survival and stasis in a constant environment (note that the nonstandardised MPM yields
population growth). The dashed line indicates a reference projection
of flowering plants. The elasticity of λ1 to change in
of population dynamics from an initially stable stage structure, and
this life cycle transition was highly positive (eqn. 11.4),
the solid black line indicates projection from a colonising stage
indicating that enhancements would significantly structure without any perturbations to the vital rates. The light-grey
increase λ1 . The elasticity of population inertia to line refers to projection from the colonising stage structure, with a
change in this vital rate was also positive (eqn. 11.6), permanent 10% decrement to the life cycle transition from stage 5 to
and thus an enhancement would also increase the 6 in an attempt to reduce attenuated inertial effects of the initial
numerical value of inertia and dampen the degree of stage structure, but with an inadvertent effect of decreasing long-term
attenuation caused by a colonising stage structure on population growth rate that also affects long-term abundance. The
orange line refers to projection from the colonising stage structure,
population density (Figure 11.2, orange line).
with a permanent 10% enhancement of survival and stasis of
The attenuated inertia for polar bears was most flowering plants within stage 6, which reduces the severity of
sensitive to proportional changes in the survival of attenuated population inertia and enhances the long-term population
2-year-old subadults (row 2, column 1 in eqn. 11.7): growth rate. Orchid image sampled from Hardy Perennials and
Old-fashioned Garden Flowers (Wood, 1884, public domain).
 
0 0 0 0 0 0.13
 0 
 1.10 0 0 0 0  attenuation in short-term density. For some popula-
 
 0 0.43 0 0 0 0  tions, such perturbations could drive them to the brink
EIAbear = 
 0 0 −0.37 −0.58 −0.15 0.29 
  of extinction, again highlighting the care that must be
 0 0 0 −0.55 −0.03 0  taken when interpreting the transient and asymptotic
0 0 0 0 −0.27 0 impacts of manipulating a species’ vital rates.
(11.7) Another option not presented here would be to
consider direct disturbances to stage structure by, for
This elasticity is positive, indicating that increasing
example, (re)introducing individuals in key stages of
the survival of young subadults would be effective
development (Koons et al. 2007). R code is provid-
at increasing the numerical value of inertia and thus
ed in Appendix 11.2 (please go to www.oup.com/
reducing its magnitude of attenuation. The inertia
companion/SalgueroGamelonDM) for conducting the
elasticity associated with survival of adults without
asymptotic and transient analyses presented above.
offspring (row 4, column 4 in eqn. 11.7) and their
chance of surviving to eventually produce and care for
11.2.4 Population dynamics in time-varying
cubs (row 5, column 4) are fairly negative, however
(eqn. 11.7), meaning that although an increase in these
environments
vital rates would have a strong positive effect on λ1 Though highly useful, metrics of population dynamics
(eqn. 11.5), it would also reduce the numerical value derived from a time-invariant MPM cannot describe
of population inertia and enhance the magnitude of the effects of environmental variability (Cohen
1977a; Tuljapurkar and Orzack 1980). MPMs can Figure 11.3a). A great deal of theoretical and applied
be /nobreakextended to accommodate the reality work focuses on aspects of the environment and
that temporal variation in vital rates persistently relationships with demographic parameters that affect
changes stage structure, nt+1 = At nt , and that both λs (e.g. Tuljapurkar 1982; Doak et al. 2005; Boyce
the vital rates and stage structure contribute to et al. 2006; Koons et al. 2009; Lawson et al. 2015; Iles
changes in realised population growth rates over et al. 2019), factors that contribute to differences in λs
time: λreal,t = ∥∥nnt+1t ∥∥ = ∥∥Ant nt ∥t ∥ . Most of the theoretical between life histories and populations (e.g. Koons et al.
and applied work on stochastic population dynam- 2008; Davison et al. 2019), and the sensitivity of λs to
ics concerns cases in which vital rates fluctuate with proportional changes in both the mean and temporal
stationary distributions, leading to stationary (but fluc- variance of vital rates and underlying environmental
tuating) distributions of stage structure (Figure 11.3a) drivers (e.g. Tuljapurkar et al. 2003; Lee 2017).
and realised population growth rate (Figure 11.3b). Lacking more appropriate methods for retrospective
The stochastic population growth rate (λs ) measures perturbation analysis in time-varying environments,
the long-term geometric average of the realised growth past studies have applied traditional, asymptotic life
rates once stage structure has reached a multivariate table response experiments (LTREs) with a ‘random
QT−1 1/T design’ to decompose vital-rate contributions to vari-
stationary distribution: λs = limT→∞ t=0 λreal,t ,
P T − 1 ance in the asymptotic λ1 associated with a set of
log λs = limT→∞ T1 t=0 log λreal,t (Lewontin and
deterministic MPMs, that is, the dominant eigenval-
Cohen 1969; Cohen 1976, 1977a, 1977b; Tuljapurkar
ues. Such analyses can be useful for comparisons of λ1
and Orzack 1980). Analogous to the asymptotically
across populations, such as spatial locations, but are
stable stage structure of a deterministic MPM for
not appropriate for retrospective analysis in a time-
constant environments, the stationary distribution
varying environment (Koons et al. 2016). Though tradi-
of fluctuating population structure in a time-varying
tional random-design LTREs have often been applied
environment means that the proportion of individuals
to time series of MPMs (reviewed by Bassar et al.
in each stage has a stable mean, variance, and other
2010; see also Table 11.1), an analysis of the dominant
moments over time (Cohen 1976, 1977a, 1977b) (see
(a)
0.0
0.8
Stage distributions
0.6
0.4
Figure 11.3 (a) Projections of polar bear
0.2 population dynamics over time in a
simulated stochastic environment,
0.0 whereby the population was initiated with
0 20 40 60 80 100 100% of the individuals in stage 1 (black
line in a), then projected over time with
(b) random selection of an MPM at each time
step from one of the five MPMs published
1.1 by Hunter et al. (2010). The distributions
of abundance in stages 2 through 6 are
1.0 shown in blue, orange, dark grey, firebrick
red, and light grey. Note that eventually
λreal
the means and variances of these

0.9
distributions stabilise to ‘stationary
distributions’. (b) As this happens, the
0.8 temporal mean and variance of the
realised population growth rates (λreal,t )
0.7 also stabilise, providing the theoretical
0 20 40 60 80 100 basis for computing λs , as described in the
Time (years) main text.
(a)
0.25
% Flowering
0.15
0.05
–0.05
0 1 2 3 4 5
(b)
5
3
λ
0
0 1 2 3 4 5
Time
Figure 11.4 Projections of realised lady orchid population dynamics (solid symbols) from an initial stage structure with 100% of the individuals
in stage 1 and the observed time series of MPMs (Jacquemyn et al. 2010; see www.oup.com/companion/SalgueroGamelonDM). (a) Realised
fraction of the population in the flowering stage (solid circles) compared to the asymptotically stable proportion associated with each individual
MPM (open circles). (b) Depicts the realised population growth rates (solid triangles) associated with the short-term projection of population
dynamics compared to the asymptotic population growth rates associated with each individual MPM (open triangles), which are markedly
different at most time steps. Orchid image sampled from Hardy Perennials and Old-fashioned Garden Flowers (Wood, 1884, public domain).
eigenvalues provides nonfactual inference about the quantify the demographic contributions to ‘differ-
demographic drivers of population dynamics in a ences’ in λs between populations (Davison et al.
time-varying environment because this essentially 2019). But stochastic LTREs do not quantify the demo-
assumes that an asymptotically stable stage structure graphic determinants of change in λreal,t over time
associated with the MPM at time t can transform into in a single population, which are of high interest
the stable structure associated with a new MPM at to managers and conservationists (Williams et al.
t+1 (Figure 11.4a, open circles). This simply does not 2002).
occur when stage structure is appropriately projected The dependence of λs on a stationary stage struc-
by time-varying MPMs in a recursive fashion: nt+1 = ture moreover assumes that temporal distributions of
At nt , (Figure 11.4a, closed circles), and the difference vital rates (and the environmental drivers thereof) are
between λreal and λ1 at each time step can be substantial stationary (e.g. Tuljapurkar 1990). Many contempo-
(Figure 11.4b). rary environments are not stationary. For example,
Relatively recently, LTREs have been developed changes in climate, land use, and hydrology are direc-
for the stochastic growth rate that appropriately tional, creating ‘nonstationary’ environments for much
accommodate the manner in which population dynam- of the world’s biota that are not likely to stabilise
ics operate in time-varying environments (Caswell any time soon (Wolkovich et al. 2014). Nonstation-
2010; Davison et al. 2010). These ‘stochastic LTREs’ ary environments are defined by trends in average
conditions, environmental variance, or both, which on realised population growth rates. These collective
keep populations in a perpetual transient state because developments of prospective and retrospective tools
they never have a chance to converge to a station- for examining transients in time-varying environments
ary stage structure (Hastings 2004; Wolkovich et al. have relinquished practitioners from the need to use
2014). Fortunately, the same MPM projection equation deterministic or stochastic perturbation analyses.
that is used to study λs (nt+1 = At nt ) can be used to To illustrate the utility of transient LTREs for gaining
study population dynamics in nonstationary environ- a better understanding of population dynamics amidst
ments (Gotelli and Ellison 2006). Alongside analyses of changing environmental conditions, we again use the
λs , Hunter et al. (2010) provide an elegant example of lady orchid and polar bear case studies. For consisten-
doing just this to project the impacts of forecasted cli- cy, we created scenarios with initial stage structures
mate change on polar bear population viability in the resembling colonisation of the youngest stage: n0 =
southern Beaufort Sea. [1, 0, . . . , 0]. For each population, we then projected the
initial stage structure forward in time using the empir-
11.2.5 Transient perturbation analysis in ically estimated sequence of MPMs over 6 or 5 years
time-varying environments respectively. At each time step of projection we kept
track of the normalised stage structure (i.e. the vector of
In many cases, the observed population dynamics relative stage-specific abundances that sum to one) and
will be transient because either the environment is the realised annual population growth rate, λreal,t . We
not stationary or too little time will have elapsed used this information to implement a transient LTRE
for environmental conditions and demographic stage to measure the contributions of past change in both
structure to have converged upon stationary distri- the vital rates and stage structure to temporal vari-
butions. Rather than assuming these properties and ance of the realised population growth rates var(λreal,t )
proceeding with a long-term stochastic demograph- (see Box 11.1). We also kept track of the asymptotic
ic analysis, one can instead focus on the abundances λ1 associated with the (nonstandardised) MPM at each
and realised population growth rates produced by an time step of study in order to compare results from the
observed sequence of time-varying environments and transient LTRE to the asymptotic LTRE with a random
corresponding MPMs (see Figure 11.4). Only recent- design. We should note that a minimum of 3 years (or
ly have perturbation analyses been developed for other time step) of data are needed to implement either
examining transients in time-varying environments. of these LTREs because one needs to be able to estimate
Caswell (2007) developed prospective sensitivity and the temporal process ‘variance’ of each demographic
elasticity tools for predicting the change in transient parameter.
demographic outcomes that would result from any The asymptotic LTRE with a random design for lady
specified change in the parameters that determine orchid indicated that variation in the rate of transition
those outcomes. into and stasis in the flowering stage contributed (C)
One outcome that is appealing for the analysis of most to the variance in λ1 :
transient dynamics in time-varying environments is  0 0 0 0 0 −0.0001

the realised population growth rate (λreal,t ) because
 0.0009 0 0 0 0 0

it accounts for the manner in which vital rates and var(λ1 )
CA orchid
=

0
0
0.0073
0
0
0.0002
0
−0.0004
0
0
0
0


stage structure jointly determine population growth in 0 0 0 0.0004 −0.0020 −0.0018
0 0 0 0 0.0023 0.0021
any type of environment, remembering that λreal,t =
∥nt+1 ∥ (11.8)
∥ nt ∥
= ∥∥Ant nt ∥t ∥ (Haridas et al. 2009). Koons et al. (2016)
developed a set of ‘retrospective’ transient LTREs that However, the sum of these contributions severe-
decompose an observed sequence of realised popu- ly underestimated the actual temporal variance of
lation growth rates into contributions from observed λreal,t (relative difference on log scale = −6.41; see
changes in vital rates and stage structure. Accordingly, Appendix 11.2 for calculation; please go to www.oup.
transient LTREs provide deep insight into the demo- com/companion/SalgueroGamelonDM). This discre-
graphic pathways through which populations change pancy arises because asymptotic LTREs with a ran-
in potentially unstable time-varying environments. dom design do not account for the highly unstable
Maldonado-Chaparro et al. (2017) extended the meth- initial stage structure, nor the subsequent fluctuations
ods of Koons et al. (2016) (1) to continuously structured in stage structure driven by temporal variation in the
IPMs, and (2) to measure explicit effects of past change vital rates. In contrast, while the first-order approxima-
in specific environmental variables (e.g. temperature) tion (see Box 11.1) of the transient LTRE was not perfect,
Box 11.1 The transient life-table response experiment for decomposing temporal variability in
realised population growth rates into contributions from vital rates and components of stage
structure.
To decompose var(λreal,t ) over time, Koons et al. (2016) Each term in the equation above is the contribution to
used methods derived from the definition of variance for a temporal variance in λreal,t from the covariance among each
linear combination of variables (see Chapter 10 of Caswell pair of parameters (which includes stage structure and vital
2001). Without affecting the measurement of λreal,t all rates). A measure of the contribution of variation in each θi to
nt should be normalised to sum to one in order to avoid var(λreal,t ) is obtained by focusing on θi specifically (i.e. drop-
issues with numerical computation. All components of nt ping the first summation) and summing over the covariances
and each parameter comprising the elements of the MPM with all other θj :
should be placed into a vector Θt , and then the sensitivity of X ∂λreal,t ∂λreal,t
λreal,t to change in each underlying demographic parameter χθi ≈ cov θi,t , θj,t θ̄ij
(∂λreal,t /∂θi,t ) should be calculated using either symbolic ∂θi,t ∂θj,t
j
calculus (which we demonstrate in Appendix 11.2; please
The magnitude of χθi may be small because λreal,t is insen-
go to www.oup.com/companion/SalgueroGamelonDM),
sitive to changes in θi,t , because θi exhibits little variability
numerical differentiation, or matrix calculus (see Caswell
over time, or because negative covariation with other θj nul-
2007). These sensitivities are then used along with covari-
lifies its direct effect. A transient analysis therefore reveals
ance among all elements of Θt to obtain the first-order
how strongly fluctuations in population growth rate have
approximation of temporal variation in λreal,t :
been driven by the direct effects of each vital rate, indirect
effects channelled through components of stage structure,
XX
var(λreal,t ) ≈ cov θi,t , θj,t and sums of χθi across joint parameters of interest. It is
i j nevertheless important to note that a decomposition of ‘vari-
ance’ in λreal,t would be most relevant to questions about
∂λreal,t ∂λreal,t
θ̄ij , population dynamics in stationary environmental conditions
∂θi,t ∂θj,t when one suspects that stage structure has not had enough
time to reach a stationary distribution (Maldonado-Chaparro
where the sensitivities are evaluated at the mean of Θt (as et al. 2017). Other properties of change in λreal,t over time
implied by the notation right of the vertical bar) across a would be more relevant in a nonstationary environment,
realised time series (note that italic theta refers to a single and Koons et al. (2016) provide several, somewhat more
element of Θt ). complicated, transient LTREs for such conditions.
it provided a much more accurate decomposition of structure, especially the proportion of the population
temporal variation in realised population growth rates in the flowering stage (last entry of far-right vector in
over the 6-year study (relative difference on the log eqn. 11.9). The greatest direct effect of any vital rate
scale was only 0.75): was that of fecundity (eqn. 11.9, upper-right corner
 0 0 0 0 0 0.19
 of matrix), thus indicating quite different parameters
 −0.01 0 0 0 0 0
 as being the dominant contributors to variation in
var(λ )
CAorchidreal,t =  0 −0.02 0 0 0 0 
 0 0 −0.08 0.15 0 0  population growth rates during the study compared to
0 0 0 −0.15 0.04 0.01
0 0 0 0 −0.05 −0.01
the asymptotic LTRE with a random design (eqn. 11.8).
 0.20
 Not only do these differing results hinder understand-
ing of past population dynamics, but also application
 0.94

+

0.27
0.08

 of inappropriate retrospective tools in time-varying
−0.67 environments (i.e. the asymptotic LTRE with a random
6.21
design) can impair near-term predictions about future
(11.9)
population dynamics that are informed by the past
The transient LTRE moreover revealed that the largest (see below).
contributions to variance in realised population The asymptotic LTRE with a random design for
growth rates were caused by fluctuations in stage polar bears indicated that variance in λ1 was primarily
driven by variation in the chance of adults without off- the key difference is that retrospective analyses com-
spring surviving and then producing and caring for pare the effects of ‘unequal changes’ in demographic
cubs (bold entry in eqn. 11.10): parameters on demographic outcomes observed in
the past. On their own, retrospective analyses do not
 
0 0 0 0 0 0.0018 tell us anything about what will happen in the future
 0.0010 0 0 0 0 0
 or what should be changed to most efficiently affect
Cvar(λ
Abear
1)
=

0
0
0.0010
0
0
0.0010
0
−0.0080
0
−0.0014
0
0.0018


0 0 0 0.0183 0.0015 0
future demographic outcomes (Caswell 2000). But
0 0 0 0 0.0060 0 when implemented in an iterative, adaptive manner,
(11.10) our response to the second question is that transient
LTREs (a retrospective analysis) can serve as extremely
Though the LTRE happened to identify the same
valuable tools for making predictions about future
matrix entry as being most influential on temporal vari-
population dynamics that are informed by the past.
ation in past population growth rates as the transient
These predictions can be gauged against future realisa-
LTRE (eqn. 11.11), the sum of the contributions overes-
tions and iteratively honed until practitioners become
timated the actual variance of λreal,t over time (relative
sound ecological forecasters in the nonstationary
difference on log scale = −0.14). In contrast, the tran-
world that we and other organisms currently live in
sient LTRE provided a highly accurate decomposition
(see Dietze 2017a, 2017b; Dietze et al. 2018).
of the demographic parameter contributions to tempo-
As is often the case (e.g. Gaillard et al. 2000, 2003),
ral variation in λreal,t (relative difference on the log scale
the inference gleaned from our transient LTRE case
was only −0.005):
studies was different than that attained from the
0 0 0 0 0 0.0004
 prospective elasticities (compare eqns. 11.9 and 11.11 to
 00 0 0 0 0 0
 corresponding prospective results in eqns. 11.4–11.7).
=  
var(λ ) 0.0034 0 0 0 0
CAbear real,t 0 0 0.0034 −0.0053 −0.0003 0.0003  This should not be surprising because (1) retrospective
0 0 0 0.0115 0.0003 0 perturbation analyses do not force the perturbations to

0
0
0
 0 0 0.0010 0
be equivalent across demographic parameters, and (2)
−0.0007
  the parameters with the greatest elasticity sometimes
+

−0.0007
−0.0002

 change the least in response to environmental change
0.0001
−0.0009
(i.e. are the most buffered; Boyce et al. 2006; Koons et al.
(11.11) 2009). When used carefully, the full arsenal of prospec-
With its full demographic accountability for tive and retrospective transient perturbation tools can
decomposing retrospective variation in realised pop- be useful in preparing and planning for the future over
ulation growth rates, the transient LTRE is more reli- near-term time-scales.
able for identifying the most influential demographic
parameters and is currently the preferred tool for retro-
spective analyses of transients in time-varying environ-
11.3 Conclusions
ments (Koons et al. 2016; Maldonado-Chaparro et al. We have provided a gentle introduction to the concepts
2017). R code is provided in Appendix 11.2 (please go and analysis of transient dynamics with MPMs in both
to www.oup.com/companion/SalgueroGamelonDM) constant and time-varying environments. The study
for conducting the LTRE analyses presented above of transients that arise from environmental distur-
and in Box 11.1. bances and perturbations in otherwise constant envi-
The collective results above remind us of two ronments has matured nicely since Caswell’s seminal
questions we are often asked by students: (1) Why do book (2001), and the reader should now be prepared
the results from retrospective perturbation analyses to dive into thorough reviews on these topics (e.g.
differ from those of a prospective analysis? and (2) Ezard et al. 2010; Stott et al. 2011; Stott 2016). But the
How can retrospective analyses be applied to conser- study of transients in time-varying environments is
vation and management? To address the first question, in its infancy, and there are ample opportunities to
prospective perturbation analyses compare the effects build theoretical foundations for ‘realised transients’ in
of ‘equivalent changes’ in demographic parameters today’s nonstationary world.
on demographic outcomes (e.g. population growth The large amount of work on transients of deter-
rate, population inertia). The derivatives used in ministic models provides a platform for developing
a prospective perturbation analysis also appear in needed theory in time-varying environments. There
retrospective perturbation analyses (see Box 11.1), but are notions of population inertia in slowly changing
environments (Tulajpurkar and Lee 1997), but ana- References

logues to transient bounds of a transient envelop have
Bassar RD, A López-Sepulcre, MR Walsh, MM Turcotte,
not been developed (see Figure 11.1 for a determin- M Torres-Mejia & DN Reznick. 2010. Bridging the
istic example). Fluctuating stage structure in a time- gap between ecology and evolution: integrating density
varying environment plays an important role in either regulation and life-history evolution. Annals of the New
buffering or amplifying the effects of vital rate vari- York Academy of Sciences 1206:17–34.
ation on population dynamics (Ellis and Crone 2013; Baxter PWJ, MA McCarthy, HP Possingham, PW Menkhorst
Gamelon et al. 2016; Koons et al. 2016; McDonald et al. & N McLean. 2006. Accounting for management costs in
2016). Continued efforts to determine how transients sensitivity analyses of matrix population models. Conser-
alter population sensitivity to ongoing environmental vation Biology 20:893–905.
Boyce MS, CV Haridas & CT Lee. 2006. Demography in
change are needed. Theoretical studies of what is pos-
an increasingly variable world. Trends in Ecology and
sible across life histories and various environmental
Evolution 21:141–148.
conditions would provide a beacon of light, much like Brown D & N Alexander. 1991. The analysis of the variance
the study of stochastic demography in stationary envi- and covariance of products. Biometrics 47:429–444.
ronments progressed from a practice of theoretical to Caswell H. 1989. The analysis of life table response exper-
empirical studies. iments. I. Decomposition of treatment effects on popula-
Perturbation analyses now exist for transients in tion growth rate. Ecological Modelling 46:221–237.
time-varying environments but have been applied to Caswell H. 2000. Prospective and retrospective perturba-
only a handful of studies (e.g. Freckleton et al. 2018; tion analyses: their roles in conservation biology. Ecology
Taylor et al. 2018; Layton-Matthews et al. 2019; Paquet 81:619–627.
Caswell H. 2001. Matrix Population Models: Construction,
et al. 2019). Transient LTREs are first-order approx-
Analysis and Interpretation, 2nd edition. Sinauer Asso-
imations to decomposing past change in realised
ciates, Sunderland, MA.
population growth rates that seem to work well Caswell H. 2007. Sensitivity analysis of transient population
for a variety of life histories and conditions (Koons dynamics. Ecology Letters 10:1–15.
et al. 2016; Maldonado-Chaparro et al. 2017), but we Caswell H. 2010. Life table response experiment analysis of
should not be surprised if they perform poorly in the stochastic growth rate. Journal of Ecology 98:324–333.
some situations. In rapidly changing environments Caughley G. 1994. Directions in conservation biology. Jour-
there is a need for more accurate solutions to transient nal of Animal Ecology 63:215–244.
LTREs that incorporate skewness and higher moments Cohen JE. 1976. Ergodicity of age structure in popula-
of demographic time series (Rice et al. 2011). These tions with Markovian vital rates, I: countable states.
Journal of the American Statistical Association 71:
solutions should be broadly applicable to any type
335–339.
of structuring variable (not just age structure; Brown
Cohen JE. 1977a. Ergodicity of age structure in populations
and Alexander 1991) and other types of demographic with Markovian vital rates, II: general states. Advances in
outcomes (e.g. lifetime reproductive success) and be Applied Probability 9:18–37.
relatively easy to implement like the current first-order Cohen JE. 1977b. Ergodicity of age structure in popula-
approximations. Fervent investment in the study of tions with Markovian vital rates, III: finite-state moments
transients in time-varying environments will rejuve- and growth rate; an illustration. Advances in Applied
nate an otherwise mature field of population biology, Probability 9:462–475.
and it will hone our abilities to forecast population Davison R, H Jacquemyn, D Adriaens, O Honnay, H
dynamics in an ever-changing world (Dietze 2017a, deKroon & S Tuljapurkar. 2010. Demographic effects of
2017b; Dietze et al. 2018). extreme weather events on a short-lived calcareous grass-
land species: stochastic life table response experiments.
Journal of Ecology 98:255–267.
Davison R, M Stadman & E Jongejans. 2019. Stochastic
Acknowledgements effects contribute to population fitness differences. Eco-
We thank A. Compagnoni and P. Capdevila for their logical Modelling 408:108760.
helpful comments on an earlier draft of the chapter. Dietze MC. 2017a. Ecological Forecasting. Princeton Univer-
sity Press, Princeton, NJ.
DNK was supported by the James C. Kennedy Wetland
Dietze MC. 2017b. Prediction in ecology: a first-principles
& Waterfowl Conservation endowment. We dedicate
framework. Ecological Applications 27:2048–2060.
this chapter to friend and colleague Markus Dyck,
Dietze MC, A Fox, LM Beck-Johnson, JL et al. 2018.
whose life was tragically snatched from this world Iterative near-term ecological forecasting: needs, opportu-
while doing what he did best, studying polar bears in nities, and challenges. Proceedings of the National Acade-
the Arctic. my of Sciences 115:1424–1432.
Doak DF, WF Morris, CP Fister, BE Kendall & EM growth: prospective and retrospective analyses. In: S
Bruna. 2005. Correctly estimating how environmental Tuljapurkar and H Caswell (eds). Structured-Population
stochasticity influences fitness and population growth. Models in Marine, Terrestrial and Freshwater Systems
The American Naturalist 166:E14–E21. (pp. 247–271). Chapman and Hall, New York.
Easterling MR, SP Ellner & PM Dixon. 2000. Size-specific Hunter CM, H Caswell, MC Runge, EV Regehr,
sensitivity: applying a new structured population model. SC Amstrup & I Stirling. 2010. Climate change threat-
Ecology 81:694–708. ens polar bear populations: a stochastic demographic
Ellis MM & EE Crone. 2013. The role of transient dynamics analysis. Ecology 91:2883–2897.
in stochastic population growth for nine perennial plants. Iles DT, R Salguero-Gómez, PB Adler & DN Koons.
Ecology 94:1681–1686. 2016. Linking transient dynamics and life history to
Ellner SP, DZ Childs & M Rees. 2016. Data-Driven Modelling biological invasion success. Journal of Ecology 104:
of Structured Populations. Springer, Cham. 399–408.
Ezard THG, JM Bullock, HJ Dalgleish, et al. 2010. Matrix Iles DT, RF Rockwell & DN Koons. 2019. Shifting vital
models for a changeable world: the importance of tran- rate correlations alter predicted population responses to
sient dynamics in population management. Journal of increasingly variable environments. The American Natu-
Applied Ecology 47:515–523. ralist 193:E57–E64.
Freckleton RP, HL Hicks, D Comont, et al. 2018. Mea- Jacquemyn H, R Brys & E Jongejans. 2010. Seed limitation
suring the effectiveness of management interventions restricts population growth in shaded populations of a
at regional scales by integrating ecological monitor- perennial woodland orchid. Ecology 91:119–129.
ing and modelling. Pest Management Science 74: Jenouvrier S, M Holland, J Stroeve, et al. 2012. Effects of
2287–2295. climate change on an emperor penguin population: anal-
Gaillard J-M & NG Yoccoz. 2003. Temporal variation in sur- ysis of coupled demographic and climate models. Global
vival of mammals: a case of environmental canalization? Change Biology 18:2756–2770.
Ecology 84:3294–3306. Keyfitz N. 1971. On the momentum of population growth.
Gaillard J-M, M Festa-Bianchet, NG Yoccoz, A Loi- Demography 8:71–80.
son & C Toigo 2000. Temporal variation in fitness Koons DN, JB Grand, B Zinner & RF Rockwell. 2005.
components and population dynamics of large herbi- Transient population dynamics: relations to life history
vores. Annual Review in Ecology and Systematics 31: and initial population state. Ecological Modelling 185:
367–393. 283–297.
Gamelon M, O Gimenez, E Baubet, T Coulson, S Tuljapurkar Koons DN, RF Rockwell & JB Grand. 2006a. Population
& J-M Gaillard. 2014. Influence of life-history tactics momentum: implications for wildlife management. Jour-
on transient dynamics: a comparative analysis across nal of Wildlife Management 70:19–26.
mammalian populations. The American Naturalist 184: Koons DN, JB Grand & JM Arnold. 2006b. Population
673–683. momentum across vertebrate life histories. Ecological
Gamelon M, J-M Gaillard, O Gimenez, T Coulson, S Tul- Modelling 197:418–430.
japurkar & E Baubet. 2016. Linking demographic respons- Koons DN, RR Holmes & JB Grand. 2007. Population inertia
es and life history tactics from longitudinal data in mam- and its sensitivity to changes in vital rates and population
mals. Oikos 125:395–404. structure. Ecology 88:2857–2867.
Gotelli NJ & AM Ellison. 2006. Forecasting extinction risk Koons DN, CJE Metcalf & S Tuljapurkar. 2008. Evolution
with nonstationary matrix models. Ecological Applica- of delayed reproduction in uncertain environments: a life
tions 16:51–61. history perspective. The American Naturalist 172:797–
Haridas CV & S Tuljapurkar. 2005. Elasticities in varying 805.
environments: properties and implications. The American Koons DN, S Pavard, A Baudisch & CJE Metcalf. 2009.
Naturalist 166:481–495. Is life-history buffering or lability adaptive in stochastic
Haridas CV, S Tuljapurkar & T Coulson. 2009. Estimat- environments? Oikos 118:972–980.
ing stochastic elasticities directly from longitudinal data. Koons DN, DT Iles, M Schaub & H Caswell. 2016. A life his-
Ecology Letters 12:806–812. tory perspective on the demographic drivers of structured
Hastings A. 2004. Transients: the key to long-term ecologi- population dynamics in changing environments. Ecology
cal understanding? Trends in Ecology and Evolution 19: Letters 19:1023–1031.
39–45. Lawson CR, Y Vindenes, L Bailey & M van de Pol. 2015.
Hastings A, KC Abbott, K Cuddington, et al. 2018. Transient Environmental variation and population responses to
phenomena in ecology. Science 361:eaat6412. global change. Ecology Letters 18:724–736.
Heppell S. 1998. Application of life-history theory and pop- Layton-Matthews K, MJJE Loonen, BB Hansen, CFD Coste,
ulation model analysis to turtle conservation. Copeia B-E Sæther & V Grøtan. 2019. Density-dependent
2:367–375. population dynamics of a high Arctic capital breed-
Horvitz C, DW Schemske & H Caswell. 1997. The rel- er, the barnacle goose. Journal of Animal Ecology 88:
ative ‘importance’ of life-history stages to population 2181191–1201.
Lee CT. 2017. Elasticity of population growth with respect Stott I. 2016. Perturbation analysis of transient population
to the intensity of biotic or abiotic driving factors. Ecology dynamics using matrix projection models. Methods in
98:1016–1025. Ecology and Evolution 7:666–678.
Lewontin RC & D Cohen. 1969. On population growth Stott I, M Franco, D Carslake, S Townley & DJ Hodgson.
in a randomly varying environment. Proceedings of the 2010a. Boom or bust? A comparative analysis of transient
National Academy of Sciences USA 62:1056–1060. population dynamics in plants. Journal of Ecology 98:
Lotka AJ. 1924. Elements of Physical Biology. Williams 302–311.
and Wilkins, Baltimore, MD. (Reprinted 1956 by Dover Stott I, S Townley, D Carslake & DJ Hodgson. 2010b.
Publications, New York as Elements of Mathematical On reducibility and ergodicity of population projection
Biology.) matrix models. Methods in Ecology and Evolution 1:
Maldonado-Chaparro AA, DT Blumstein, KB Armitage 242–252.
& DZ Childs. 2017. Transient LTRE analysis reveals Stott I, S Townley & DJ Hodgson. 2011. A framework
the demographic and trait-mediated processes for studying transient dynamics of population projection
that buffer population growth. Ecology Letters 21: matrix models. Ecology Letters 14:959–970.
1693–1703. Taylor LU, BK Woodworth, BK Sandercock & NT Wheel-
May RM. 1981. Theoretical Ecology, Principles and Applica- wright. 2018. Demographic drivers of collapse in an island
tions. Blackwell, Oxford. population of tree swallows. The Condor: Ornithological
McDonald JL, I Stott, S Townley & DJ Hodgson. 2016. Applications 120:828–841.
Transients drive the demographic dynamics of plant pop- Townley S & D Hodgson. 2008. Erratum et addendum:
ulations in variable environments. Journal of Ecology transient amplification and attenuation in stage-
104:306–314. structured population dynamics. Journal of Applied
Neubert MG & H Caswell. 1997. Alternatives to resilience Ecology 45:1836–1839.
for measuring the responses of ecological systems to per- Townley S, D Carslake, O Kellie-Smith, D McCarthy & D
turbations. Ecology 78:653–665. Hodgson. 2007. Predicting transient amplification in per-
Nichols JD & JE Hines. 2002. Approaches for the direct turbed ecological systems. Journal of Applied Ecology
estimation of λ, and demographic contributions to λ, 44:1243–1251.
using capture–recapture data. Journal of Applied Statis- Tuljapurkar S & SH Orzack. 1980. Population dynam-
tics 29:539–568. ics in variable environments I: long-run growth
Paquet M, D Arlt, J Knape, M Low, P Forslund & T Pärt. rates and extinction. Theoretical Population Biology
2019. Quantifying the links between land use and popu- 18:314–342.
lation growth rate in a declining farmland bird. Ecology Tuljapurkar S. 1982. Population dynamics in variable envir-
and Evolution 9:868–879. onments. II. Correlated environments, sensitivity analysis
Rice SH, A Papadopoulos & J Harting. 2011. Stochas- and dynamics. Theoretical Population Biology 21:
tic processes driving directional evolution. In: P. Pon- 114–140.
tarotti (ed.), Evolutionary Biology—Concepts, Biodiversi- Tuljapurkar S. 1990. Population Dynamics in Variable Envi-
ty, Macroevolution and Genome Evolution (pp. 21–33). ronments. Springer, New York.
Springer, Berlin. Tuljapurkar S & R Lee. 1997. Demographic uncertainty and
Salguero-Gómez R, OR Jones, CR Archer, et al. 2015. the stable equivalent population. Mathematical and Com-
The COMPADRE Plant Matrix Database: an open online puter Modelling 26:39–56.
repository for plant demography. Journal of Ecology Tuljapurkar S, CC Horvitz & JB Pascarella. 2003. The
103:202–218. many growth rates and elasticities of populations in
Salguero-Gómez R, OR Jones, CR Archer, et al. 2016. random environments. The American Naturalist 162:
COMADRE: a global data base of animal demography. 489–502.
Journal of Animal Ecology 85:371–384. Williams BK, JD Nichols & MJ Conroy 2002. Analysis and
Schaub M, H Jakober & W Stauber. 2013. Strong contribution Management of Animal Populations. Academic Press, San
of immigration to local population regulation: evidence Diego, CA.
from a migratory passerine. Ecology 94:1828–1838. Wolkovich EM, BI Cook, KK McLauchlan & TJ Davies. 2014.
Shigesada N & K Kawasaki. 1997. Biological Invasions: Temporal ecology in the Anthropocene. Ecology Letters
Theory and Practice. Oxford University Press, Oxford. 17:1365–1379.
CHAPTER 12
Individual-based models
Viktoriia Radchuk, Stephanie Kramer-Schadt, Uta Berger, Cédric Scherer,
Pia Backmann, and Volker Grimm
12.1 Introduction can be important, or even mandatory, to use even

though they require more effort than more aggregat-
Individual-based models (IBMs, also often referred to
ed models. We will then characterise the kind of data
as agent-based models, ABMs) represent a broad class
required for parameterising IBMs and present three
of simulation models that depict relevant processes
examples to demonstrate how IBMs look, work, and
at the individual level, thereby allowing higher-level
help answering research questions.
processes such as population and community dynam-
ics to emerge. They are used whenever there is reason
to believe that one or more of the following aspects 12.2 What are individual-based models
are critical to consider: intraspecific trait variation (e.g.
and how do we document them?
regarding age, size, or personality), local interactions
(in contrast to global interactions assumed in nonspa- IBMs are simulation models that iteratively apply the
tial models), heterogeneous environments, nonlinear rules describing individual-level processes in discrete
feedbacks, and adaptive behaviour (DeAngelis and time. For a long time the only precise and complete
Grimm 2014). documentation of an IBM was the computer program
In the 1990s, when computing power for running implementing it, but since IBMs are implemented in
IBMs became generally available, IBMs started to be all kinds of programming languages and on different
used in different disciplines dealing with autonomous computer types and operation systems, no standard
agents, for example ecology, geography, economics, way of communicating IBMs existed. This severely
and social sciences (Vincenot 2018). Nowadays, limited the use of IBMs as a scientific tool because
IBMs/ABMs are established tools in virtually every repeatability is a key element of the scientific method.
discipline investigating systems comprised of agents. To overcome this situation, a standard format for
Earlier differences between IBMs, which focused on describing IBMs, the ODD (overview, design concepts,
trait variation and local interactions, for example in and details) protocol (Grimm et al. 2006, 2010), was pro-
forest models, and ABMs, which focused on adaptive posed and is now widely used (Vincenot 2018). Here
behaviour, are fading away (Railsback and Grimm we use it to introduce seven main elements of an IBM
2019), so that both terms can be used interchangeably. in general. Examples of full ODD model descriptions
Demography is an explicit aspect of IBMs that deal can be found in the online supplements that are linked
with the full life cycle of their agents. Such IBMs distin- to our example IBMs described in Section 12.6.
guish among agents by differentiating their age, traits,
(1) Purpose and patterns: IBMs are, like any other mod-
and other variables. They thus complement more tra-
el, developed for a certain purpose. This purpose
ditional modelling approaches in demography and are
to a large degree determines which aspects of real-
particularly suitable for addressing the questions about
ity are included in an IBM. Additionally to the
the response of agent-based systems to some kind of
model purpose, a recent update of ODD (Grimm
change (e.g. in physical or social environment). In the
et al. 2020) requires listing the criteria that are
following we first briefly characterise all elements of
used to judge a model to be realistic enough for
an IBM. Then we discuss in more detail when IBMs
Viktoriia Radchuk et al., Individual-based models. In: Demographic Methods Across the Tree of Life.
its intended purpose. For applied problems, this The ODD protocol facilitates understanding what
can be a set of observed specific patterns (‘pattern- a model is and does. It should include the factual
oriented modelling’; Grimm and Railsback 2012); description of its elements and the rationales behind
for more theoretical models, it can be more generic the model’s assumptions and design. ODD is inde-
patterns, for example the very fact that a popula- pendent of scientific discipline, model purpose and
tion does not go extinct. complexity, and implementation details. Still, like any
(2) Entities, state variables, and scales: An IBM has verbal description, it can contain ambiguities that
entities, which are characterised by a set of state may prevent full replication. Therefore, ideally, ODDs
variables. Typical entities are agents (e.g. plants, would contain links to chunks of their corresponding
animals, humans), spatial units (e.g. grid cells computer codes; the rationale behind such interlinks is
of a certain size), and the overall environment. that at the level of a few code lines, most programming
Agents are often characterised by age, size, wealth, languages can be understood without being familiar
or behavioural types; spatial units by soil type, with them. In addition to ODD, the communication
moisture, plant cover, habitat suitability, or price; of IBMs is facilitated by diagrams based on Unified
the environment by temperature, precipitation, or Modelling Language (UML). UML diagrams are an
demand. IBMs have a certain temporal extent and approved ISO standard for the visualisation of process-
resolution, that is, length of the time step by which based software. Although they are not yet commonly
the model proceeds. Spatially explicit IBMs are used for IBMs, this promising tool may facilitate the
additionally characterised by a spatial extent and communication of IBMs in the future.
resolution (cell size).
(3) Process overview and scheduling: Processes consti-
tuting the core of the model are executed either
12.3 Why individual-based models?
each time step or depending on certain conditions. IBMs are indispensable to analyse and predict the
To understand how IBMs work it is important to development of small populations (as required, for
realise that most of these processes are not execut- example, in population viability analyses [PVA]), in
ed by the modeller (also called the ‘observer’ in which the fate of each individual is of particular impor-
individual-based modelling) but by the entities of tance and to a large degree affected by demograph-
the model: for example, individuals grow, repro- ic and environmental noise (Caughley 1994). It may
duce, and die. Thus, in the process overview or seem, however, that IBMs are less needed when study-
central schedule of an IBM, the question is: Who is ing large populations and communities. In such sys-
doing what, at what time, and in what sequence? tems, one would assume mean field approximations,
(4) Design concepts: IBMs are developed according to differential equations, or matrix model notations to be
important design concepts (Railsback 2001). The more suitable because the price of describing ‘averaged
most important ones are: Which behaviours of the individuals’ seems to be paid off by the tractability of
agents are imposed, and which emerge from their the models. However, this is not generally true. Matrix
own decision-making (‘emergence’; for details see models grouping individuals with similar character-
Section 12.3.1)? What do agents want to achieve istics into separate stages might be feasible in sim-
(‘objectives’)? What do agents sense and know ple cases, but even those models become analytically
about themselves and their environment (‘sens- intractable when the number of individual characteris-
ing’)? How do agents interact (‘interaction’)? How tics increases.
does the modeller observe the simulated system
(‘observation’)?
(5) Initialisation: With what parameter settings is the
12.3.1 Emergent behaviour vs. imposed
model initialised? demographic rates
(6) Input data: Are there any input data imported from Matrix and many other aggregated models are based
external files or other models which describe envi- on ‘demographic thinking’ (Grimm and Berger 2016)
ronmental variables that drive the dynamics of the where population growth is the result of demographic,
system? or vital, rates, that is, fecundity and mortality. If these
(7) Submodels: How, in detail, are the processes list- rates are known, for example from mark–recapture
ed in schedule (3) implemented in terms of algo- studies (see Chapter 13), they indeed allow us to pre-
rithms and equations? What are the parameters, dict population growth for the environmental condi-
their values, their units, and their uncertainty, and tions and population structure under which they were
where do they come from? measured. However, if the task is to predict how a
I N D I V I D UA L - B A S E D M O D E L S 215
population responds to changes in the environment explicitly code multilocus systems for each trait that is
and population structure, it can become critical to transmissible. Such an approach was used to investi-
let those demographic rates emerge from what the gate the effects of climate change on the microevolution
individual organisms are actually doing, that is, their of Atlantic salmon (Piou and Prévost 2012a, 2012b)
behaviour and life cycle (Radchuk et al. 2019). In and of stocking and barriers on the population persis-
the Anthropocene, characterised by multiple environ- tence of brown trout (Frank and Baret 2013). The level
mental factors changing at unprecedented rates, pre- of detail with which genes are mapped to the traits
dictions of response to such changes have become a or behaviours they code for depends on the specific
mandatory task for ecological modelling. question for which the model is developed.
Basing models on demographic rates means to In IBMs, density dependence is not imposed, for
impose the performance of individuals instead of letting example by making a vital rate depend on population
it emerge from their adaptive behaviour. ‘Full-fledged’ density, but emerges from the individuals compet-
IBMs therefore usually let one or a few key behaviours ing for resources. For example, Stephens et al. (2002)
of the individuals, for example foraging, emerge from showed that four alternative model versions, vary-
their adaptive decision-making, while processes to ing in how they implemented density dependence
which the population dynamics is less sensitive are still of alpine marmots, differed greatly in the resulting
imposed, for example some background mortality. population dynamics. All model versions, including a
The key element of such ‘next-generation’ IBMs matrix model (Chapter 9), resulted in similar popula-
(Grimm and Berger 2016), adaptive behaviour, requires tion dynamics around the equilibrium densities. How-
representing the trade-offs that individuals have to ever, only a behaviour-based IBM version, in which
consider (DeAngelis and Diaz 2019), for example the density dependence at the population level emerged
trade-off between predation and starvation risk for ani- from the lower-level individual interactions, could
mals or the allocation of resources to the shoots or roots capture the processes at lower population densities,
for plants. Adaptive behaviour ensures individual and including positive density effects.
hence species persistence in a changing world and
therefore is crucial to be accounted for in the modern
research on demography.
12.3.2 Local interactions and nonlinear
For example, including adaptive habitat selection in feedbacks
a fish population dynamics model allowed assessing Another key element of most IBMs is local interac-
the effects of multiple environmental factors on the tions, which often have important implications for the
population persistence (Harvey and Railsback 2007). demography of the focal species and the ones sharing
Results of such an assessment would have been biased resources with it. Such interactions are easy to capture
had the model failed to consider adaptive behaviour, in IBMs, but it is difficult to do so using mean field
because individual responses to each environmental approaches, especially when space is heterogeneous.
factor would have been imposed according to fixed Donalson and Nisbet (1999) investigated the effects of
phenomenologically measured relations, failing to con- relaxing two main assumptions of a classical Lotka–
sider that such relations may change as an organ- Volterra model: (1) a well-mixed population and (2) the
ism adapts to environmental changes. Such adaptive absence of demographic stochasticity. The results of
responses may occur via microevolution, phenotypic population persistence were strikingly different, even
plasticity, or both (Gienapp et al. 2008). Arguably, such in homogeneous environments.
eco-evolutionary dynamics is much easier to imple- Finally, the responses of individuals to environment
ment and interpret using IBMs compared to other and to individuals of con- and hetero-specifics are often
modelling approaches (e.g. Dunlop et al. 2009). nonlinear (Harvey and Railsback 2007). Such nonlin-
IBMs of eco-evolutionary dynamics differ widely earities at the individual level can further propagate to
with regard to how they depict evolutionary processes the population level, resulting in unexpected or coun-
in the model. Thus, some models directly hard-wire a terintuitive outcomes. This is especially the case when
mean phenotypic trait value that undergoes evolution individuals experience the impact of multiple factors,
due to mutation and drift, without explicit considera- as successfully demonstrated by an IBM applied to the
tion of the underlying genetic structure. For example, population dynamics of cutthroat trout subjected to
this is the way in which ecological traits are evolving in elevated turbidity, elevated stream temperature, and
the asexual version of the model developed by Dieck- reduced pool frequency (Harvey and Railsback 2007).
mann and Doebeli (1999) to investigate the potential of Similarly, an IBM of a freshwater detritivore revealed
sympatric speciation. At the same time, other models that multiple environmental stressors interacted in
their effects on population viability, but implications matrix models. They are harder to develop, param-
were much stronger for the ecosystem functioning eterise, and understand than more aggregated mod-
delivered by this detritivore (Galic et al. 2017). els. Understanding model behaviour, in particular,
requires considerable time and effort. Indeed, a com-
parison of resources needed to construct and analyse a
12.4 The pros and cons of IBMs
matrix model vs. an IBM used for the PVA of a butter-
Generally, IBMs are not bound to a particular model fly species (Radchuk 2012) showed that IBMs require
structure like some other model types (e.g. matrix mod- twice as much time for model development and anal-
els; Chapter 9) and instead are very flexible, allowing ysis and have much longer running time (Table 12.1).
a focus on the processes that are especially impor- Likewise, higher complexity implies more parameters
tant for answering a specific research question. For (Table 12.1) and hence more data to parameterise them.
example, if individual growth and resource allocation Such data are not always available, so that parameters
are essential for understanding competition among remain uncertain.
plants, these processes can be included in the model Careful exploration of the sensitivity of the model
(Backmann et al. 2019; see Section 12.6). Additionally to output to this uncertainty (‘parameter uncertainty’) is
the research question, the choice of an IBM’s structure one of the key aspects of individual-based modelling.
is determined by the availability of data to parame- Similarly, it is critical that modellers assess the sensitiv-
terise the required processes (Radchuk et al. 2016b). ity of choices made about the structure of the includ-
Thanks to their flexibility, IBMs are used for ed submodels (‘structural uncertainty’; Railsback and
understanding a wide range of phenomena and Grimm 2019). Given the complexity of IBMs, it is essen-
systems, including bird flocking (Hemelrijk and tial that all the choices made during iterative model
Hildenbrandt 2015), viability of small populations development and testing are well reported and docu-
(Beissinger and Westphal 1998), disease dynamics and mented (Grimm et al. 2014).
persistence (Lange and Thulke 2017), functioning of Together with sensitivity analysis, model valida-
socio-ecological systems (Schlüter et al. 2017), and the tion is another key tool in model analysis. Validation
structure and dynamics of plant (May et al. 2009) and in its strict sense reflects the ability of a model to
microbial communities (Hellweger et al. 2016) and reproduce the patterns in data that were not used for
even entire ecosystems (Fu et al. 2018). model development. Validation is thus the ultimate
However, this flexibility is also associated with costs. test of model credibility (Grimm et al. 2014). Impor-
IBMs usually have richer structure than, for example, tantly, in more aggregated models, such as matrix
Table 12.1 Comparison of the resources needed to construct and run two model types: a matrix model and an IBM. Both models were used to
assess the population viability of a vulnerable butterfly species, Boloria eunomia (the bog fritillary butterfly), in the south of Belgium. This table
is modified after Radchuk (2012).
Resources Matrix model IBM
Data analyses Capture–mark–recapture data: Capture–mark–recapture data:

12 days × 7 years 12 days × 7 years
Survival experiments: Survival experiments:
Eggs: 5 working days Eggs: 5 working days
Larva 1: 7 working days Larva 1: 7 working days
Larva 2: 12 working days Larva 2: 12 working days
Pupa: 10 working days Pupa: 10 working days
Fecundity experiments: 8 working days Fecundity experiments: 12 working days
Habitat mapping: 17 working days Habitat mapping: 17 working days
Model development and analysis 160 working days 380 working days
Running time for baseline 10 min 27 s 12 h 24 min 19 s
scenario (10 runs)
Number of parameters 21 + 800 of average and SD dispersal rates1 47 + 800 of average and SD dispersal rates1
Skills Knowledge of a programming language Knowledge of a programming language
1
Dispersal rates are drawn from the phenomenologically obtained empirical distributions, reflecting the probability of emigration in this 20-patch meta-population system.
models, validation is possible at the highest organisa- The following categories of species data can be dis-
tion level only—the population level—for example by tinguished: (1) individual-level data (Chapter 5) may
hindcasting the population sizes to the past. In con- include rules describing the energy budget of an organ-
trast, the hierarchical structure of IBMs that are based ism (e.g. consumption rates, allocation rates), vital
on individual-level processes from which the higher- rates, decision rules describing sociality (Chapter 3)
level phenomena are emerging allows us to conduct and territoriality if necessary, and rules for habitat
validation at several organisational levels, given the selection and mate choice; (2) movement and dis-
respective data are available. For example, if demo- persal data may be represented by emigration and
graphic rates or emigration rates are emerging from the immigration rates (in the case of aggregated dispersal
underlying individual behaviour, they can be used as description) or by detailed movement rules and param-
patterns for validation, in addition to the distribution eters (in the case of fine-grained movement descrip-
of the population sizes. tion); (3) genetic data (Chapter 1) may include muta-
To sum up, IBMs, like other model types, have their tion rate, heritability, and so on. If a model represents
pros and cons. They complement more aggregated more than one species, species data are needed for all
modelling approaches by capturing the mutual and species considered, in addition to the rules describing
inseparable micro–macro link between the behaviour the interactions of hetero-specific individuals.
of individuals and the structure and dynamics of the Environmental data (Chapter 6) can be categorised
systems they comprise. into two main classes: climate data (describing the
climatic variables that are essential to the viability of
the population) and landscape data represented by
12.5 Types of data used to build an IBM
maps of habitat availability and quality (if several
The data used to build an IBM for demography can be different resources are distinguished they should all
grouped into two main classes: species data and envi- be depicted by a map). Of course, only those envi-
ronmental data (Radchuk et al. 2016b) (Figure 12.1). ronmental data that either modulate the effect of the
Types of species data will vary depending on the habitat, for example soil moisture, on individuals, or
purpose of the model, the structure of the model, have direct effects on individual processes should be
and the detail at which processes are represented. considered.
Environmental data
Species data
Demographic Dispersal data Genetic data

data
Rules about Movement Mutation rate,

Biotic
sociality and rules, heritability
interactions Climate
territoriality, emigration
Predation, Weather
mate choice, rates, rules
parasitism, descriptors
energy budget about
facilitation
rules, vital rates settlement
Landscape
Space representation
Habitat
heterogeneity
Figure 12.1 Schematic representation of the types of data that can be used in demographic IBMs (modified after Radchuk et al. 2016b). Species
data are shown on the white background and can be categorised into three groups. Individuals of each species closely interact with their
environment, and therefore species data are ‘embedded’ in a block representing environmental data (shown on the grey background). We
distinguish three groups of environmental data. Examples of data for each group are given in italics. Environmental effects on species are shown
with dashed arrows. Not only different environmental aspects can affect one or several aspects of species viability (i.e. effects on demography,
dispersal, genetics), but also they may interact in their effects on species (depicted by dashed lines between aspects of environment and the arrows
going from these lines).
If environmental data are included, there should although by itself it is insufficient to reproduce the
also be rules and parameters defining the effect cycles observed in natural populations. At the same
of the environment (climate and/or habitat) on the time, empirical evidence shows that population cycles
individual-level processes (e.g. behaviour, resource are affected by intrinsic factors, such as dispersal
acquisition, metabolism, survival, fecundity, move- and sociality. The importance of interactions between
ment, and/or dispersal). For example, Meli et al. (2014) intrinsic and extrinsic factors was highlighted a long
included linear regressions parameterising the effect of time ago (Lidicker 1988), and recently a conceptual
the toxicant’s concentration on the survival and fecun- framework was proposed (Andreassen et al. 2013) that
dity of individuals as well as their avoidance of the cycles are driven by a combination of two intrinsic
polluted grid cells. (sociality and negative density-dependent dispersal)
Importantly, for the purpose of demographic IBMs, and one extrinsic (predation) factor.
not all the above-mentioned data types have to be The purpose of the vole–mustelid model present-
included in the model. As explained at the beginning, ed here (Radchuk et al. 2016a) was to assess which
the data demand is driven by the model structure, factors (or their combinations) lead to the popula-
which is dictated by the research question and the tion cycles characteristic of natural vole populations.
ecology of the species in consideration. Therefore, if Radchuk et al. (2016a) contrasted alternative mod-
the spatial aspect is considered unnecessary to answer el versions by ‘switching off/on’ submodels to sys-
the research question and to appropriately describe the tematically investigate their ability to reproduce a
demography of the studied species, the landscape data, set of observed patterns (‘pattern-oriented modelling’;
as well as dispersal data, may be left out. Grimm and Railsback 2012). An IBM was needed main-
ly because the vole life cycle, in particular aspects
related to sociality, could not be accurately depicted
12.6 Case studies with more aggregated models.
Below we briefly present three IBMs as examples of The modelled system imitates the experimental sys-
different aspects and uses of IBMs. These address pop- tem at Evenstad Research Station (Norway) from
ulation cycles in rodents, the persistence of wildlife which most of the parameters on root voles (Microtus
diseases in a given landscape, and the emergence oeconomus) were derived. The system consists of
of delayed chemical defence of plants against insect six habitat patches embedded into a hostile matrix
herbivores. The models are quite similar in terms of (Andreassen et al. 2013). Individual voles are char-
complexity, while much simpler but also much more acterised by age (weeks), stage (weanling: <3 weeks
complex IBMs exist. They are also all rooted in empir- old; subadult: 3–4 weeks old; and adult: >4 weeks),
ical patterns but still include a large proportion of sex (female or male), family (ID of the family an indi-
stylised representations of processes; entirely stylised vidual belongs to), maturity state (mature/not mature),
or highly detailed and realistic IBMs also exist. The and reproductive state for mature individuals (repro-
ODD model descriptions and the corresponding codes ductive/not reproductive). Model time step is 1 week.
for all three models are provided in Section 12.8 (Sup- The duration of each simulation is 1,020 weeks, cor-
plementary materials 1–5). Additionally, we upload- responding to 30 average generations across the prey
ed the descriptions and codes of all models to the and predator species. Since seasonality is important
CoMSES model library. Demography is implicit in all in the vole life cycle, we distinguish two seasons:
these models, since age structure is important in all summer (weeks 17–43) and winter (weeks 1–16 and
three cases. Moreover, sex differences and belonging 44–52).
to a certain family group are crucial in the models of The following processes are modelled: reproduc-
vole cycles and disease persistence. Taken together, tion, survival, dispersal, maturation, and predation
these individual-level differences generate differences (Figure 12.2). Two processes occur only in summer,
in population structure that have importants implica- namely reproduction and dispersal. The number of
tions for population dynamics. weanlings is positively affected by the number of
females in the patch (maximum 2), reflecting the
kinship effect. For dispersal, the emigration proba-
12.6.1 Vole–mustelid model bility is negatively density dependent and sex- and
Rodent population cycles have attracted much research stage-specific. Density dependence, as here, is still
attention. Predation, an extrinsic factor, was suggested often imposed in IBM but only when referring to spa-
as one of the main drivers behind population cycles, tial units that are small enough to assume that local
(a) Week 17 – 44 (b) Week 44

Summer Summer (c) Week 1 – 16 & 45 – 52
Winter
Survival Predation
Survival Dispersal
Survival
Ageing Dispersal
Ageing Reproduction
Reproduction Ageing
(d) Ageing Maturation Reproduction (e) Predation

1.0
Predator growth
7 15
Proportion of
that mature
0.8
individuals
rate, Rpred
Number of
wearings
5 10
0.6
4 (f)
0.4 3
5 Ageing
0.2 2
0.0 1 Aget = Aget–1+1
0 0
0 10 20 30 40 50 0 50 0 50 100 150 200 250
Week Home range Prey abundance, ind.
overlap, %
Dispersal
Number of prey to be
Males Females 250
0.5 0.10 Survival 200 Survival
killed, nkill
0.08
Male turnover
Emigration probability
0.4
Weekly survival rate
Adults
1.0 150
0.988 0.992 0.996

0.06
Winter survival
0.3 100
0.04
0.8
0.2 0.02 50
0 10 20 30 0 10 20 30
Abundance, ind. Abundance, ind. 0.6 0
0.4 0.3 0 50 100 150 200 250
0.4 Prey abundancy, ind.
Subadults
0.3 0.2
5 10 15 20 25
gs
ea s
Fe ls
W ale
Predator abundance, ind.

di ll
ua
in
in A
nl
m
Individuals in the
vid
0.2 0.1 0.01 1.01 2.51 5.51 patch

0 10 20 30 0 10 20 30 0.51 1.51 3.51 7.51
Abundance, ind. Abundance, ind.
Figure 12.2 Schematic representation of the processes included in the vole-mustelid model and their schedule (reproduced from Radchuk et al.
2016a). (a)–(c) Schedules of processes taking place during different times of the year. Processes taking place in the summer season are shown on a
white background, and those taking place in winter are shown on a grey background. (d)–(f) Details on parameterisation and functional relations
used to implement the modelled processes: maturation, dispersal, survival, and reproduction in summer (d), predation occurring in week 44 (e),
and ageing and survival in winter (f).
density indeed affects all involved individuals via local density is higher than the critical one. As for the
interactions. Each emigrant incurs dispersal mortality most important design concept, emergence, virtually
and ends up in one of the habitat patches that do not all behaviours of the individuals were imposed via
contain an adult of the same sex as its own. In winter, empirical rules or rates. For the purpose of the model,
survival is negatively density dependent according to explaining the emergence of population cycles in a
empirical findings (Aars and Ims 2002), and in sum- given environment, this was sufficient.
mer it is fixed irrespective of the stage, age, and sex Radchuk et al. (2016a) compared the population
of individuals. If the sociality submodel is activated, dynamics of empirically observed natural vole pop-
then survival probability of a female and her weanlings ulations with those obtained using four alternative
declines in case of a male turnover. models representing studied combinations of extrinsic
The predation by a specialist predator (i.e. weasels) and intrinsic factors. Four metrics (so-called patterns
is modelled using the predator–prey model commonly in pattern-oriented modelling: Grimm and Railsback
used for voles (see Turchin and Hanski 1997). Predation 2012) were used to describe the vole population cycles
is modelled once a year in week 44 and consists of and to contrast the models with real populations:
two steps. In the first step, predator population den- autumn population density, periodicity, amplitude,
sity is updated using the predator population growth and population growth rate. Only the full model
rate. In the second step, the number of prey to be that incorporated predation and two intrinsic factors
killed by predator is determined according to the reproduced all four patterns the closest to the empirical
type 2 functional response only if the vole population ones (Figure 12.3). Periods were too short if one of
(a) (b) (c) (d)

0.6 0.4 1.0
0.04
full
0.3 0.2 0.5
0.02
0.00 0.0 0.0 0.0
0.6 0.4 1.0
0.04
pred
0.3 0.2 0.5
Frequency
0.02
0.00 0.0 0.0 0.0
0.6 0.4 1.0
pred+disp
0.04
0.02 0.3 0.2 0.5
0.00 0.0 0.0 0.0
0.6 0.4 1.0
pred+soc
0.04
0.3 0.2 0.5
0.02
0.00 0.0 0.0 0.0
0 100 200 300 400 0 1 2 3 4 5 6 7 0 10 20 30 40 50 –5.0 –2.5 0.0 2.5 5.0
Population density Period (years) Amplitude In(yearly growth rate)
(ind. per ha)
Figure 12.3 Performance of the four models against the four patterns summarising the empirical vole population cycles: (a) autumn population
density, individuals (ind.) per ha; (b) cycle period; (c) amplitude; and (d) yearly population growth rates (natural-log transformed). Black lines
denote the means of the patterns observed in simulations under each model; grey dashed lines show the means of the empirically observed
patterns; and grey dotted lines show the 5 and 95 percentiles of the empirical distributions (except for (a), where dotted lines show the observed
range of mean autumn population abundances across studies and locations). Models reflecting the prevailing hypotheses about the drivers of vole
cycles are: full (predation and both intrinsic factors), pred (predation only), pred + disp (predation and dispersal), and pred + soc (predation and
sociality). Only the full model version yields the amplitude and period of the cycles closest to those characterising natural populations. (This figure is
adapted from Radchuk et al. (2016a))
the intrinsic factors was left out. Similarly, observed species (e.g. seasonal changes in birth and survival
population densities were too low for the models with rates) and the underlying landscape (Lloyd-Smith
predation only or for a combination of predation and et al. 2005). The consideration of such spatiotemporal
sociality. The model with predation only failed to heterogeneities in host densities and their effects on
capture the multimodal distribution of growth rates as disease persistence is facilitated by spatially explicit
observed in natural populations. IBMs that allow for a detailed representation of (1) host
This example IBM demonstrates the interplay of demography, (2) the transmission characteristics of the
empirical knowledge, data, and modelling. Direct pathogen, and (3) the underlying resource landscape.
observations to determine which mechanisms are Further, such models enable making quantitative pre-
responsible for the observed population cycles were dictions, which are important for disease management
virtually impossible. Still, each observed pattern con- (Eisinger and Thulke 2008).
tained information about the underlying mechanisms, The purpose of the host–pathogen model originally
and the IBM could be used to inversely ‘decode’ developed by Kramer-Schadt et al. (2009) is to under-
this information (Railsback and Grimm 2019). IBMs stand how the interaction between virus transmission
therefore are a powerful tool for inferring mechanisms characteristics, host demography, and landscape fea-
from observed patterns. tures determine the persistence of a disease in a given
landscape. The model is a combination of a spatial-
ly explicit, stochastic, individual-based demographic
12.6.2 Wild boar–classical swine fever (CSF)
model of wild boars (Sus scrofa L.) and an epidemi-
virus model ological model for the classical swine fever (CSF)
The ability of a pathogenic agent to successfully invade virus.
and avoid extinction—termed disease persistence— The model entities are individual wild boars and
depends on both the transmission characteristics of square spatial units or habitat cells. A wild boar is char-
the pathogen and the mixing and hence contact rates acterised by sex and age: piglet (<34 weeks), subadult
of hosts (Hagenaars et al. 2004). Heterogeneities in (between 34 weeks and <1 year for females and <2
host mixing, namely spatiotemporal differences in years for males), and adult. Further, it is described
host densities, arise from the demography of the host by an epidemiological status: susceptible, transiently
infected, lethally infected, and immune. Females that plant’s growth and, more generally, fitness, many
are at least subadult may be assigned as breeders species use induced defences, which are triggered by
according to the habitat quality of their cell. Each grid an actual herbivore attack. One might assume that
cell is characterised by habitat quality that determines plants should react as quickly as possible after the
its maximum group size. Each cell represents an area herbivore attack with the onset of chemical defence
of 2 km × 2 km encompassing an average home range to prevent tissue loss to the hebivores (Karban et al.
area of a wild boar group. All processes take place on 1999). However, a wide variety of delays is observed
a spatial grid of 100 × 25 cells. The model proceeds in in nature (Mathur et al. 2011). Backmann et al. (2019)
weekly time steps that correspond to the approximate hypothesised that, if the attacked plants grow in dense
CSF incubation time. cohorts with strong competition, it may be beneficial
The IBM’s demographic model mimics seasonal for a plant to delay its chemical response until the
reproduction, natal dispersal, and mortality of wild herbivore larvae are mobile enough to move away and
boars, while the epidemiological model describes the also to damage neighbouring plants, thus reducing
effects of virus on the wild boar population density the competition with the focal plant. Thus, delayed
and structure via virus-induced mortality and litter responses could reduce the competitive pressure on
size depression (Figure 12.4). The pathogen spread the focal plant (Figure 12.6).
from infected to susceptible individuals is modelled via The purpose of the model by Backmann et al. (2019)
locally density-dependent within- and between-group was to explore this hypothesis. The developed IBM
infection rates. Infected individuals either die after a depicted plant–herbivore interactions using as a model
given infectious period (mue) or gain lifelong immunity system interactions between the wild tobacco species
depending on the case fatality ratio (cfr, defined as pro- Nicotiana attenuata and its specialised invertebrate her-
portion of disease-induced deaths). The habitat quality bivore, caterpillars of the moth Manduca sexta. This is
of each cell is assigned randomly (Figure 12.5a). The one of the best-studied systems for induced defences,
virus is released in a random week of year 2 to a cer- with a large amount of data collected both in the
tain boar group, and simulations are run as long as any field and in the laboratory on the relevant aspects
infected individuals are present. (from physiology and chemical responses to ecological
Using a slightly adopted version of this model (Sciai- dynamics), allowing for model parameterisation.
ni et al. 2018); Scherer (2019) demonstrated that disease The IBM is based on an existing model of intraspecif-
persistence was more likely to occur in case of mod- ic plant competition (May et al. 2009) that was further
erate case fatality ratios and long-lasting infectious developed to include plant defences, fitness costs, and
periods (Figure 12.5b). Longer infectious period (e.g. detailed representation of the herbivore’s demogra-
mean of 9 weeks) led to higher number of infected indi- phy. The model’s entities are tobacco plants, larvae,
viduals (Figure 12.5c, A+B) compared to shorter infec- and grid cells. Plants are characterised by their posi-
tious periods (e.g. mean of 4 weeks; Figure 12.5c, C+D). tion, above- and belowground biomass, their defence
Moderate case fatality ratio allowed for a large popu- level, and the delay time (time between a larva attack-
lation of susceptible and immune individuals, thereby ing the plant and the plant starting defence produc-
favouring a sustained transmission chain and annual tion). Larvae are characterised by their biomass, age,
cycles of infection due to sufficient contacts and the and position. The model represents an area of 250 ×
birth of susceptible piglets (Figure 12.5c, A+C). In con- 250 square grid cells that are used to model zones
trast, high case fatality ratios caused a high probability of influence in a discretised way (May et al. 2009).
of pathogen extinction after the initial spread due to Time step is 4 hours; simulations were run for 27 days,
overexploitation of the host by disease-induced deaths corresponding to a caterpillar’s life cycle.
leading to stochastic fade-outs (Figure 12.5c, B+D). In each time step the following processes were
Insights obtained with such epidemiological IBMs are scheduled for the plants: growth, competition for
essential for the planning of diseasse prevention and resources, production of defence compounds if
control measures. attacked, and death as a result of being completely
eaten by a caterpillar. The underlying model of plant
competition is based on a two-layer zone of influ-
12.6.3 Wild tobacco–insect model (TIMELY) ence (ZOI) approach, in which one layer represents
Many plants can produce toxic chemicals to defend belowground competition of plants for nutrients
themselves against insect herbivores. Since producing and water (size-symmetric) and the second layer
these chemicals is costly and thus might reduce a represents aboveground competition for light mainly
initalisation
set week = week + 1
TRUE
week % 52 == 0 calculate survival probabilities
FALSE
assign breeding females
TRUE
week == week_Release pathogen release
FALSE
disease transmission
TRUE
week == 17 natal dispersal males
FALSE
TRUE
week == 29 assign dispersing females
FALSE
natal dispersal females
reprodution
mortality
Figure 12.4 Flowchart depicting main

ageing and processes and their timing in the wild
disease course boar—CSF virus model.
(a) (b) (c)

cfr = 0.3, mue = 9 cfr = 0.9, mue = 9
9
Infectious period (mue)

B
A
8 30,000
7
20,000
Number of individuals
6
5
10,000
4 C D
3 0
2
1 cfr = 0.3, mue = 4 cfr = 0.9, mue = 4
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
30,000
Case fatality ratio (cfr) 20,000 Health class:
Persistence pobability: susceptible
10,000 infected
immune
0
0.00
0.25
0.50
0.75
1.00
0 104 208 312 416 520 624 0 104 208 312 416 520 624
Week
Figure 12.5 (a) Heterogeneous resource landscape depicting the mean number of reproducing wild boar females per raster cell (i.e. the darker
the grey scale, the higher host density). (b) Effects of the infectious period (mue, in weeks) and case fatality ratio (cfr) on disease persistence and
(c) exemplary temporal dynamics of the three health classes (i.e. susceptible, infected, immune) for specific parameter combinations. (b) Disease
persistence was estimated using a 10-year threshold, that is, the disease was defined as ‘persistent’ only in those simulation runs in which any
infected individuals were present 10 years after the pathogen was released to the naive population. Probability of persistence for each parameter
combination was calculated as the proportion of runs that were classified as ‘persistent’ out of 200 simulation runs. Letters indicate the respective
parameter combination used in (c). (c) Number of individuals per health class (mean ± sd), estimated for each week of the simulation run (i.e. as
long as any infected individuals were present after the initial infection). One wild boar group was initially infected during the second year (indicated
by the vertical dotted line). The rugs on the x-axis visualise pathogen extinction events.
(size-asymmetric). The radii of ZOI were allometrically plants were assigned the same delay time or they were
scaled to the respective above- and belowground plant assigned random delay times. Plants were assigned to
biomass. When, after an insect attack, a plant is random positions. Larvae were placed on plants ran-
induced it starts investing a proportion (30%) of its domly, and several levels of larvae densities were test-
metabolic resources into defence production; the rest ed. The mean number of larval host switches peaked
is invested into growth and reproduction. at the intermediate time delay levels (3–5 days). Inter-
Larvae initially stayed on their host plant, where estingly, by using a genetic algorithm, where the suc-
they fed and grew. After having reached a certain cess of having a certain delay time in terms of the
age and weight, they became mobile enough to leave final biomass determined the proportion of this delay
the host plant if it initiated chemical defence. Follow- time in the following generation, the same delay times,
ing processes are modelled for larvae in each time 3–5 days, emerged as being dominant. These find-
step: growth, mortality, and host switching. Caterpil- ings demonstrate that in contrast to former assump-
lars grew exponentially and consumed plant mass pro- tions, under heavy competitive pressure, intermediate
portionally to their weight. When the larvae reached delay times are more favourable than an instanta-
their maximum size, they pupated and were inactive in neous defence reaction. Under high larval densities
the model. Caterpillar growth and instar progression longer delay times were favoured, whereas under low
were negatively affected by the plant defence com- densities the delay times decreased as the number of
pounds. Caterpillars’ per-time-step mortality scaled generations passed (Figure 12.7).
proportionally with their current host plant’s defence This model demonstrates several features of IBMs:
level and inversely with the caterpillar’s body mass. the reuse of an existing model, which indicates a
If the plant defence level reached a certain threshold, maturation of the IBM approach, as we increasing-
caterpillars could switch the host if they were heavy ly can base IBMs on existing building blocks instead
enough to move to another plant. The next host was of always having to start from scratch (Thiele and
chosen randomly among all plants within the move- Grimm 2015). Furthermore, IBMs can combine both
ment radius of the larva. Larvae switching the hosts realistic setting and highly stylised representations
incurred higher mortality. to explore ideas and test hypotheses. Moreover,
After testing the model by exploring the interaction IBMs can, in combination with genetic algorithms or
of two plants, 400 plants were simulated. Either all 400 neural networks, simulate the emergence of certain
competition
(a) (b) (c)
No response Immediate response Delayed response
predation Figure 12.6 Schematic representation of

the possible pathways in the TIMELY
model. Plants with induced chemical
defence are violet; plants without defence
are green. Three possible ways a plant can
react to the attack by a larva: (a) the plant
move
does not produce defence compounds and
as a result of herbivory it is dead; (b) the
plant reacts immediately by producing
defence compounds, which leads to the
death of the larva but also lowers final
biomass of the plant compared to its
competitor; (c) the plant reacts after a
smaller certain delay time, causing the larva to
move to the neighbouring plant and as a
dead result having higher final biomass.
Pathway (c) was the one that emerged in
simulations based on genetic algorithm.
(The figure is adapted from Backman et al.
(2019))
behaviours and how they depend on environmen-

10 tal conditions and the trade-offs individual organisms
9 should consider.
8
Delay time (days)
7
6 12.7 Conclusions
5
4 IBMs describing the demography of populations, that
3 is, considering age-dependent vital rates, were devel-
2 oped for a wide variety of species differing in their
1 ecology and life history (e.g. Stephens et al. 2002;
0 Radchuk 2012). It was suggested, however, that IBMs
are more beneficial for species with complex life his-
0 100 300 500
tories, for example social or territorial species (e.g.
Number of larvae
wild dogs: Gusset et al. 2009; root voles: Radchuk
Figure 12.7 Violin plots of the final delay times obtained under et al. 2016a; marmots: Stephens et al. 2002) or com-
different initial larvae number. The genetic algorithm started with plex movement behaviour (e.g. Eurasian lynx: Kramer-
randomly assigned delay times in the first generation and ran for 300 Schadt et al. 2004; lizard Tiliqua rugosa: Malishev et al.
generations. (The figure is adapted from Backman et al. (2019)) 2018; Fender’s blue butterfly: McIntire et al. 2007).
On the other hand, species with a comparatively sim- integral-projection models (Chapter 12) or matrix pro-
ple ecology can usually be modelled sufficiently well jection models. By doing so systematically across
with matrix projection models (see e.g. Caswell 2002; species with different ecologies and for models dif-
Chapter 9) as long as local interactions and adaptive fering in their purposes we would be able to derive
behaviour can be neglected. However, as we demon- a set of species characteristics and model purposes
strate in this chapter, this is rarely the case under for which one or the other model type may be a
natural conditions. Therefore, models that fail to con- better trade-off between the costs of model develop-
sider the consequences of these processes for popula- ment and the insight gained. However, we are still
tion dynamics may result in distorted projections of the behind in terms of such comparative analyses. Until
population dynamics (Meli et al. 2014), and the IBM then, given the necessary data are available, IBMs
approach is often beneficial for making demographic should be preferred if the system is characterised by
projections. nonlinear dynamics, when modelling the demogra-
In our examples of IBMs we focused mainly on ani- phy of small or declining populations, if inter- and
mal systems, but it does not mean that these tools intraspecific interactions in time and space are cru-
are used exclusively in zoology. In fact, IBMs inves- cial for capturing the ecology of the species, and if a
tigating demography were applied to diverse taxa: species is characterised by a non-negligible adaptive
plants (Berger and Hildenbrandt 2000; May et al. 2009), potential.
mammals (Stephens et al. 2002; Kramer-Schadt et al.
2004; Gusset et al. 2009; Radchuk et al. 2016a), birds
(Green et al. 1996; Penteriani et al. 2005), reptiles
(Malishev et al. 2018), microbes (Kreft et al. 2013; Banitz
12.8 Note on Supplementary Comparison
et al. 2015; Hellweger et al. 2016), fish (Harvey and
Railsback 2007; Piou and Provost 2012a, 2012b; Frank
Supplementary list
and Baret 2013), and arthropods (McIntire et al. 2007; The following supplementary materials for this
Radchuk 2012; Meli et al. 2014). chapter are available to download for free at the com-
IBMs are indispensable to properly depict the panion website:https://www.oup.com/companion/
demography of species characterised by complex inter- SalgueroGamelonDM
actions in time and space, to flexibly handle manage-
Supplementary Material 1. ODD of vole–mustelid
ment scenarios, or if asking eco-evolutionary ques-
model.
tions. IBMs are in general needed if a model is sup-
posed to predict the response of a population to new Supplementary Material 2. NetLogo code of vole–
conditions because they allow for emergence of demo- mustelid model.
graphic rates from the behaviour of individuals instead Supplementary Material 3. ODD of wild boar–CSF
of imposing them (Radchuk et al. 2019). These advan- virus model.
tages of IBMs, however, come with the costs associated
Supplementary Material 4. NetLogo code of wild
with the data required to parameterise them and time
boar–CSF virus model.
needed to develop and properly test such complex
models. These costs can be partially alleviated giv- Supplementary Material 5. ODD of TIMELY model.
en the increase in computer power in recent decades, Supplementary Material 6. NetLogo code of TIMELY
alternative ways to collect the data for model param- model.
eterisation (GPS tracking of many individuals, remote
sensing, next generation sequencing), and if the mod- NetLogo codes and ODDs of all models are
ules of already developed models would be reused uploaded to the CoMSES model library (https://
when developing a new one. Indeed, a large share of www.comses.net/) and can be found at the following
time that goes into model developing and testing can links: https://www.comses.net/codebases/4b484186
be saved if modellers would reuse the existing mod- -d8fb-4307-a710-fc05daa36afa/releases/1.0.0/ (vole–
el blocks developed for the same processes in species mustelid model),https://www.comses.net/codebases
with similar ecologies (Thiele and Grimm 2015). /82f2b53e-ae0e-4ac0-9e06-d541ecb4d1b6/releases/
Better understanding of the factors affecting the 1.0.0/(wild boar–CSF virus model), and https://www
demography of species can be achieved by con- .comses.net/codebases/a6ba9582-da40-4323-b4c9
trasting IBMs with other model types, for example -3a45def89db1/releases/1.0.0/(TIMELY model).
References combined effects of warming and toxicants. Global Change

Biology, 23, 2973–89.
Aars, J. & Ims, R.A. (2002) Intrinsic and climatic determi-
Gienapp, P., Teplitsky, C., Alho, J.S., Mills, J.A., & Merilä, J.
nants of population demography: the winter dynamics of
(2008) Climate change and evolution: disentangling envi-
tundra voles. Ecology, 83, 3449–56.
ronmental and genetic responses. Molecular Ecology, 17,
Andreassen, H.P., Glorvigen, P., Rémy, A., & Ims, R. A.
167–78.
(2013) New views on how population-intrinsic and
Green, R.E., Pienkowski, M.W., & Love, J.A. (1996) Long-
community-extrinsic processes interact during the vole
term viability of the re-introduced population of the
population cycles. Oikos, 122, 507–15.
white-tailed eagle Haliaeetus albicilla in Scotland. Journal
Backmann, P., Grimm, V., Jetschke, G., et al. (2019) Delayed
of Applied Ecology, 33, 357–68.
chemical defense: timely expulsion of herbivores can
Grimm, V. & Berger, U. (2016) Structural realism, emer-
reduce competition with neighboring plants. The American
gence, and predictions in next-generation ecological mod-
Naturalist, 193, 125–39.
elling: synthesis from a special issue. Ecological Modelling,
Banitz, T., Gras, A., & Ginovart, M. (2015) Individual-based
326, 177–87.
modeling of soil organic matter in NetLogo: transpar-
Grimm, V. & Railsback, S.F. (2012) Pattern-oriented mod-
ent, user-friendly, and open. Environmental Modelling and
elling: a ‘multi-scope’ for predictive systems ecology.
Software, 71, 39–45.
Philosophical Transactions of the Royal Society B: Biological
Beissinger, S.R. & Westphal, M.I. (1998) On the use of demo-
Sciences, 367, 298–310.
graphic models of population viability in endangered
Grimm, V., Berger, U., Bastiansen, F., et al. (2006) A standard
species management. Journal of Wildlife Management, 62,
protocol for describing individual-based and agent-based
821–41.
models. Ecological Modelling, 198, 115–26.
Berger, U. & Hildenbrandt, H. (2000) A new approach to spa-
Grimm, V., Berger, U., DeAngelis, D.L., Polhill, J.G., Giske, J.,
tially explicit modelling of forest dynamics: spacing, age-
& Railsback, S.F. (2010) The ODD protocol: a review and
ing and neighbourhood competition of mangrove trees.
first update. Ecological Modelling, 221, 2760–8.
Ecological Modelling, 132, 287–302.
Grimm, V., Augusiak, J., Focks, A., et al. (2014) Towards bet-
Caswell, H. (2002) Matrix Population Models: Construction,
ter modelling and decision support: documenting model
Analysis and Interpretation. Sinauer Associates, Sunder-
development, testing, and analysis using TRACE. Ecolog-
land, MA.
ical Modelling, 280, 129–39.
Caughley, G. (1994) Directions in conservation biology. Jour-
Grimm, V., Railsback, S.F., Vincenot, C.E., et al. (2020) The
nal of Animal Ecology, 63, 215–44.
ODD protocol for describing agent-based and other sim-
DeAngelis, D.L. & Diaz, S.G. (2019) Decision-making in
ulation models: a second update to improve clarity, repli-
agent-based modeling: a current review and future
cation, and structural realism. Journal of Artificial Societies
prospectus. Frontiers in Ecology and Evolution, 6, 237.
and Social Simulation, 23, 7.
DeAngelis, D.L. & Grimm, V. (2014) Individual-based
Gusset, M., Jakoby, O., Muller, M.S., Somers, M.J., Slo-
models in ecology after four decades. F1000Prime Reports,
tow, R., & Grimm, V. (2009) Dogs on the catwalk: mod-
6, 39.
elling re-introduction and translocation of endangered
Dieckmann, U. & Doebeli, M. (1999) On the origin of species
wild dogs in South Africa. Biological Conservation, 142,
by sympatric speciation. Nature, 400, 354–7.
2774–81.
Donalson, D.D. & Nisbet, R.M. (1999) Population dynam-
Hagenaars, T.J., Donnelly, C.A., & Ferguson, N.M. (2004)
ics and spatial scale: effects of system size on population
Spatial heterogeneity and the persistence of infectious
persistence. Ecology, 80, 2492–507.
diseases. Journal of Theoretical Biology, 229, 349–59.
Dunlop, E.S., Heino, M., & Dieckmann, U.L.F. (2009) Eco-
Harvey, B.C. & Railsback, S.F. (2007) Estimating multi-factor
genetic modeling of contemporary life-history evolution.
cumulative watershed effects on fish populations with an
Ecological Applications, 19, 1815–34.
individual-based model. Fisheries, 32, 292–8.
Eisinger, D. & Thulke, H.-H. (2008) Spatial pattern forma-
Hellweger, F.L., Van Sebille, E., Calfee, B.C., et al. (2016)
tion facilitates eradication of infectious diseases. Journal of
The role of ocean currents in the temperature selection of
Applied Ecology, 45, 1321–9.
plankton: insights from an individual-based model. PLOS
Frank, B.M. & Baret, P.V. (2013) Simulating brown trout
One, 11, e0167010.
demogenetics in a river/nursery brook system: the
Hemelrijk, C.K. & Hildenbrandt, H. (2015) Diffusion and
individual-based model DemGenTrout. Ecological Mod-
topological neighbours in flocks of starlings: relating a
elling, 248, 184–202.
model to empirical data. PLOS One, 10, e0126913.
Fu, C., Travers-Trolet, M., Velez, L., et al. (2018) Risky busi-
Karban, R., Agrawal, A.A., Thaler, J.S., & Adler, L.S. (1999)
ness: the combined effects of fishing and changes in prima-
Induced plant responses and information content about
ry productivity on fish communities. Ecological Modelling,
risk of herbivory. Trends in Ecology and Evolution, 14,
368, 265–76.
443–7.
Galic, N., Grimm, V., & Forbes, V.E. (2017) Impaired ecosys-
Kramer-Schadt, S., Revilla, E., Wiegand, T., &
tem process despite little effects on populations: modeling
Breitenmoser, U. (2004) Fragmented landscapes, road
mortality and patch connectivity: modelling influences population persistence and microevolution of Atlantic
on the dispersal of Eurasian lynx. Journal of Applied salmon. Global Change Biology, 19, 711–23.
Ecology, 41, 711–23. Radchuk, V. (2012) Dealing with Biological Complexity in
Kramer-Schadt, S., Fernández, N., Eisinger, D., Grimm, V., Population Viability Analysis: Lessons from Two Endan-
& Thulke, H.H. (2009) Individual variations in infectious- gered Butterfly Species. Univertsité Catholique de Louvain,
ness explain long-term disease persistence in wildlife Louvain-la-Neuve.
populations. Oikos, 118, 199–208. Radchuk, V., Ims, R.A., & Andreassen, H.P. (2016a) From
Kreft, J.U., Plugge, C.M., Grimm, V., et al. (2013) Mighty individuals to population cycles: the role of extrinsic
small: observing and modeling individual microbes and intrinsic factors in rodent populations. Ecology, 97,
becomes big science. Proceedings of the National Academy 720–32.
of Sciences, 110, 18027–8. Radchuk, V., Oppel, S., Groeneveld, J., Grimm, V., &
Lange, M. & Thulke, H.H. (2017) Elucidating transmission Schtickzelle, N. (2016b) Simple or complex: relative
parameters of African swine fever through wild boar car- impact of data availability and model purpose on the
casses by combining spatio-temporal notification data and choice of model types for population viability analyses.
agent-based modelling. Stochastic Environmental Research Ecological Modelling, 323, 87–95.
and Risk Assessment, 31, 379–91. Radchuk, V., Kramer-Schadt, S., & Grimm, V. (2019)
Lidicker, W.Z. (1988) Solving the enigma of Microtine cycles. Transferability of mechanistic ecological models is
Journal of Mammalogy, 69, 225–35. about emergence. Trends in Ecology & Evolution, 34,
Lloyd-Smith, J.O., Schreiber, S.J., Kopp, P.E., & Getz, W.M. 487–8.
(2005) Superspreading and the effect of individual varia- Railsback, S.F. (2001) Concepts from complex adaptive sys-
tion on disease emergence. Nature, 438, 355–9. tems as a framework for individual-based modelling.
Malishev, M., Bull, C.M., & Kearney, M.R. (2018) An Ecological Modelling, 139, 47–62.
individual-based model of ectotherm movement integrat- Railsback, S.F. & Grimm, V. (2019) Agent-Based and
ing metabolic and microclimatic constraints. Methods in Individual-Based Modelling: A Practical Introduction. Prince-
Ecology and Evolution, 9, 472–89. ton University Press, Princeton, NJ.
Mathur, V., Ganta, S., Raaijmakers, C.E., Reddy, A.S., Vet, Scherer, C. (2019) Infection on the Move: Individual Host Move-
L.E.M., & Van Dam, N.M. (2011) Temporal dynamics of ment Drives Disease Persistence in Structured Landscapes.
herbivore-induced responses in Brassica juncea and their University of Potsdam, Potsdam.
effect on generalist and specialist herbivores. Entomologia Schlüter, M., Baeza, A., Dressler, G., et al. (2017) A frame-
Experimentalis et Applicata, 139, 215–25. work for mapping and comparing behavioural theories in
May, F., Grimm, V., & Jeltsch, F. (2009) Reversed effects models of social-ecological systems. Ecological Economics,
of grazing on plant diversity: the role of below-ground 131, 21–35.
competition and size symmetry. Oikos, 118, 1830–43. Sciaini, M., Fritsch, M., Scherer, C., & Simpkins, C.E. (2018)
McIntire, E.J.B., Schultz, C.B., & Crone, E.E. (2007) Designing NLMR and landscapetools: an integrated environment for
a network for butterfly habitat restoration: where indi- simulating and modifying neutral landscape models in R.
viduals, populations and landscapes interact. Journal of Methods in Ecology and Evolution, 9, 2240–8.
Applied Ecology, 44, 725–36. Stephens, P.A., Frey-Roos, F., Arnold, W., & Sutherland, W.J.
Meli, M., Palmqvist, A., Forbes, V.E., Groeneveld, J., & (2002) Model complexity and population predictions. The
Grimm, V. (2014) Two pairs of eyes are better than one: alpine marmot as a case study. Journal of Animal Ecology,
combining individual-based and matrix models for eco- 71, 343–61.
logical risk assessment of chemicals. Ecological Modelling, Thiele, J. & Grimm, V. (2015) Replicating and breaking mod-
280, 40–52. els: good for you and good for ecology. Oikos, 124, 691–6.
Penteriani, V., Otalora, F., Sergio, F., & Ferrer, M. (2005) Turchin, P. & Hanski, I. (1997) An empirically based model
Environmental stochasticity in dispersal areas can explain for latitudinal gradient in vole population dynamics. The
the ‘mysterious’ disappearance of breeding populations. American Naturalist, 149, 842–74.
Proceedings of the Royal Society B: Biological Sciences, 272, Vincenot, C.E. (2018) How new concepts become univer-
1265–9. sal scientific approaches: insights from citation network
Piou, C. & Prévost, E. (2012a) A demo-genetic individual- analysis of agent-based complex systems science. Pro-
based model for Atlantic salmon populations: mod- ceedings of the Royal Society B: Biological Sciences, 285,
el structure, parameterization and sensitivity. Ecological 20172360.
Modelling, 231, 37–52.
Piou, C. & Prévost, E. (2012b) Contrasting effects of cli-
mate change in continental vs. oceanic environments on
CHAPTER 13
Survival analyses
Sarah Cubaynes, Simon Galas, Myriam Richaud, Ana Sanz Aguilar,
Roger Pradel, Giacomo Tavecchia, Fernando Colchero, Sebastien Roques,
Richard Shefferson, and Carlo Giovanni Camarda
13.1 Introduction eight case studies of survival analyses in lab organisms,

free-ranging animal and plant populations, and human
Fitness differences among individuals are the bedrock
populations. While it is relatively straightforward to
of ecological and evolutionary dynamics (Stearns
gather survival data and assess survival using simple
1992). Survival is without doubt one major component
models under controlled lab conditions (Klein 2016;
of fitness, which makes survival analyses a key tool for
see section 13.2), monitoring survival in free-ranging
demographers, ecologists, and evolutionary biologists
populations often requires more sophisticated capture–
(Metcalf and Pavard 2007).
mark–recapture (CMR) techniques to deal with imper-
Assessing survival is not always an easy task. This
fect detection of individuals (Williams et al. 2002; see
is because individuals die only once in a lifetime,
section 13.3). Indeed, survival data are often ‘incom-
therefore precluding repeated measurements on the
plete’ in free-ranging populations, timing and cause
same individual, and death is rarely directly observed.
of death can be hard to assess, and multiple environ-
However, a diversity of monitoring techniques exists
mental factors are at play in influencing survival (see
to gather survival data for species across the Tree
section 13.3; see also Chapters 4, 5, and 6). Humans are
of Life, from short-living lab organisms and plants
an exception, with the existence of several consequent
exhibiting dormancy, to long-living and elusive wild
databases with perfect knowledge of age and cause
vertebrates (see Chapters 4 and 5). Each monitoring
of death for several human populations, such as the
technique leads to a peculiar type of survival data,
Human Mortality Database (University of California,
and a plethora of methods exist to estimate survival,
Berkeley (USA), and Max Planck Institute for Demo-
from simple nonparametric estimators to more or
graphic Research (Germany) (2019). A great diversity
less complex semiparametric and fully parametric
of mortality models has been developed specifically to
models fitted in a continuous or in a discrete manner
analyse age at death data in human demography (e.g.
(reviewed in e.g. Wienke 2010; Miller 2011; Klein
Canudas-Romo et al. 2018; see section 13.4).
et al. 2016; Canudas-Romo et al. 2018; Cox 2018). The
choice of method is guided by the type of survival
data (e.g. collecting age at death or monitoring live
individuals with perfect or imperfect detection), the 13.2 What is survival analysis and why
species life history (e.g. single or numerous stages do we need it?
or ages), and the environment it experiences (e.g.
13.2.1 Time matters
controlled conditions vs. variable environments).
The aim of this chapter is not to provide another Survival analysis is used to analyse the time up to a spe-
extensive review of the existing techniques for sur- cific event that is going to occur. The event is generally
vival analyses, but rather to illustrate and contrast death, but it can be of another kind (e.g. reproduc-
the most commonly used methods to estimate sur- tion, migration, or exposure to a pathogen). Survival
vival across the Tree of Life. The focus is placed on data usually involve following a set of individuals over
Sarah Cubaynes et al., Survival analyses. In: Demographic Methods Across the Tree of Life.
a specific period of time and recording the time of In both cases, survival or mortality can be modelled
occurrence of the event of interest. using a nonparametric (see example in section 13.2),
semiparametric (see example in section 13.4), or fully
parametric (see examples in sections 13.3 and 13.4)
13.2.2 Censoring and missing data
model.
By nature, survival data have some particularities: sur-
vival cannot take a negative value, survival data are
often censored, and missing data are frequent. Right-
13.3 Survival analyses in the lab:
censoring occurs when an individual remains alive when it’s (almost) all under control
after the end of the study or drops out of the study,
13.3.1 Monitoring survival in the lab
for example due to permanent emigration (Klein et al.
2006). Left-censoring occurs when we cannot observe In the lab, gathering survival data usually involves
the time when the event occurred: for example, an counting the number of individuals alive at different
individual was already exposed to a pathogen before times or simply collecting the exact time at death. Col-
the study started (Klein et al. 2006). Other types of lecting survival data is generally easier than in the wild
missing data are common, because individuals cannot because individuals are available for monitoring at all
always be observed at all occasions (see Chapter 5 and times during the study, and the influence of external
section 13.3). We need survival analyses to avoid intro- variables is limited and often controlled. A common
ducing bias, because ordinary linear regression cannot objective is to analyse differences in survival functions
effectively handle the censoring of observations and between groups of individuals that present different
missing data (but see section 13.3). characteristics: for example various genotypes; groups
of individuals exposed to various treatments; and age,
sex, and life history differences.
13.2.3 Known fate data versus imperfect
detection
13.3.2 Kaplan–Meier estimator
In the presence of data referred to as ‘known fate’
data, survival can be modelled in a continuous man-
and log-rank tests
ner using continuous distributions (e.g. Gompertz, Laboratory experiments generate data sometimes
Weibull, Makeham, Siler). The survival function S(t) referred to as ‘known fate’ data, in which the prob-
is the probability that an individual survives up to a ability of detection is 1 and thus does not need to
certain time t. The hazard function h(t) represents the be accounted for when estimating survival. The sur-
instantaneous event rate for an individual who has vival function S(t) is usually estimated using the non-
already survived to time t. Both functions are relat- parametric Kaplan–Meier (KM) estimator (Kaplan and
ed: the hazard relates to the death rate, while sur- Meier 1958). For each time interval, it is calculated as
vival reflects its cumulative nonoccurrence. One might the number of individuals surviving divided by the
choose to model survival when the data involve counts number of individuals at risk. Censored individuals
of individuals alive at different points in time, while who have dropped out of the study are not counted in
mortality models can be preferred to model age at the denominator. The cumulative survival probability
death data. is calculated by multiplying probabilities of surviving
Alternatively, CMR models are used to estimate sur- from one interval to the next. A plot of the KM survival
vival when the data involve individuals missing at probability against time provides a useful summary
certain monitoring occasions and unknown time at that can be used to estimate parameters such as medi-
death (Lebreton et al. 1992; see Chapter 5). CMR mod- an survival time (Figure 13.1). The smaller the time
els involve estimating survival probability in between intervals, the smoother the survival curve.
consecutive occasions as a function of a set of covari- Graphically, a vertical gap between the survival
ates, while accounting for the imperfect detection of curves of different groups means that at a specific time
individuals at each occasion (e.g. a marked individual point one group has a greater fraction of individuals
might be present in the study area at a given occa- surviving, while a horizontal gap means that it took
sion but missed). Fully parametric or semiparametric longer for one group to experience a certain fraction of
approaches are available to model survival, transi- deaths (Figure 13.1). The log-rank test (Peto et al. 1977)
tion, or detection in CMR models (see examples in is often used to compare the survival distributions of
Section 13.3). two or more groups. It is a nonparametric test based
S U R V I VA L A N A LYS E S 231
Strata Male Female
1.00
0.75
0.50
0.25
p = 0.0013
Figure 13.1 Example of KM survival
curves for male and female patients with
0.00 advanced lung cancer (data from Loprinzi
et al. 1994). KM estimator (plain line) with
0 250 500 750 1000 95% confidence intervals (shaded area) is
Time provided for each group against time. The
Number at risk number of patients at risk at different
138 62 7 2 times is given for each group. Vertical lines
Strata
Male 20
represent censored individuals. The
Female 90 53 21 3 0
p-value of the log-rank test indicates a
0 250 500 750 1000 significant difference between male and
Time female patients’ survival rates.
on a chi-square statistic, which makes no assumptions group over another group is constant over time. Pro-
about the survival distributions. It assumes that the portionality tests are used to evaluate this assumption
groups have the same survival as the null hypothesis. (e.g. Miller 2011; see Supplementary Information S1
When the log-rank statistic is large, it is evidence for a for an example; please go to www.oup.com/comp
significant difference in the survival times between the anion/SalgueroGamelonDM). Models allowing for
groups. different assumptions, such as accelerated failure time
KM curves and log-rank tests are very useful in (AFT) models, are discussed for example in Kirkwood
assessing whether a categorial covariate (e.g. treatment and Sterne (2003) and Klein et al. (2016).
A vs. treatment B; males vs. females) affects survival.
However, it does not allow investigating the effects 13.3.4 Example of a stress assay in an
of multiple or continuous covariates (e.g. weight, age)
unconventional resistant organism: the
and to know how much more at risk one group is than
another.
tardigrade
Tardigrades hold their own phylum that stands in
between the phylum Arthropoda and Nematoda
13.3.3 Cox’s proportional hazards to adjust
(Figure 13.2). They are renowned organisms for their
for covariates ability to cope with the harshest environments, such
An alternative method is Cox’s proportional hazards as exposure to organic solvents, extreme tempera-
regression analysis (Cox 1972). It is a semiparametric tures (from −272 to 151◦ C), or high radiation doses,
method which can be used to assess simultaneously at any stage of their life (e.g. Jönsson et al. 2005).
the effect of several risk factors (both categorial and Most notably, some tardigrades have survived a 10-
continuous covariates) on survival time. It is analo- day flight in the vacuum of space and high-pressure
gous to a multiple regression in which the response equivalent to a depth of 180 km below the surface of
variable is the hazard measuring the instantaneous the earth (Ono et al. 2016). However, it seems that
rate of the event. It assumes a constant proportional the genetic ‘toolbox’ that ensures the uncommon resis-
hazard across groups over time; that is, the ratio of tance of these organisms to extreme stress is not shared
the risk of dying at a particular point in time in one by all tardigrade species. The resistance of tardigrade
13.4.1 Overview of CMR models

Principle
Within a typical CMR protocol designed to estimate
survival probabilities, individuals are sampled on dis-
crete occasions (often ≥ three occasions), at which they
may be detected or not. Data collected in the field are
then encoded into encounter histories organised by
individual or cohort (see details in White and Burn-
ham 1999). The simplest way to codify the encounter
histories is by using binary codes ‘1’ and ‘0’ specifying
if the individual has been detected or not. Howev-
er, when additional information reflecting the state
of the individual (e.g. breeder vs. nonbreeder, differ-
ent resighting sites, epidemiological states) is collected,
encounter histories may include additional codes (see
multistate and multi-event capture–recapture frame-
works below). Moreover, individual covariates can
also be recorded and included at the end of the indi-
Figure 13.2 Laser confocal image of an adult tardigrade Hypsibius vidual encounter histories either in order to categorise
exemplaris. (Photo M. Richaud.) individuals (e.g. males and females, age class, different
populations) or by indicating a particular quantita-
tive individual trait (e.g. size) (White and Burnham
1999).
species can be assessed and compared under controlled Models for the analysis of CMR data are classical-
conditions in the lab. ly based on multinomial distributions that describe on
In this example, we illustrate the use of the KM sur- the one hand the biological processes and on the oth-
vival curve and log-rank test to evaluate the effect of er hand the observational processes, conditional to the
the chemical stressor on tardigrades survival time. We biological ones. In the simplest case, the probabilities
use a Cox Proportional Hazard model to assess the involved are ϕi (probability of surviving the time inter-
effect of the chemical stressor while taking into account val i to i+1, and pi (probability to detect a live individ-
tardigrades age. The full procedure including data and ual at occasion i). More complex models (see sections
R script to run the analysis using the R package survival 13.3.4.2 and 13.3.4.3) actuate multinomial rather than
is provided in Supplementary Information S1. binomial biological processes and observational pro-
cesses, but the basic structure remains the same.
13.4 Survival analyses in the wild:
Single-state models
dealing with uncertainty and variable
The Cormack–Jolly–Seber (CJS) model was the
environments first CMR model (with an earlier formulation: the
In free-ranging animal and plant populations, CMR Jolly–Seber (JS) model [Jolly 1965; Seber 1965]) to
field methods are generally used to gather survival allow the estimation of demographic parameters
data (procedure described in Chapter 5). A pletho- under the assumption of an open population (i.e. open
ra of CMR models exist to estimate demographic to birth, immigration, death, and emigration, a.k.a.
parameters (including survival, access to reproduction, BIDE. models). Under the CJS approach, apparent
dispersal) while accounting for imperfect detection of survival (ϕi ) is the probability that an individual alive
individuals (e.g. Lebreton et al. 1992; Williams et al. at time i will be alive at time i+1, while resighting
2002; Schaub et al. 2004; Pradel et al. 2005). (pi ) is the probability that an individual alive and
Hereafter, we introduce the principle of CMR anal- present just before time i is seen (and marked) on that
ysis and the main types of CMR models. We then occasion (Lebreton et al. 1992). Apparent survival is
provide an overview of their use in animal and plant generally not referred to simply as ‘survival’ because
demography with step-by-step study cases. it may be confounded by permanent emigration of
marked individuals out of the study site and by monitoring occasion i survives to monitoring occasion
long-term vegetative dormancy in plants. Further, i+1, irrespective of the state of the individual at the
apparent survival is the probability to survive during latter occasion. State-transition (Ψ i jk ) is the probabil-
the interval between two monitoring occasions and ity, conditional on survival, of changing or moving
thus not at a specific monitoring occasion. In contrast, from state j at time i to state k at time i+1, assuming
resighting is an estimator of detection of previously that the individual survived the interval between the
seen individuals during the monitoring occasion. two occasions. Additionally, multistate models esti-
CMR models rely on several assumptions, the most mate resighting (pi j ), defined as the probability that an
important of which are that marked individuals are individual alive and in state j at monitoring occasion i
independent, tags are not lost, and the past histo- is also observed in that occasion.
ry does not influence the fate of the individual (i.e. This multisite–multistate formulation allows to
no trap response or negative effect of capture on address questions on survival and movement proba-
survival). Pollock et al. (1990) developed a series of bilities in metapopulation systems (e.g. Balkiz et al.
goodness-of-fit tests based on contingency tables for 2010), but it also provides a suitable framework for the
the CJS model to assess the validity of the assump- study of between-state transitions to study evolution-
tions. Later, directional tests for the detection of spe- ary trade-offs (Nichols and Kendall 1995) and recruit-
cific effects were derived (trap-dependence, Pradel ment probability (e.g. Jenouvrier et al. 2008) or to mix
1993; transients, Pradel et al. 1997). The availability of information of different types, that is, recoveries and
informative goodness-of-fit tests makes the CJS model recaptures (e.g. Lebreton et al. 1999). The multisite–
a common ‘umbrella’ model in model selection proce- multistate and robust design models have also been
dure (Lebreton et al. 1992). The CJS model can be easi- successfully applied to model dispersal to unobserv-
ly expanded to age-dependent parameters when new able sites or transition to unobservable states—that is,
animals are released at different ages or in multiple places or states through which animals can move but
cohorts, or reduced assuming parameters are constant in which they cannot be seen—or unobservable dor-
over time (examples in Lebreton et al. 1992). mancy states in plants (Kendall and Nichols 2002). In
The CJS model, and its extensions, have been, and the unobservable site/state, probability of detection is
still are, extensively used in the ecological literature, fixed to zero, and for this reason they are often referred
as they provide a suitable analytical framework to to as ‘ghost’ sites/states (Jenouvrier et al. 2008; Balk-
address multiple questions that tackle variability in iz et al. 2010). Grosbois and Tavecchia (2003) applied
survival over time (Lebreton et al. 1992), with external this idea to unobservable transitions. They consid-
covariates (e.g. Grobois et al. 2008), changes in recruit- ered the probability of dispersal as a two-step process,
ment probabilities (e.g. Pradel and Lebreton 1999), one accounting for the probability of leaving a given
recapture processes (e.g. Sanz-Aguilar et al. 2010), or site and a second, conditional to this, incorporating
evolutionary trade-offs (Tavecchia et al. 2001). How- the probability of settling into a new site. A similar
ever, the CJS model is based on capture–recapture approach was used by Schaub and Pradel (2004) to
data obtained from a single population and cannot estimate the relative importance of different causes
explicitly frame observations of animals moving of death. However, multisite–multistate models have
across multiple sites or between relevant biological an important limitation: they assume that the state or
states. the site in which an individual is observed is always
certain.
Multi-state models
Mutli-state models extend the CJS model by making Multi-event models
the parameters state-specific (Arnason 1973; Schwartz State uncertainty is a general problem in CMR models,
1993). Thus, apparent survival (ϕi jk ) is now the prob- but it might be particularly relevant in some studies.
ability that an individual alive in state j on monitor- For example, the study of the evolutionary trade-off
ing occasion i survives to occasion i+1 and, during between survival and reproduction relies on the fact
the latter occasion, transits to state k. This parameter that the breeding state of the individuals observed
may be decomposed into two parameters unseen in is always determined correctly. This can be true in
the CJS model: state-specific survival (Si j ) and state- most cases, but sometimes it can be difficult to deter-
transition (Ψ i jk ). Here, state-specific survival (Si j ) is mine with certainty whether an animal or plant is
the probability that an individual alive in state j at breeding or not, is healthy or affected by a particular
disease, or even if it is a male or a female. Pradel (2005) 13.4.2 Environmental variability and individual
solved this problem by generalising the multisite– heterogeneity in CMR models
multistate model into a multi-event framework. In
Within a CMR model, survival, transition, and/or
this new framework, individuals are still assumed
detection parameters can be assumed to be a function
to move across different states through survival and
of external covariates (Lebreton et al. 1992). This
transition processes, but a new parameter, the ini-
formulation allows the inclusion of environmental
tial state probability, appears. Usually, field observa-
effects in a regression-like framework, by including,
tions are not fully informative of biological processes.
for example, climatic variables, or individual time-
By formally separating the ‘real’ state process from
invariant characters such as genotype. Multifactorial
the observational process events, Pradel included a
effects of environmental covariates can also be mod-
parameter to account for state uncertainty, that is,
elled using hierarchical models (King et al. 2009), for
the probability to not assign or erroneously assign a
example to study evolutionary processes in the wild
given state to an individual. Multi-event models pro-
(Cubaynes et al. 2012). However, time-varying individ-
vide a solution to estimate sex- and/or age-dependent
ual covariates, such as body weight, are tricky because
survival in species with cryptic or little age and/or
when an individual is not detected, the value of the
sexual dimorphism (see section 13.3.4.1). Transients
covariate is unknown. Inference can be based on a con-
can be considered as a particular initial state and
ditional likelihood approach using only the observed
modelled directly using multi-event models (Genovart
covariate values (trinomial approach; Catchpole et al.
et al. 2012; Santidrian et al. 2017). More than uncer-
2008), or missing values can be imputed from an under-
tainty about the state of an individual, multi-event
lying distribution (e.g. multiple imputation; Worthing-
models allow to explicitly model unobserved hetero-
ton et al. 2015). However, methods of imputation are
geneity among individuals. Indeed, latent state mix-
sensitive to the underlying model and the number of
ture models can be easily implemented within the
missing values (Langrock and King 2013). One possi-
multi-event framework, a model particularly suitable
bility is to discretise the covariate and use a multistate
to frame individual heterogeneity or frailty (Gimenez
model (Fernández-Chacón et al. 2015; Gimenez et al.
et al. 2018a). Other approaches, for example using
2018a).
individual random effects, can be used to implement
Unfortunately, we often do not measure all covari-
frailty (e.g. Cam et al. 2016; Hamel et al. 2018; see also
ates influencing demographic parameters. Also
section 13.3.2).
individuals may react in different ways to envi-
Multi-event and multistate models can be used to
ronmental variation depending on unobservable
address many other ecological and evolutionary ques-
individual states. This leads to unobserved (latent)
tions. If individual states can change over time, multi-
individual heterogeneity (see Chapter 9). Ignoring
event also allows modelling the transition dynamics
latent individual heterogeneity may lead to flawed
between, for example, breeding states (Desprez et al.
inference about the ecological or evolutionary pro-
2013) and epidemiological status (Benhaiem et al.
cesses at hand (Cam et al. 2016; Hamel et al. 2018),
2018). Lagrange et al. (2014) developed multi-event
such as senescence patterns (Cam and Monnat 2000;
models able to study dispersal among numerous sites
Service 2000; Peron et al. 2010). Latent individual
for birds and amphibians. Tavecchia et al. (2012)
heterogeneity can be framed using finite mixtures
modelled mortality due to different causes of mor-
or as individual random effects (see Gimenez et al.
tality of radiotagged individuals, while accounting
2018a for a review about how to implement individual
for the loss of the radio signal (see section 13.3.4.2).
heterogeneity in CMR models).
Multi-event models have been used to model sur-
vival when marks identifying individuals are lost (even
13.4.3 Inference framework
totally; see Badia-Boher et al. 2019). The multi-event
approach can be also used to exploit supplementary Implementation of CMR models can be carried out
information and estimate survival, dispersal, and/or using either a frequentist or Bayesian approach. A dif-
recruitment in partially monitored populations (Sanz- ferent philosophy stands behind each approach, and
Aguilar et al. 2016; Tavecchia et al. 2016). Finally, trap- there is a long-standing debate about whether ecolo-
responses and memory effects can be modelled into the gists should use one or the other (e.g. Lele and Dennis
multi-event approach (Rouan et al. 2009; Pradel and 2009). While the frequentist approach may be faster,
Sanz-Aguilar 2012). the Bayesian approach allows a great flexibility in the
model writing, which can be useful to tackle analyti- E-SURGE and R package marked (Laake et al. 2013)
cal complexity, such as choosing underlying distribu- and in a Bayesian framework using package BaSTA
tions for model parameters or fitting temporal random and rjags.
effects (e.g. Kery and Schaub 2011). Another appeal of
the Bayesian approach is the possibility to include prior 13.4.5 Case studies in animal demography
knowledge on biological parameters to facilitate the
Example 1: Estimating sex-dependent survival
estimation, for example information on body weight or
when sex assignment is uncertain: a multi-event
survival of a closely related species (MacCarthy et al.
model of the Balearic wall lizard (Podarcis lilfordi)
2005). The Bayesian approach of CMR models uses the
state-space modelling (SSM) formulation that clearly In this example, we consider the possibility of erro-
distinguishes the observation process (detection) from neous assignment of sex to newly captured individu-
the underlying demographic process of interest (tran- als, a situation common to the monitoring of species,
sition between states; e.g. Gimenez et al. 2007; Royle for example, with little sexual dimorphisms. In CMR
2008; Kery and Schaub 2011), the observation process analyses, erroneously assigning sex at the beginning of
being conditional to the state process. The SSM for- the capture history leads to bias in the estimated sur-
mulation therefore allows to easily implement complex vival difference between the sexes. Here, our aim is
multifactorial observation processes and combine mul- to estimate sex-specific survival rates in the Balearic
tiple sources of information (e.g. Buoro et al. 2012; see wall lizard, a small species endemic to the Balearic
Supplementary Information S5 for an SSM formulation archipelago, Spain. Immature males are sometimes dif-
of the JS model). ficult to be sexed in the field and can be confounded
Prior to model fitting, goodness-of-fit tests are gen- with mature females of similar size. Sex-specific sur-
erally performed to check the validity of the assump- vival rates can be estimated by accounting for the
tions behind a CMR model, for example using the uncertainty on sex assignment using a multi-event
R2ucare package (Gimenez et al. 2018b). In the fre- CMR model to separate the ‘real’ sex of the individu-
quentist approach, model implementation can be car- als (state) from the ‘apparent’ (observed) sex (event). In
ried out using program MARK and the widely used this model, we considered four events (type of obser-
RMark package (White and Burnham 1999; Laake vations) which code three states (‘real’ state of the indi-
2013), marked (Laake 2013), or E-SURGE (Choquet vidual). Data, together with step-by-step instructions
et al. 2009). Common tools for model comparisons to implement the models in E-SURGE and interpret the
include the Akaike information criterion (AIC) and its results, are provided in Supplementary Information S2.
variants (AICc, QAIC, wAIC; Burnham and Anderson
2002), which serves to rank the models and calculate Example 2: Survival and the issue of tag-loss:
weights of evidence for each of them or for a partic- study case of the red kite (Milvus milvus)
ular effect (Burnham and Anderson 2002). Analysis Tag-loss is a common issue in wildlife monitoring of
of Deviance (Anodev) is also used to calculate the marked individuals and can lead to underestimated
proportion of variance explained by a specific covari- survival (Arnason and Mills 1981). To cope with this
ate (Grosbois et al. 2008). In the Bayesian approach, issue, ecologists have developed advanced method-
models can be implemented using program JAGS ological tools: from multiple-marking to advanced
(Plummer 2003), R packages such as rjags (Plummer statistical methods to integrate tag-loss in the individ-
et al. 2018), or Bayesian Survival Trajectory Analy- ual state (Cowen and Schwarz 2006; Tavecchia et al.
sis (BaSTA) (Colchero et al. 2012). Posterior predictive 2012). Here is one case study to integrate the loss of a
checks can be used for performing model assessment remote tracking device in the multi-event modelling
(Chambert et al. 2014), and information criterion such framework.
as deviance information criterion (DIC) (Spiegelhalter The use of remote tracking devices (radio-satellite
et al. 2002) or wAIC are often used for model compar- and GSM/GPS transmitters) to collect detailed indi-
ison (Hooten and Hobbs 2015). Further details about vidual history data is increasingly common in the eco-
implementation in both a frequentist and Bayesian logical literature (see Chapters 3 and 5). A problem in
framework can be found in McCrea and Morgan (2014). estimating survival from tracking data is that the life
Hereafter, we develop study cases in animal and span of the remote signal is commonly shorter than the
plant demography showing how to implement CMR life span of the individual that carries the device. In this
models in a frequentist framework using program case the survival probability refers to the life span of
(a) (b) (c) (d)

Mortality rate, µ(x)
Figure 13.3 Mortality, μX (x|θ), resulting

from the four basic models included in
BaSTA: (a) exponential; (b) Gompertz;
(c) Weibull; and (d) logistic. The three
different lines in each plot (except in a)
show examples of the shapes that can be
Age, x tested with BaSTA, namely ‘simple’,
simple Makeham bathtub ‘Makeham’, and ‘bathtub’.
the radio signal and not to the one of the animal. When capture occasions (ϕi ) is considered as the remaining
animals are marked with tags or rings in addition to probability at the stopover place, and thus 1-ϕi is the
the radio device, their encounter history can follow the departure probability between two occasions. As indi-
loss of the radio signal. In this example, we illustrate viduals may have arrived in the stopover area before
how multi-event models can accommodate the loss of first capture, the model needs to not be conditional
the signal and provide unbiased estimate of survival on the first capture (as in the CJS model). Thus, we
in the presence of radio-loss or radio-failure using a can use the JS model parametrised with entry proba-
real data set on red kite on the island of Mallorca (data bilities noted as ηi for the probability to entry in the
provided bt T. Muñoz, GOB-Mallorca). Data, together stopover area between time i and i+1 if not previously
with step-by-step instructions to implement the mod- entered (Schwarz and Arnason 1996). This SSM formu-
els in E-SURGE and interpret the results, are provided lation of the JS model allows an easy implementation
in Supplementary Information S3. in the Bayesian framework and a straightforward com-
putation of the stopover duration (Lyons et al. 2016).
Bayesian implementation using the R package We can also easily incorporate the effect of a weather
BaSTA covariate on the departure probability. The implemen-
The R package BaSTA (Colchero et al. 2012) provides tation of this example with R and JAGS is provided in
a set of tools that complement other CMR methods Supplementary Information S5.
when users want to estimate age-specific mortality
from CMR data sets where times of birth are known
13.4.6 Case studies in plant demography
only for a few individuals (or none). Several parametric
mortality models are available in BaSTA, including the Plants do not move, but certain aspects of their ecolo-
exponential, Gompertz, Weibull, logistic, Makeham, gies, such as vegetative dormancy, variable sprouting
and Gompert–Makeham models (Figure 13.3). In order times, and complex growth patterns, can make them
to include all records in the analysis, BaSTA estimates just as challenging to work with as animals.
the missing ages at birth and at death, which reduces
the bias in the estimation of the mortality and cumula- Example 1: Linear modelling of plant survival
tive survival functions. An example of implementation The most common method to analyse survival using
is provided in Supplementary Information S4. plant resighting data sets is using linear analysis
under a logistic, generalised linear model (GLM)
Bayesian state-space formulation of the JS model or generalised linear mixed model (GLMM) frame-
to study stopover decisions of migratory birds work, with survival modelled assuming a binomial
using JAGS distribution (e.g. Salguero-Gómez et al. 2012). This
Migratory birds cannot realise their journey between method assumes that, at the very least, the resighting
breeding and wintering areas in a single flight of thou- of previously observed individuals is nearly per-
sands of kilometres and usually stop-over at places fect, because any phenomenon decreasing redetec-
where they can replenish their energy reserves. At tion would be observed as mortality and yield-biased
these stopover places individuals are not easy to detect. survival estimates. In cases where redetection is not
Studying the stopover decisions of migratory birds is perfect, some have argued that this approach is still
a typical case where the detectability needs to be tak- useful provided that monitoring data sets are par-
en into account to be able to make strong ecological ticularly long and large and that redetection is still
inference. In this example the survival between two above 90% (Shefferson et al. 2018). However, in studies
of vegetative dormancy-prone perennials, dormancy the problem of unobserved life stages (Kendall and
will increasingly be confused with mortality as study Nichols 2002).
length decreases. In the final year of a study, there will In this example, we also use the Cypripedium parv-
be no ability to differentiate the dead from the dor- iflorum data set, in combination with the R package
mant. This suggests that the final 1–3 years of data in a marked (Laake et al. 2013), to investigate costs of repro-
study should be used simply to assign dormancy with- duction using CMR models. Full procedure including
in the remaining data and that survival should not be data and R script to run the analysis are provided in
estimated for those years. This loss of estimable years Supplementary Information S7.
adds value to long data sets, particularly those over 10
years long.
We illustrate the use of linear modelling for survival 13.5 Mortality analysis in human
analysis in plants using a case study on Cypripedium populations
parviflorum, the North American small yellow lady’s
Modelling mortality in human populations is relative-
slipper, using data collected from 1994 to 2003 within
ly easier than in nonhuman ones. For a given group of
a larger population from Illinois in the United States.
individuals, we often know their age-at-death, calen-
These data were previously used in, for example,
dar year of the event, and their sex. Thanks to these
Shefferson et al. (2018). Using the R packages lme4 for
reliable data sources, methodological advances have
model fitting and MuMin for model comparison, we
been produced since De Moivre (1725) and Gompertz
show that nonflowering plants have decreasing sur-
(1825). These long-standing demographic and statis-
vival with increasing size, while flowering plants have
tical developments have been often drawn by politi-
increasing survival with increasing size. Full proce-
cal, military, and economic reasons. In the following
dure including data and R script to run the analysis are
sections, we present a brief overview of the most com-
provided in Supplementary Information S6.
mon models used to describe mortality patterns over
Example 2: CMR survival analysis for plants age and/or time on human mortality data.
Plant population ecologists have long used field meth-

ods that may be considered in the same vein as mark– 13.5.1 Human data and assumptions
recapture methods in wildlife ecology. However, the
application of CMR methods to plant population ecol- For a given sex, we usually have deaths and expo-
ogy is very recent. In one of the first studies to use sures to the risk of death arranged in two matrices,
mark–recapture analyses in plants, Alexander et al. whose rows and columns are classified by age at death
(1997) faced all of these problems in a population of and year of death. The stochastic assumption behind
the Mead’s milkweed, Asclepias meadii. This population mortality has a central role in modelling it. The most
consists of plants that grow in high densities and do suitable distribution when we observe mortality data
not always produce aboveground tissue in a growing is the Poisson distribution. The aim of any mortality
season. Closed population mark–recapture analysis model is to seek for a parsimonious, yet satisfactory
allowed them to produce estimates of population size description of the so-called force of mortality μi,j , giv-
unbiased by these challenges. Expanding on this work, en observed deaths di,j , and exposures ei,j . One could
Shefferson et al. (2001) proposed the use of open popu- estimate force of mortality in a fully nonparametric
d
lation mark–recapture models to estimate annual sur- framework computing the death rates μi,j ≈ ei,ji,j . Sim-
vivorship in populations in which living individuals ple plots of rates over age and/or time are good tools
do not always sprout in a growing season. Since then, for a first presentation of mortality development. Com-
CMR studies have blossomed in plant population ecol- monly, rates are plotted on a logarithmic scale to bet-
ogy, with extensions into the estimation of transition ter acknowledge differences (Figure 13.4). A relatively
rates among life history stages (e.g. Shefferson et al. strong assumption behind this is that within the Pois-
2003), estimation of the demographic impacts of her- son distribution mean and variance are equal. When
bivory (e.g. Kéry and Gregg 2004), investigations into the observed variance is larger than the theoretical one,
relationships among life history traits (e.g. Shefferson we often attribute this feature either to overdispersion
et al. 2003), tests of correlation with climatic factors or to some hidden patterns in the data. Specific meth-
(e.g. Shefferson and Tali 2007), estimation of minimal ods for coping with this issue have been proposed in
recruitment levels necessary to sustain populations the literature for all models below, though they will
(Slade et al. 2003), and theoretical papers inspired by not be presented in this chapter. For a comprehensive
1
1960 1970 1980 1990 2000 2010
0.1
Mortality (log−scale) 0.01
0.001
0.0001
0 20 40 60 80 100
Ages
Figure 13.4 Death rates in log-scale over ages for all available years. Japanese females, 1960–2016, ages 0–100.
overview of them see, for example, Cameron and Trive- of the Gompertz–Makeham models. A logistic func-
di (2013). Furthermore, binomial distribution could be tion to model the late-life mortality deceleration can
used when we deal with probability of dying (deaths begiven by
divided by persons-at-risk), and the multinomial dis-
tribution has been suggested for modelling mortality aebi
μi = c + (13.3)
from a cohort perspective (Canudas Romo et al. 2018). 1 + αebi
where α captures the mortality deceleration at old-
13.5.2 Parametric models over age est ages. This law of mortality can be derived when
heterogeneity is assumed in a proportional hazard set-
Parametric models for a suitable representation of the ting. Commonly called the Gamma–Gompertz model,
variation of mortality over age have been proposed eqn. (13.3) is the hazard at a population level when
since Gompertz (1825). He observed that after a cer- standard mortality is described as a Gompertz and
tain age, a ‘law of geometric progression pervades, frailty values are assumed to be Gamma distributed
in an approximate degree, large portions of different (Wienke 2010).
tables of mortality’ (Gompertz 1825, p. 514). He thus A simplified version of the previous logistic law has
suggested representing force of mortality as: been proposed by Thatcher et al. (1998):
μi = aebi (13.1) aebi
μi = c + (13.4)
where a represents the mortality at time zero (usually 1 + aebi
age 30/40) and b is the rate of increase of mortality and Heligman and Pollard (1980) derived a descriptive
is frequently used as a measure of the rate of ageing. model, covering the whole age range. Here we propose
Makeham (1860) extended Gompertz’s equation by a version for death rates:
adding a constant, an age-independent term, c >0, to
GHi
μi = A(i+B) + De−E(lni−lnF) +
C 2
account for risks of death that do not depend on age: (13.5)
1 + GHi
bi
μi = c + ae (13.2)
where A, B, . . ., H are the eight parameters in the model.
Human mortality often shows a levelling-off above Each component aims to describe mortality at child-
certain ages (often 80) (Vaupel 1997). Logistic models hood, during young-adult ages, and at older ages. It is
have been proposed to portray this feature. Perks easy to see that such parameterisation can cause diffi-
(1932) was the first to propose a logistic modification culties in the estimation procedure. Moreover, it would
be hard to disentangle the physical meaning of each In the last decades, further variants of this model
parameter (Booth and Tickle 2008). have been proposed for enhancing several features of
Another three-component, competing-risk mortali- the model (e.g. Booth et al. 2002).
ty model has been proposed by Siler (1983). Initially
developed for animal survival data, this model has
been recently used in human demography, especially 13.5.4 Semiparametric models: the example
for simulating possible scenarios in mortality devel- of P-splines
opments (Canudas-Romo 2018). This model aims at
A compromise between simple parametric and over-
portraying the whole of the age range with five param-
parametrised models could be found in the area of
eters:
semiparametric statistics. Without the aim of pro-
ducing estimated parameters with physical meanings
μi = a1 e−bi i + a2 + a3 eb3 i (13.6)
and searching for a good fit to the data, smoothing
approaches have been lately developed in the study
13.5.3 Overparametrised models: the example of nonlinear phenomena. Among numerous options,
of the Lee–Carter model we will mention here (and illustrate with more details
in Supplementary Information S7) a methodology that
Moving to a two-dimensional perspective and giv-
is particularly suitable for the analysis of mortality
en the wealth of data, traditional demographic meth-
developments: B-splines with penalties, known as P-
ods for analysing mortality surfaces, that is, data on
splines. In a unidimensional setting, applications to
deaths and exposures cross-classified by age and year
mortality have been proposed by Currie et al. (2004)
of occurrence, tend to apply a high number of parame-
and Camarda (2008). Estimation of a model for the
ters, leading to all but parsimonious models. A typical
complete mortality surface will be achieved by a spe-
example of this kind is the widely used model intro-
cific R package: MortalitySmooth (Camarda 2012). It
duced by Lee and Carter (1992). In its original formula-
is noteworthy that, being an extremely flexible tool,
tion, this approach reduces the complexity of the whole
this methodology can be easily implemented for the
surface by introducing the following bilinear model for
analysis of demographic phenomena other than mor-
the log-death-rates:
tality. The main idea of the P-splines is to construct an
intentionally overparametrised model and simultane-
ln mij = αi + βi κj + εij , i = l, . . . , m (13.7) ously to restrict, via a penalty, all redundant features
for achieving a wisely parsimonious description of
j = l, . . . , n the data. Instead of smoothing a given structure, this
approach used local supports such as equally spaced
B-splines over ages and/or years and penalty term(s)
where αi, βi, and κj are vectors of parameters to be esti-
on the associated coefficients.
mated, and εij represents the error term. Interpretation
of the parameters is straightforward: αi and βi describe
the general shape of mortality and the fixed rate of 13.5.5 Implementation example
mortality improvement at age i respectively, and κj is
a time-varying index which captures the general level We use data from the Human Mortality Database
of mortality. The variance εij in Lee and Carter (1992) (2019) on Japanese females from the years 1960 to 2016
is assumed to be constant for all i and j. As present- and from ages 0 to 100 to fit the models described
ed, the Lee–Carter (LC) model is underdetermined and above. Guidance with R script is provided in Supple-
requires additional constraints on the parameters to be mentary Information S8.
successfully estimated. Usually, the model is centred
P P
by βi = 1, κj = 0.
As pointed out in section 13.4.1, a Poisson assump-
13.6 Conclusions
tion is more suitable for estimating mortality models, This chapter has illustrated how the choice of a specific
and a further development of the LC model was devot- method for survival analysis is driven by the research
ed to this issue (Brouhns et al. 2002). In Supplementary question (e.g. comparing survival across groups vs.
Information S7 we provide computational details for assessing the effects of environmental predictors), the
estimating the LC model within a Poisson framework. species life history (more or less easy to monitor), and
its living environment (controlled vs. stochastic envi- Brouhns, N., Denuit, M., and Vermunt, J. K. (2002). A Pois-
ronments), which determine the type of survival data son log-bilinear regression approach to the construction
(e.g. proportion of individuals alive, CMR data, or age of projected lifetables. Insurance: Mathematics and Eco-
at death). nomics 31, 373–393.
Buoro M., Prévost, E. and Gimenez, O. (2012). Digging
With the increase in new monitoring techniques
through model complexity: using hierarchical models to
allowing the gathering of more and more detailed data
uncover evolutionary processes in the wild. Journal of
at the individual level (see Chapter 5) has come the
Evolutionary Biology 25(10), 2077–2090.
development of advanced statistical tools for survival Burnham, K.P. and Anderson, D.R. (2002). Model Selec-
analyses, including multi-event models and hierarchi- tion and Multimodel Inference: A Practical Information-
cal models. More than simply assessing survival to Theoretic Approach, 2nd edition. Springer, New York.
feed population projection models (see Chapters 9, 10, Cam, E., and Monnat, J. Y. (2000). Apparent inferiority of
11, and 14), modern survival analyses have addressed first-time breeders in the kittiwake: the role of heterogene-
questions about evolutionary trade-offs (e.g. Nichols ity among age classes. Journal of Animal Ecology, 69(3),
and Kendall 1995), static versus dynamic heterogene- 380–394.
ity in demographic parameters (e.g. Cam et al. 2016; Cam, E., Aubry, L. M., and Authier, M. (2016). The conun-
drum of heterogeneities in life history studies. Trends in
Gimenez et al. 2018a), assessing senescence (e.g. Peron
Ecology and Evolution, 31(11), 872–886.
et al. 2010), and quantifying the heritability of demo-
Camarda, C. G. (2008). Smoothing Methods for the Analysis
graphic parameters in the wild (e.g. Papaïx et al. 2010). of Mortality Development. PhD thesis, Programa de Doc-
Addressing more and more complex questions has torado en Ingeniería Matemática. Universidad Carlos III,
brought new challenges to the field of survival anal- Departamento de Estadística, Madrid.
yses. Among others, current methodological develop- Camarda, C. G. (2012). MortalitySmooth: an R package for
ments deal with methods to implement models with smoothing Poisson counts with P-splines. Journal of Sta-
numerous states, predict age at death, and consider tistical Software 50, 1–24.
dependence among individuals when estimating sur- Cameron, A. C., and Trivedi, P. K. (2013). Regression Anal-
vival. ysis of Count Data (Vol. 53). Cambridge University Press,
Cambridge.
Canudas-Romo, V., Mazzuco, S., and Zanotto, L. (2018).
Measures and models of mortality. In: C.R. Rao (ed.),
References Handbook of Statistics (Vol. 39, pp. 405–442). Elsevier,
Alexander, H.M., Slade, N.A. and Kettle, W.D. (1997). Appli- Amsterdam.
cation of mark–recapture models to estimation of the Catchpole, E. A., Morgan, B. J., and Tavecchia, G. (2008).
population size of plants. Ecology, 78, 1230–1237. A new method for analysing discrete life history data
Arnason, A. N. (1973). The estimation of population size, with missing covariate values. Journal of the Royal Sta-
migration rates and survival in a stratified population. tistical Society: Series B (Statistical Methodology), 70(2),
Researches on Population Ecology, 15, 1–8. 445–460.
Arnason, A. N., and Mills, K. H. (1981). Bias and loss of Chambert, T., Rotella, J. J., and Higgs, M. D. (2014). Use of
precision due to tag loss in Jolly–Seber estimates for mark– posterior predictive checks as an inferential tool for inves-
recapture experiments. Canadian Journal of Fisheries and tigating individual heterogeneity in animal population
Aquatic Sciences, 38(9), 1077–1095. vital rates. Ecology and Evolution, 4(8), 1389–1397.
Badia-Boher, J. A., Sanz-Aguilar, A., De La Riva, M., et al. Choquet, R., Rouan, L., and Pradel, R. (2009). Program
(2019). Evaluating European LIFE conservation projects: E-SURGE: a software application for fitting multi-event
improvements in survival of an endangered vulture. Jour- models. In: D.L. Thomson, E.G. Cooch, M.J. Conroy, (eds),
nal of Applied Ecology, 56(5), 1210–1219. Modeling Demographic Processes in Marked Populations
Balkız, O., Béchet, A., Rouan, L., et al. (2010). Experience- (pp. 845–865). Springer, Boston, MA.
dependent natal philopatry of breeding greater flamingos. Colchero, F., Jones, O. R., and Rebke, M. (2012). BaSTA:
Journal of Animal Ecology 79, 1045–1056. an R package for Bayesian estimation of age-specific
Benhaiem, S., Marescot, L., Hofer, H., et al. (2018). Robust- survival from incomplete mark–recapture/recovery data
ness of eco-epidemiological capture–recapture parameter with covariates. Methods in Ecology and Evolution, 3(3),
estimates to variation in infection state uncertainty. Fron- 466–470.
tiers in Veterinary Sciences 5, 197. Cowen, L., and Schwarz, C. J. (2006). The Jolly–Seber model
Booth, H., and Tickle, L. (2008). Mortality modelling and with tag loss. Biometrics, 62(3), 699–705.
forecasting: a review of methods. Annals of Actuarial Cox, D. R. (1972). Regression models and life-tables. Journal
Science, 3(1–2), 3–43. of the Royal Statistical Society: Series B (Methodological),
Booth, H., Maindonald, J., and Smith, L. (2002). Applying 34(2), 187–202.
Lee–Carter under conditions of variable mortality decline. Cox, D. R. (2018). Analysis of Survival Data. Chapman and
Population Studies 56, 325–336. Hall/CRC Press, Boca Raton.
Cubaynes, S., Doutrelant, C., Grégoire, A., Perret, P., Faivre, Jenouvrier, S., Tavecchia, G., Thibault, J., Choquet, R., and
B., and Gimenez, O. (2012). Testing hypotheses in evo- Bretagnolle, V. (2008). Recruitment processes in long-lived
lutionary ecology with imperfect detection: capture– species with delayed maturity: estimating key demo-
recapture structural equation modeling. Ecology, 93(2), graphic parameters. Oikos, 117, 620–628.
248–255. Jolly, G. M. (1965). Explicit estimates from capture–recapture
Currie, I. D., M. Durbán, and P. H. C. Eilers (2004). Smooth- data with both death and immigration-stochastic model.
ing and forecasting mortality rates. Statistical Modelling, Biometrika, 52(1/2), 225–247.
4, 279–298. Jönsson, K. I., Harms-Ringdahl, M., and Torudd, J. (2005).
De Moivre, A. (1725). Annuities Upon Lives: Or, The Valua- Radiation tolerance in the eutardigrade Richtersius coro-
tion of Annuities Upon Any Number of Lives. NewsBank nifer. International Journal of Radiation Biology, 81(9),
Readex. 649–656.
Desprez, M., McMahon, C. R., Hindell, M. A., Harcourt, R., Kaplan, E. L., and Meier, P. (1958). Nonparametric estima-
and Gimenez, O. (2013). Known unknowns in an imper- tion from incomplete observations. Journal of the Ameri-
fect world: incorporating uncertainty in recruitment esti- can Statistical Association, 53(282), 457–481.
mates using multi-event capture–recapture models. Ecol- Kendall, W. L. and Nichols, J. D. (2002). Estimating
ogy and Evolution, 3, 4658–4668. state-transition probabilities for unobservable states
Fernández-Chacón, A., Genovart, M., Álvarez, D. et al. using capture–recapture/resighting data. Ecology, 83,
(2015). Neighbouring populations, opposite dynamics: 3276–3284.
influence of body size and environmental variation on Kéry, M. and Gregg, K. B. (2004). Demographic analysis of
the demography of stream-resident brown trout (Salmo dormancy and survival in the terrestrial orchid Cypripedi-
trutta). Oecologia, 78, 379–389. um reginae. Journal of Ecology, 92, 686–695.
Genovart, M., Pradel, R., Oro, D. (2012). Exploiting uncertain Kéry, M., and Schaub, M. (2011). Bayesian Population Analy-
ecological fieldwork data with multi-event capture– sis Using WinBUGS: A Hierarchical Perspective. Academ-
recapture modelling: an example with bird sex assign- ic Press, Cambridge, MA.
ment. Journal of Animal Ecology, 81, 970–977. King, R., Morgan, B., Gimenez, O., and Brooks, S. (2009).
Gimenez, O., Rossi, V., Choquet, R., et al. (2007). State- Bayesian Analysis for Population Ecology. Chapman and
space modelling of data on marked individuals. Ecolog- Hall/CRC Press, Boca Raton.
ical Modelling, 206(3–4), 431–438. Kirkwood B. R., and Sterne J. A. C. (2003). Essential Medical
Gimenez, O., Cam, E., and Gaillard, J. M. (2018a). Individual Statistics, 2nd edition. Blackwell, Oxford.
heterogeneity and capture–recapture models: what, why Klein, J. P., and Moeschberger, M. L. (2006). Survival analy-
and how? Oikos, 127(5), 664–686. sis: techniques for censored and truncated data. Springer
Gimenez, O., Lebreton, J. D., Choquet, R., and Pradel, R. Science + Business Media, Berlin.
(2018b). R2ucare: an r package to perform goodness-of-fit Klein, J. P., Van Houwelingen, H. C., Ibrahim, J. G. and
tests for capture–recapture models. Methods in Ecology Scheike, T. H. (eds). (2016). Handbook of Survival Anal-
and Evolution, 9(7), 1749–1754. ysis. CRC Press, Boca Raton.
Gompertz, B. (1825). XXIV. On the nature of the func- Laake, J. L., Johnson, D. S., and Conn, P. B. (2013). marked:
tion expressive of the law of human mortality, and on an R package for maximum likelihood and Markov Chain
a new mode of determining the value of life contingen- Monte Carlo analysis of capture–recapture data. Methods
cies. In a letter to Francis Baily, Esq. FRS. Philosophi- in Ecology and Evolution, 4(9), 885–890.
cal Transactions of the Royal Society of London, (115), Lagrange, P., Pradel, R., Bélisle, M., and Gimenez, O.
513–583. (2014). Estimating dispersal among numerous sites
Grosbois, V. and Tavecchia, G. (2003). Modeling dispersal using capture–recapture data. Ecology, 95, 2316–
with capture–recapture data: disentangling decisions of 2323.
leaving and settlement. Ecology 84, 1225–1236. Langrock, R., and King, R. (2013). Maximum likelihood esti-
Grosbois, V., Gimenez, O., Gaillard, J.-M., et al. (2008). mation of mark–recapture–recovery models in the pres-
Assessing the impact of climate variation on survival in ence of continuous covariates. Annals of Applied Statis-
vertebrate populations. Biological Reviews, 83, 357–399. tics, 7(3), 1709–1732.
Hamel, S., Gaillard, J. M., Douhard, M., Festa-Bianchet, M., Lebreton, J.-D, Burnham, K. P., Clobert, J., and Ander-
Pelletier, F., and Yoccoz, N. G. (2018). Quantifying indi- son, D. R. (1992). Modeling survival and testing bio-
vidual heterogeneity and its influence on life-history tra- logical hypotheses using marked animals—a unified
jectories: different methods for different questions and approach with case-studies. Ecological Monographs, 62,
contexts. Oikos, 127(5), 687–704. 67–118.
Heligman, L., and Pollard, J. H. (1980). The age pattern Lebreton, J.-D., Almeras, T., and Pradel, R. (1999). Com-
of mortality. Journal of the Institute of Actuaries, 107(1), peting events, mixtures of information and multistrata
49–80. recapture models. Bird Study, 46,S39–S46.
Hooten, M. B., and Hobbs, N. T. (2015). A guide to Bayesian Lee, R. D., and Carter, L. R. (1992). Modeling and fore-
model selection for ecologists. Ecological Monographs, casting US mortality. Journal of the American Statistical
85(1), 3–28. Association, 87(419), 659–671.
Lele, S. R., and Dennis, B. (2009). Bayesian methods for hier- Pollock, K., Nichols, J., Brownie, C., and Hines, J. (1990).
archical models: are ecologists making a Faustian bargain? Statistical inference for capture–recapture experiments.
Ecological Applications, 19(3), 581–584. Wildlife Monographs, 107, 3–97.
Loprinzi, C. L., Laurie, J. A., Wieand, H. S., et al. (1994). Pradel, R. (1993). Flexibility in survival analysis from recap-
Prospective evaluation of prognostic variables from ture data: handling trap-dependence. In: J.-D. Lebre-
patient-completed questionnaires. North Central Cancer ton and P. M. North (eds), Marked Individuals in
Treatment Group. Journal of Clinical Oncology. 12(3), the Study of Bird Population (pp. 29–37). Birkhäuser,
601–607. Basel.
Lyons, J. E., Kendall, W. L., Royle, J. A., Converse, S. J., Pradel, R. (2005). Multi-event: an extension of multi-state
Andres, B. A., and Buchanan, J. B. (2016). Population capture–recapture models to uncertain states. Biometrics,
size and stopover duration estimation using mark–resight 61, 442–447.
data and Bayesian analysis of a superpopulation model. Pradel, R. and Lebreton, J.-D. (1999). Comparison of dif-
Biometrics, 72(1), 262–271. ferent approaches to the study of local recruitment of
Makeham, W. M. (1860). On the law of mortality and the breeders. Bird Study, 46, 74–81.
construction of annuity tables. Journal of the Institute of Pradel, R. and Sanz-Aguilar, A. (2012). Modeling trap-
Actuaries, 8(6), 301–310. awareness and related phenomena in capture–recapture
McCarthy, M.A. and Masters, P. I. P. (2005). Profiting from studies. PLOS One, 7, e32666.
prior information in Bayesian analyses of ecological data. Pradel, R., Hines, J. E., Lebreton, J.-D., and Nichols, J. D.
Journal of Applied Ecology, 42(6), 1012–1019. (1997). Capture–recapture survival models taking account
McCrea, R. S., and Morgan, B. J. (2014). Analysis of of transients. Biometrics, 53, 60–72.
Capture–Recapture Data. Chapman and Hall/CRC Press, Rouan, L., Choquet, R., and Pradel, R. (2009). A general
Boca Raton. framework for modeling memory in capture–recapture
Metcalf, C. J. E., and Pavard, S. (2007). Why evolutionary data. Journal of Agricultural Biological and Environmen-
biologists should be demographers. Trends in Ecology tal Statistics, 14, 338–355.
and Evolution, 22(4), 205–212. Royle, J. A. (2008). Modeling individual effects in the
Miller Jr, R. G. (2011). Survival Analysis (Vol. 66). Cormack–Jolly–Seber model: a state–space formulation.
John Wiley & Sons, New York. Biometrics, 64(2), 364–370.
Nichols, J. D. and Kendall, W. L. (1995). The use of multi- Salguero-Gómez, R., Siewert, W., Casper, B.B. and
state capture–recapture models to address questions in Tielbörger, K. (2012). A demographic approach to study
evolutionary ecology. Journal of Applied Statistics, 22, effects of climate change in desert plants. Philosophical
835–846. Transactions of the Royal Society B: Biological Sciences,
Ono, F., Mori, Y., Takarabe, F. A, et al. (2016) Effect of ultra- 367, 3100–3114.
high pressure on small animals, tardigrades and Artemia. Santidrián Tomillo, P., Robinson, N., et al. (2017). High and
Cogent Physics, 3(1). variable mortality of leatherback turtles reveal possible
Papaïx, J., Cubaynes, S., Buoro, M., Charmantier, A., Perret, anthropogenic impacts. Ecology, 98, 2170–2179.
P., and Gimenez, O. (2010). Combining capture–recapture Sanz-Aguilar, A., Tavecchia, G., Mínguez, E., et al. (2010).
data and pedigree information to assess heritability of Recapture processes and biological inference in monitor-
demographic parameters in the wild. Journal of Evolu- ing burrowing nesting seabirds. Journal of Ornithology,
tionary Biology, 23(10), 2176–2184. 151, 133–146.
Perks, W. (1932). On some experiments in the graduation of Sanz-Aguilar, A., Igual, J.M., Oro, D., Genovart, M., and
mortality statistics. Journal of the Institute of Actuaries, Tavecchia, G. (2016). Estimating recruitment and survival
63(1), 12–57. in partially monitored populations. Journal of Applied
Peron, G., Crochet, P.A., Choquet, R., Pradel, R., Lebreton, Ecology, 53, 73–82.
J.-D. and Gimenez, O. (2010). Capture–recapture models Schaub, M. and Pradel, R. 2004. Assessing the relative
with heterogeneity to study survival senescence in the importance of different sources of mortality from recov-
wild. Oikos, 119, 524–532. eries of marked animals. Ecology, 85, 930–938.
Peto, R., Pike, M. C., Armitage, P., et al. (1977). Design and Schaub, M., Gimenez, O., Schmidt, B. R., and Pradel, R.
analysis of randomized clinical trials requiring prolonged (2004). Estimating survival and temporary emigration in
observation of each patient. II. Analysis and examples. the multi-state capture–recapture framework. Ecology,
British Journal of Cancer, 35(1), 1–39. 85(8), 2107–2113.
Plummer, M. (2003). JAGS: a program for analysis of Schwarz, C. J., and Arnason, A. N. (1996). A general method-
Bayesian graphical models using Gibbs sampling. Pro- ology for the analysis of capture–recapture experiments in
ceedings of the 3rd International Workshop on Distributed open populations. Biometrics, 52, 860–873.
Statistical Computing, 124(125.10). Schwarz, C. J., Schweigert, J. F., and Arnason, A. N. (1993).
Plummer, M., Stukalov, A., Denwood, M., and Plummer, M. Estimating migration rates using tag-recovery data. Bio-
M. (2018). Package ‘rjags’. update, 16, 1. metrics, 49, 177–193.
Seber, G. A. (1965). A note on the multiple-recapture census. cost of first reproduction in greater flamingos. Ecology, 82,
Biometrika, 52(1/2), 249–259. 165–174.
Service, P. M. (2000). Heterogeneity in individual mor- Tavecchia, G., Adrover, J., Navarro, A. M., and Pradel, R.
tality risk and its importance for evolutionary stud- (2012). Modelling mortality causes in longitudinal data
ies of senescence. The American Naturalist, 156(1), in the presence of tag loss: application to raptor poison-
1–13. ing and electrocution. Journal of Applied Ecology, 49,
Shefferson, R. P. and Tali, K. (2007). Dormancy is associated 297–305.
with decreased adult survival in the burnt orchid, Neotinea Tavecchia, G., Sanz-Aguilar, A., and Cannell, B. (2016). Mod-
ustulata. Journal of Ecology, 95, 217–225. elling survival and breeding dispersal to unobservable
Shefferson, R. P., Sandercock, B. K., Proper, J., and nest sites. Wildlife Research, 43, 411–417.
Beissinger, S. R. (2001). Estimating dormancy and sur- Thatcher, R., Kannisto, V., and Vaupel, J. W. (1998).
vival of a rare herbaceous perennial using mark–recapture The force of mortality at ages 80 to 120. Mono-
models. Ecology, 82, 145–156. graphs on Population Aging, 5. Odense University Press,
Shefferson, R. P., Proper, J., Beissinger, S. R., and Simms, Odense.
E. L. (2003). Life history trade-offs in a rare orchid: the University of California, Berkeley (USA), and Max Planck
costs of flowering, dormancy, and sprouting. Ecology, 84, Institute for Demographic Research (Germany). (2019).
1199–1206. Human Mortality Database. www.mortality.org or
Shefferson, R. P., Kull, T., Hutchings, M. J., et al. (2018). www.humanmortality.de (accessed 2019).
Drivers of vegetative dormancy across herbaceous peren- Vaupel, J. W. (1997). Trajectories of Mortality at Advanced
nial plant species. Ecology Letters, 21, 724–733. Ages. Between Zeus and the Salmon: The Biodemog-
Siler, W. (1983). Parameters of mortality in human pop- raphy of Longevity, 17–37. National Academies Press,
ulations with widely varying life spans. Statistics in Washington, DC.
Medicine, 2(3), 373–380. White, G. C. and Burnham, K. P. (1999). Program MARK:
Slade, N. A., Alexander, H. M., and Kettle, W. D. (2003). survival estimation from populations of marked animals.
Estimation of population size and probabilities of sur- Bird Study, 46, S120–S139.
vival and detection in Mead’s milkweed. Ecology, 84, Wienke, A. (2010). Frailty Models in Survival Analysis.
791–797. Chapman and Hall/CRC Press, Boca Raton, FL.
Spiegelhalter, D. J., Best, N. G., Carlin, B. P., and van der Williams, B. K., Nichols, J. D., and Conroy, M. J. (2002). Anal-
Linde, A. (2002). Bayesian measures of model complexity ysis and Management of Animal Populations. Academic
and fit (with discussion). Journal of the Royal Statistical Press, Cambridge, MA.
Society B, 64, 583–639. Worthington, H., King, R., and Buckland, S. T. (2015).
Stearns, S. C. (1992). The Evolution of Life Histories (No. 575 Analysing mark–recapture–recovery data in the presence
S81). Oxford University Press, London. of missing covariate data via multiple imputation. Journal
Tavecchia, G., Pradel, R., Boy, V., Johnson, A. R., and Cezilly, of Agricultural, Biological, and Environmental Statistics,
F. (2001). Sex- and age-related variation in survival and 20(1), 28–46.
CHAPTER 14
Efficient use of demographic data:

integrated population models
Marlène Gamelon, Stefan J. G. Vriend, Marcel E. Visser,
Caspar A. Hallmann, Suzanne T. E. Lommen, and Eelke Jongejans
14.1 Introduction: why use integrated to the integrated approach, which would otherwise not
population models? be possible due to missing data. Fourth, changes in
population structure and interactions at the population
To understand the dynamics of wild populations, var- (e.g. population counts) and at the individual level (e.g.
ious types of demographic data can be collected in the CMR and fecundity data) are modelled (Plard et al.
field, including population counts, productivity sur- 2019a). Fifth, separating the true pattern with obser-
veys, capture–mark–recapture (CMR), mark–recovery vation error and reconciling across data types allows
(MR) data, and telemetry data (see Chapters 4 and more robust matching of structural demography with
5). As seen in previous chapters, it is possible to gain observed patterns. Thus, resulting population models
an understanding of population sizes from population are more likely to follow the observed temporal fluc-
counts (see Chapter 7). Likewise, from CMR data, sur- tuations in population size than when such models are
vival probabilities may be estimated (see Chapter 13). based on vital-rate data only.
Sometimes, all these demographic data are collected in The use of IPMs began in the 1980s in fisheries (see
the same population. Considered separately, each of Maunder and Punt 2013 for a review). Twenty years
these data sets provides useful information on popu- later, IPMs were developed in terrestrial systems (see
lation size and vital rates such as survival, fecundity, Besbeas et al. 2002 for a study on birds), and from
or immigration/emigration, but inference is limited 2002 onwards the number of studies applying IPMs has
when component data sets are compared. Indeed, pop- markedly increased in the literature (see Zipkin and
ulation sizes fluctuate from year to year due to all Saunders 2018 for a review). For a long time, most of
of these processes together: the loss and gain of indi- the studies using IPMs have focused on the statistical
viduals through survival, fecundity, immigration, and developments of methods. For instance, Abadi et al.
emigration. Consequently, these data sources share (2010a) explored to what extent parameter estimates
common demographic information about the studied are more accurate and precise under varying sample
population. sizes. But in recent years, IPMs have become an impor-
Integrated population models (IPMs) (Besbeas et al. tant tool in ecological studies to obtain more precise
2002) use different types of demographic data by joint- and accurate estimates of vital rates (see e.g. Lee et al.
ly analysing them. For instance, census data and CMR 2015 for a case study on the Svalbard reindeer Rangifer
data can be analysed jointly. This integrated approach tarandus platyrhynchus) or to estimate vital rates that
has several advantages. First, parameter estimates such are difficult to estimate with classical approaches such
as population sizes or survival rates, shared among dif- as capture–mark–recapture (CMR) models (see Abadi
ferent data sets, are more precise and accurate. Second, et al. 2010b; van Oosten et al. 2015 for an estimation of
imperfect detection and observation error inherently immigration rate). IPMs also appear to be useful tools
associated with data sampled in the field are accounted in retrospective analyses to study transient dynamics
for. Third, some parameter values can be estimated due
Marlène Gamelon et al., Efficient use of demographic data: integrated population models. In: Demographic Methods Across the Tree of Life.
(see Chapter 11) and make inference on the demo- 14.2.1 Population model
graphic and structural drivers of realised population
The first step to build an IPM is to define a popu-
growth rate (Koons et al. 2016), as well as in prospective
lation model that links the vital rates (e.g. survival,
analyses (Oppel et al. 2014). IPMs offer many other
fecundity) to population sizes. This is basically a
possibilities by, for instance, including spatially explic-
(st)age-structured matrix population model that allows
it CMR data (Chandler and Clark 2014) or two sexes
(st)age-specific population sizes at year t+1 to be esti-
(Tenan et al. 2016).
mated from (st)age-specific population sizes at year t
Because of all the possibilities offered by IPMs, they
and vital rates (see Chapter 9 for an introduction to
have also started to be widely used for conservation
matrix population models). Showing these transitions
purposes (see Zipkin and Saunders 2018 for a review).
between (st)ages with a life cycle graph based on the
In declining populations, the data collected are gen-
life history of the studied species may help to construct
erally scarce due to low sample sizes and because
the population model.
researchers are reluctant to catch and mark individuals
in vulnerable populations. In such cases, IPMs are use-
ful tools; by making efficient use of all data available, 14.2.2 Individual likelihoods
key vital rates that drive the dynamics of declining The second step in building an IPM is to define the
populations are identified and appropriate conserva- likelihood functions for each data set (e.g. CMR data,
tion actions can be advised (see e.g. Lieury et al. 2015, census data) separately, which describe the proba-
2016). IPMs also are relevant tools in a management bility of an observed outcome (i.e., the data) condi-
context, where they provide important insights on the tional on particular parameter values. For instance,
dynamics of exploited populations and thus help man- for population count data, state-space models (de
agers to conduct habitat manipulations and develop Valpine and Hastings 2002) are often used. They
relevant harvest strategies (Arnold et al. 2018). consist of a process model describing how the true
More generally, IPMs are appropriate tools to population size and structure change while account-
test hypotheses in life history theory, conservation, ing for observation error over years (Besbeas et al.
and eco-evolutionary biology (Plard et al. 2019a), 2002). The process model corresponds to the popu-
as the submodels are amenable to common regres- lation model described in section 14.2.1. The obser-
sion structures. Recent methodological advances, for vation model describes the link between the pop-
instance the introduction of Markov Chain Monte Car- ulation counts and the true population sizes. For
lo (MCMC) methods in ecology, have highly contribut- CMR data, among the variety of likelihoods avail-
ed to the expansion of this approach that can be imple- able for survival analyses and depending on data
mented in freely available software (e.g. JAGS, STAN, structure and research question(s), Cormack–Jolly–
NIMBLE, or BUGS). Moreover, even if implementing Seber models (Lebreton et al. 1992) can be used,
IPMs can be complex, example codes can easily be and a multinomial likelihood links the data to
found in the pivotal Kéry and Schaub (2012) book and survival and recapture probabilities. For fecundity
in many published studies that provide the annotated data, a simple Poisson regression model is generally
scripts (see e.g. Gamelon et al. 2016, 2019). used.
14.2.3 Joint likelihood

14.2 How to build an IPM? The third and last step to build an IPM is to define the
The three main steps required to build an IPM are: (1) joint likelihood, that is, the likelihood of the integrat-
define the population model that best represents the ed model. Assuming independence among the data
ecology of the species and links the vital rates to pop- sets, the likelihood of the IPM is given by the prod-
ulation sizes; (2) define the likelihoods for each data uct of the likelihoods of the different data sets. But
set to estimate the vital rates (e.g. survival, fecundity); often, the same individuals are found in different data
and (3) define the joint likelihood of all data sets com- sets, and so the assumption of independence might be
bined. Here, we provide a brief overview of these three violated. While the data types (e.g. CMR data) might
steps. Note that we will not mention any goodness- determine the extent of the independence issue, a sim-
of-fit tests in this chapter as we are still at early days ulation study of IPM performance has shown that the
for goodness-of-fit tests for IPMs (Besbeas and Morgan violation of the assumption of independence does not
2014). This is currently the main limitation associated have a strong impact on the performance of the estima-
with the use of the IPMs. tors and highlighted a gain in precision and accuracy
E F F I C I E N T U S E O F D E M O G R A P H I C D ATA : I N T E G R AT E D P O P U L AT I O N M O D E L S 247
when data sets such as population counts, fecundity, influence population density. As the study area is
and CMR data are analysed simultaneously into an surrounded by a matrix of potential suitable habi-
IPM (Abadi et al. 2010a). Note, however, that in the tats for blue tits, the population is open to immi-
case of JAGS/BUGS/NIMBLE/STAN there is no need gration and emigration. Females usually start repro-
to state the likelihood product explicitly. ducing at age 1 (i.e. in the second calendar year of
life). The number of recruits, that is, the number of
females produced that survive and breed in the nest
14.3 Case studies boxes the next year, is a common measure of breeding
In this section, we illustrate how IPMs can be used to success. Three types of demographic data were col-
address major questions in demography in both ani- lected in this population between 1990 and 2017: (1)
mals and plants. The blue tit is our first case study. CMR data, (2) population counts, and (3) fecundity
This is a short-lived bird species, abundant in Euro- data.
pean gardens and woodlands as year-round residents. Beech mast is an important food resource for blue
From a unique long-term monitoring of a Dutch blue tits, especially during winter when other resources are
tit population, we show how different data sources can scarce. It is also indicative of seed production of other
be analysed simultaneously and how such an integrat- tree species (Perdeck et al. 2000). The beech crop index
ed analysis might help us to obtain a comprehensive (BCI), measured as the net weight of all nuts per square
picture of its dynamics. While most of the IPMs pub- metre, was recorded from 1990 to 2017.
lished so far have been built for animal species, our CMR data
second case study on common ragweed shows that
IPMs may also be applicable and very useful in the Each year, nest boxes were visited at least once per
plant kingdom. week during the breeding season (April to June), and
all young were ringed. The ringed adult females (pre-
viously caught and marked at the nest as young) were
14.3.1 Blue tit identified, and the unringed adult females were giv-
en a ring, allowing future identifications. Unringed
Main questions tackled adult females were assumed to have immigrated into
Different questions can be addressed in a Dutch blue tit the population in the current year. It was possible
Cyanistes caeruleus population while using an IPM. We to exactly age most of these immigrants. Overall,
provide below a (nonexhaustive) list of questions: CMR data were available for 708 breeding females of
known age (locally fledged females plus females that
• What are the ‘true’ annual population sizes? fledged elsewhere but immigrated to the study area to
• Do the ‘true’ annual population sizes differ from the breed).
annual population counts?
• What are the annual age-specific numbers of Population counts
females/survival rates/recruitment rates/recapture The second type of demographic data available was
rates? an estimate of the total number of counted breeding
• How does beech crop size influence age-specific sur- females each year t (Ct ). As females start to reproduce
vival rates/recruitment rates? at age 1 with a high breeding rate, breeding population
size is a good proxy for the total number of females
Data collection of age 1 or older. The population count Ct at year
The data analysed come from a long-term study of t corresponds to the sum of the number of breeding
a blue tit population at Hoge Veluwe National Park females of known and unknown age at year t and an
in the Netherlands (52◦ 02′N, 5◦ 51′E). The blue tit is estimate of the number of breeding females not caught
a small short-lived sedentary passerine bird species. (and thus not identified nor aged) because they have
This cavity-nester nests in boxes, making it possi- deserted their clutches early in the breeding attempt
ble to monitor the entire breeding population as at year t. The number of females of the latter type is
long as more nest boxes than required are provided. the number of clutches found without the mother being
Only a few females in the study area bred in natu- identified.
ral cavities, partly because 50 out of 450 nest boxes
have a small entrance hole only accessible to blue Fecundity data
tits. More nest boxes than required were provided The last type of data was age-class-specific counts
to ensure that the availability of nest boxes did not of breeding females and the age-class-specific
contribution to recruitment. The observed number of F4

female fledglings in year t that successfully became F3
F1
a first-year breeding female in year t+1 corresponds F2
to the recruitment for year t ( Jt ). This recruitment N1 n1 N2 n2 n3 N4 n4
N3
could be broken down by the age class of the adult A1 I1 S1 A2 I2 S2 A3 I3 S3 A4 I4
female: first-year breeder, second-year, and so on.
This provided estimates of the number of recruits
S4
by adult females of age class i in year t ( Ji,t ). We also
recorded the total number of breeding females of
Figure 14.1 Life cycle of Hoge Veluwe blue tit population. Each
each age class i in year t (Ri,t ). In total, 89 daughters arrow describes an annual transition from age i to age i + 1. Ni is the
of known age adult females were locally recruited age-specific number of breeding females, which is the sum of the
during the study period. Note that in most of the number of local breeding females (ni ), the number of immigrant
examples available in the literature, fecundity and females of known age (Ai ), and the number of immigrant females of
first-year survival are structural factors, whereas in the unknown age (Ii ); Si is the age-specific survival; and Fi is the
current example they are combined in the recruitment age-specific recruitment.
rate.
The integrated population model immigrants of age class i (added as a deterministic

number), and Ii,t was the number of immigrants of
We aimed to estimate the annual age-specific survival
age class i but for which it was impossible to give an
and recruitment rates as well as the true annual age-
exact age.
specific numbers of females. Population counts con-
To account for demographic stochasticity, especially
tain several sources of observation error. There were
important in small populations (Lande et al. 2003), we
females for which the age class was unknown (some
used Poisson and binomial processes to describe the
immigrants). Moreover, not all females were recap-
number of local breeding females in each age class
tured, resulting in errors and uncertainty in the number
(ni,t ) : n1,t+1 ~ Poisson(N1,t × F1,t + N2,t × F2,t + N3,t × F3,t +
of females in the different age classes. This also leads
N4,t × F4,t ) , n2,t + 1 ~ Binomial (N1,t ,S1,t ), n3,t + 1 ~ Binomial
to errors and uncertainty in the estimates of survival
(N2,t ,S2,t ) and n4,t+1 ~Binomial(N3,t ,S3,t ) + Binomial(N4,t ,
and recruitment. In addition, there was a possibility
S4,t ),where Fi and Si were the age-specific recruitment
of double counts: for instance, if one female produced
and survival rates, respectively. For the immigrant
two clutches but was only identified in one of them
females Ii,t for which it was impossible to give an exact
because she had deserted one of the clutches. Lastly,
age, we considered that each year t, a random pro-
some clutches could have been missed because females
portion of them are females of age class i. Modelling
bred in natural cavities within the population instead
the number of immigrants as a function of population
of breeding in nest boxes. To account for these issues,
size would have led to biased estimates (Schaub
we jointly analysed CMR data of known-age females,
and Fletcher 2015). This also allowed considering
population counts (Ct ), and fecundity data ( Ji,t and Ri,t )
uncertainty on the age of these immigrant females
using an IPM.
without making any explicit assumption about their
exact age. Females were classified as immigrants only
Population model in their year of arrival into the study population and
Based on the life history of blue tits, we considered a then joined the local population. The true number of
prebreeding age-structured female model with four breeding females in the population at year t (Ntot,t )
age classes: age class 1 corresponded to the first year was then ΣNi,t (Figure 14.1).
of breeding, age class 2 to the second year, age class
3 to the third year, and age class 4 to older breeding Likelihood functions separately for each data set
females. A very low proportion of individuals were The likelihood functions were specified for the differ-
in the oldest age class (about 13%). We assumed that ent data sets as follows. For CMR data of breeding
reproduction started when females were 1 year old. females of known age, we used the Cormack–Jolly–
The total number of breeding females of age class i in Seber model (Lebreton et al. 1992) which allowed esti-
year t was denoted Ni,t and defined as Ni,t = ni,t + Ai,t + mation of annual survival between age class i and
Ii,t , where ni,t was the number of local females in each i+1 (Si,t ) and recapture (pt ) probabilities. For fecun-
age class i in year t, Ai,t was the number of known age dity data, we assumed that the observed number of
female recruits per age class i ( Ji,t ) was Poisson dis- Gamelon et al. 2016 for a similar approach). We used (1)
tributed with Ji,t ~ Poisson(Ri,t × Fi,t ), where Ri,t was recruitment specified on the log scale log(Fi,t ) = μF i +
the observed number of breeding females of age class εF i,t , where μF i was the mean contribution of adult
i and Fi,t was the recruitment rate of females of age females of age class i to recruitment next year and εF i,t
class i at year t. Fi,t was the term we were estimating was the age-specific temporal residual; (2) the logit link
and was the contribution of adult females of age for survival so that logit(Si,t ) = μS i + εS i, . ε was the
class i to recruitment into the age class 1 breeding matrix including the temporal residuals of the eight
class next year. Note that emigration (which mostly vital rates (i.e. one temporal residual for recruitment
occurs within the first year of age) and mortality in rate εF i,t and one temporal residual for survival rate
the first year of life were included in the recruitment εS i,t per age class i). We used a multivariate normal
estimate. (MVN) distribution ε ~ MVN(0, Σ), where Σ was the
For the population count data, we used a state-space variance–covariance matrix allowing correlated vari-
model (de Valpine and Hastings 2002) which consisted ability among vital rates; and (3) recapture probability,
of a process model (i.e. the population model described assumed to be age-independent, modelled with ran-
in the previous section) and an observation model. dom time variation as well as log(pt ) = μP + εP t .
This latter described the link between the population The IPM was fit within a Bayesian framework,
counts Ct and the true number of females in the popu- and noninformative priors were specified for all the
lation (Ntot,t ). We assumed that Ct ~ Normal(Ntot,t , σ 2 c ) parameters. MCMC simulation was used for param-
truncated to positive values, which incorporated obser- eter estimation. To assess convergence, we ran four
vation error of the counts. The observation model independent chains with different starting values for
accounted for both count errors (females unobserved 200,000 MCMC iterations, with a burn-in of 150,000
or double-counted) and lack of fit of the state equations iterations, thinned every 100th observation, resulting
to the true dynamics of the population (Schaub and in 2,000 posterior samples. We used the Brooks and
Abadi 2011). Gelman diagnostic R̂ to assess the convergence of
the simulations and used the rule R̂ < 1.02 to deter-
Joint likelihoods mine whether convergence was reached (Brooks and
Assuming independence among the data sets, the like- Gelman 1998). The analyses were implemented using
lihood function of the IPM was the product of the JAGS (Plummer 2003) called from R (R Development
likelihoods specified for each of the three different data Core Team 2017) with package R2jags (Su and Yajima
sets, that is, population counts, fecundity data, and 2012).
CMR data (Figure 14.2). Our aim was to estimate annu- As mentioned earlier, IPMs are appropriate tools for
al vital rates, which were thought to originate from a hypothesis testing on all vital rates. As an example, out-
random process with a common mean and constant side the IPM, we then explored the effect of beech crop
temporal variance (see e.g. Schaub et al. 2012, 2013; resources BCI on age-specific survival and recruitment
Count data
State-space model
CMR data Fecundity data Figure 14.2 Directed acyclic graph of

Cormack-Jolly-Seber model Poisson model the IPM. Squares represent the data,
circles represent the parameters to be
p Si ω Fi Ri estimated, and dashed lines represent the
individual submodels. Arrows represent
m Ji dependencies. Three types of data are
collected: CMR data (m), count data (C),
and fecundity data (number of daughters
N=
ΣNi
locally recruited (J) and number of mothers
(R) in each age class i). Estimated
parameters are the recapture probability p,
age-specific survival Si , age-specific
c recruitment Fi , immigration rate ω, and
age-specific numbers Ni .
rates. We modelled the effect of BCI on survival and population growth rate were variable, the mean pop-
recruitment as follows: ulation growth rate was 1.024 (95% CRI 1.010, 1.036),
indicating an overall increasing population trend. The
logit (Si,t ) = αSi + βSi × BCIt + resSi,t , population is dominated by recruits (age class 1),
log (Fi,t ) = αFi + βFi × BCIt + resFi,t , which show high temporal variation in their numbers
(Figure 14.3b). The remaining age classes (2 to 4) are
where α were the intercepts, res the residuals of the approximately equally represented and show less tem-
regressions, and β i the age-specific effect of BCI on poral variation (Figure 14.3b).
survival and recruitment. Annual survival estimates vary among ages and
The data and the JAGS code for the analyses are show high temporal variation (Figure 14.3c). Mean sur-
freely available. vival probability increased with age: from 0.277 (95%
CRI 0.234, 0.327) in age class 1 to 0.432 (95% CRI 0.348,
Results 0.520) in age class 2, 0.440 (95% CRI 0.311, 0.573) in
The estimated total number of breeding females (i.e. age class 3, and 0.504 (95% CRI 0.282, 0.675) in age
the ‘true’ annual population sizes) correlates well class 4. Annual recruitment estimates also show vari-
with the population counts (Pearson correlation coef- ation among ages and through time (Figure 14.3d).
ficient r = 0.782 [95% CRI 0.720, 0.834]), and fluc- Mean recruitment rate is the lowest for age class 1
tuates between 25.91 (95% CRI 18.00, 35.00) and (0.130; 95% CRI 0.097, 0.169), higher for age class 2
122.69 (95% CRI 106.98, 140.00) over the study period (0.222; 95% CRI 0.151, 0.311) and age class 3 (0.209;
(Figure 14.3a). Although the annual estimates of the 95% CRI 0.089, 0.395), and the highest for age class 4
(a) (b) Age class

150 Estimated Population counts 80
1 2 3 4+
60
100
40
N
50
20
0 0
90
95
00
05
10
15
90
95
00
05
10
15
19
19
20
20
20
20
19
19
20
20
20
20
Year Year
(c) Age class (d) Age class (e)

1.0 1.2 1.0
1 2 3 4+ 1 2 3 4+
1.0
Recapture probability
0.8
Recruitment rate
0.8
0.8
0.6 0.6
0.6
0.4 0.4
0.4
0.2 0.2 0.2
0.0 0.0 0.0

90
95
00
05
10
15
90
95
00
05
10
15
90
95
00
05
10
15
19
19
20
20
20
20
19
19
20
20
20
20
19
19
20
20
20
20
Year Year Year
Figure 14.3 Annual estimates for the Hoge Veluwe blue tit population resulting from the IPM between 1990 and 2017. (a) Annual population
counts (blue dots) and annual posterior means and associated 95% CRI of the estimated total number of breeding females (grey area).
(b) Posterior means of the annual age-specific numbers of breeding females. (c) Posterior means of the annual age-specific survival probabilities.
(d) Posterior means of the annual age-specific recruitment rates. (e) Posterior means and associated 95% CRI of the annual recapture probabilities.
(0.315; 95% CRI 0.137, 0.569). Annual estimates of and spring seed bank densities. We explain each of the
the recapture probability were approximately constant collected data sets below.
throughout the study period (Figure 14.3e), with a
mean recapture probability of 0.745 (95% CRI 0.647, Seedling data
0.828). On 12 June 2014, the number of seedlings was recorded
We generally find a positive effect of BCI on survival in 13 square 0.25 m2 plots. The timing of this first cen-
and recruitment rates, and this effect is age-specific for sus was chosen such that the peak of germination had
survival. In particular, β S is estimated to be 0.028 (95% already taken place, but later germination did occur.
CRI 0.007, 0.049) for age class 1, 0.039 (95% 0.018, 0.061) Per plot, 4–15 seedlings were marked with aluminium
for age class 2, 0.048 (95% CRI −0.011, 0.109) for age tags to follow their survival and maturation till seed
class 3, and 0.002 (95% CRI −0.029, 0.031) for age class set.
4. For recruitment rates, we find that β F equals 0.127
(95% CRI 0.065, 0.189) for age class 1, 0.104 (95% CRI Plant data
0.049, 0.158) for age class 2, 0.112 (95% CRI 0.052, 0.170) On 16 September 2014, the number of plants per
for age class 3, and 0.107 (95% CRI 0.053, 0.161) for age plot (including both individuals that were already
class 4. present in the first census and ones that had germi-
nated since the first census) was determined. Whether
14.3.2 Common ragweed individually marked plants had survived was also
recorded.
Data collection
The second case study is the estimation of seed bank Seed bank data
density, germination, and survival rates of the inva-
At the time of the second census, 16 September 2014,
sive plant Ambrosia artemisiifolia, also known as com-
soil samples were taken to estimate the number of
mon ragweed. Common ragweed is an annual plant
seeds per square metre (see protocol outlined in Lom-
that flowers in early autumn and overwinters as seed
men et al. 2018a). As the soil samples were taken before
in the soil seed bank. Seeds in the soil seed bank
seed rain, the soil samples did not contain fresh seeds
can stay alive for decades. The species is problematic,
but only 1-year-old or older seeds.
as high densities of the plant can reduce agricultur-
al crop yield and the pollen are allergenic to humans Main questions tackled
(Essl et al. 2015). To study the spatial and tempo-
While Integrated Population Models are rarely used in
ral variation in population dynamics, Lommen et al.
plant demography, mechanistic models fitted to mul-
(2018b, 2018c) designed a demographic field study that
tiple data sets are very useful for estimating vital rates
involved multiple European countries. Here we use
that have not been estimated directly. Here we aim to
data from one field season (2014) in Magnago, Italy, as
use an IPM, and the data introduced above, to answer
an example.
the following questions:
Two population censuses (after the majority of the
seedlings has established: mid-June census, and at • What is the seed density (number per square metre)
seed set at the end of the summer: mid-September in the soil in spring (i.e. just before germination
census) are performed in 13 square 0.25 m2 plots. starts)?
The total number of plants is counted at each cen- • What is the per-seed establishment rate by the time
sus, but only a random subset of the young plants of the first census?
is marked to record their fate at the end of summer. • What is the per-seed establishment rate between the
Additionally, soil samples are taken adjacent to half two censuses?
of the plots at the end of summer (before seed rain) • What is the probability that a seedling survives up
to monitor seed bank density. The major challenge to flowering (i.e. the second census)?
with this field approach is that in order to estimate
seedling establishment rates (an important aspect in Estimation of these vital rates is key to the
the population dynamics of common ragweed), seed construction of periodic population models, and this
densities at the start of the growing season need to estimation can only be properly done by combining
be estimated. Our proposed model here integrates all three data sets. The IPM could be extended by
seasonal plant data and seed bank data before seed incorporating plant growth and size-dependent seed
rain at the end of summer to provide simultaneous production (and specifically the influx of fresh seeds
estimates of germination rates, seedling survival rate, into the soil seed bank), which would also allow the
June September
census census
S S
Eearly Elate
(1– Eearly ) (1– Elate ) Figure 14.4 Schematic of the population

Spring Autumn dynamics of common ragweed during the
Time growing season.
estimation of seed survival over winter. However, to soil samples in September (i.e. before influx of fresh
keep things tractable, we here focus on the processes seeds) follows a binomial distribution, with DSep ~
that happen during the growing season. Binomial(DsoilSampleSpring , (1–Eearly ) × (1–Elate )). In this
function, DsoilSampleSpring is the unmeasured number
The integrated population model of seeds in spring in a soil area (1/62.5 m2 ) equal to
In order to simultaneously estimate seedling establish- that of the soil samples collected in September. For
ment rates, seedling survival, and spring seed densi- the realisation of DsoilSampleSpring we assume a Poisson
ties, we constructed a simple model of the dynamics distribution: DsoilSampleSpring ~ Poisson(DSpring /62.5).
of a common ragweed population over summer. As For the observed seedling densities in June (LJune ) we
shown in Figure 14.4, our model keeps track of the assumed a binomial process: LJune ~ Binomial(0.25 ×
number of seeds, seedlings, and plants over the course DSpring , Eearly ). The function for the third data set,
of the growing season. In spring the entire population the number of plants per plot in September, is
is composed of seeds in the soil, but by mid-June a more complex because three different groups of
proportion of those seeds have germinated and suc- plants could have contributed to the total number
cessfully established themselves as seedlings: LJune = of plants in the plots: PSep ~ Poisson(Pearly,marked +
0.25 × DSpring × Eearly , where Eearly is the probability Pearly,unmarked + Plate,unmarked ). Plants of these three
of seedling establishment per seed, DSpring the densi- groups differed in whether they were marked in
ty of seeds in the soil (per square metre), and LJune the June or not and whether unmarked plants had
number of seedlings per 0.25 m2 plot during the first germinated early or late. The latter distinction cannot
census. A similar process is modelled for late seedling be made when observing the plants in September,
establishment (even though seedlings that established stressing once more the usefulness of an IPM. For
after mid-June were not recorded until the final census each of the three contributions to PSep we assumed a
in September): Llate = 0.25 × DSpring × (1–Eearly ) × Elate . binomial process: Pearly,marked ~ Binomial(LJune,marked , S);
After establishment, both early and late seedlings have Pearly,unmarked ~ Binomial(LJune,unmarked , S); and Plate,unmarked
a certain probability of surviving and being counted as ~ Binomial(DSpring /4 – LJune , Elate × S). This combi-
plants in September: PSep = (Learly +Llate ) × S, in which nation of a Poisson process with three underlying
S is the seedling survival rate and PSep the number of binomial processes gives the optimisation algorithm a
plants per plot in September. lot of flexibility to find the most likely set of parameter
values given these three data sets.
Likelihood functions Please note that we made a few simplifying assump-
We wrote a mechanistic JAGS model (R code provided) tions in this modelling exercise. For instance, we
based on the population model shown in Figure 14.4 assumed that seeds do not die during the growing sea-
in order to estimate the unknown parameters DSpring , son. Soil samples were taken close to the demographic
Eearly , Elate , and S using the combined information in plots (which were operated for 3 years), so we assumed
the soil, seedling, and plant data sets. We assumed that seed densities are homogeneous over such short
that the observed number of viable seeds in the distances (but see Detto et al. 2019 for cautionary
(a) (b)
25
20 20
Density
15 15
Density
10 10
5 5
0 0
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
Establishment rate before Establishment rate after
June census June census
(c) (d)
2000
10 Seed per m2 in spring
1000 Figure 14.5 Posterior distributions of
8 500 the estimated parameters of the Ambrosia
Density
6 artemisiifolia IPM. (a) Eearly , the seedling

200
4 establishment rate before the June
100
census; (b) Elate , the seedling establish-
2 50 ment rate after the June census; (c) S, the
0 20 seedling-to-plant survival rate; and
0.5 0.6 0.7 0.8 0.9 1.0 1 3 5 7 9 11 13 (d) DSpring , the seed density in the soil in
Seedling-to-plant survival rate Plot number spring (i.e. before germination).
advice on such matters). Also, we fitted the same sur- new generation of IPMs includes both an integral
vival rate to seedlings prior and after the June census projection model (Chapter 10) and an integrated pop-
(i.e. no phenological dependent survival rates). The ulation model (Plard et al. 2019b). Here, the integrated
IPM was fitted in a very similar way as the blue tit mod- approach was especially useful because in the Septem-
el above (see technical details in the provided R code). ber plant count it was impossible to tell individuals
apart that had established before or after the June cen-
Results sus, and because soil seed density estimates were only
The posterior distributions of the fitted model param- available for 5 of the 13 plots and not for spring. When
eters are narrow, suggesting that the IPM has con- comparing these Eearly and Elate estimates with those of
verged on a set of parameter values that is very like- other populations or years, one has to realise that the
ly, given the observations. This notion is confirmed timing of the first census can have a large effect on ratio
by looking at the stability and similarity of the three of these two seedling establishment probabilities. Now,
separate chains that were run. Early seedling estab- Eearly was larger than Elate , meaning that the moment of
lishment, Eearly , was estimated at 0.122 (95% CRI 0.091, the census was correctly chosen according to protocol:
0.160), which was higher than the probability of late after the germination peak.
establishment, Elate : 0.055 (95% CRI 0.028, 0.091). Sub-
sequent seedling survival, S, was estimated at 0.795
(95% CRI 0.720, 0.859). The plot-specific estimates of Acknowledgements
soil seed densities in spring varied between 159 and We are grateful to all those who have collected data
1,367, with a median of 546 seeds per m2 . The average on blue tits Cyanistes caeruleus at Hoge Veluwe as well
(among the plots) standard deviation was 151, indicat- as the board for their permission to work in their
ing that although variable, also the seed densities were forest. We are also grateful to all those who have
estimable (Figure 14.5). collected data on Ambrosia artemisiifolia in Magnago,
Italy. We acknowledge financial support from the EU
Discussion of the plant example COST Action FA1203 ‘Sustainable management of
These results can be used to parameterise pop- Ambrosia artemisiifolia in Europe (SMARTER)’ and from
ulation projection models (see Chapter 9) or to the Research Council of Norway through its Centres of
parametrise integrated integral projection models. This Excellence funding scheme (project number 223257).
References Lebreton, J.-D., K. P. Burnham, J. Clobert, and D. R.

Anderson. 1992. Modeling survival and testing biological
Abadi, F., O. Gimenez, R. Arlettaz, and M. Schaub. 2010a. An
hypotheses using marked animals: a unified approach
assessment of integrated population models: bias, accu-
with case studies. Ecological Monographs 62:67–118.
racy, and violation of the assumption of independence.
Lee, A. M., E. M. Bjørkvoll, B. B. Hansen, et al. 2015. An
Ecology 91:7–14.
integrated population model for a long-lived ungulate:
Abadi, F., O. Gimenez, B. Ullrich, R. Arlettaz, and M. Schaub.
more efficient data use with Bayesian methods. Oikos
2010b. Estimation of immigration rate using integrat-
124:806–816.
ed population models. Journal of Applied Ecology 47:
Lieury, N., M. Gallardo, C. Ponchon, A. Besnard, and
393–400.
A. Millon. 2015. Relative contribution of local demogra-
Arnold, T. W., R. G. Clark, D. N. Koons, and M. Schaub. 2018.
phy and immigration in the recovery of a geographically-
Integrated population models facilitate ecological under-
isolated population of the endangered Egyptian vulture.
standing and improved management decisions. Journal of
Biological Conservation 191:349–356.
Wildlife Management 82:266–274.
Lieury, N., A. Besnard, C. Ponchon, A. Ravayrol, and
Besbeas, P., and B. J. T. Morgan. 2014. Goodness-of-fit of A. Millon. 2016. Geographically isolated but demographi-
integrated population models using calibrated simula- cally connected: immigration supports efficient conserva-
tion. Methods in Ecology and Evolution 5:1373–1382. tion actions in the recovery of a range-margin population
Besbeas, P., S. N. Freeman, B. J. T. Morgan, and of the Bonelli’s eagle in France. Biological Conservation
E. A. Catchpole. 2002. Integrating mark–recapture– 195:272–278.
recovery and census data to estimate animal Lommen, S., C. A. Hallmann, B. Chauvel, et al. 2018a.
abundance and demographic parameters. Biometrics 58: Field survey of the population dynamics of common
540–547. ragweed (Ambrosia artemisiifolia). Protocols, doi: dx.doi.
Brooks, S. P., and A. Gelman. 1998. General methods org/10.17504/protocols.io.mmyc47w.
for monitoring convergence of iterative simulations. Lommen, S. T. E., C. A. Hallmann, E. Jongejans, et al.
Journal of Computational and Graphical Statistics 7: 2018b. Explaining variability in the production of seed
434–455. and allergenic pollen by invasive Ambrosia artemisiifolia
Chandler, R. B., and J. D. Clark. 2014. Spatially explicit across Europe. Biological Invasions 20:1475–1491.
integrated population models. Methods in Ecology and Lommen, S. T. E., E. Jongejans, M. Leitsch-Vitalos, et al.
Evolution 5:1351–1360. 2018c. Time to cut: population models reveal how to
De Valpine, P., and A. Hastings. 2002. Fitting population mow invasive common ragweed cost-effectively. NeoBio-
models incorporating process noise and observation error. ta 39:53–78.
Ecological Monographs 72:57–76. Maunder, M. N., and A. E. Punt. 2013. A review of integrated
Detto, M., M. D. Visser, S. J. Wright, and S. W. Pacala. 2019. analysis in fisheries stock assessment. Fisheries Research
Bias in the detection of negative density dependence in 142:61–74.
plant communities. Ecology Letters 22:1923–1939. Oppel, S., G. Hilton, N. Ratcliffe, et al. 2014. Assessing pop-
Essl, F., K. Biró, D. Brandes, et al. 2015. Biological flora of ulation viability while accounting for demographic and
the British Isles: Ambrosia artemisiifolia. Journal of Ecology environmental uncertainty. Ecology 95:1809–1818.
103:1069–1098. Perdeck, A. C., M. E. Visser, and J. H. van Balen. 2000. Great
Gamelon, M., V. Grøtan, S. Engen, E. Bjørkvoll, M. E. Visser, tit Parus major survival and the beech-crop cycle. Ardea
and B. Sæther. 2016. Density dependence in an age- 88:99–106.
structured population of great tits: identifying the critical Plard, F., R. Fay, M. Kéry, A. Cohas, and M. Schaub.
age classes. Ecology 97:2479–2490. 2019a. Integrated population models: powerful methods
Gamelon, M., S. J. G. Vriend, S. Engen, et al. 2019. Account- to embed individual processes in population dynamics
ing for interspecific competition and age structure in models. Ecology 100:e02715.
demographic analyses of density dependence improves Plard, F., D. Turek, M. U. Grüebler, and M. Schaub. 2019b.
predictions of fluctuations in population size. Ecology IPM2: toward better understanding and forecasting of
Letters 22:797–806. population dynamics. Ecological Monographs 89:e01364.
Kéry, M., and M. Schaub. 2012. Bayesian Population Analy- Plummer, M. 2003. JAGS: a program for analysis of Bayesian
sis using WinBUGS: A hierarchical perspective. Academic graphical models using Gibbs sampling. Pp. 20–22 in Pro-
Press, Boston. ceedings of the 3rd International Workshop on Distributed
Koons, D. N., D. T. Iles, M. Schaub, and H. Caswell. 2016. Statistical Computing. Hornik K, Leisch F, Zeileis A (eds),
A life-history perspective on the demographic drivers Vienna, Austria.
of structured population dynamics in changing environ- R Development Core Team. 2017. R: a language and
ments. Ecology Letters 19:1023–1031. environment for statistical computing. https://www.
Lande, R., S. Engen, and B.-E. Sæther. 2003. Stochastic Pop- R-project.org/.
ulation Dynamics in Ecology and Conservation. Oxford Schaub, M., and F. Abadi. 2011. Integrated population
University Press, Oxford. models: a novel analysis framework for deeper insights
into population dynamics. Journal of Ornithology 152: Su, Y., and M. Yajima. 2012. R2jags: a package for running
227–237. jags from R. R package version 0.03–08.
Schaub, M., and D. Fletcher. 2015. Estimating immigration Tenan, S., A. Iemma, N. Bragalanti, et al. 2016. Evaluat-
using a Bayesian integrated population model: choice of ing mortality rates with a novel integrated framework
parametrization and priors. Environmental and Ecologi- for nonmonogamous species. Conservation Biology 30:
cal Statistics 22:535–549. 1307–1319.
Schaub, M., T. S. Reichlin, F. Abadi, M. Kéry, L. Jenni, and van Oosten, H. H., C. van Turnhout, C. A. Hallmann, et al.
R. Arlettaz. 2012. The demographic drivers of local pop- 2015. Site-specific dynamics in remnant populations of
ulation dynamics in two rare migratory birds. Oecologia northern wheatears Oenanthe oenanthe in the Netherlands.
168:97–108. Ibis 157:91–102.
Schaub, M., H. Jakober, and W. Stauber. 2013. Strong con- Zipkin, E. F., and S. P. Saunders. 2018. Synthesizing mul-
tribution of immigration to local population regulation: tiple data types for biological conservation using inte-
evidence from a migratory passerine. Ecology 94:1828– grated population models. Biological Conservation 217:
1838. 240–250.
PART 3
Applications
CHAPTER 15
Spatial demography
Guillaume Péron
15.1 Introduction rate is thus key to understanding the spatial func-

tioning of populations, pinpoint the areas deserving
Eco-evolutionary processes are rooted in the way fit-
of conservationists’ attention, infer whether dispers-
ness varies across space and the way individuals imple-
ing individuals are using the population density or
ment dispersal strategies in response to that variation.
breeding performance of their conspecifics to decide
As climate and land use changes redistribute the abi-
where to settle, and determine gene flow from source–
otic drivers that influence demography, the impor-
sink dynamics. I will first review the generic nonpara-
tance of spatial aspects in eco-evolutionary dynamics
metric statistical tools to estimate spatial variation in
is increasingly obvious (Thuiller et al. 2013). In partic-
any variable (section 15.2.1) and provide a nonexhaus-
ular, the balance between extinctions and colonisations
tive review of recent nonparametric analyses pertain-
contributes to range shifts in metapopulations (Hans-
ing to spatial demography (section 15.2.2). Paramet-
ki and Gyllenberg 1997; Gandon and Michalakis 1999;
ric approaches will only be briefly discussed (section
McCauley et al. 2013). In the broader community con-
15.2.3) because they have been reviewed elsewhere and
text, the balance between local adaptation (Bohonak
extensively (e.g., Dormann et al. 2007; Beale et al. 2010).
1999; McRae et al. 2008; Burton et al. 2010) and dis-
persal/movement also creates ways for species and
individuals to coexist by implementing different space 15.2.1 Statistical tools for nonparametric
use tactics (Leibold et al. 2004; Wolf et al. 2007; Péron inference about spatial variation
et al. 2019).
The objective in this section is to provide an overview
This chapter reviews the analysis of demograph-
of often complex statistical techniques and point the
ic data for inference about spatial variation in fitness
reader in the right direction, hopefully.
components and about dispersal strategies (Table 15.1).
First, at the data exploration stage, users may want
It also includes a section about prospective analysis,
to use Moran’s index of autocorrelation to deter-
that is, the study of model properties for predictive
mine whether there is spatial structure in their data
inference (section 15.4: the feedback between space use
(Moran 1950; Cressie 1993). Moran’s index detects
and fitness).
the presence of a systematic similarity between data
points that are close together compared to data points
15.2 Spatial variation in that are further apart. In practice, one may use the
demographic rates function Moran from the raster package for R if
the data are in raster format and assuming a linear
The intrinsic population growth rate, the balance between
decay in autocorrelation strength with Euclidian dis-
local births and local deaths, documents whether a
tance; or the function moran from the spdep package
place acts as a population source that contributes to the
to accommodate less regular samples (Bivand et al.
overall increase in population (Pulliam 1988) or a pop-
2013).
ulation sink likely to attract more immigrants than it
I will now review the three main options for the
emits emigrants (Novaro et al. 2005). Mapping out
analysis of spatial variation in demographic data. Note
the spatial variation in intrinsic population growth
Guillaume Péron, Spatial demography. In: Demographic Methods Across the Tree of Life.
Table 15.1 List of the spatial demography methods reviewed in this chapter. Note that this chapter does not consider
movement tracking data or genetic data, only demographic data in the sense of information about vital rates and
population abundance.
Spatial variation in demographic rates
Method name Objectives Section
Splines Continuous spatial variation in a demographic rate, nonparametric 15.2.1

(no environmental predictor needed)
Autoregressive models Continuous variation, nonparametric 15.2.1
Random forest algorithm Continuous variation, nonparametric 15.2.1
Linear models Continuous or discrete variation, parametric (environmental 15.2.3
predictor needed)
Multisite capture–recapture Discrete spatial variation in survival probability (i.e. analysis of the 15.3.2
model differences between predefined population units)
Integrated metapopulation model Same as above but ability to estimate fecundity too 15.3.2
Dispersal fluxes
Method name Objectives Section
Integrated population model The net flux of individuals into or outside of a focal population 15.3.1
Multisite capture–recapture The rates at which individuals disperse between discrete population 15.3.2
model units
Integrated metapopulation model Same as above but increased statistical power and spatial extent 15.3.2
Diffusion equation Landscape resistance to population homogenisation in a 15.3.3
continuously distributed population
that these methods can be applied directly to spatial- new objects into bags without a priori knowledge of the
ly explicit sets of fecundity data or population survey dependent variable and only knowing the features of
data, but, to apply them to capture–recapture data for the object (e.g. ‘determine palatability based on colour
inference about survival and lifespan (cf. Chapter 13), and shape’). The main issue with this type of approach
they must first be incorporated into the appropriate is overfitting, for example, when the tree uses more fea-
state-space model that deals with imperfect detection tures than supported by the training data and identifies
of individuals (Gimenez et al. 2006; Péron et al. 2011). small-sample artefacts rather than meaningful trends.
To prevent overfitting without losing information, a
Random forest algorithms random forest is a group of decision trees each based
Also known as hierarchical clustering algorithms, this on a reduced subset of features. The prediction is the
class of method belongs to the data mining and consensus between all the trees in the forest. For the
machine learning movement. It owes its name to its random forest to perform better than a single deci-
building blocks which are decision trees (Breiman sion tree, we simply need some of the features to be
2001). A decision tree is a sequence of multiple-choice meaningful, that is, related to the predicted variable,
questions based on clustering criteria or features (e.g. ‘Is and that the predictions made by individual trees have
the object red, blue, or green?’, ‘Is it a square or a cir- low correlations with each other (Breiman 2001). The
cle?’). The training dataset is grouped into bags accord- latter condition is usually enforced by training each
ing to these features (e.g. bag 1 = ‘green squares’). The tree in the forest with a different subset of the data
value of the dependent variable in each bag is comput- and by selecting the features of each tree at random
ed from the training data (e.g. ‘bag 1 was good to eat’). irrespective of their predictive power.
The decision tree is fitted to the training data using tra- When applied to spatial demography, the key
ditional statistics (e.g. least square criterion or receiver strength of the random forest framework is the ability
operating characteristic curve). The interest then lies in to input ancillary information in the form of spatial
the predictive power of the tree, that is, its ability to bin covariates. If the covariates are meaningful and well
S PAT I A L D E M O G R A P H Y 261
chosen, the algorithm makes it possible to downscale the fact that the criterion C (b, β, λ) is the same as that
or disaggregate the demographic data at the resolution of a linear mixed model with predictor X, fixed-effect
of these ancillary covariates (Stevens et al. 2015). How- regression coefficient β, design matrix Z e = Z · Ω−1/2 ,
ever, the random forest algorithm also performs ade- Gaussian random effects u = Ω · b, and individual-
1/2
quately if provided with nothing more than the geo- ly and independently distributed Gaussian error term
graphical coordinates of the samples, in which case it ε such that Var (ε) = λ2 Var (u). This means that we
will look for clusters of adjacent samples with similar can fit the model using any software designed to fit
demographic properties (Brickhill et al. 2015; Rushing linear mixed effect models, as long as the software
et al. 2016). In practice, the random forest algorithm interface allows us to specify a custom design matrix.
for spatial data is implemented in, for example, the Implementations using Bayesian software are available
ranger package for R (Wright and Ziegler 2017). (Crainiceanu et al. 2005), including within capture–
recapture models that account for imperfect detection
Splines of individuals (Gimenez et al. 2006; Péron et al. 2011).
This section is going to be more technical than the rest
of the chapter. The principle of this family of tech- Autoregressive models
niques is to approximate the spatial variation in a vari- This section is also going to be more technical than
able Y by a sum of K unimodal functions with desirable the rest of the chapter. Note, however, that autoregres-
mathematical properties, which are termed the splines sive models are readily available in R packages spdep
(Ruppert et al. 2003). The main feature of the splines is and nlme (Pinheiro et al. 2013) and are extensively
their location, that is the position of their knot κk , which used in ecology (Dormann et al. 2007). The underlying
is typically decided using a data-driven space-filling assumption of this class of model is that the value of the
algorithm (Nychka and Saltzman 1998). The analytical target variable Y at location (x, y) can be predicted from
shape of the spline is less influencial. Hereafter, I will the value at nearby location (x+dx, y+dy) and inversely.
consider radial splines denoted D (κk , ·), where D is the There are two nonexclusive options to implement that
distance operator. The dependent variable Y at loca- principle (Cressie 1993).
tion (x, y), is then modelled through the link function ℓ,
P
as ℓ Y (x, y) = β0 + β1 x + β2 y + Kk=1 bk · D (κk , (x, y) ). • In a spatial lag model, the observations are directly
The b parameters represent the weights associated to regressed against each other: ℓ (Y) (x, y) = µ (x, y) +
each of the K splines. The β parameters represent clinal ρWℓ (Y) + ε (x, y), where µ is a mean term that usual-
variation. To fit that model, we need to minimise the ly depends on spatially explicit linear predictors, ρ is
least-square criterion C (b, β) = ∥ℓ (Y) − X · β − Z · b∥2 , the autocorrelation coefficient, W is a weight matrix
where Y contains the N observations of the focal vari- typically corresponding to the Euclidian distances
able, X is a N×3 matrix containing the coordinates at between samples, and ε contains individually and
the sampling locations preceded by a 1 for the inter- independently distributed (IID) error terms.
cept, and Z is a N×K matrix, with Zi,k the value of spline • In a spatial error model, the spatial autocorrelation
k at sampling location i. affects the error term only: ℓ (Y) (x, y) = µ (x, y) +
The main issue with this basic implementation of the e (x, y) with e (x, y) = λWe + ε (x, y). This type of
spline model is, like in the decision tree case, overfit- model is typically used when the sampling noise is
ting. To avoid giving too much weight to idiosyncratic spatially autocorrelated. If this autocorrelation in the
patterns occurring at the sampled locations, one typi- sampling error was not accounted for, the estimation
cally incorporates a penalisation term designed to force of µ, that is, the ecological process of how the depen-
the sum of the squared bs to stay below a thresh- dent variable Y depends on environmental covari-
old (Ruppert et al. 2003, p. 65). The criterion to be ates at location (x, y), would be biased, sometimes
minimised then becomes of the form C (b, β, λ) = severely (Beale et al. 2010). Importantly, however,
∥ℓ (Y) − X · β − Z · b∥2 +λ2 · bT · Ω · b. Following Ruppert spatial error models work under the assumption that
et al. (2003, p. 73), we use for Ω the K×K distance matrix only the sampling error is autocorrelated. If it is not
with Ωk,l = D (κk , κl ). λ is the smoothing parameter con- the case, that is, there is a biological process of spa-
trolling how much the idiosyncrasies will be smoothed tial autocorrelation, using a spatial error model will
out, to be estimated directly from the data themselves. assign too much spatial variance to the error and not
λ is akin to the bandwidth parameter in kernel den- enough variance to the mean.
sity estimators or the penalisation coefficient in lasso
regularised regressions. To estimate the parameters, In terms of R implementation, one may use the
following Ruppert et al. (2003, p. 108ff), we exploit functions lagsarlm for the spatial lag model and
errorsarlm for the spatial error model, both from Even survival studies are quite rare to the best of
package spdep. The spatial lag and spatial error mod- my knowledge. Saracco et al. (2010) used autoregres-
els can also be merged together to obtain the simulta- sive models to generate maps of survival probability
neous autoregressive models (SAR) that features both for the wood thrush (Hylocichla mustelina), a migratory
autocorrelation structures. There is also the condition- passerine. The way survival increased in the northern
al autoregressive model (CAR) or autologistic model, a and especially northwesternmost sections of the range
variant of the spatial lag model. CAR is often preferred is congruent with the observed increase in abundance
in ecological applications because of its flexibility and in the northwestern part of the range (Rushing et al.
ability to accommodate missing data (Yackulic et al. 2016) and aligns with predictions from climate change.
2012; Péron et al. 2016). CAR is obtained by designing The particularity of the Saracco et al. study is that they
the weight matrix according to user-specified lists of accommodated transience at the bird ringing locations,
‘neighbours’ for each sample, for example, the adjacent that is, an excess of individuals captured only once
cells in a raster. We specify W as the adjacency matrix (Pradel et al. 1997)—an important nuisance parameter
with a 1 if two samples are neighbours and 0 otherwise, to accommodate when individuals are not site-faithful.
and in addition, D is the normalisation matrix, a diago- Péron et al. (2011) used the spline method to delin-
P
nal matrix with Dii = j Wij . Then the CAR model is a eate population sinks as areas where the predicted
Gaussian random field (Gelfand and Vounatsou 2003), overwinter survival probability of Eurasian wood-
of mean µ and variance σ 2 (D − ρW)−1 (Cressie 1993). cocks (Scolopax rusticola) fell below the population
Following Besag (1974), there is an analytical solution renewal rate, itself a function of spatially invariant
under some constraints. In practice, implementations fecundity and summer survival rates. Their results con-
like the function spautolm from the spdep package firm the additive nature of hunting mortality (Péron
rely on numerical optimisation to offer more flexibili- 2013) and imply that the dispersal of juveniles into pop-
ty regarding what goes in µ and how the W matrix is ulation sinks is critical to the persistence of the species
defined, at the cost of increased risk of numerical errors in the sinks.
and optimisation failures. Many ecological applica- Campbell et al. (2018) also used the autoregressive
tions furthermore include the CAR structure within method to perform a spatially explicit population via-
a more complex hierarchical structure, for example, bility analysis of the Sonoran desert tortoise (Gopherus
a partially observed colonisation/extinction process morafkai) based on spatially explicit survival rates.
(Yackulic et al. 2012; Péron et al. 2016). In this desert-adapted species, their results provide
a demographic explanation for the ‘abundant centre
hypothesis’ (Péron and Altwegg 2015b), that is, demo-
15.2.2 Recent implementations in graphic performance was best at the core of the species’
range and worst near the edge of the species’ range.
spatial demography
Interestingly, a decrease in the age at maturity near
Spatially explicit matrix population models the edge of the range failed to compensate for the
By ‘spatially explicit matrix population model’ I mean decreased survival rates.
an array of grid cell-specific matrix population models To my knowledge, there is no study to date using the
(cf. Chapter 9), parameterised using the nonparametric random forest method to estimate spatial variation in
estimation techniques outlined in section 15.2.1. They survival rates. This is because the cross-validation part
describe the spatial variation in intrinsic population of the random forest algorithm is not straightforwardly
growth rate but do not account for dispersal. Actual- compatible with the state-space modelling framework
ly, I will only review the studies of spatial variation in required to estimate survival from capture–recapture
survival rate, not in intrinsic population growth rate data with imperfect detection.
per se. This is because I could not locate any study of
the spatial variation in fecundity that used a nonpara- Modelling the population growth rate from
metric technique. However, see data sets of passerine count data
clutch size and nest fate (Baillie 1990; Brickhill et al. When the vital rates themselves are not available, and
2015; Eglington et al. 2015), acorn production in oak if detailed time-specific and spatially-explicit popu-
(Quercus sp.) forests (Koenig and Knops 2013; Touzot lation count data are available, these counts may be
et al. 2018), and egg or foetus mass in harvested interpolated both spatially and temporally to compute
individuals from exploited species like commercial fish the spatial variation in the population growth rate
(Bell et al. 1992; Kraus et al. 2000; Stige et al. 2017) and directly (Renner et al. 2013; Rushing et al. 2016). How-
game (Karns 2014; Gamelon et al. 2018). ever, in most cases, data sets have proven too sparse
and authors either separated the spatial and temporal natural spatial component to them (Schwartz et al.
components or they pooled the data at a coarse ecore- 2010; Basille et al. 2013; Péron et al. 2017). However,
gion level. In the latter case, random forest algorithms researchers should be acutely aware that, especially
appear well suited to delineate these ecoregions based in our data-rich times, some climate and vegetation
on the demographic data themselves (Brickhill et al. covariates may adequately predict the observed spa-
2015; Rushing et al. 2016). In any case, as reviewed tial variation in demography, but without capturing
further in section 15.3, in vertebrates it is extremely any significant biological mechanism, in other words,
rare to collect exact population counts at a fine spatial by coincidence (Journé et al. 2019). Such correlations
resolution and over large geographical areas. What is are potentially great for interpolating gaps in cover-
typically available instead is crowd-sourced, rasterised age, but they should not be overinterpreted or used for
presence/absence or minimum count data (e.g. Péron predictions and extrapolations. For example, within a
and Altwegg 2015a). Using the link between the over- population of red-billed chough (Pyrrhocorax pyrrhoco-
all abundance in the landscape and the probability rax), the between-site variation in survival probability
that any given location is occupied, or alternatively the was caused by the different natal origins of the indi-
link between the local abundance and the probabili- viduals in the different sites, not by the environmental
ty that at least one individual is detected (Royle and attributes of the sites (Reid et al. 2006). In an archipela-
Nichols 2003), we may use these data as a proxy for go population of great tits (Parus major), the spatial
the local abundance (Williams et al. 2017) and com- variation in demography was driven by an evolution-
pute the spatial variation in the population growth rate ary ‘island syndrome’, not directly by the variation
that way. in climate between islands and the continent (Postma
and van Noordwijk 2005). To avoid the pitfall of over-
interpreting correlations, one may define broad ecore-
15.2.3 Discussion gions and compute spatial variation in demographic
rate at this resolution. Alternatively, all the nonpara-
Empirical discrepancies between the three
metric approaches described in section 15.2.1 can be
nonparametric regression methods
made into semiparametric tools that combine a regres-
Empirically, the outputs of the three nonparametric sion against spatial covariates and a nonparametric
methods are often significantly different (e.g. ‘RF’ vs. spatial structure in the error term of the regression
‘MARS’ in Figure 6 of Renner et al. 2013). This is in (Gimenez et al. 2006).
part because of differences in the amount of smooth-
ing. As illustrated by Kie (2013) in the context of ker- On spatial capture–recaptures
nel density estimators, the smoothing parameter is of
paramount importance (see also ‘GAM’ vs. ‘MARS’ in In the context of spatial demography, beware that
Figure 6 Renner et al. 2013; and ‘KDE’ vs. ‘KDEr’ and the phrase ‘spatial capture–recaptures’ (Borchers and
vs. ‘AKDE’ in Figure 1 of Péron 2019b). In other words, Efford 2008; Royle et al. 2013) usually refers to a type
the differences in output at least in part stem from of model that is indeed spatially explicit but does not
the different amounts of smoothing. If the noise/signal typically yield an estimate of spatial variation in sur-
ratio is large, that is, if nonspatial variance exceeds vival. Spatial capture–recapture models are designed
spatial variance in the data, the random forest is the to improve the estimation of the population size by
most likely to overfit the data, and researchers should accommodating the decrease in individual capture
therefore prefer either the spline or the autoregressive probability with the distance between the home range
method. But by contrast, if there are discontinuities or centroid of the animal and the capture locations. They
locally sharp gradients in the focal variable, the ran- thereby provide a better fit to the capture–recapture
dom forest method is the most relevant because other data. They can in some configurations inform spatial
methods will smooth out these discontinuities. variation in population density (Gardner et al. 2010),
but in most cases the spatial variation in survival prob-
What about fully parametric approaches? ability remains out of bounds for this type of model
and data.
Instead of a nonparametric, descriptive approach,
researchers are often keen on using spatial covariates
to predict spatial demography and infer the underlying
ecological mechanisms (Beale et al. 2010; Germain et al.
15.3 Estimating dispersal fluxes
2018). This is especially the case when extrapolating Dispersal is defined as movement that has the potential
mortality factors from telemetry data which have a to lead to gene flow (Clobert et al. 2001). Movements
within individual home ranges and seasonal migra- precise ageing of the juveniles upon capture to sep-
tions are not considered dispersal. In many taxa, adults arate local recruits from immigrants during the post-
are site-faithful either by constraint (plants, sessile ani- dispersal capture events of the same year. Peery et al.
mals) or due to philopatric behaviour (Greenwood (2006) combined the observed recruitment rate of local-
1980), and thereby gene flow mostly occurs through ly born chicks of marbled murrelets Brachyramphus
natal dispersal, that is, the movements of immatures or marmoratus, with Pradel’s (1996) estimate of anterior-
propagules (Greenwood 1980). ity within the breeding adults, in order to compute the
Importantly, especially with a vertebrate focus, dis- immigration rate as the difference between local and
persal is the result of a series of decisions—from the total recruitment. Note that both of the latter approach-
decision to leave the current location to the choice of the es only quantify immigration; emigration is confound-
new location among the locations that were explored ed with mortality in these two approaches. Potential
(Grosbois and Tavecchia 2003). These different deci- users should furthermore carefully consider the model
sions may use different environmental cues, and dif- assumptions and the fieldwork requirements that the
ferent individuals may make different decisions based above authors listed.
on the same cues. These decisions, summed over time
and across individuals, yield the dispersal kernel: the
probability that an individual at location (x1 , y1 ) at time
15.3.2 Multisite capture–recapture models
t1 will disperse to location (x2 , y2 ) at time t2 . One also and integrated metapopulation models
refers to ‘dispersal fluxes’ when the population is spa- In these methods, individuals are monitored over
tially structured (the number of individuals dispersing discrete population units or ‘sites’, such as seabird
from one site to another site, per unit of time) or to colonies or forest fragments. When an individual is
‘dispersal rates’ (the number of individuals dispers- recorded to move from one site to another, this rep-
ing from one site to another, per unit of time and per resents a recorded dispersal event. The data are pro-
individual in the site of origin) (Table 15.1). cessed through a multisite capture–recapture model
Dispersal may be recorded directly by documenting to allow for missing observations of some individu-
the changes in individual locations from one breeding als in some years (Lebreton et al. 2009), as described
attempt to the next or from the place of birth to the in Chapter 13, yielding an estimate of dispersal rates.
first breeding attempt (Doligez et al. 2002; Ducros et al. To further study the dispersal decision process, the dis-
2019). But this type of data remains rare and therefore persal rates may be decomposed into the probability
I will mostly review less direct methodologies. of site-fidelity and the site selection conditioned on the
lack of site-fidelity (Grosbois and Tavecchia 2003). The
model may feature a ‘memory effect’ of the location of
15.3.1 Integrated population models
birth (Péron et al. 2010b) or of the previous dispersal
In this method, the target variable is the net flux of decision (Lagrange et al. 2014).
individuals that exit or enter a given study region In addition, to accommodate the fact that researchers
over a given period of time, which is the difference in the field may not be able to monitor all the sites
between the number of emigrants and the number in a metapopulation, a ‘ghost’ population unit may
of immigrants. The net flux is estimated as the dif- be incorporated in the model, representing unmoni-
ference between the overall population growth (∆N) tored population units where an individual may tem-
and the intrinsic population growth (b – d, where b is porarily reside and therefore temporarily escape detec-
the number of births and d is the number of deaths). tion (Schaub et al. 2004). Statistical inference in this
In practice, the best approach is to jointly analyse case may be facilitated if the unmonitored popula-
capture–recapture data, fecundity survey data, and tion units that form the ghost state can be surveyed
population survey data in an integrated population for population size. These data may be jointly anal-
model (Abadi et al. 2010), as described in Chapter 14. ysed with the capture–recapture data in an ‘integrat-
Because it remains extremely challenging to collect ed metapopulation model’ (Péron et al. 2010a). The
these three types of data together and at the same time, key benefit of the integrated approach is improved
examples are still rare in the literature (Millon et al. accuracy of parameter estimates and the ability to
2019). Importantly, however, several less-data-hungry extrapolate the dispersal information from a subset of
approaches have been proposed for special cases. For sites, where marked individuals are monitored, to a
example, Nichols and Pollock (1990) described how to larger ensemble of sites where only population count
use a ‘robust design’ of the capture–recapture proto- data are collected (Péron et al. 2010a). Lastly, when
col for voles (Microtus pennsylvanicus) that relies on the the interest lies only in the decision to leave or stay
in the current site (Schmidt 2004), the dispersal infor- variation in the rasterised population abundance N
mation may be simplified into a binary variable— may be modelled as a partial differential equation or
site-faithful or site-unfaithful—thereby discarding the diffusion equation (Skellam 1951; Okubo and Levin 2001;
information about the site identities (Lagrange et al. Ovaskainen et al. 2008; Foltête and Giraudoux 2012; see
2014). also Neubert and Caswell 2000; Jongejans et al. 2011).
As a final note the dispersal rates in these models That framework allows an elegant partition of the local
do not account for permanent emigration out of the population growth rate dN dt
into the intrinsic population
survey area, which is confounded with death-but may growth on the one hand and a diffusion term rep-
be age-specific which researchers should be aware of resenting the movements of individuals from or into
when studying age-specific dispersal. neighbouring locations on the other hand:
In practice, multisite capture–recapture models intrinsic
z }|
{ z diffusion
are best implemented using dedicated software as }| {
dN N
described in Chapter 13. Integrated metapopulation =R·N· 1− + ∇2 (D · N)
dt K
models are best implemented in the Bayesian frame-
work (e.g. Péron et al. 2012 and references therein), The diffusion parameters in the matrix in D directly
although initial implementations were in the frequen- quantify the resistance surface. R and K represent the
tist framework using the Kalman filter to compute the population growth rates and carrying capacities, and
likelihood of the population count data (Péron et al. ∇2 is the Laplace operator that computes the sum of
2010a and references therein). the second partial derivatives relative to geographical
coordinates. In practice the diffusion term is approx-
imated by its rasterised formulation, with adjacency
15.3.3 Dispersal as a diffusion process along matrix W (as explained in section 15.2.1) and raster grid
resistance surfaces resolution h:
The main drawback of the two previous methods is X (D · N)j − (D · N)i
∇2 (D · N)i ≈
that they require clear boundaries delineating popu- h2
j|Wij =1
lation units. In most cases, these do not exist. The
population is instead distributed in a continuous man- The rasterisation transforms the diffusion equation
ner along ecological gradients. The dispersal process into a set of ordinary differential equations, making it
is then more akin to a diffusion process than to a dis- solvable within any numerical solver that can manage
crete state-switching process (Ovaskainen et al. 2008; large systems of equations. However, such a disper-
Jongejans et al. 2011; Foltête and Giraudoux 2012; sal kernel does not feature any preferential dispersal
Williams et al. 2017). Nevertheless, dispersal may still direction. In wind-dispersed plants (Jongejans et al.
be spatially structured by landscape features (McRae 2011), and in the theoretical situation where individ-
et al. 2008; Spear et al. 2010; Berthier et al. 2013; Hanks uals would have been selected by climate change to
and Hooten 2013) or by the social cues that the individ- systematically disperse in a particular direction, the
uals use to decide where to move, such as conspecific contributions to the sum over j in the above equation
population density and conspecific breeding success could be weighted according to the dispersal direction.
(Doligez et al. 2002; Péron et al. 2010b). Another major caveat, which is not always recog-
Traditionally, this type of ecological landscape struc- nised, is that the diffusion equation as written above
ture is documented either by landscape genetics, where- assumes that movements are most likely between adja-
by the genetic structure of the population is compared cent grid cells. The framework does not really accom-
to a landcover map, mostly to infer the barriers to modate long-distance dispersal. Depending on the
dispersal but also the dispersal corridors that promot- behaviour of the species at stake, this model may there-
ed genetic homogenisation (McRae et al. 2008; Spear fore grossly misrepresent the biological process. This
et al. 2010; Hanks and Hooten 2013), or by movement may explain some apparent lack of fit of that model
data analysis, whereby the tracked movements of focal to the patterns that occur along the colonisation front
individuals are extrapolated to infer the barriers to of species that routinely perform long-distance dis-
movement and the corridors that would facilitate the persal (Louvrier et al. 2019). Alternative formulations
movements of potential dispersers (Ovaskainen 2004; of the diffusion term in the above equation therefore
Panzacchi et al. 2016; Scharf et al. 2018; Zeller et al. 2016; appear warranted. Avenues of research in this direc-
Wang 2020). The result is a resistance surface. tion include habitat- rather than distance-based adja-
The resistance surface may also be estimated cency matrices, but also maybe the Riesz fractional
from demographic data. The spatial and temporal derivatives, as used to represent the infamous (for
movement ecologists) ‘Lévy flight’ (Mandelbrot 1982; not conceptually very different from unidimensional
Sims et al. 2008; Çelik and Duman 2012). IPMs designed to study the dynamics of quantitative
Eventually, the main issue with this approach is traits like body mass (Lewis et al. 2006; Jongejans et al.
going to be the data requirements, but a new era seems 2011; Merow et al. 2014). The basic principle is to define
to have opened in that respect. Recently, Williams a kernel function k representing the probability that
et al. (2017) proposed to leverage the link between the an individual at location (x1 , y1 ) at time t1 will yield
abundance of a species and the probability that it is an individual at location (x2 , y2 ) at time t2 . The ker-
detected during any one visit to a location of occur- nel function combines all the information about how
rence (Royle and Nichols 2003). This makes it possible the individual survived, where it moved, whether it
to fit the above model to presence/absence data, reproduced, and where the offspring moved. Further
which are much easier to obtain nowadays by crowd- defining the function to function operator K (ν) (x2 , y2 ) =
RR
sourcing than population survey data which would Ω
ν (x1 , y1 ) k (x1 , y1 , x2 , y2 ) dx1 dy1 , where ν represents
require an intense professional field effort. The param- a distribution function, the IPM equation becomes a
eters of the dynamic occupancy model (colonisation single line equation n (t2 ) = K (n (t1 )), where n (t)
rates, extinction rates, and detection probabilities; denotes the population distribution at time t. In the fol-
Yackulic et al. 2012; Péron et al. 2016) are expressed lowing I will outline how to analyse the properties of
as functions of R, K, and D, allowing the estimation of K, but first I will review a lingering question in spatial
the latter. Recent examples include the description of ecology, that is, the fundamental difference between
the colonisation of a newly ice-free bay by sea otters contingency and agency.
(Enhydra lutris) (Williams et al. 2019) and the recoloni-
sation of south-eastern France by wolves (Canis lupus)
(Louvrier et al. 2019). The main limiting assumption
15.4.2 Space use or resource selection?
of that presence–absence framework is that the pop- One of the critical decisions that eventual applications
ulation must remain well below carrying capacity. In of the above principles will have to make is whether
other words, the framework applies to rare species or to model space use per se—for example, the focal vari-
species that are currently expanding. When the pop- able is the geographical coordinates of the individual
ulation enters the density-dependent regime, most of (Lewis et al. 2006; Jongejans et al. 2011)—or whether
the study area is occupied, the detection of at least one to model a behavioural trait that governs space
individual does not depend on abundance anymore use—for example, resource selection coefficients, per-
(Royle and Nichols 2003), and the intrinsic growth rate sonality traits, or ‘movement syndromes’. ‘Move-
is mostly influenced by K rather than R. This makes the ment syndromes’ refer to covariations between move-
model unidentifiable, that is, the parameters cannot be ment rates and between movement rates and other
separately estimated in this biological configuration. behavioural traits, for instance explorative or compet-
This is called weak identifiability. itive personality (Wolf et al. 2007; Spiegel et al. 2017).
Variability in these hard-wired syndromes between
15.4 The feedback between space use species and individuals is known to foster coexistence
and fitness (Vanak et al. 2013; Péron et al. 2019) and population via-
bility in fragmented landscapes (Zheng et al. 2009; Zera
In the previous sections, I described how to estimate 2017).
the link between the geographic location of an In the first case (space use), the model describes the
individual, its demographic performance, and its dis- interaction between geographical coordinates and fit-
persal kernel. Now I will outline how to use this infor- ness. For geographical coordinates, one may use the
mation to parameterise an integral projection model centroid of an animal’s home range. One may also
(IPM) to forecast the future distribution of the popu- replace the coordinates with the proportion of a giv-
lation based on its spatial demography. en landcover in the home range (Péron 2019b) or the
environmental covariates where a plant is growing
(Merow et al. 2014), thereby swapping the geograph-
15.4.1 Principles of integral projection models
ic coordinates for the ecological coordinates. In this
as pertaining to spatial demography type of application, the IPM will become a relatively
In IPM parlance (cf. Chapter 10), natal dispersal corre- phenological model, forecasting the population distri-
sponds to ‘inheritance’ and adult or breeding dispersal bution if current reaction norms are maintained into
corresponds to ‘growth’. Besides that, spatial IPMs are the future (Neubert and Caswell 2000; Lewis et al.
(a) (b)
Seed production: 26 Seed production: 53
1.0 1.0
Density (rescaled)
Stable
0.8 At birth 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0.0 0.0
−1.0 −0.5 0.0 0.5 1.0 −1.0 −0.5 0.0 0.5 1.0
Aridity at growing location x Aridity at growing location x
Figure 15.1 Simulation inspired by Rebarber et al. (2012) of a perennial plant population with no seed bank and random seed rain, faced with
two scenarios of aridity. (a) a drier scenario. Seeds are more likely to fall in an arid spot (cf. grey curve). Compared to the distribution among
seedlings (grey curve), the stable population distribution is shifted towards wetter locations (black curve), mostly because the seedlings that grow
in the driest locations fail to survive. This means that over time the population is expected to shift its range centroid towards wetter locations, at a
pace that can be estimated, but that there will always be a few individuals growing in arid locations because that is where they fell as a seed. (b) a
wetter scenario. The stable population distribution is barely different from the expectation from the seed rain. This means that the population range
is not expected to change much over time. Note also among the model outputs the forecast of the average population fitness and reproductive
rates. In detail, the model was specified by the survival function s (x) = logit−1 (2 − 2x), the fecundity function c (x) = exp (2 − 2x), and the
density-dependent recruitment g (C) = C−0.33 .
2006; Merow et al. 2014). However, that approach inheritance component corresponds to physical barri-
will describe but not explain the spatial variation in ers that constrain the distance between offspring and
demography. parents.
In the second case (resource selection/movement
syndrome), one first estimates behavioural traits and in
a second step links these traits to spatial demography. 15.4.3 Formal analysis of IPMs
But in this case, researchers may need to incorporate IPMs, once parameterised with field data, can be for-
plastic reaction norms describing how the behavioural mally analysed for inference about equilibrium distri-
trait changes depending on the context. For example, bution, spatially explicit population growth, sensitivity
resource selection coefficients depend on what is avail- of the population growth rate to variation in model
able for selection, often in a nonlinear way (Forester parameters, individual variation, transitory dynamics
et al. 2009). A typical example is roe deer (Capreo- (Ellner and Rees 2006), and the speed of any shift in
lus capreolus), which today are commonly found in the population geographic distribution, under some
open agricultural landscapes but will revert to select- additional conditions (Lewis et al. 2006). As an illus-
ing forest edges if given the opportunity. The realised tration, I will focus on a well-worked example that
resource selection coefficients may therefore need to pertains to monocarpic perennial plants (Rose et al.
be interpreted with caution and may perhaps best be 2005; Ellner and Rees 2006; Rebarber et al. 2012).
replaced by alternative behavioural metrics that cap- The keys to this model are that (1) the distribution
ture the plasticity of resource selection by each individ- of the focal trait among propagules is independent
ual (Bonnot et al. 2018). from the trait value of their parents; and (2) density
Importantly, the two approaches are not mutually dependence occurs among seedlings/propagules only,
exclusive. They are only opposed here to reflect on meaning that the recruitment rate declines with the
the underlying philosophy and challenges in terms of number of candidates to recruitment, not the num-
data acquisition. As a final note, the interpretation of ber of adults. There is usually a stable distribution
spatial IPMs in terms of evolutionary dynamics is not of the trait in such a system (Ellner and Rees 2006;
always straightforward. Similar to the way that off- Rebarber et al. 2012) which can be computed as a func-
spring may inherit the large or small body mass of tion of the trait-specific survival function s, the trait
their mother because they feed on the same resource, distribution at birth b, the trait-specific fecundity func-
not because they share the same genotype, the inher- tion c, and the density-dependent recruitment term g
itance parameter in spatial IPMs combines true her- (Figure 15.1; Appendix 1; please go to www.oup.com/
itability and nongenetic inheritance. The nongenetic companion/SalgueroGamelonDM).
Density at time t Survival Fecundity Density at time t+1
Natal dispersal Breeding dispersal

rules rules
Figure 15.2 Principles of the simulation approach to analyse complex IPMs. The simulated animals are drawn from the density distribution at
time t, made to perform and move according to the IPM kernel function, yielding the density distribution at time t+1. The process may be iterated a
large number of times to obtain the asymptotic behaviours of the system. The kernel function may incorporate more complex rules than depicted,
for example, dependency on nonstationary environmental covariates and interactions between the model individuals. Importantly, the kernel
functions may depend on heritable individual traits that govern the spatial behaviour, rather than directly on the geographical coordinates. For
example, data permitting, the trait under study could be the preferential direction of the natal dispersal kernel, and the model would provide
inference about the selection pressure on dispersal direction.
Now, to make this model a model of spatial demog- reproduction. The simulation parameters are informed
raphy, the trait may, for example, represent the aridity by analyses from section 15.2 about spatially explic-
at the specific location where the plant is growing. it survival and fecundity and section 15.3 about dis-
Then, b represents the relative frequency of arid and persal kernels. The IPM properties are computed as
wet places in the landscape where seeds are randomly emergent properties of the individual-based simula-
dispersed, by wind, for example; that is, b represents tion (e.g. Wiens et al. 2017). The movement mod-
the environmental constraints. The stable distribution el does not need to be more complex than a step
represents the eventual population distribution along selection model (Signer et al. 2017; Péron 2019a).
the aridity gradient, as caused by the demographic However,the interest of the framework lies in its
performance of the individuals (Figure 15.1). There is, flexibility, for example, to incorporate interactions
however, no evolution involved: at each generation the between individuals, to take into account the cost of
seedlings are redistributed at random along the aridi- movement (Péron 2019b), to add heritable individu-
ty gradient (using the grey curves in Figure 15.1). As al covariates that influence dispersal decisions, and
a side note, the complex mathematics underlying the to outfit the individuals with the ability to gather
formal analysis of density-dependent IPMs (Rebarber information about the spatial variation in fitness
et al. 2012) render the framework somewhat costly to (Doligez et al. 2002) (Figure 15.2).
tailor to specific situations, which might restrict the Technical bottlenecks that currently hamper the
type of study systems for which it is relevant and more widespread empirical implementation of spatial
the type of evolutionary questions that can be tackled IPMs to vertebrates include the need to track the move-
with it. ments and demographic performance of a large num-
ber of individuals with known parentage and over long
15.4.4 Individual-based simulation approach enough periods of time. This is because the repeatabili-
ty, reaction norms, and age specificity of their space use
As highlighted in section 15.4.3, the formal analysis of
patterns and fitness components need to be estimated.
IPMs may not offer enough flexibility in practice to
Despite the democratisation of geotracking technolo-
accommodate the complexity of ecological processes
gies, these conditions remain hard to meet.
and to easily tailor the IPMs to specific case studies.
In the context of spatial demography, a more flexible
alternative is individual-based modelling (Chapter 12).
15.5 Conclusions
The simulation focuses on individual movement deci- Although spatial demography is firmly located at the
sions and their consequences in terms of survival and core of eco-evolutionary thinking, it does not appear
to have given rise to a rich empirical literature yet, Brickhill, D., P. G. H. Evans, and J. M. Reid. 2015. Spatio-
even if many methods are now readily available. The temporal variation in European starling reproductive
reasons include on the one hand the prohibitive data success at multiple small spatial scales. Ecology and Evo-
requirements and on the other hand the relative lack of lution 5:3364–3377.
Burton, O. J., B. L. Phillips, and J. M. J. Travis. 2010. Trade-
communication between method-heavy ‘quantitative
offs and the evolution of life-histories during range expan-
ecology’ projects and concept-heavy theoretical initia-
sion. Ecology Letters 13:1210–1220.
tives. This overall makes it challenging to adopt a com-
Campbell, S. P., E. R. Zylstra, C. R. Darst, R. C. Averill-
prehensive empirical outlook on spatial demography. Murray, and R. J. Steidl. 2018. A spatially explicit hierar-
The increasing availability of crowd-sourced species chical model to characterize population viability. Ecolog-
distribution data and the democratisation of geotrack- ical Applications 28:2055–2065.
ing technologies represent exciting opportunities. Çelik, C. and M. Duman. 2012. Crank–Nicolson method
The fact that spatial demography bridges some cur- for the fractional diffusion equation with the Riesz
rently isolated subfields like movement ecology and fractional derivative. Journal of Computational Physics
evolutionary biodemography may also open fruitful 231:1743–1750.
interfaces. Clobert, J., E. Danchin, A. A. Dhondt, and J. D.
Nichols. 2001. Dispersal. Oxford University Press,
New York.
References Crainiceanu, C. M., D. Ruppert, and M. P. Wand. 2005.
Abadi, F., O. Gimenez, B. Ullrich, R. Arlettaz, and Bayesian analysis for penalized spline regression
M. Schaub. 2010. Estimation of immigration rate using using WinBUGS. Journal of Statistical Software 14:
integrated population models. Journal of Applied Ecology 1–24.
47:393–400. Cressie, N. A. C. 1993. Statistics for Spatial Data. Wiley,New
Baillie, S. R. 1990. Integrated population monitoring of York.
breeding birds in Britain and Ireland. Ibis 132:151–166. Doligez, B., É. Danchin, and J. Clobert. 2002. Public infor-
Basille, M., B. Van Moorter, I. Herfindal, et al. 2013. Selecting mation and breeding habitat selection in a wild bird
habitat to survive: the impact of road density on survival population. Science 297:1168–1170.
in a large carnivore. PLOS One 8:e65493. Dormann, C., J. McPherson, M. Araújo, et al. 2007. Meth-
Beale, C. M., J. J. Lennon, J. M. Yearsley, M. J. Brewer, and ods to account for spatial autocorrelation in the analy-
D. A. Elston. 2010. Regression analysis of spatial data. sis of species distributional data: a review. Ecography
Ecology Letters 13:246–264. 30:609–628.
Bell, J. D., J. M. Lyle, C. M. Bulman, K. J. Graham, Ducros, D., N. Morellet, R. Patin, et al. 2019. Beyond dis-
G. M. Newton, and D. C. Smith. 1992. Spatial varia- persal versus philopatry? Alternative behavioural tactics
tion in reproduction, and occurrence of non-reproductive of juvenile roe deer in a heterogeneous landscape. Oikos
adults, in orange roughy, Hoplostethus atlanticus Collett 129(1):81–92.
(Trachichthyidae), from south-eastern Australia. Journal Eglington, S. M., R. Julliard, G. Gargallo, et al. 2015. Latitu-
of Fish Biology 40:107–122. dinal gradients in the productivity of European migrant
Berthier, K., S. Piry, J.-F. Cosson, et al. 2013. Dispersal, land- warblers have not shifted northwards during a period
scape and travelling waves in cyclic vole populations. of climate change. Global Ecology and Biogeography 24:
Ecology letters 17:53–64. 427–436.
Besag, J. 1974. Spatial interaction and the statistical analysis Ellner, S. P. and M. Rees. 2006. Integral projection models for
of lattice systems. Journal of the Royal Statistical Society: species with complex demography. American Naturalist
Series B 36:192–225. 167:410–428.
Bivand, R. S., E. Pebesma, and V. Gómez-Rubio. 2013. Foltête, J.-C. and P. Giraudoux. 2012. A graph-based app-
Applied Spatial Data Analysis with R, 2nd edition. roach to investigating the influence of the landscape
Springer, New York. on population spread processes. Ecological Indicators
Bohonak, A. J. 1999. Dispersal, Gene Flow, and Population 18:684–692.
Structure. University of Chicago Press. Forester, J. D., H. K. Im, and P. J. Rathouz. 2009. Account-
Bonnot, N. C., M. Goulard, A. J. M. Hewison, et al. 2018. ing for animal movement in estimation of resource selec-
Boldness-mediated habitat use tactics and reproductive tion functions: sampling and data analysis. Ecology
success in a wild large herbivore. Animal Behaviour 90:3554–3565.
145:107–115. Gamelon, M., T. Gayet, E. Baubet, et al. 2018. Does
Borchers, D. L. and M. G. Efford. 2008. Spatially explicit max- multiple paternity explain phenotypic variation
imum likelihood methods for capture–recapture studies. among offspring in wild boar? Behavioral Ecology 29:
Biometrics 64:377–385. 904–909.
Breiman, L. 2001. Random forests. Machine Learning Gandon, S. and Y. Michalakis. 1999. Evolutionarily stable
45:5–32. dispersal rate in a metapopulation with extinctions and
kin competition. Journal of Theoretical Biology 199:275– Leibold, M. A., M. Holyoak, N. Mouquet, et al. 2004. The
290. metacommunity concept: a framework for multi-scale
Gardner, B., J. Reppucci, M. Lucherini, and J. A. Royle. 2010. community ecology. Ecology Letters 7:601–613.
Spatially explicit inference for open populations: estimat- Lewis, M. A., M. G. Neubert, H. Caswell, J. S. Clark, and
ing demographic parameters from camera-trap studies. K. Shea. 2006. A guide to calculating discrete-time inva-
Ecology 91:3376–3383. sion rates from data. Pages 169–192 in M. W. Cadotte,
Gelfand, A. E. and P. Vounatsou. 2003. Proper multivariate S. M. McMahon, and T. Fukami, editors. Conceptual Ecol-
conditional autoregressive models for spatial data analy- ogy and Invasion Biology: Reciprocal Approaches to
sis. Biostatistics 4:11–15. Nature. Springer, Dordrecht, NL.
Germain, R. R., R. Schuster, C. E. Tarwater, W. M. Louvrier, J., J. Papaïx, C. Duchamp, and O. Gimenez.
Hochachka, and P. Arcese. 2018. Adult survival and 2019. A mechanistic–statistical species distribution mod-
reproductive rate are linked to habitat preference in terri- el to explain and forecast wolf (Canis lupus) coloniza-
torial, year-round resident song sparrows Melospiza melo- tion in South-Eastern France. https://arxiv.org/abs/
dia. Ibis 160:568–581. 1912.09676.
Gimenez, O., C. Crainiceanu, C. Barbraud, S. Jenou- Mandelbrot, B. B. 1982. The Fractal Geometry of Nature.
vrier, and B. J. T. Morgan. 2006. Semiparametric W. H. Freeman and Co., New York.
regression in capture-recapture modeling. Biometrics 62: McCauley, S. J., C. J. Davis, E. E. Werner, and M. S. Robeson.
691–698. 2013. Dispersal, niche breadth and population extinc-
Greenwood, P. J. 1980. Mating systems, philopatry and tion: colonization ratios predict range size in North
dispersal in birds and mammals. Animal Behaviour American dragonflies. Journal of Animal Ecology 83:
28:1140–1162. 858–865.
Grosbois, V. and G. Tavecchia. 2003. Modeling dispersal McRae, B., B. Dickson, T. Keitt, and V. Shah. 2008. Using cir-
with capture–recapture data: disentangling decisions of cuit theory to model connectivity in ecology, evolution,
leaving and settlement. Ecology 84:1225–1236. and conservation. Ecology 89:2712–2724.
Hanks, E. M. and M. B. Hooten. 2013. Circuit theory and Merow, C., A. M. Latimer, A. M. Wilson, S. M. McMahon,
model-based inference for landscape connectivity. Journal A. G. Rebelo, and J. A. Silander. 2014. On using integral
of the American Statistical Association 108:22–33. projection models to generate demographically driv-
Hanski, I. and M. Gyllenberg. 1997. Uniting two general pat- en predictions of species’ distributions: development
terns in the distribution of species. Science 275:397–400. and validation using sparse data. Ecography 37:1167–
Jongejans, E., K. Shea, O. Skarpaas, D. Kelly, and S. P. 1183.
Ellner. 2011. Importance of individual and environmen- Millon, A., X. Lambin, S. Devillard, and M. Schaub. 2019.
tal variation for invasive species spread: a spatial integral Quantifying the contribution of immigration to popu-
projection model. Ecology 92:86–97. lation dynamics: a review of methods, evidence and
Journé, V., J. Barnagaud, C. Bernard, P. Crochet, and perspectives in birds and mammals. Biological Reviews
X. Morin. 2019. Correlative climatic niche models predict 94:2049–2067.
real and virtual species distributions equally well. Ecology Moran, P. A. P. 1950. Notes on continuous stochastic phe-
101:e02912. nomena. Biometrika 37:17–23.
Karns, G. R. 2014. Spatiotemporal Breeding Strategies within Neubert, M. G. and H. Caswell. 2000. Demography and
a High Density, Male-Skewed White-Tailed Deer Popula- dispersal: calculation and sensitivity analysis of inva-
tion. Auburn University. sion speed for structured populations. Ecology 81:
Kie, J. G. 2013. A rule-based ad hoc method for select- 1613–1628.
ing a bandwidth in kernel home-range analyses. Animal Nichols, J. D. and K. H. Pollock. 1990. Estimation of recruit-
Biotelemetry 1:13. ment from immigration versus in situ reproduction using
Koenig, W. D. and J. M. H. Knops. 2013. Large-scale spa- Pollock’s robust design. Ecology 71:21–26.
tial synchrony and cross-synchrony in acorn production Novaro, A., M. Funes, and R. Walker. 2005. An empirical test
by two California oaks. Ecology 94:83–93. of source–sink dynamics induced by hunting. Journal of
Kraus, G., A. Muller, K. Trella, and F. W. Kouster. 2000. Applied Ecology 42:910–920.
Fecundity of Baltic cod: temporal and spatial variation. Nychka, D. and N. Saltzman. 1998. Design of Air-Quality
Journal of Fish Biology 56:1327–1341. Monitoring Networks. Pages 51–76. Springer Nature,
Lagrange, P., R. Pradel, M. Bélisle, and O. Gimenez. Cham.
2014. Estimating dispersal among numerous sites using Okubo, A. and S. A. Levin, editors. 2001. Diffusion and
capture–recapture data. Ecology 95:2316–2323. Ecological Problems: Modern Perspectives. Springer,
Lebreton, J. -D., J. D. Nichols, R. J. Barker, R. Pradel, New York.
and J. A. Spendelow. 2009. Modeling Individual Ani- Ovaskainen, O. 2004. Habitat-specific movement parame-
mal Histories with Multistate Capture–Recapture Models. ters estimated using mark–recapture data and a diffusion
Advances in Ecological Research 41:87–173. model. Ecology 85:242–257.
Ovaskainen, O., H. Rekola, E. Meyke, and E. Arjas. 2008. Péron, G., C. Bonenfant, R. Gagnon, and C. T. Mabika.
Bayesian methods for analyzing movements in hetero- 2019. The two oxpecker species reveal the role of move-
geneous landscapes from mark–recapture data. Ecology ment rates and foraging intensity in species coexistence.
89:542–554. Biology Letters 15:20190548.
Panzacchi, M., B. Van Moorter, O. Strand, et al. 2016. Pre- Pinheiro, J., D. Bates, S. DebRoy, D. Sarkar, and the R
dicting the continuum between corridors and barriers Development Core Team. 2013. nlme: linear and non-
to animal movements using step selection functions and linear mixed effects models. R package version 3.1-111.
randomized shortest paths. Journal of Animal Ecology http://cran.r-project.org/package=nlme.
85:32–42. Postma, E. and A. J. Van Noordwijk. 2005. Gene flow main-
Peery, M. Z., B. H. Becker, and S. R. Beissinger. 2006. tains a large genetic difference in clutch size at a small
Combining demographic and count-based approaches to spatial scale. Nature 433:65–68.
identify source–sink dynamics of a threatened seabird. Pradel, R. 1996. Utilization of capture–mark–recapture for
Ecological Applications 16:1516–1528. the study of recruitment and population growth rate.
Péron, G. 2013. Compensation and additivity of anthro- Biometrics 52:703.
pogenic mortality: life-history effects and review of meth- Pradel, R., J. E. Hines, J. -D. Lebreton, and J. D. Nichols.
ods. Journal of Animal Ecology 82:408–417. 1997. Capture–recapture survival models taking account
Péron, G. 2019a. Modified home range kernel density esti- of transients. Biometrics 53:60–72.
mators that take environmental interactions into account. Pulliam, H. R. 1988. Sources, sinks, and population regula-
Movement Ecology 7:16. tion. The American Naturalist 132:652–661.
Péron, G. 2019b. The time frame of home-range stud- Rebarber, R., B. Tenhumberg, and S. Townley. 2012. Global
ies: from function to utilization. Biological Reviews 94: asymptotic stability of density dependent integral popu-
1974–1982. lation projection models. Theoretical Population Biology
Péron, G. and R. Altwegg. 2015a. Twenty-five years of 81:81–87.
change in southern African passerine diversity: non- Reid, J. M., E. M. Bignal, S. Bignal, D. I. McCracken, and
climatic factors of change. Global Change Biology 21: P. Monaghan. 2006. Spatial variation in demography and
3347–3355. population growth rate: the importance of natal location.
Péron, G. and R. Altwegg. 2015b. The abundant centre syn- Journal of Animal Ecology 75:1201–1211.
drome and species distributions: insights from closely Renner, M., J. K. Parrish, J. F. Piatt, K. J. Kuletz,
related species pairs in southern Africa. Global Ecology A. E. Edwards, and G. L. Hunt. 2013. Modeled dis-
and Biogeography 24:215–225. tribution and abundance of a pelagic seabird reveal
Péron, G., P.-A. Crochet, P. F. Doherty Jr, and J.-D. Lebre- trends in relation to fisheries. Marine Ecology Progress
ton. 2010a. Studying dispersal at the landscape scale: Series 484:259–277.
efficient combination of population surveys and capture– Rose, K. E., S. M. Louda, and M. Rees. 2005. Demograph-
recapture data. Ecology 91:3365–3375. ic and evolutionary impacts of native and invasive insect
Péron, G., J.-D. Lebreton, and P.-A. Crochet. 2010b. Breed- herbivores on Cirsium canescens. Ecology 86:453–465.
ing dispersal in black-headed gull: the value of familiarity Royle, J. A., and J. D. Nichols. 2003. Estimating abundance
in a contrasted environment. Journal of Animal Ecology from repeated presence–absence data or point counts.
79:317–326. Ecology 84:777–790.
Péron, G., Y. Ferrand, F. Gossmann, C. Bastat, M. Guenezan, Royle, J. A., R. B. Chandler, K. D. Gazenski, and
and O. Gimenez. 2011. Nonparametric spatial regression T. A. Graves. 2013. Spatial capture–recapture models
of survival probability: visualization of population sinks for jointly estimating population density and landscape
in Eurasian woodcock. Ecology 92:1672–1679. connectivity. Ecology 94:287–294.
Péron, G., C. A. Nicolai, and D. N. Koons. 2012. Demograph- Ruppert, D., M. P. Wand, and R. J. Carroll. 2003. Semi-
ic response to perturbations: the role of compensatory parametric Regression. Cambridge University Press,
density dependence in a North American duck under Cambridge.
variable harvest regulations and changing habitat. Journal Rushing, C. S., T. B. Ryder, A. L. Scarpignato, J. F. Sarac-
of Animal Ecology 81:960–969. co, and P. P. Marra. 2016. Using demographic attributes
Péron, G., R. Altwegg, G. A. Jamie, and C. N. Spot- from long-term monitoring data to delineate natural pop-
tiswoode. 2016. Coupled range dynamics of brood ulation structure. Journal of Applied Ecology 53:491–500.
parasites and their hosts responding to climate and Saracco, J. F., J. A. Royle, D. F. DeSante, and B. Gardner. 2010.
vegetation changes. Journal of Animal Ecology 85: Modeling spatial variation in avian survival and residency
1191–1199. probabilities. Ecology 91:1885–1891.
Péron, G., C. H. Fleming, O. Duriez, et al. 2017. The energy Scharf, A. K., J. L. Belant, D. E. Beyer, M. Wikelski, and
landscape predicts flight height and wind turbine collision K. Safi. 2018. Habitat suitability does not capture the
hazard in three species of large soaring raptor. Journal of essence of animal-defined corridors. Movement Ecology
Applied Ecology 54:1895–1906. 6:18.
Schaub, M., O. Gimenez, B. R. Schmidt, and R. Pradel. 2004. Vanak, A. T., D. Fortin, M. Thaker, et al. 2013. Moving to stay
Estimating survival and temporary emigration in the mul- in place: behavioral mechanisms for coexistence of African
tistate capture–recapture framework. Ecology 85:2107– large carnivores. Ecology 94:2619–2631.
2113. Wang, I. J. 2020. Topographic path analysis for modelling
Schmidt, K. A. 2004. Site fidelity in temporally correlated dispersal and functional connectivity: calculating topo-
environments enhances population persistence. Ecology graphic distances in the topoDistance R package. Methods
Letters 7:176–184. in Ecology and Evolution 11:265–272.
Schwartz, C. C., M. A. Haroldson, and G. C. White. 2010. Wiens, J. D., N. H. Schumaker, R. D. Inman, T. C. Esque,
Hazards affecting grizzly bear survival in the Greater K. M. Longshore, and K. E. Nussear. 2017. Spatial demo-
Yellowstone ecosystem. Journal of Wildlife Management graphic models to inform conservation planning of golden
74:654–667. eagles in renewable energy landscapes. Journal of Raptor
Signer, J., J. Fieberg, and T. Avgar. 2017. Estimating uti- Research 51:234–257.
lization distributions from fitted step-selection functions. Williams, P. J., M. B. Hooten, J. N. Womble, G. G. Esslinger,
Ecosphere 8:e01771. M. R. Bower, and T. J. Hefley. 2017. An integrated data
Sims, D. W., E. J. Southall, N. E. Humphries, et al. 2008. model to estimate spatiotemporal occupancy, abundance,
Scaling laws of marine predator search behaviour. Nature and colonization dynamics. Ecology 98:328–336.
451:1098–1102. Williams, P. J., M. B. Hooten, G. G. Esslinger, J. N. Womble,
Skellam, J. G. 1951. Random dispersal in theoretical popula- J. L. Bodkin, and M. R. Bower. 2019. The rise of an apex
tions. Biometrika 38:196–218. predator following deglaciation. Diversity and Distribu-
Spear, S. F., N. Balkenhol, M.-J. Fortin, B. H. McRae, and tions 25:895–908.
K. Scribner. 2010. Use of resistance surfaces for landscape Wolf, M., G. S. Van Doorn, O. Leimar, and F. J. Weissing.
genetic studies: considerations for parameterization and 2007. Life-history trade-offs favour the evolution of ani-
analysis. Molecular Ecology 19:3576–3591. mal personalities. Nature 447:581–584.
Spiegel, O., S. T. Leu, C. M. Bull, and A. Sih. 2017. What’s Wright, M. N. and A. Ziegler. 2017. ranger: a fast implemen-
your move? Movement as a link between personality and tation of random forests for high dimensional data in C++
spatial dynamics in animal populations. Ecology Letters and R. Journal of Statistical Software 77 (2017).
20:3–18. Yackulic, C. B., J. Reid, R. Davis, J. E. Hines, J. D. Nichols,
Stevens, F. R., A. E. Gaughan, C. Linard, and A. J. Tatem. and E. Forsman. 2012. Neighborhood and habitat effects
2015. Disaggregating census data for population mapping on vital rates: expansion of the barred owl in the Oregon
using random forests with remotely-sensed and ancillary Coast Ranges. Ecology 93:1953–1966.
data. PLOS One 10:e0107042. Zeller, K. A., K. McGarigal, S. A. Cushman, P. Beier,
Stige, L. C., N. A. Yaragina, Ø. Langangen, B. Bogstad, T. W. Vickers, and W. M. Boyce. 2016. Using step and
N. Chr Stenseth, and G. Ottersen. 2017. Effect of a fish path selection functions for estimating resistance to move-
stock’s demographic structure on offspring survival & ment: pumas as a case study. Landscape Ecology 31:
sensitivity to climate. Proceedings of the National Acade- 1319–1335.
my of Sciences of the United States of America 114: Zera, A. J. 2017. The biochemical basis of life history adap-
1347–1352. tation: gryllus studies lead the way. Pages 229–243 in
Thuiller, W., T. Münkemüller, S. Lavergne, et al. 2013. H. Noji, W. Horch, T. Mito, A. Popadićhideyo, and O.
A road map for integrating eco-evolutionary processes Sumihare, editors. The Cricket as a Model Organism.
into biodiversity models. Ecology Letters 16(Suppl 1): Springer, Tokyo.
94–105. Zheng, C. Z., O. Ovaskainen, and I. Hanski. 2009. Modelling
Touzot, L., M. C. Bel-Venner, M. Gamelon, et al. 2018. single nucleotide effects in phosphoglucose isomerase
The ground plot counting method: a valid and reliable on dispersal in the Glanville fritillary butterfly: coupling
assessment tool for quantifying seed production in tem- of ecological and evolutionary dynamics. Philosophical
perate oak forests? Forest Ecology and Management 430: Transactions of the Royal Society B: Biological Sciences
143–149. 364:1519–1532.
CHAPTER 16
Evolutionary Demography
Shripad Tuljapurkar and Wenyun Zuo
16.1 Introduction we discuss how these theoretical methods are relat-

ed to analyses and theories of post-reproductive life,
Evolutionary demography has grown rapidly in recent
via the general concept of “borrowing fitness”. Finally,
years, as witness this volume, and now covers most
we discuss nonlinear models of mutation and selection
topics at the intersection of demography and evolution.
and density-dependent models.
Particular impetus has come from work on senescence
in humans and then in other species, aided by the
rapid growth of databases such as the HMD (Uni- 16.2 The early theory
versity of California, Berkeley (USA) and Max Planck
Medawar (1957) and Hamilton (1966) are foundation-
Institute for Demographic Research (Germany) 2019),
al to the theory senescence and well worth a read. Both
HFD (Max Planck Institute for Demographic Research
start with an age-structured population in demograph-
(Germany) and Vienna Institute of Demography (Aus-
ic equilibrium, whose individuals have a survival rate
tria) 2019), COMPADRE (Salguero-Gómez et al. 2015),
px and a fertility rate mx that both depend on age x. At
and COMADRE (Salguero-Gómez et al. 2016). The bio-
demographic equilibrium the population changes at a
logical topics of life history evolution and evolution
constant rate. Usually only one sex is counted, typically
in population with complex life cycles have benefit-
females, although extensions are possible and are dis-
ted from and contributed to this newer and broader
cussed briefly later. Ages can be discrete or continuous:
focus on evolutionary biodemography. The goals of
here, we use discrete age and time (the discrete time
this chapter are to provide a critical summary of the
interval used may be a day or a month, etc.). Newborns
central ideas and methods. We emphasise theoretical
counted at the beginning of a discrete time interval are
methods, starting with the main ideas that have attract-
labelled as being in age class 1 (although some prefer
ed attention in the field, the assumptions behind these,
age 0), meaning that they are between exact age 0 and
and efforts to relax those assumptions, and provide
any age up to 1 when counted. For brevity, we often
a short account of some new directions. Only a few
use “age” interchangeably with “age class.” In discrete
empirical findings are discussed. The reader will find
time, the growth rate of the population, called λ, is geo-
more extensive accounts of empirical work elsewhere
metric per interval of time. It is often convenient to use
in this volume. We are deliberately not historically
the equivalent exponential rate r = logλ. Then r solves
complete or comprehensive, but give references that
the famous Lotka equation,
the reader can use as points of departure and discovery.
We begin with the classic work of Peter Medawar X
β
and William Hamilton but in somewhat modern form. 1= e−rx l(x) m(x), (16.1)
α
Their ideas and results are basic in evolutionary
demography, and we discuss briefly the connections, and using the age-specific survival probabilities
applications, assumptions, and limitations. Then we
px = survival rate from age x to age x + 1,
discuss extensions to variable (random) environments
and the large body of theory around that topic. Next l1 = 1, lx = p1 p2 . . . px−1 , x ≥ 2.
Shripad Tuljapurkar and Wenyun Zuo, Evolutionary Demography. In: Demographic Methods Across the Tree of Life.
In this equation the symbols α and β are the first and balancing the arrival rate against the corresponding
last ages of reproduction respectively. The approach we force of selection is
have used here is suited to humans, but details may
ν
differ for other species (Caswell 2001). q* = . (16.3)
|smx |
The Lotka equation makes clear that r (or the
equivalent one-period growth rate λ = er ) is a func- Key assumptions here are that mutations occur at a
tion of α, β, px , and mx —and these parameters make single autosomal locus in a large population (so drift
up a life history. This equation allows one to analyse is not an issue) and that mutations have an additive
the relationship between life history components, but effect. Generalisations of eqn. (16.3) to consider effects
of course the equation, by itself, cannot tell what the on multiple ages and correlations across ages are dis-
values of these quantities can or should be. Rather, cussed by Charlesworth (2001).
this is a “stone tablet” equation. To use it, we must Important aspects and implications of these
be given (as if on a tablet) a range of life histo- equations are as follows:
ries, usually defined by life history parameters, and
(a) The sensitivities above have been explored in
for each parameter set within this range the Lot-
detail, as also have the corresponding elasticities
ka equation yields corresponding values of λ. How
that are defined as follows: given spx , the corre-
then should we use the Lotka equation to understand
sponding elasticity is
evolution?
The first task is to identify fitness, and Hamilton (and epx = (px /λ) spx = (∂log λ/∂log px ) . (16.4)
presumably Medawar) assumed that fitness is given by
λ in eqn. (16.1). This idea goes back to Fisher (1930) An example of these is given for red deer (Cervus
and the foundations of evolutionary theory. One can elephas, Benton et al. (1995) for data) in Figure 16.1.
now work “locally”, as Hamilton did, and compute the Clearly we can define sensitivities and elasticities
changes in the fitness λ produced by small changes in for changes in any kind of life cycle, for example,
the life history. If the small changes are assumed to be one defined by stage rather than age, or one that
due to mutations, the changes in fitness can be inter- requires age+stage. This approach has been used
preted as selection acting on the mutations, and we to gain many ecological and evolutionary insights
can think about local evolutionary change of the life (e.g. cases in this book). See Caswell et al. (2018)
history. In our view, Medawar’s arguments are essen- for recent developments and extensions, as well as
tially similar. This local approach has been extensively references to earlier work. We should note (Caswell
developed and expounded in ecological and evolution- 2001) that selection pressures are directly related to
ary theory (Charlesworth 1994; Caswell 2001). Similar sensitivities, not elasticities.
arguments have been used to study random environ- (b) It should be obvious that the elasticities are zero
ments and other factors. at ages where the vital rates are currently zero.
The local question that Hamilton asked is: What is This is always true when fertility or survival are
the force of selection against a mutation that causes set at zero by the biology. For example, fertility is
a small change in survival or fertility at age x? The always zero at ages x past the maximum potential
answer lies in the partial derivatives of λ, also known reproductive age, such as age at menopause for
as sensitivities: humans. Figure 16.2 shows the difference between
sensitivities and elasticities to fertility and survival
spx = (∂λ/∂px ) and smx = (∂λ/∂mx ) . (16.2) for a modern human population. The age limit of
reproduction is critical to any work that attempts
Thus, if a change in only fertility mx at age x is caused to apply Hamilton-type theories on senescence to
by a mutation acting only at that age, we can interpret species with post-reproductive life. In particular
smx as the force of selection acting on that mutation. note that the Lotka eqn. (16.1) shows that λ only
Of course the forces of selection in eqn. (16.2) are depends on reproduction and survival between
negative when the mutations decrease fertility or the ages of first and last reproduction. Some
survival, as expected of the vast majority of mutations previous work ignores this limitation or tries to
that are likely to be deleterious. If mutations reducing circumvent this point in unconvincing ways. For
fertility at age x occur at a constant rate ν, and if the example, the otherwise elegant analysis in Abrams
effect of mutations on fertility is simply additive, the and Ludwig (1995) is only relevant when there is
equilibrium frequency of mutations maintained by no upper limit to the age of reproduction.
E V O L U T I O N A RY D E M O G R A P H Y 275
0.15 (a) (b)

0.15
Sensitivities to survival
Elasticities to survival
0.1
0.1
0.05
0.05
0 0
(c) (d)
0.02
0.1
Sensitivities to fertility
Elasticities to fertility
0.06
0.01
0.02
Figure 16.1 The sensitivities and
0 0
elasticities to survival and fertility for Red
5 10 15 5 10 15 deer (Cervus elephas). (Benton et al.
Age (years) Age (years) (1995) for data).
0.04 0.04
(a) (b)
Sensitivities to survival
Elasticities to survival
0.03 0.03
0.02 0.02
0.01 0.01
0 0
0.003 (c) (d)
0.03
Sensitivities to fertility
Elasticities to fertility
0.002
0.02
0.001 0.01
Figure 16.2 The sensitivities and

0 0 elasticities to survival and fertility for Japan
0 10 20 30 40 50 0 10 20 30 40 50 at 2009. (Data provided by van Daalen
Age (years) Age (years) and Caswell (2017).)
(c) The reader will note that all sensitivities decrease eqn. (16.3) increases at old age. And once repro-
with increasing age (after the age of first reproduction stops, mutational frequencies should
duction), a general feature of age-structured life increase towards 1 (i.e. fixation). In such cases,
histories (Caswell 2001). In consequence, the some deleterious mutations will always get in, and
equilibrium frequency of mutations implied by more of them get in at late age, so that mutations
should “chip” away at late-age reproduction. Thus generally, the Hamilton theory does not apply
the Hamilton theory by itself must surely lead to directly in variable environments, and we discuss
an evolutionarily unstable age of last reproduc- stochastic extensions of the theory below.
tion. This begs the question, why are we all not like
salmon (Tuljapurkar 1997)? Wachter et al. (2013) One can also work more “in the large,” asking
call this limit, more strikingly, a “wall of death.” whether trade-offs and constraints may be used to
(d) The preceding point also poses a fatal obstacle to define a region of fitness space in which some set
Williams’s celebrated view that we should think of life histories is more fit than alternatives. These
about ageing as a result of antagonistic pleiotropy. constraints may be energetic, as in arguments that
This view argues that vital rates at early ages are metabolic rate reflects the “pace of life.” The produc-
high because late-age vital rates are low, and that tion rate scales with body mass in the same way as
such trade-offs are genetic, hence the antagonism metabolic rate, so that, per unit mass, large sized organ-
(Williams 1996). But such antagonism can only isms have slower rates of production than small sized
strengthen the instability of life histories (with ones (Brown and Sibly 2006). On the other hand, the
a defined age of last reproduction), as discussed variations in lifestyles and environmental conditions
above. Certainly cases of antagonistic pleiotropy also affect the evolution of body size and produc-
can be found – but antagonistic pleiotropy cannot tion rates, thus directly affecting energy consumption
provide a general explanation for senescence. and allocation (Sibly and Brown 2007). Constraints
(e) One direction of improvement of the theory is to may also be biomechanical. Biomechanical designs of
analyse mutations that affect a range of ages, so organisms are critical to performance of organisms
that the selection pressure does not depend only which affects their fitness, such as the effect of fish
on one focal late age. In any model using an age morphology on swimming performance (Langerhans
range, mutations are selected against over the and Reznick 2009; Higham et al. 2016) and plant traits
entire age-range as in Charlesworth (2009). This on support and anchorage (Niklas 1992, 1994; Read
is, of course, an example of correlated effects of and Stokes 2006). Finally, factors such as energetics,
deleterious mutations across several ages, which biomechanics, and genetic correlations also act at the
is an example of positive rather than antagonistic population level and can be based on representations
pleiotropy. of the ecological dynamics that affect life histories, as
(f) Another and even more important direction in the arguments about “r-K” selection and the fast-
recognises that existing mutations must decrease or slow-paced lifestyle (Stearns 1992). The latter are
individual fitness. As a result, the growth rate λ of discussed later in this chapter.
a population depends on the distribution of muta-
tions over individuals in the population, which of
course requires a nonlinear theory, which has been
developed by Evans et al. (2013). In a nonlinear
16.3 Randomly varying environments
theory, mutations can affect several ages, the Unpredictable environmental changes are so ubiqui-
effect of mutations on fitness is not additive, and tous that it makes sense to replace the Lotka eqn. (16.1)
different individuals may carry different numbers by equations that include variability. In the pres-
of mutations. See the reference for more details. ence of temporal variation, a life history necessarily
(g) The great increase in human life span in the past depends on several factors – the average vital rates,
century has mostly occurred at older ages, that variances and covariances of vital rates at any one time,
is, ages past menopause. This change cannot be and serial correlations between times now and in the
explained by the application of traditional Hamil- future (or past). The latter quantities tell us how envi-
ton theory, or even some improved versions of it. ronments vary and how vital rates respond to that
A different approach is needed and is discussed variation.
below. The relevant theory originated with Furstenberg
(h) The Hamilton theory assumes the population is in and Kesten (1960) and Cohen (1976), and later devel-
demographic equilibrium with constant vital rates. opments and other references are discussed in Tul-
In cases where density is assumed to maintain japurkar (1990). The theory has since been extensively
an equilibrium population, Caswell (2001) shows explained, developed, and applied. Let nt denote the
how a density-independent analysis is relevant to vector of population numbers at any time t. Then the
density-dependent models at equilibrium. More population between times (t−1) and t is governed by a
random vital rate matrix At , and the population vector with respect to changes in the average b(i, j) of the (i, j)
at time t is given by element.
Assume the random matrix Ht has stationary
nt = At nt−1 . (16.5)
variance-covariance at all lags (e.g., when the
In this interval, the total population sizes change from environments are stationary and ergodic). At each
Nt − 1 at (t − 1) to Nt with (time-varying) growth rate time t, so at time lag zero, the variance-covariance
Nt matrix has elements
λt = . (16.6)
Nt−1 E{H1 (i, j) H1 (k, l)}, (16.12)
At each time, the environmental state determines the
and at times separated by a time lag M > 0, the serial
projection matrix that applies; for example, if the envi-
variance-covariance matrix has elements
ronment is in state x at time t, that state determines that
the projection matrix is Bx . The long-run population E{HM+1 (i, j) H1 (k, l)}. (16.13)
growth rate converges to a (deterministic) limit. Over
When the scaling factor ϵ is small, Tuljapurkar (1982)
time, environments, or equivalently matrices, occur in
showed that the stochastic growth rate is given by the
a random sequence that we denote by the symbol ω.
general small-noise expansion
Over such a sequence, the cumulative growth rate from
t = 0 to t = T is the product of single-period growth a ≈ log λ − W1 + W2 , (16.14)
rates,
where both W 1 and W 2 are O(ϵ2 ).
The term W 1 is due
ΛT (ω) = λT λT−1 . . . , λ1 . (16.7) to interannual variability (i.e., variability within each
The theory ensures that every random sequence time interval) whereas W 2 is due to autocorrelation
(except sequences of zero probability) has the same between distinct times. The contribution W 1 is
stochastic growth rate 1 X
W1 = sij sk,l Cov{H1 (i, j) H1 (k, l)}. (16.15)
2λ2
1 ij,kl
a = lim log ΛT . (16.8)
T→∞ T
The effects of serial correlations and the nature of W 2
Existence of this limit occurs under wide conditions are discussed by Tuljapurkar and Haridas (2006) and
(Tuljapurkar 1990). In brief, the requirements are weak more generally by Tuljapurkar et al. (2009).
demographic ergodicity and mixing of the stochastic Important implications of these equations are as
process (a finite number of matrices is not required). follows:
The limit in eqn. (16.8) clearly equals an average over
sequences, (a) A basic question: Does the relative fitness of a life
history depend on the amount of environmental-
1
a = lim E [log ΛT (ω)]. (16.9) ly driven variability in vital rates? The answer is
T→∞ T
a decided yes. We illustrate with a simple two-
Fitness is now the stochastic growth rate a defined by stage life history with delays possible at juvenile
eqn. (16.9), from which one can compute a growth rate and/or adult stages (Tuljapurkar and Wiener 2000).
λS = ea . These can be simply represented using the projec-
Now separate the average projection matrix and the tion matrix
random bit by writing
p Rt (1 − q)
At = . (16.16)
At = B + ϵ Ht , (16.10) (1 − p) q
The extent of delay at each stage is measured
E (Ht ) = 0, (16.11)
by parameters (p, q). Thus if (p,q)=(0,0) then all
where ϵ > 0 is a scale factor that determines the mag- individuals live exactly two time-steps, one as an
nitude of the random effects, and E indicates an immature and one as an adult producing Rt off-
expectation (average) over random environments. The spring. If p > 0, then immature may remain in
projection matrix B contains average vital rates and that state for several time-steps, and if q > 0 then
has a dominant eigenvalue λ. This theory applies to adults can “spread” their mean production of Rt
age+stage models, so in our discussion we take the offspring over several time-steps. Note that these
projection matrix At to have elements At (i, j) in row i delay parameters are always fixed here. Stochastic-
and column j. The average matrix B has sensitivities si,j ity enters via the reproductive rate Rt which varies
0.08 0.05
0.06 0.04
a 0.04 a 0.03
0.02 0.02
0.00 0.01
1.0 1.0
0.8 1.0 0.8 1.0
0.8 0.6 0.8
0.6
0.6 0.6
q 0.4 0.4 q 0.4 0.4
p 0.2 p
0.2 0.2 0.2
0.0 0.0 0.0 0.0
Figure 16.3 Fitness surface for constant environment. a: Stochastic Figure 16.4 Fitness surface for variable environment. a: Stochastic
growth rate (see eqn. 16.14). p: Extent of delay at juvenile stage (see growth rate (see eqn. 16.14). p: Extent of delay at juvenile stage (see
eqn. 16.16). q: Extent of delay at adult stage (see eqn. 16.16). eqn. 16.16). q: Extent of delay at adult stage (see eqn. 16.16).
0.02
around an average R so that 0.00
a −0.02
Rt = R + Ht . (16.17) −0.04
−0.06
Now suppose there is a constant environment, in 1.0 0.8 0.6 0.4 0.2 0.8 1.0
0.0 0.0 0.2 0.4 0.6
which case the growth rate is the logarithm of the q p
dominant eigenvalue λ of the average projection
matrix. Either kind of delay slows down the life Figure 16.5 Fitness surface for variable environment with a serial
history, so logλ decreases as either p or q increas- autocorrelation. a: Stochastic growth rate (see eqn. 16.14). p: Extent
es, as shown in Figure 16.3. Therefore in a con- of delay at juvenile stage (see eqn. 16.16). q: Extent of delay at adult
stage (see eqn. 16.16).
stant environment we do not expect delayed life
cycles to evolve. But in a variable environment the
projection matrix (eqn. 16.16) is randomly varying. covariance elements in eqns. 16.12 and 16.13. As a
The stochastic growth rate a in the presence of vari- first step, we can simply look at changes in average
ability, computed by simulation, shows (Figure16.4) rates and compute elasticities such as
that a small amount of variability causes the growth
∂a
rate to increase sharply as the amount of delay (of .
either kind) increases, and then a falls when delay is ∂log b(i, j)
long. Clearly this is an argument for the evolution Caswell (2001) discusses how the small-noise
of life cycle delays and would not hold in a con- approach can be used to understand these. But
stant environment. We note also that, in this model, more generally, there are many elasticities of a,
serial autocorrelation of reproduction Rt can pro- as explained with examples by Tuljapurkar et al.
duce a positive feedback effect on fitness and thus (2003). In that paper, we point out that a general
on evolution. As seen in Figure 16.5, autocorrelation decomposition of any change in vital rates can be
leads to a greater increase of fitness when just one analysed by combining the effects on fitness when
type of delay increases. Evolution is then expected one changes just the mean or just the variance of
to lead to life histories in which one or the other kind vital rates. This decomposition is applied to life
of delay predominates. This feedback effect is seen history evolution in Haridas and Tuljapurkar (2005).
more clearly in a contour plot, Figure 16.6. (c) More generally we can examine how a changes
(b) Fitness a in a variable environment depends on with respect to the the variation in the environment
all aspects of how the vital rates vary in response and/or the vital rates. A given environmental
to environments. So fitness depends not just on pattern leads to environmental sequences, and
average vital rates but also on all of the variance- we can change the vital rates associated to each
0.
00
fitness in deterministic environments (Orzack 1985;
0.
01 50 Tuljapurkar and Wiener 2000).
00
0.
02
0.8 00
16.4 Post-reproductive life
0.0302
Post-reproductive life is fairly common among female
0.0301
0.6 mammals (Cohen 2004; Ellis et al. 2018). In humans, the
q
end of female reproduction is marked by menopause at
1
age around 50 years. There has been far greater down-
0
0.03
0.4 ward secular change in the age at menarche (Frisch
and Revelle 1971; Prentice et al. 2010) than in the age
0.
0.030
at menopause over the past century. Yet this period
02
00
0.2 0
0. .010 2 has seen a dramatic increase, essentially a doubling,
00 0
50
in human lifespan (Tuljapurkar and Boe 1998). Gurven
and Kaplan (2007) even present data suggesting that
early humans lived well past the age of menopause.
0.2 0.4 0.6 0.8
In recent humans, extended life is unlikely (probably
p
impossible) to be the result of genetic change (Zuo
Figure 16.6 Contours for variable environment with a serial et al. 2018), which means that (1) post-reproductive life
autocorrelation. The contours show the stochastic growth rate a (see is a very plastic trait, and (2) in evolutionary terms,
eqn. 16.14). p: Extent of delay at juvenile stage (see eqn. 16.16). selection must have made this plasticity possible by
q: Extent of delay at adult stage (see eqn. 16.16). preventing the accumulation of mutations that would
make rapid increases impossible. Even a short human
environmental state – which affects average rates life after menopause needs evolutionary explanation,
or the deviations in eqn. (16.10). Alternatively, we as more broadly discussed by Levitis et al. (2013).
can maintain the association between vital rates We address here one general class of evolution-
and environments, but change the rules governing ary arguments about post-reproductive life: that older
environmental sequences. As a result we have two individuals “borrow fitness” from younger females
kinds of sensitivities: derivatives of a with respect who are reproductively capable. Such borrowing
to vital rates and derivatives of a with respect to occurs when old individuals (who cannot reproduce)
environmental pattern. The paper by Steinsaltz take actions that increase the fitness of younger indi-
et al. (2011) provides fast numerical methods of viduals (who can reproduce). As a result, selection
computing a and both kinds of derivatives. in favour of reproduction by younger females will
(d) The importance of autocorelation in understanding act indirectly to maintain the longer life of older
evolutionary demography was analysed in detail in individuals. One version of this is provided by the cele-
an important paper by Orzack (1993). New analyses brated grandmother hypothesis (Hawkes et al. 1998), a
in Paniw et al. (2018) present analyses of the quali- second is the older-male hypothesis (Tuljapurkar et al.
tative and quantitative effects of serial correlation. 2007), and the most general is the transfer hypothe-
More general perspectives on the importance of sis (Lee 2003). Lee’s analysis is framed in the context
environmental variability in life history evolution of economic transfers but applies to transfers of any
are given by Tuljapurkar et al. (2009). kind including the provision of care, the sharing of
(e) Empirical studies of serial correlation are unfor- knowledge or information, and so on, and is the most
tunately still rare. However, eqn. (16.14) provides general formulation (Tuljapurkar 2019). Useful matrix
useful ways of analysing data to separate the contri- models of such models have been analysed by Pavard
butions of variability and correlation, as shown by and Branger (2012).
Davison et al. (2010). The small-noise method can It is important to note that an analysis of one set
also be used in combination with randomization of historical data by Moorad and Walling (2017) does
to understand whether a data set evidences serial not support these hypotheses. However, Moorad and
autocorrelation (Jongejans et al. 2010). Walling use quantitative genetic models to translate the
(f) Finally we note the robust conclusion that environ- evolutionary hypotheses into hypotheses about cor-
mental variability allows for the coexistence of very relations in the data. It may be useful to reexamine
different life histories that would have unequal the relevant arguments and look at other data sets.
However, some new evolutionary hypotheses may be provide examples of unstable life histories that, pre-
needed (Cohen 2017)! sumably, are not held together by positive pleiotropy.
The field of evolutionary demography has received This is a surprise because earlier additive theories
considerable impetus from work on human senescence for mutation accumulation assumed that positive
and human health in old age. The existence of extend- pleiotropy was a way of transferring fitness across
ed human lives is, as we have seen, a challenge for the life cycle. The demographic analysis may seem
evolutionary theory. But for evolutionary arguments technically formidable, but the examples in the paper
to bear on human health, beyond the usual “just-so” (Wachter et al. 2013) show that numerical implemen-
stories, is even more of a challenge. Metabolic theo- tation should be relatively straightforward. These non-
ry has been of value in many ecological settings (as linear analyses are important for many questions in the
illustrated by Brown and Sibly (2006); Sibly and Brown evolutionary demography of complex life cycles, and
(2007)), and a different kind of metabolic theory has extensions and developments should be valuable.
been important in human health and healthy aging
(Steptoe et al. 2015; Camandola and Mattson 2017). The
time is ripe for a synthesis of these streams of thought
16.6 Density-dependent models
and may open a useful wedge into the effects of evo- Finally, we consider density dependence canonical-
lution on health. There is of course a large literature ly exemplified for one species by the logistic model
on evolutionary medicine, some of which is reviewed (Verhulst 1838), is widely used in population and com-
by Ewald (2018), but we do not believe this is useful munity ecology and describes the population by
or relevant in the context of the evolution of late-age

mortality. Nt
Nt+1 − Nt = rNt 1− (16.18)
K
16.5 Nonlinear models
where N is the population size, t is time, K is the carry-
for mutation–selection ing capacity (also known as equilibrium density), and r
A distinct and important issue is the correct represen- is the per capita growth rate (also known as the intrin-
tation of mutation–selection balance. This is essentially sic rate of increase). In continuous time, the left side
the same issue raised by Kimura and Maruyama (1966) of eqn. (16.18) is a derivative; here we ignore differ-
who considered the distribution of mutations in a pop- ences between continuous and discrete time versions
ulation. Since mutations affect individuals, any pop- (that may be very large; see e.g. Royama (1992)). In
ulation will contain individuals who carry different eqn. (16.18), a population starting at N0 < K grows until
numbers of mutations (0, 1, 2, . . . mutations). As Kimu- it reaches carrying capacity K, which is the maximum
ra and Maruyama (1966) pointed out, we can expect number of individuals the environment can support.
that selection will result in an equilibrium distribution The growth rate r is the difference between the per
of mutations. In the simplest situation, there will be capita birth rate and per capita death rate.
a random number M of mutations in any individual, MacArthur (1962) and MacArthur and Wilson (1967)
and the probability distribution p(m) = Pr{M = m} will proposed the theory of r- and K-selection, the first of
be maintained by a mutation–selection balance. This is its kind addressing density-dependent life history evo-
quite different from the simple analysis that leads to lution. The fitness in the case of r-selection is just r.
eqn. (16.3) as the number of mutations per individual. In harsh environments (such as extreme seasonality,
In the present setting, individuals will carry differ- fires, floods, or storms), populations decline because
ent numbers of mutations, and individual and average of increased density-independent mortality, then re-
fitness will depend on the relationship between muta- expand. In this scenario, selection would favour indi-
tions and individual-level vital rates. viduals that have high productivity as measured by r.
The general setting and suitable tools are described Alternatively, the fitness in the case of K-selection is the
by Evans et al. (2013). The mathematical treatment carrying capacity, K. In stable environments, popula-
here is beyond that usually found in biology, but the tions stay close to carrying capacity, so selection would
book begins with a useful informal description that favour individuals with high efficiency of resource
motivates and explains the theory. A demographic utilisation. Later work by Roughgarden (1971) united
implementation of this nonlinear theory, along with these two theories of evolution (r- and K-selection) with
a variety of pleiotropic effects of mutation, has been application to evolution in environments of interme-
applied by Wachter et al. (2013). This theory does diate harshness. Charlesworth (1971) generalised this
by assuming that fitness is a strictly decreasing func- (d) impose a consumer foraging time effect on the
tion of population size. He showed that Ki (the carrying functional responses (Abrams 1982);
capacity of genotype i) determines whether a stable (e) incorporate nonlinear density-dependence growth
polymorphism exists and ri (the per capita growth rate of the resource population (Abrams 2001, 2009a).
of genotype i) controls the composition of the polymor-
Such models have been applied to interactions among
phism. Moreover, he demonstrated that the amount of
different numbers of consumers and/or resources
genetic load in the population affected the mean sur-
(Abrams 1977, 1980, 1998, 2009b).
vival of genotypes within a generation, producing an
Any life history is directly or indirectly affected
effectively density-independent force of selection on
by traits such as body mass (Dieckmann et al. 1995;
both the r and K. In recent years, this form of selec-
Brown and Sibly 2006) and locomotion (Langerhans
tion theory has been criticised as too simplistic (Nosil
and Reznick 2009). Traits that suit environments and
et al. 2018).
thus increase individual fitness are called adaptive
Intraspecific dynamics may not be enough to under-
traits; in the opposite case we have maladaptive traits.
stand or predict life history evolution. Interspecific
At the individual level, selection presumably favours
interactions such as competition and predation may
adaptive traits. Surprisingly, adaptive traits do not
have to be included to get close to reality. When a pair
always increase population growth or numbers, and
of species (e.g., competitors, or a predator–prey pair) or
vice versa. These counter-intuitive situations have been
a multispecies system (such as a food web) is consid-
both termed as “adaptive decline” (Abrams 2019;
ered, intraspecific interactions may lead to predictions
Brady et al. 2019b). When and how often does adap-
inconsistent with single species models (Wootton 1994;
tive decline happen? In eqn. (16.19), the individual
Abrams 2007, 2019). A relevant read on the evolution
fitness of a consumer is (bcNt −d), where b, c, and d
of predator-prey interactions is Abrams (2000).
are as previously defined, but now the aim is to con-
A standard model for “consumer-resource” interac-
sider their evolution. When the consumer population
tions is the Lotka–Volterra. Incorporating a density-
is at equilibrium, the equilibrium population size of
dependent resource (MacArthur 1970, 1972) leads to
resource is
the model
d
N* = . (16.21)
Nt bc
Nt+1 − Nt = rNt 1 − − cNt Pt (16.19)
K Since natural selection should always favour increas-
ing individual fitness, adaptive evolution at con-
Pt+1 − Pt = Pt (bcNt − d). (16.20) sumer individual level should increase b and c and/or
decrease d whenever possible. But any such change
Here N and P are population sizes of resource and reduces the equilibrium resource population (eqn.
consumer respectively, t is time, r and K describe the (16.21)). Substituting eqn. (16.21) into eqn. (16.20)
resource, c is the capture rate by an average consumer, yields the equilibrium population size of the consumer
b is the conversion efficiency of captured resource into

new consumers, and d is the per capita death rate of r d
P* = 1− , (16.22)
consumers. The consumer species has linear functional c Kbc
and numerical responses, and the resource has logistic
which indicates that adaptive evolution on b, c, and
growth in the absence of the consumer.
d may affect P* in either direction. However density-
Common modifications of eqns. 16.19 and 16.20 are:
independence resource growth, where (1/K)≈ 0 in eqn.
(a) replace a linear functional response (the capture (16.19), guarantees adaptive decline (see eqn. (16.22)).
rate by an average consumer) with nonlinear ver- Abrams (2019) points out that the special case may
sions (Holling 1959; Abrams 1980, 1990, 2009a; actually be common, which implies frequent adaptive
Abrams and Ginzburg 2000); decline. This may explain why maladaptation is often
(b) replace a linear numerical response (the relation- observed (Brady et al. 2019a, 2019b). For more complex
ship between the amount eaten and the per capita consumer–resource systems, such as food webs, these
consumer reproductive rate) with a nonlinear ver- changes have yet to be explored.
sion (Abrams 2009a); In this way, models of life history evolution built
(c) impose a negative effect of consumer density on on density independence may be usefully extended
the numerical or functional responses (Abrams to incorporate the effects of certain kinds of density
1993); independence.
References Brown, J.H. and R.M. Sibly (2006). “Inaugural Article:

Life-History Evolution Under a Production Constraint”.
Abrams, P.A. (1977). “Density-Independent Mortality and
Proceedings of the National Academy of Sciences 103.47,
Interspecific Competition: A Test of Pianka’s Niche Over-
p. 17595.
lap Hypothesis”. The American Naturalist 111.979, pp.
Camandola, S. and Mark P Mattson (2017).“ Brain
539–552.
Metabolism in Health, Aging, and Neurodegeneration”.
Abrams, P.A. (1980). “Consumer Functional Response and
The EMBO Journal 36.11, pp. 1474–1492.
Competition in Consumer Resource Systems”. Theoretical
Caswell, H. (2001). Matrix Population Models: Construction,
Population Biology 17.1, pp. 80–102.
Analysis, and Interpretation. 2nd edition. Sunderland, MA:
Abrams, P.A. (1982). “Functional Responses of Optimal For-
Sinauer Associates.
agers”. The American Naturalist 120.3, pp. 382–390.
Caswell, H. et al. (2018). “Age × Stage-Classified Demo-
Abrams, P.A. (1990). “The Effects of Adaptive Behavior
graphic Analysis: A Comprehensive Approach”. Ecolog-
on the Type-2 Functional Response”. Ecology 71.3, pp.
ical Monographs 88.4, pp. 560–584.
877–885.
Charlesworth, B. (1971). “Selection in Density-Regulated
Abrams, P.A. (1993). “Why Predation Rate Should Not
Populations”. Ecology 52.3, pp. 469–474.
be Proportional to Predator Density”. Ecology 74.3,
Charlesworth, B. (1994). Evolution in Age-Structured Popula-
pp. 726–733.
tions. Cambridge: Cambridge University Press.
Abrams, P.A. (1998). “High Competition with Low Similari-
Charlesworth, B. (2001). “Patterns of Age-Specific Means
ty and Low Competition with High Similarity: Exploita-
and Genetic Variances of Mortality Rates Predicted by
tive and Apparent Competition in Consumer-Resource
the Mutation-Accumulation Theory of Ageing”. Journal of
Systems”. The American Naturalist 152.1, pp. 114–128.
Theoretical Biology 210.1, pp. 47–65.
Abrams, P.A. (2000). “The Evolution of Predator-Prey Inter- Charlesworth, B. (2009). “Effective Population Size and
actions: Theory and Evidence”. Annual Review of Ecology Patterns of Molecular Evolution and Variation”. Nature
and Systematics 31.1, pp. 79–105. Reviews Genetics 10.3, pp. 195–205.
Abrams, P.A. (2001). “The Effect of Density-Independent Cohen, A.A. (2004). “Female Post-Reproductive Lifespan:
Mortality on the Coexistence of Exploitative Competitors A General Mammalian Trait”. Biological Reviews 79.4,
for Renewing Resources”. The American Naturalist 158.5. pp. 733–750.
PMID: 18707301, pp. 459–470. Cohen, A.A. (2017). “The Mystery of Life Beyond
Abrams, P.A. (2007). “Defining and Measuring the Impact Menopause”. Nature Ecology and Evolution 1.11, p. 1604.
of Dynamic Traits on Interspecific Interactions”. Ecology Cohen, J.E. (1976). “Ergodicity of Age Structure in Popu-
88.10, pp. 2555–2562. lations with Markovian Vital Rates, I: Countable States”.
Abrams, P.A. (2009a). “Determining the Functional Form Journal of the American Statistical Association 71.354, pp.
of Density Dependence: Deductive Approaches for Con- 335–339.
sumer Resource Systems Having a Single Resource”. The Davison, R. et al. (2010). “Advances in Plant Demogra-
American Naturalist 174.3, pp. 321–330. phy Using Matrix Models”. Journal of Ecology 98.2, pp.
Abrams, P.A. (2009b). “The Implications of Using Multiple 255–267.
Resources for Consumer Density Dependence”. Evolution- Dieckmann, U., Paul Marrow, and Richard Law (1995).
ary Ecology Research 11, pp. 517–540. “Evolutionary Cycling in Predator-Prey Interactions: Pop-
Abrams, P.A. (2019). “How Does the Evolution of Univer- ulation Dynamics and the Red Queen”. Journal of Theoret-
sal Ecological Traits Affect Population Size? Lessons from ical Biology 176.1, pp. 91–102.
Simple Models”. The American Naturalist 193.6. PMID: Ellis, S. et al. (2018). “Postreproductive Lifespans are Rare in
31094600, pp. 814–829. Mammals”. Ecology and Evolution 8.5, pp. 2482–2494.
Abrams, P.A. and Lev R. Ginzburg (2000). “The Nature of Evans, S.N., David Steinsaltz, and Kenneth W Wachter
Predation: Prey Dependent, Ratio Dependent or Neither?” (2013). A Mutation–Selection Model with Recombination
Trends in Ecology & Evolution 15.8, pp. 337–341. for General Genotypes. Vol. 222. Providence, RI: Ameri-
Abrams, P.A and Donald Ludwig (1995). “Optimality The- can Mathematical Society.
ory, Gompertz’law and the Disposable Soma Theory of Ewald, P.W. (2018). Darwinian Medicine: Evolutionary
Senescence”. Evolution 49.6, pp. 1055–1066. Approaches to Disease. In: H. Callan (ed.) The Internation-
Benton, T. G., A. Grant, and T. H. Clutton-Brock (1995). al Encyclopedia of Anthropology, pp. 1–9. John Wiley & Sons,
“Does Environmental Stochasticity Matter? Analysis of Hoboken.
Red Deer Life-Histories on Rum”. Evolutionary Ecology 9.6, Fisher, R. A. (1930). The Genetical Theory of Natural Selection.
pp. 559–574. Oxford: Clarendon Press.
Brady, S.P. et al. (2019a). “Causes of Maladaptation”. Evolu- Frisch, R.E. and Roger Revelle (1971). “Height and Weight
tionary Applications 12.7, pp. 1229–1242. at Menarche and a Hypothesis of Menarche”. Archives of
Brady, S.P. et al. (2019b). “Understanding Maladaptation by Disease in Childhood 46.249, pp. 695–701.
Uniting Ecological and Evolutionary Perspectives”. The
American Naturalist 194.4. PMID: 31490718, pp. 495–515.
Furstenberg, H. and H. Kesten (1960). “Products of Ran- Medawar, P.B. (1957). An Unsolved Problem of Biology.
dom Matrices”. The Annals of Mathematical Statistics 31.2, In: P.B. Medawar (ed.) The Uniqueness of the Individual.
pp. 457–469. London: Routledge.
Gurven, M. and Hillard Kaplan (2007). “Longevity Among Moorad, J.A. and Craig A Walling (2017). “Measuring Selec-
Hunter-Gatherers: A Cross-Cultural Examination”. Popu- tion for Genes that Promote Long Life in a Historical
lation and Development Review 33.2, pp. 321–365. Human Population”. Nature Ecology & Evolution 1.11,
Hamilton, WD (1966). “The Moulding of Senescence by p. 1773.
Natural Selection”. Journal of Theoretical Biology 12.1, Niklas, K.J. (1992). Plant Biomechanics: An Engineering
pp. 12–45. Approach to Plant Form and Function. Chicago, IL: Univer-
Haridas, C. V. and S. Tuljapurkar (2005). “Elasticities in sity of Chicago Press.
Variable Environments: Properties and Implications”. The Niklas, K.J. (1994). “Morphological Evolution Through
American Naturalist. 166.4, pp. 481–95. Complex Domains of Fitness”. Proceedings of the National
Hawkes, K. et al. (1998). “Grandmothering, Menopause, and Academy of Sciences 91.15, pp. 6772–6779.
the Evolution of Human Life Histories”. Proceedings of the Nosil, P. et al. (2018). “Natural Selection and the Predictabil-
National Academy of Sciences 95.3, pp. 1336–1339. ity of Evolution in Timema Stick Insects”. Science 359.6377,
Higham, T.E. et al. (2016). “Speciation Through the Lens of pp. 765–770.
Biomechanics: Locomotion, Prey Capture and Reproduc- Orzack, S. H. (1985). “Population Dynamics in Variable
tive Isolation”. Proceedings of the Royal Society B: Biological Environments. V. The Genetics of Homeostasis Revisit-
Sciences 283.1838, p. 20161294. ed”. The American Naturalist, pp. 550–572.
Holling, C. S. (1959). “The Components of Predation as Orzack, S. H. (1993). “Life History Evolution and Population
Revealed by a Study of Small-Mammal Predation of the Dynamics in Variable Environments: Some Insights from
European Pine Sawfly”. The Canadian Entomologist 91.5, Stochastic Demography”. Lecture Notes in Biomathematics
293–320. 98, pp. 63–104.
Jongejans, E., Hans De Kroon, S. Tuljapurkar, and Katriona Paniw, M., Arpat Ozgul, and Roberto Salguero-Gomez
Shea (2010). “Plant Populations Track Rather than Buffer (2018). “Interactive Life-History Traits Predict Sensitivi-
Climate Fluctuations”. Ecology Letters 13.6, pp. 736–743. ty of Plants and Animals to Temporal Autocorrelation”.
Kimura, M. and T. Maruyama (1966). “The Mutational Load Ecology Letters 21.2, pp. 275–286.
with Epistatic Gene Interactions in Fitness”. Genetics 54.6, Pavard, S. and Frédéric Branger (2012). “Effect of Mater-
pp. 1337–1351. nal and Grand Maternal Care on Population Dynamics
Langerhans, R. B. and David N. Reznick (2009). Ecology and and Human Life-History Evolution: A Matrix Projec-
Evolution of Swimming Performance in Fishes: Predicting tion Model”. Theoretical Population Biology 82.4, pp.
Evolution with Biomechanics. Fish Locomotion: An Etho- 364–376.
Ecological Perspective. Ed. by P. Domenici and B.G. Kapoor. Prentice, S. et al. (2010). “Evidence for a Downward Secular
Boca Raton, FL: Taylor & Francis Group. Chap. 7, pp. Trend in Age of Menarche in a Rural Gambian Popula-
200–248. tion”. Annals of Human Biology 37.5, pp. 717–721.
Lee, R.D. (2003). “Rethinking the Evolutionary Theory of Read, J. and Alexia Stokes (2006). “Plant Biomechanics in
Aging: Transfers, not Births, Shape Senescence in Social an Ecological Context”. American Journal of Botany 93.10,
Species”. Proceedings of the National Academy of Sciences pp. 1546–1565.
100.16, pp. 9637–9642. Roughgarden, J. (1971). “Density-Dependent Natural Selec-
Levitis, D.A., Oskar Burger, and Laurie Bingaman Lackey tion”. Ecology 52.3, pp. 453–468.
(2013). “The Human Post-Fertile Lifespan in Comparative Royama, T. (1992). Analytic Population Dynamics. London:
Evolutionary Context”. Evolutionary Anthropology: Issues, Chapman and Hall.
News, and Reviews 22.2, pp. 66–79. Salguero-Gómez, R. et al. (2015). “The compadre Plant
MacArthur, R.H. (1962). “Some Generalised Theorems of Matrix Database: An Open Online Repository for Plant
Natural Selection”. Proceedings of the National Academy of Demography”. Journal of Ecology 103.1, pp. 202–218.
Sciences 48.11, pp. 1893–1897. Salguero-Gómez, R. et al. (2016). “COMADRE: A Glob-
MacArthur, R.H. (1970). “Species Packing and Competi- al Data Base of Animal Demography”. Journal of Animal
tive Equilibrium for Many Species”. Theoretical Population Ecology 85.2, pp. 371–384.
Biology 1.1, pp. 1–11. Sibly, R. M. and James H. Brown (2007). “Effects of Body
MacArthur, R.H. (1972). Geographical Ecology. Princeton, NJ: Size and Lifestyle on Evolution of Mammal Life Histo-
Princeton University Press. ries”. Proceedings of the National Academy of Sciences 104.45,
MacArthur, R.H. and E. O. Wilson (1967). The Theory of Island pp. 17707–17712.
Biogeography. Princeton, NJ: Princeton University Press. Stearns, S. C. (1992). The Evolution of Life Histories. Oxford:
Max Planck Institute for Demographic Research Oxford University Press.
(Germany) and Vienna In stitute of Demography Steinsaltz, D., S. Tuljapurkar, and C. Horvitz (2011). “Deriva-
(Austria) (2019). Human Fertility Database. Available at tives of the Stochastic Growth Rate”. Theoretical Population
www.humanfertility.org. Biology 80.1, pp. 1–15.
Steptoe, A., Angus Deaton, and Arthur A Stone (2015). Random Environments”. The American Naturalist 162.4,
“Subjective Wellbeing, Health, and Ageing”. The Lancet pp. 489–502.
385.9968, pp. 640–648. Tuljapurkar, S., J-M. Gaillard, and T. Coulson (2009). “From
Tuljapurkar, S (1982). “Population Dynamics in Variable Stochastic Environments to Life Histories and Back”.
Environments. III. Evolutionary Dynamics of r-Selection”. Philosophical Transactions of the Royal Society of London.
Theoretical Population Biology 21.1, pp. 141–165. Series B, Biological Sciences 364.1523, pp. 1499–1509.
Tuljapurkar, S (1990). Population Dynamics in Variable Envi- University of California, Berkeley (USA) and Max
ronments. New York: Springer. Planck Institute for Demographic Research (Germany)
Tuljapurkar, S. (1997). The Evolution of Senescence. In: (2019). Human Mortality Database. Available at
K.W. Wachter and C.E. Finch (eds) Between Zeus and www.mortality.org or www.humanmortality.de.
the Salmon—The Biodemography of Longevity, pp. 65–77. van Daalen, S.F. and Hal Caswell (2017). “Lifetime Repro-
Washington, DC: National Academy Press. ductive Output: Individual Stochasticity, Variance, and
Tuljapurkar, S. (2019). The Challenges of Evolutionary Sensitivity Analysis”. Theoretical Ecology 10.3, pp. 355–
Biodemography and the Example of Menopause. In: 374.
O. Burger, R. Lee, and R. Sear (eds) Human Evolutionary Verhulst, Pierre-François (1838). “Notice sur la loi que la
Demography, pp. 319–334. Cambridge: Open Book Population Suit dans son Accroissement”. Correspondance
Publishers. Mathématique et Physique. 10, pp. 113–126.
Tuljapurkar, S. and Carl Boe (1998). “Mortality Change and Wachter, K.W., Steven N Evans, and David Steinsaltz (2013).
Forecasting: How Much and How Little do we Know?” “The Age-Specific Force of Natural Selection and Biode-
North American Actuarial Journal 2.4, pp. 13–47. mographic Walls of Death”. Proceedings of the National
Tuljapurkar, S. and C. V. Haridas (2006). “Temporal Auto- Academy of Sciences 110.25, pp. 10141–10146.
correlation and Stochastic Population Growth”. Ecology Williams, G.C. (1996). Adaptation and Natural Selection: A Cri-
Letters 9.3, pp. 327–337. tique of Some Current Evolutionary Thought. Princeton, NJ:
Tuljapurkar, SD, Cedric O Puleston, and Michael D Princeton University Press.
Gurven (2007). “Why Men Matter: Mating Patterns Wootton, J. T. (1994). “The Nature and Consequences of Indi-
Drive Evolution of Human Lifespan”. PLOS One 2.8, rect Effects in Ecological Communities”. Annual Review of
e785. Ecology and Systematics 25, pp. 443–466.
Tuljapurkar, S. and P. Wiener (2000). “Escape in Time: Stay Zuo, W. et al. (2018). “Advancing Front of Old-Age Human
Young or Age Gracefully?” Ecological Modelling 133.1–2, Survival”. Proceedings of the National Academy of Sciences
pp. 143–159. 115.44, pp. 11209–11214.
Tuljapurkar, S., C. Horvitz, and J Pascarella (2003). “The
Many Growth Rates and Elasticities of Populations in
CHAPTER 17
Reproductive value and analyses

of population dynamics of
age-structured populations
Bernt-Erik Sæther and Steinar Engen
17.1 Introduction stochasticity. We will illustrate the practical use of

this approach by analyses of the dynamics of a small
Many populations show large temporal variation in
island population of moose (Alces alces). Finally, we
age structure, often related to fluctuations in the envi-
will show that these methods can be particularly use-
ronment (Coulson et al. 2001; Planque et al. 2010; Vlam
ful in viability analyses of small populations of vul-
et al. 2017) or the effects of density regulation (Lande
nerable or endangered species by estimating extinc-
et al. 2006). In particular, exploited populations tend
tion probabilities of Eurasian lynx (Lynx lynx) in
to show large changes in age distributions caused by
Scandinavia.
variation in the rate of harvesting (Toresen and Østvedt
2000; Rouyer et al. 2011; Pukk et al. 2013). These fluctua-
tions generate variation in population size, which often 17.2 Stochastic theory
introduce time-lags in the population dynamics (Lande
In populations with no age structure but overlapping
et al. 2002). This makes it difficult to identify the under-
generations, the contribution W from a female to the
lying factors causing variation in population growth.
next generation is her number of female offspring sur-
Thus, the effects of fluctuations in age distribution are
viving to the next census (usually at time steps of one
important to include in analyses of how populations
year) plus one when mother survives herself. During
are expected to respond to variation in the environment
a time step the population then changes by ∆N =
(Perry et al. 2010). P
W − N, where N is the population size. Writing Z
The purpose of this chapter is to summarise the use
for a vector of variables expressing the environment
of an approach originally suggested by Fisher (1930)
that the population experience, the variance of the con-
based on the concept of the total reproductive value,
tributions can then according to Engen et al. (1998) be
that is, the sum of all individual reproductive val-
decomposed into stochastic components as
ues, to study the dynamics of populations. Because
fluctuations in total reproductive value approximates
var(W) = Evar(W|Z) + varE(W|Z) = σd2 + σe2 ,
well to temporal variation in population size even for
species with complex life histories (Engen et al. 2007),
where σd2 = Evar(W|Z) is the variance of individual
this variable provides an approach to account for the
contributions within years averaged over time, called
effects of stochastic fluctuations in the age distribu-
the demographic variance, while σe2 = varE(W|Z) is the
tion on the population dynamics. Another advantage
temporal variance of the mean individual contribution
of this approach is that the use of this concept enables
within years, called the environmental variance. The
decomposition of the long-run growth rate (Lande
expected change in population size is then
et al. 2003) into stochastic components caused by age-
specific variation in demographic and environmental E(∆N|N) = NEW − N = (λ − 1)N,
Bernt-Erik Sæther and Steinar Engen, Reproductive value and analyses of population dynamics of age-structured populations.
where λ = EW and the variance of change in population 2013). As in the case of no age structure, the vari-
size is var(∆N|N) = σd2 N + σe2 N2 (Engen et al. 1998). ance of the yearly changes in total reproductive value
We then consider the corresponding stochastic theo- can be decomposed into additive environmental and
ry for an age-structured model with projection matrix L demographic components, thus:
fluctuating among years with no temporal autocorrela-
tions so that the population size next year is n+∆n = Ln. var(∆V|V) ≈ σe2 N2 + σd2 N ≈ σe2 V2 + σd2 V.
Here n is the column vector (n1 , n2 . . . nk )T , where ni is To analyse population growth and extinction, one com-
the number of individuals in age class i (Caswell 2001). monly works on the logarithmic scale (Lewontin and
Under a prebreeding census scenario, the first row of Cohen 1969; Cohen 1977, 1979; Tuljapurkar 1982; Lande
L is the mean fecundities in the age classes includ- et al. 2003; Engen et al. 2009b), which shows that
ing first-year survival, and the subdiagonal defines the
fractions surviving to the next year. σe2 σd2 −1
var(∆ ln V|V) ≈ + V
Let the temporal expectation of the stochastic pro- λ2 λ 2
jection matrix be l = EL and let λ now denote the real and hence the expected growth on log scale, to the same
dominant eigenvalue of l. The associated right and order of approximation, is
left eigenvectors u (column vector) and v (row vector)
are then defined by lu = λu and vl = λv, respective- σe2 σ2
E(∆ ln V|V) ≈ r − 2
+ d2 V−1
ly. In mathematical literature these vectors are usually 2λ 2λ
scaled to have unit length, while in demography it is where r = ln λ. When working on the log-scale the fac-
P P
more appropriate to require ui = 1 and vu = ui vi = 1 tors λ−2 are often included in the definitions of demo-
(Engen et al. 2009b), where u is the stable age distribu- graphic and environmental variances (Lande et al.
tion in the deterministic model with projection matrix 2003).
l. The components of the vector v are called the repro- Under small and moderate stochastic fluctuations in
ductive values of the different age classes (Caswell age structure the process V can be approximated by a
1978), a concept originating from Fisher (1930) when diffusion with infinitesimal mean and variance μ(v) =
dealing with age-structured models in continuous age λv and ν(v) = σe2 v2 + σd2 v, respectively. When the pop-
and time. The age-specific distribution of reproductive ulation size is large enough to ignore the demographic
P
values scaled so that ui vi = 1 is termed the Fisherian noise, then ln V is approximately a Brownian motion
stable age distribution (Engen et al. 2011) and express- with drift parameter r − σe2 /2, which is often called the
es how the total reproductive value V is distributed long-run growth rate, and variance σe2 .
among age classes if the population is at the stable age Many species subject to population viability analy-
distribution u. Sæther et al. (2013) empirically showed ses have complex life histories. Such complexities make
that in several species of birds and mammals the Fish- it difficult to provide reliable population trajectories
erian stable age distribution is generally characterised and hence obtain estimates of the risk of extinction
by a large contribution of younger age classes to V, (Fieberg and Ellner 2001). A useful metric in such anal-
with values declining with age. The total reproduc- yses may be the probability of ultimate extinction, that
P
tive value V = vn = ni vi equals the population size is, the probability the population will reach a certain
P
N= ni if the population is exactly at its stable age lower threshold (e.g. 0 or 1 individual) (Sæther and
distribution, ni = Nui . In the average environment (no Engen 2003). Using the diffusion approximation one
stochasticity) V + ∆V = v(n+∆n) = vln = λvn = λV (Fish- can show using the general formula for the first pas-
er 1930; Crow and Kimura 1970; Charlesworth 1994; sage time to 1 (or infinite) (Karlin and Taylor 1981) that
Lion 2018), so that the total reproductive value has a the probability of ultimate extinction for a process ini-
constant multiplicative growth rate exactly equal to λ. tiated at a total reproductive value of V0 and extinction
Engen et al. (2005), (2007) showed that in a fluc- barrier at 1 is approximately
tuating environment the age distribution will fluctu-
!2r/σe2 −1
ate around u, whereas the total reproductive value V σd2 + σe2
fluctuates around N with no temporal autocorrelation σd2 + σe2 V0
to the first-order approximation. This means that the
noise in the process fluctuates independently of years, Thus, the probability of ultimate extinction shows a
which also has been found in several populations of strong increase with increasing σd2 (Sæther and Engen
birds and mammals (Engen et al. 2013; Sæther et al. 2019).
R E P R O D U C T I V E VA L U E A N D P O P U L AT I O N DY N A M I C S 287
17.3 The concept of individual mean over all individual reproductive values is accord-
P
reproductive value ingly EW̄i = λv̄i = λ ui vi = λ. Thus, the individual
reproductive value can be decomposed into additive
The statistical analysis based on individual data on sur- components Ji vi+1 and Bi v1 due to individual varia-
vival and reproduction at different ages is significantly tion in survival and fecundity schedules, respectively
simplified by introducing the concept of individual (Box 17.1).
reproductive value, defined as each individual’s contri- In contrast to the general decrease with age in the
bution to the total reproductive value of the population age-specific components of the total reproductive
the next year (Engen et al. 2009b). Consider first indi- value found in several populations of birds and mam-
viduals in a given age class i with reproductive value mals, there were much larger interspecific differences
vi . One of these individual’s contributions to the total in the pattern of age-specific variation in individual
reproductive value next year is reproductive value (Sæther et al. 2013). In six of the
Wi = Ji vi+1 + Bi vi , populations included in the study of Sæther et al.
(2013) maximum individual reproductive value was
where Ji is an indicator variable for survival (equals 1 achieved at some intermediate age (Figure 17.1).
for survival and otherwise 0) and Bi is the number of Consequently, those age classes contributing most to
offspring. Wi are stochastic quantities assumed to have the total reproductive value of the population may
the same distribution for all ni individuals in age class not necessarily be those consisting of individuals
i. Then EWi = λvi because the total reproductive value with large contributions, because the former is more
of any subpopulations grows with expected rate λ. The strongly dependent on the stable age distribution of
Box 17.1 Decomposing individual reproductive value into components due to reproduction and
survival
P
First, let us ignore the environmental stochasticity. In this Lij nj vi . Under small or moderate fluctuations in the age
2 ij
case, the variance of reproductive value next year is σdi =
var(Wi ) = sdi v2i+1 + fbi v21 + 2cdi vi+1 v1 , where cdi = distribution so that nj can be approximated by Nuj , the
cov [Ji , Bi ] and pi and sdi = pi (1 − pi ) are the mean and variance in the change of the total reproductive value is
X
variance of Ji , respectively, while fdi is the variance of Bi . The var(V + ∆V|V) ≈ N2 vi vk uj ul cov(Lij , Lkl ) = N2 σe2
demographic variance Pof the contribution to the next year by ij
this age class is then ni var(Wi ). Using the approximation
ni ≈ Nui , the total demographic variance is so that the environmental variance is
X
X
k σe2 = vi vk uj ul cov(Lij , Lkl ).
var(V + ∆V|V) ≈ N ui [sdi v2i+1 + fdi v21 + 2cdi vi+1 v1 ] ij
i=1
X Defining the survival probabilities seij = cov(Li+1,i , Lj,j+1 )
2
=N ui σdi = Nσd2 and fecundities feij = cov(L1i , L1j ), and the covariance
ceij = cov(Li+1,i , L1,j ), we obtain the decomposition of the
where environmental variance into components from reproduction
X
k X and survival, thus
σd2 = ui [sdi v2i+1 + fdi v21 + 2cdi vi+1 v1 ] = 2
ui σdi X
σe2 = ui uj [seij vi+1 vj+1 + feij v21 + ceij vi+1 v1 ]
i=1
ij
Here 2
σdi = var(Wi ) is the within-year variance of individual X
= ui uj τeij .
reproductive value among individuals of age i.
ij
Second, let us consider the environmental variance by
assuming a very large population, where the demograph- In analogy with the model without any age structure we
ic noise can be ignored. In this case we have n + ∆n = can now add the demographic and environmental vari-
L(Z )n, where the projection matrix L(Z ) is now deter- ance terms, provided that the demographic components are
mined primarily by the temporal fluctuations in the environ- replaced by their temporal mean values under environmental
ment Z . Then we get V + ∆V = v (n + ∆n) = vLn = fluctuations.
Bighorn sheep Blackbrowed albatross Colombian ground squirrel

0.25
0.15
0.06
0.20
Fisherian age
Fisherian age
Fisherian age
distribution
0.05
distribution
distribution
0.10 0.15
0.04
0.03 0.10
0.05
0.02 0.05
0.01
0.00 0.00
5 10 15 5 10 15 20 2 4 6 8 10
Age Age Age
Common tern Mauritius kestrel Red deer

0.10 0.12
0.08 0.20 0.10

Fisherian age
Fisherian age
Fisherian age
distribution
distribution
distribution
0.08
0.06 0.15
0.06
0.04 0.10 0.04
0.02 0.02
0.05
0.00 0.00
5 10 15 20 2 4 6 8 5 10 15
Age Age Age
Roe deer Southern fulmar Peregrine falcon

0.30 0.14
0.06
0.25 0.12
0.05
Fisherian age
Fisherian age
Fisherian age
distribution
distribution
0.20 distribution 0.10

0.04
0.15 0.08
0.03
0.10 0.02 0.06
0.05 0.01 0.04
0.00 0.00 0.02
2 4 6 8 10 12 0 5 10 15 20 25 30 2 4 6 8 10 12
Age Age Age
Wandering albatross White stork Yellow−bellied marmot

0.06 0.30
0.15
0.05 0.25
Fisherian age
Fisherian age
Fisherian age
distribution
distribution
distribution
0.04 0.10 0.20

0.03 0.15
0.02 0.05 0.10
0.05
0.01
0.00 0.00
0 5 10 15 20 25 30 2 4 6 8 2 4 6 8 10 12
Age Age Age
Figure 17.1 The Fisherian age distribution in five mammal species (bighorn sheep (Ovis canadensis), Columbian ground squirrel (Spermophilus
columbianus), red deer (Cervus elaphus), roe deer (Capreolus capreolus), and yellow-bellied marmot (Marmota flaviventris)) and seven bird species
(blackbrowed albatross (Thalassarche melanophrys), common tern (Sterna hirundo), Mauritius kestrel (Falco punctatus), peregrine falcon (Falco
peregrinus), southern fulmar (Fulmarus glaicialoides), wandering albatross (Diomedea exulans), and white stork (Ciconia ciconia). This age
P P
distribution refers to the reproductive value in age class i, scaled so that ui = 1 and vu = ui vi = 1, where u is the stable age distribution
for the deterministic model. This shows the relative contribution of the different age classes to the total reproductive value of the population
(see Sæther et al. 2013 for further details).
the population. Similarly, a curvilinear relationship λ−1 η is generated by the temporal autocorrelations.
between reproductive value and age has been found Engen et al. (2013) showed that
in four passerines (Moller and De Lope 1999; Keller X
et al. 2008; Brown and Roth 2009; Bouwhuis et al. η = cov(W̄j , nj /V)
j
2011) as well as in the sparrow-hawk (Accipiter nisus)
(Newton and Rothery 1997), but the reproductive
where W̄ is the mean individual reproductive value
value increased with age in female red-billed chough
of individuals in age class j at a given year, nj is the
(Pyrrhocorax pyrrhocorax) (Trask et al. 2017).
number of individuals in this class, and V is the total
The distribution of reproductive values throughout
reproductive value of the population. When obser-
the life cycle provides important information about
vations of these variables are available over several
how different stages contribute to the population
years, η can be estimated by the cross-product sum
growth (Engen et al. 2009b). In an analyses of birds and
over years divided by one less than the number of
mammals, Sæther et al. (2013) showed that the larg-
years.
er variation among species in the contribution to the
For further details on how to further decompose
total reproductive value of the population and in the
this into contributions from survival and fecundity see
demographic variance was found in the younger age
Engen et al. (2013).
classes (relative to generation time). Furthermore, the
contribution of earlier stages within the life cycle to
stochastic population growth rate was greater in long- 17.5 Case studies
lived than in short-lived species. In contrast, environ-
mental stochasticity at different life history stages had
17.5.1 Moose on the island of Vega
similar effects on population fluctuations per genera- Moose (Alces alces) is a sexual dimorphic species
tion. Similarly, the distribution of reproductive values that is subject to intense harvesting in most parts of
throughout the life span of plants differed dependent Scandinavia (Lavsund et al. 2003). Its life history is
on life form (Mbeau-Ache and Franco 2013). The results characterised by strong age dependency in fecundity,
of these analyses indicated that the attributes related to with high twinning rates among prime-aged females
pace were constrained by duration of life, generating (Sæther and Haagenrud 1983). In addition, several
large differences among life forms in the age-specific aspects of individual variation in reproductive success
distribution of reproductive values. of females are dependent on the body mass of females
early in life (Sæther and Haagenrud 1985; Lee et al.
2017; Markussen et al. 2018). In 1985, three moose
17.4 Temporal autocovariance in the
colonised the island of Vega in Northern Norway. In
environmental noise 1992, we used this opportunity of having access to a
Until now we have assumed that the yearly realised newly established population to initiate a long-term
Leslie matrices are temporally uncorrelated. In age- study to examine the ecological and evolutionary con-
structured models, as in models with no age structure, sequences of harvest. This involved radio-collaring of
environmental autocorrelation affects the long-run almost all individuals on the island, which facilitated
growth rate (Tuljapurkar 1982; Caswell 2001; Doak construction of the pedigree using genetic parentage
et al. 2005); Tuljapurkar and Haridas 2006; Morris et al. assignment (Markussen et al. 2019). The environmen-
2011) through its effect on the environmental noise. tal conditions for moose on the island are favourable,
The impact over a time interval is proportional to the resulting in large body masses (Solberg et al. 2008).
length t, whereas the standard deviation is proportion- The age and sex compositions of the population have
al to the square root of t. This term is usually negative in periods been altered through directional harvest
and may be smaller or larger in magnitude than σe2 , of particular categories of animals (Sæther et al. 2003,
dependent on the mean projection matrix as well as 2004).
the strength of temporal autocorrelations. In this more The reproductive value of the moose showed strong
general model, the long-run growth rate takes the form age-specific variation in both sexes, with a maximum
at age 2 years in females and 1 year later in bulls
1 −2 2
s=r− λ σe + λ−1 η (Figure 17.2). This difference was caused by partic-
2 ularly high mating success among prime-aged bulls
P
where σe2 = E(vεt u)2 = vi uj vk um cov[Lijt , Lkmt ] is the (Markussen et al. 2019). The population growth esti-
environmental variance, εt = Lt − EL, and the last term mated by means of the total reproductive value of
1.6 3.4
1.4
3.2 Prehunt
1.2
1.0
In Nt, In Vt
3.0
0.8
vx
0.6 2.8 Posthunt

0.4
2.6
0.2
0.0 2.4
1 2 3 4 5 6 7 8 9 10 11 1992 1994 1996 1998 2000 2002 2004
Age Years
Figure 17.2 Age-specific variation in the reproductive value vx of Figure 17.3 Annual variation in (log-) population size (solid lines) of
female (open squares) and male (black circles) moose (Alces alces) on female moose on the island of Vega just before (Prehunt) and after
the island of Vega in northern Norway (Kvalnes et al. 2016). This (Posthunt) the hunting season. The dashed lines indicate the (log-)
demonstrates that in both sexes the greatest contribution to the total reproductive value of the population (Sæther et al. 2007a). This
growth of this population occurs from age classes around age at shows that the total reproductive value of the population describes
maturity, which then decreases with increasing age. the population fluctuations quite well (see also Engen et al. 2007).
the population closely resembles the actual number of in the population dynamics (Caswell 2001; Lande et al.
females in the population both before and after the 2003), which can erroneously be interpreted as environ-
hunting season (Figure 17.3). This shows that V actual- mental stochasticity and result in biased predictions of
ly can describe the population fluctuations well even in future population fluctuations (Holmes 2001).
populations that have undergone large variation in age We will introduce the population prediction interval
composition through directed harvest towards certain (PPI) (Engen et al. 2001; Sæther and Engen 2002) as a
age classes. useful concept in developing reliable population pro-
jections. The PPI is defined as the stochastic interval
17.5.2 A practical application: estimating risk that includes a given population size with probabili-
ty (1 − α), where α is the probability that the variable
of extinction in harvested Scandinavian lynx
we want to predict is not contained in the stochastic
populations interval. For example, we can model a population of
Most large carnivores are subject to human exploita- initial size of 20 females for years. After 50 years, the
tion. For example, in Scandinavia, Eurasian lynx (Lynx distribution of predicted population sizes is graphed
lynx) is regularly harvested to reduce the conflict level using quantiles, showing the proportion of population
associated with depredation on free-ranging livestock size estimates in, for example, 10 percentile groups.
(e.g. domestic sheep (Ovis aries)) and semi-domestic The width (e.g. the difference between the 5% and 95%
reindeer (Rangifer tarandus). In addition, the lynx is sub- percentiles) indicates the precision of the predictions
ject to substantial rates of illegal killings (Andrén et al. and the probability that the population size will be
2006). Here we will demonstrate that analyses based below a critical threshold (often 0 or 1 individual). The
on the total reproduction of the population can be a advantage of this approach is that changes in expected
useful tool to predict the consequences of such human- population size, population variability due to demo-
induced demographic changes for the probability of graphic and environmental stochasticity, and uncer-
extinction. tainties in parameter estimates all affect the width of
Such analyses are difficult because a general char- the PPI. Large uncertainties in parameter estimates as
acteristic of the life history of large carnivores is that well as strong stochastic influences on the population
they are long lived once they have reached sexual dynamics will typically result in wide PPI. Further-
maturity. For instance, the natural annual survival more, the width of the PPI will increase through time
rate of adult female lynx in Fennoscandia was 92.3% until some stationary distribution is reached (see exam-
(Andrén et al. 2006). Harvest that induces changes in ples in Sæther et al. 2002a, 2002b, 2007b). Finally, the
age structures will therefore easily result in time lags PPI can be used to assess the quality of our predictions
(Sæther et al. 2009, 2019). If our model makes correct Sæther et al. (2010) analysed temporal variation in
predictions of future population fluctuations, then the the total reproductive value of two Scandinavian pop-
quantiles of simulated population sizes at a given time ulations of lynx (Hedmark in Norway and Sarek in
step will be uniformly distributed across the quantiles Sweden) for which long-term individual-based demo-
of the PPI. graphic data were available (Andrén et al. 2006, 2002)
An important advantage of this approach is that to estimate the population growth rate as well as the
the PPI relates directly to the International Union for influence of demographic and environmental stochas-
Conservation of Nature (IUCN)’s criteria E for quanti- ticities on the population dynamics. The estimate of the
tative classification of population extinction risk (IUCN population growth rate was significantly higher and
2012). For instance, a population is defined as vulnera- more certain in Sarek (λ = 1.09; 95% confidence interval
ble, that is, it has a 10% probability of extinction within = 1.00–1.17) than in Hedmark (λ = 1.06; 95% confidence
100 years, if the lower 10% quantile of PPI decreases interval = 0.93–1.19). The demographic variance was
below 1 before 100 years. The PPI is found by paramet- similar in the two populations (σd2 = 0.312 and σd2 =
ric bootstrapping from the sampling distributions of 0.309 in the northern and southern population, respec-
the parameter estimates, which is now commonly used tively), whereas the environmental variance σe2 was not
in population viability analyses (Holmes 2004; Holmes significantly different from 0 in any of the two pop-
et al. 2007). This approach has also previously been ulations. These estimates derived from the analyses
used to explore management strategies for wolverines of reproductive values in the two populations (Boxes
and brown bears in Scandinavia (Sæther et al. 1998, 17.1 and 17.2) can then be used to construct PPIs of
2005; Tufto et al. 1999). future population fluctuations to suggest sustainable
Box 17.2 Estimation
Estimation of variance components requires that individu- Assuming no demographic covariances (Engen et al. 1998),
al data on survival and fecundity are available for all age the demographic stochastic components are independent
classes over several years. As in models with no age struc- between individuals so that the demographic contribution
ture, estimates of demographic variance components can be to cov(Lij , Lkl ) are zero for j ̸= l and thus there are only
rather certain even if the observation period is only a few environmental contributions, that is, cov(Li+1,i , Lj+1,j ) = seij ,
years, provided that observations are available from a rather cov(L1,i , L1,j ) = feij , and cov(Li+1,i , L1,j ) = ceij . Within
large number of individuals. Estimates of environmental a column (i = j) there may be demographic components
components, however, are in general rather uncertain even due to nonzero individual covariances between survival and
if data are available over long periods of times (Sæther et al. fecundity. The components sdi /Ni , fdi /Ni , and cdi /Ni must
2008). We use a single subscript i for age, writing Ji , Bi , and therefore be added, giving, for example cov(Li+1,i , Li+1,i ) =
Wi for stochastic variables expressing age-specific survival seii + sdi /Ni .
and reproduction that are independent within years, and i The elements of the expected projection matrix l = EL
and j when two age classes are considered jointly, such as can be estimated by mean values of all records of
in the covariance of contributions of individuals to the next (Jimt , Bimt ). From this matrix we can compute the corre-
generation cov(Wi , Wj ). In a given year t data from nit indi- sponding estimate of the multiplicative growth rate λ, the
viduals numbered by m are available from age class i. The stable age distribution u, and reproductive values v. To
complete data set consists therefore of records of survival find the uncertainties it is recommended to simulate boot-
and reproduction, (Jimt , Bimt ), for ages i = 1, 2, … k, in strap replicates of the data, resampling the recordings of
years t = 1, 2, … T, and individuals m = 1, 2,… nit . When (Jimt , Bimt ) with replacement within each year.
considering two different individuals in an age class in a
given year we use subscript q as well as m for numbering
Estimating demographic variance components
individuals. Similarly, when required we use y as well as t for
time. An important advantage of this approach is that all indi-
The exact relationship between the population vector at vidual reproductive values Wimt = Jimt vi+1 + Bimt v1 can
two subsequent years is given by the Leslie matrix L with
elements that are mean survivals J̄ at the subdiagonal and
mean fecundities B̄ at the first row (Engen et al. 2005). continued
Box 17.2 Continued
be observed, provided that estimates of u and v are avail- Estimating environmental stochastic
2
able. Using the definition σdi = Evar(Wi |Z) we then first components
2
estimate σdi (Z) = var(Wi |Z) for each year (with different
Assuming that all individual demographic components are
but unknown values of Z) by the simple sum of squares
independent for a given environment Z and applying the
1 X decomposition
nit
2 2
σ̂di (Z) = (Wimt − W̄it )
nit − 1 cov(Wimt , Wjqt ) = Ecov[Wimt , Wjqt |Z]
m=1
P
nit + cov[E(Wimt |Z), E(Wjqt |Z)],
where W̄it = nit −1 Wimt . The weighted mean of the
m=1 it appears that
above estimates over years with at least two records from
cov(Wimt , Wjqt ) = cov[E(Wimt |Z), E(Wjqt |Z)] = τeij
individuals in the age class with weights nti − 1 then yields
2
an estimate σ̂di of the contribution from age class i,P
while the where m ̸= q when i and j refer to the same age class (i = j).
total demographic variance is estimated by σ̂d2 = 2
ui σ̂di . Engen et al. (2009b) showed that 12 (Wimt −Wim′ y )(Wjqt −
Bootstrap replicates should be produced as for the Leslie Wjq′ t ) has expectation τeij , so that an unbiased estimate of
matrix, although replicates for the sum of squares for year τeij is obtained as the mean value of these products over all
t and age class i should be multiplied by nit /(nit − 1) to combinations of t ̸= y and all appropriate combinations of
make them unbiased (Engen et al. 2009b). the second subscript. This step can easily be computed even
The age-specific components generated by survival and if the number of terms is large. Engen et al. (2009b) also
fecundity, sdi and fdi , can similarly be estimated by replac- showed how the survival- and fecundity-specific variance
ing Wimt by Jimt and Bimt , respectively, while the estimate components as well as the contribution from the environ-
for the covariance cdi is the corresponding sum of cross mental covariance between survival and fecundity can be
products. estimated in the same way.
harvest strategies to ensure the populations remain of the demographic parameters estimated for the two
viable. populations using the reproductive values (Box 17.2).
More specifically, Sæther et al. (2010) analysed the In the county of Hedmark in southern Norway the
effects of adopting three different harvest strategies: minimum (politically determined) threshold for the
(1) proportional harvesting (Lande et al. 1995, 1997); population size was 10 family groups. The propor-
(2) threshold harvesting (Lande et al. 1995, 1997); and tion of adult females that had kittens indicated an
(3) proportional threshold harvesting (Engen et al. adult population size of 19 adult (≥ 2-year-old) females
1997; Tufto et al. 1999). For all three harvest strategies (Andrén et al. 2002). In addition, based on the sta-
the harvested females were assumed to be chosen at tionary age distribution, eight 1-year-old females were
random, since the hunters typically are unable to visu- also assumed to be present. As expected from large
ally differentiate between the age and sex of lynx in uncertainties in the parameter estimates and the large
a hunting situation. Proportional harvesting implies demographic stochasticity, the population prediction
that a certain fraction p of the females is harvested intervals soon became wide (Figure 17.3). However, the
each year. Under threshold harvesting the population simulations showed that if the population size was har-
N1 +N2 is harvested down to a threshold c if N1 +N2 > c; vested down to the minimum number of 10 reproduc-
otherwise no hunting occurs. Proportional threshold tive adult females, the lower 10% quantile was below
harvesting has in particular been shown to be a good 1 already after 34 years. Similarly, this population will
strategy when the population estimates N̂1 and N̂2 are have a 50% probability of extinction after 273 years.
uncertain (Engen et al. 1997; Lillegård et al. 2005), in We then repeated this exercise for the population in
which case the harvest is (N̂1 + N̂2 )a when N1 + N2 > c, the county of Nordland in northern Norway, where we
and otherwise there is no harvest. assumed similar demography as in the lynx popula-
The sustainability of the current harvest strategies tion in Sarek in northern Sweden. Here, a threshold
of lynx in Norway was explored by simulating future harvest strategy with no harvest when less than 10
population fluctuations based on bootstrap replicates family groups were recorded will result in stationary
30
q90 30
25 q50
Female population size
25
CV=0.4
20 q10
Time (years)
20
15 15 CV=0.6
10 q50 10
5 q10 5
0 0
0 20 40 60 80 100 120 140 160 180 0.0 0.2 0.4 0.6 0.8 1.0
Time (years) q
Figure 17.4 Prediction of population fluctuations of lynx (Lynx lynx) Figure 17.6 The time it takes for the lower 10% quantile of the
in the Norwegian counties of Hedmark (solid lines) and Nordland population prediction interval (PPI) q10 to reach one female in relation
(dashed lines) assuming threshold harvesting. Values of q10 q50 , and to the proportion of animals above the threshold c = 23 that is
q90 denote the 10%, 50%, and 90%, respectively, quantiles of the removed in Hedmark for different uncertainties in the population
population prediction interval (PPI). For the northern population, estimates (CV). We assume an initial size of the populations of eight
demographic characteristics were assumed to be similar to those 1-year-old and nineteen adult (≥ 2-year-old) females. No harvest
estimated for the Sarek population in northern Sweden. In both areas occurs below the harvest threshold.
the threshold is set at 10 reproductive females (i.e. females recorded
during winter with kittens). The simulations were done using the
population counts in 2007 as initial population size and assuming
relatively low values of the threshold (Figure 17.5).
that harvest occurred according to the proportion of each age class in
the population. For further details see Sæther et al. (2010).
In contrast, the population in Hedmark was like-
ly to go extinct even if a large threshold is chosen
(Figure 17.5). However, assuming a proportional har-
140 vesting strategy, extinction is likely irrespective of the
demographic characteristics of the population (Sæther
120 Sarek et al. 2010). In addition, large observation errors in
100 population estimates must require more careful har-
Time (years)
80 vesting to ensure population viability. In such cases,

Engen et al. (1997) proposed that proportional thresh-
60 old harvesting should be applied. For the lynx in Hed-
40 Hedmark mark, uncertainties in the population estimates strong-
ly affected the estimates of time to extinction (Figure
20 17.6). For a coefficient of variation (CV) in the popula-
0 tion estimate of 40%, removing more than 60% of the
0 10 20 30 40 50 population above the threshold of 27 female lynx will
Threshold lead to rapid increase in the risk of extinction.
Figure 17.5 The time it takes for the lower 10% quantile of the
population prediction interval (PPI) q10 to reach one female in relation 17.6 Conclusions
to the harvest threshold of females (≥ 1 year old) c for the two
populations in Hedmark and Sarek. We assume the same initial sizes In this chapter we have shown that characterising the
of the populations (eight 1-year-old and nineteen adult (≥ 2-year-old) population dynamics through analyses of temporal
females). No harvest occurs below the harvest, and all females above variation in the total reproductive value of the pop-
the threshold are immediately harvested (p = 1.0). ulation has several advantages: (1) In a fluctuating
environment the total reproductive value V will
fluctuations between 15 and 23 1-year-old females fluctuate around N with no temporal autocorrelation
(Figure 17.4). In general, a threshold harvest strat- and will have a constant multiplicative growth rate
egy of the population in the county of Nordland exactly equal to λ. (2) The variance of yearly changes
with demographic characteristics similar to the Sarek in the total reproductive value can be decomposed into
population will ensure viable populations even for additive components caused by environmental and
demographic stochasticity. (3) The Fisherian age distri- the possibilities for analyses of selection and evolution
P
bution scaled so that ui vi = 1 expresses how the total in age-structured populations (Crow 1979; Engen et al.
reproductive value V is distributed among age classes. 2009a, 2010, 2011, 2012; Barton and Etheridge 2011) and
(4) The dynamics of the total reproductive value of provides an approach to link evolutionary changes to
the population provides information about important variation in the environment even in organisms with
characteristics of the population dynamics, such as quite complex life cycles (Kvalnes et al. 2016).
the effects of temporal autocorrelations in the environ- A complication in analyses of population dynamics
mental noise and the probability of ultimate extinction. based on the total reproductive value of the popula-
(5) The distribution of reproductive values throughout tion is introduced by density regulation. Lande et al.
the life cycle show how different stages contribute to (2006) estimated the overall strength of density depen-
the population growth, which then can be decomposed dence as well as the total stochasticity (including the
into age-specific contributions from fecundity and sur- demographic stochasticity) acting on the population
vival as well as their covariance (Box 17.1). Thus, the dynamics of red deer (Cervus elaphus) using fluctua-
application of Fisher’s (1930) concept of the total repro- tions in density-dependent reproductive values allow-
ductive value of the population provides an important ing for density dependence within and between age
tool in analyses of the population dynamics espe- groups. This approach was based on time series of fluc-
cially of species with a complex life history in which tuations in the complete age structure and showed that
fluctuations in age distributions are likely to inflate the matrix of density-dependent coefficient describing
estimation of key parameters. Our analyses of the lynx how fluctuations in the size of different age classes
population in Scandinavia illustrate that this approach affected the population growth rate despite high cen-
can be particularly useful in analyses of the viability sus accuracy became subject to substantial sampling
of populations of vulnerable species because it enables error. Such complexities in the structure of density-
us to estimate the stochastic effects on the dynamics dependent effects, which occur even in species with
as well as the uncertainty in predictions (Figure 17.4). quite simple life histories (Gamelon et al. 2016), make
This requires, however, that the reproductive success it extremely difficult to approximate the dynamics of
and survival are known for an unbiased sample of density-regulated populations by temporal variation in
individuals from the different age classes. reproductive values.
Reproductive value has since the concept was intro-
duced by Fisher (1930) been commonly used in anal-
yses of selection in age-structured populations. For Acknowledgements
example, Williams (1966) decomposed reproductive This study was funded by the Research Council of
value into components due to current reproduction Norway (SFF-III 223,257/F50). We are grateful to John
and residual reproductive value to derive the opti- Fryxell for comments on a previous version of this
mal reproductive effort. This approach provided the paper.
foundation for the maximisation principle derived by
Schaffer (1974) and Caswell (1982), who showed that
in age-structured populations a life history maximis- References
ing fitness is equivalent to one that maximises the Andrén, H., Linnell, J. D. C., Liberg, O., et al. 2002. Estimat-
reproductive value of each age class. Similarly, after ing total lynx Lynx lynx population size from censuses of
the pioneering work of Hamilton (1966), reproductive family groups. Wildlife Biology, 8, 299–306.
values have been widely applied in analyses of selec- Andrén, H., Linnell, J. D. C., Liberg, O., et al. 2006. Sur-
tion on age-specific variation in several life history vival rates and causes of mortality in Eurasian lynx (Lynx
traits covering both animals and plants (e.g. Silvertown lynx) in multi-use landscapes. Biological Conservation, 131,
et al. 2001; Brommer et al. 2007; Bouwhuis et al. 2011; 23–32.
Barton, N. H. and Etheridge, A. M. 2011. The relation
Caswell and Salguero-Gomez 2013; Mbeau-Ache and
between reproductive value and genetic contribution.
Franco 2013; Zhang et al. 2015). Application of the
Genetics, 188, 953–973.
concept individual reproductive value that includes Bouwhuis, S., Choquet, R., Sheldon, B. C., and Verhulst, S.
stochastic components facilitates further decomposi- 2011. The forms and fitness cost of senescence: age-specific
tion of the contributions to the total reproductive value recapture, survival, reproduction, and reproductive val-
due to age-specific variation in survival and fecundi- ue in a wild bird population. The American Naturalist, 179,
ty as well as their covariance (Box 17.1). This extends E15–E27.
Brommer, J. E., Wilson, A. J., and Gustafsson, L. 2007. Explor- Engen, S., Lande, R., and Sæther, B.-E. 2009a. Reproduc-
ing the genetics of aging in a wild passerine bird. The tive value and fluctuating selection in an age-structured
American Naturalist, 170, 643–650. population. Genetics, 183, 629–637.
Brown, W. P. and Roth, R. R. 2009. Age-specific reproduc- Engen, S., Lande, R., Sæther, B.-E., and Dobson, F. S. 2009b.
tion and survival of individually marked wood thrushes, Reproductive value and the stochastic demography of
Hylocichla mustelina. Ecology, 90, 218–229. age-structured populations. The American Naturalist, 174,
Caswell, H. 1978. A general formula for the sensitivity of 795–804.
population growth rate to changes in life history parame- Engen, S., Lande, R., Sæther, B.-E., and Gienapp, P. 2010.
ters. Theoretical Population Biology, 14, 215–230. Estimating the ratio of effective to actual size of an age-
Caswell, H. 1982. Optimal life histories and the maximiza- structured population from individual demographic data.
tion of reproductive value: a general theorem for complex Journal of Evolutionary Biology, 23, 1148–1158.
life-cycles. Ecology, 63, 1218–1222. Engen, S., Lande, R., and Sæther, B.-E. 2011. Evolution of a
Caswell, H. 2001. Matrix Population Models, 2nd edition. plastic quantitative trait in age-structured population in a
Sunderland, MA, Sinauer. fluctuating environment. Evolution, 65, 2893–2906.
Caswell, H. and Salguero-Gomez, R. 2013. Age, stage and Engen, S., Saether, B.-E., Kvalnes, T., and Jensen, H.
senescence in plants. Journal of Ecology, 101, 585–595. 2012. Estimating fluctuating selection in age-structured
Charlesworth, B. 1994. Evolution in Age—Structured Popu- populations. Journal of Evolutionary Biology, 25, 1487–
lations, 2nd edition. Cambridge, Cambridge University 1499.
Press. Engen, S., Sæther, B.-E., Armitage, K. B., et al. 2013. Esti-
Cohen, J. 1977. Ergodicity of age structure in populations mating the effect of temporally autoccorrelated environ-
with Markovian vital rates, III: finite-state moments and ments on the demography of density-independent age-
growth rate; an illustration. Advances in Applied Probability, structured populations. Methods in Ecology and Evolution,
9, 462–475. 4, 573–584.
Cohen, J. E. 1979. Ergodic theorems in demography. Bulletin Fieberg, J. and Ellner, S. P. 2001. Stochastic matrix models
of the American Mathematical Society, 1, 275–295. for conservation and management: a comparative review
Coulson, T., Catchpole, E. A., Albon, S. D., et al. 2001. Age, of methods. Ecology Letters, 4, 244–266.
sex, density, winter weather, and population crashes in Fisher, R. A. 1930. The Genetical Theory of Natural Selection.
Soay sheep. Science, 292, 1528–1531. Oxford, Oxford at the Clarendon Press.
Crow, J. F. 1979. Gene frequency and fitness change in an Gamelon, M., Grøtan, V., Engen, S., Bjørkvoll, E., Visser, M.
age-structured population. Annals of Human Genetics, 42, E., and Sæther, B.-E. 2016. Density-dependence in an age-
355–370. structured population of great tits: identifying the critical
Crow, J. F. and Kimura, M. 1970. An Introduction to Population age-class. Ecology, 97, 2479–2490.
Genetics Theory. New York, Harper & Row. Hamilton, W. D. 1966. The moulding of senescence by natu-
Doak, D. F., Morris, W. F., Pfister, C., Kendall, B. E., and ral selection. The Journal of Theoretical Biology 12:12–45.
Bruna, E. M. 2005. Correctly estimating how environmen- Holmes, E. E. 2001. Estimating risks in declining popula-
tal stochasticity influences fitness and population growth. tions with poor data. Proceedings of the National Academy
The American Naturalist, 166, E14–E21. of Sciences, 98, 5072–5077.
Engen, S., Lande, R., and Sæther, B.-E. 1997. Harvesting Holmes, E. E. 2004. Beyond theory to application and eval-
strategies for fluctuating populations based on uncertain uation: diffusion approximations for population viability
population estimates. Journal of Theoretical Biology, 186, analysis. Ecological Applications, 14, 1272–1293.
201–212. Holmes, E. E., Sabo, J. L., Viscido, S. V., and Fagan, W. F.
Engen, S., Bakke, Ø., and Islam, A. 1998. Demographic 2007. A statistical approach to quasi-extinction forecast-
and environmental stochasticity—concepts and defini- ing. Ecology Letters, 10, 1182–1198.
tions. Biometrics, 54, 840–846. IUCN 2012. IUCN Red List Categories and Criteria. Version 3.1,
Engen, S., Sæther, B.-E. and Møller, A. P. 2001. Stochastic 2nd edition. Gland, Switzerland, IUCN.
population dynamics and time to extinction of a declining Karlin, S. and Taylor, H. M. 1981. A Second Course in Stochastic
population of barn swallows. Journal of Animal Ecology, 70, Processes. New York, Academic Press.
789–797. Keller, L. F., Reid, J. M., and Arcese, P. 2008. Testing evo-
Engen, S., Lande, R., Sæther, B.-E., and Weimerskirch, H. lutionary models of senescence in a natural population:
2005. Extinction in relation to demographic and environ- age and inbreeding effects on fitness components in song
mental stochasticity in age-structured models. Mathemati- sparrows. Proceedings of the Royal Society London B, 275,
cal Biosciences, 195, 210–227. 597–604.
Engen, S., Lande, R., Sæther, B.-E., and Festa-bianchet, M. Kvalnes, T., Sæther, B.-E., Haanes, H., Roed, K. H., Engen, S.,
2007. Using reproductive value to estimate key parame- and Solberg, E. J. 2016. Harvest-induced phenotypic selec-
ters in density-independent age-structured populations. tion in an island population of moose, Alces alces. Evolu-
Journal of Theoretical Biology 244, 208–317. tion, 70, 1486–1500.
Lande, R., Engen, S., and Sæther, B.-E. 1995. Optimal har- Perry, R. I., Cury, P., Brander, K., Jennings, S., Mollmann,
vesting of fluctuating populations with a risk of extinc- C., and Planque, B. 2010. Sensitivity of marine systems
tion. The American Naturalist, 145, 728–745. to climate and fishing: concepts, issues and management
Lande, R., Sæther, B.-E., and Engen, S. 1997. Threshold har- responses. Journal of Marine Systems, 79, 427–435.
vesting for sustainability of fluctuating resources. Ecology, Planque, B., Fromentin, J. M., Cury, P., et al. 2010. How does
78, 1341–1350. fishing alter marine populations and ecosystems sensitiv-
Lande, R., Sæther, B.-E., Engen, S., Filli, F., Matthysen, E., ity to climate? Journal of Marine Systems, 79, 403–417.
and Weimerskirch, H. 2002. Estimating density depen- Pukk, L., Kuparinen, A., Jarv, L., Gross, R., and Vasema-
dence from population time series using demographic gi, A. 2013. Genetic and life-history changes associated
theory and life-history data. The American Naturalist, 159, with fisheries-induced population collapse. Evolutionary
321–332. Applications, 6, 749–760.
Lande, R., Engen, S., and Sæther, B.-E. 2003. Stochastic Rouyer, T., Ottersen, G., Durant, J. M., et al. 2011. Shifting
Population Dynamics in Ecology and Conservation. Oxford, dynamic forces in fish stock fluctuations triggered by age
Oxford University Press. truncation? Global Change Biology, 17, 3046–3057.
Lande, R., Engen, S., Sæther, B.-E., and Coulson, T. 2006. Sæther, B.-E. and Engen, S. 2002. Including uncertainties
Estimating density dependence from time series of in population viability analysis using population predic-
population age structure. The American Naturalist, 168, tion intervals. In: Beissinger, S. R. and Mccullough, D. R.
76–87. (eds.) Population Viability Analysis, pp. 191–212. Chicago:
Lavsund, S., Nygren, T., and Solberg, E. J. 2003. Status of University of Chicago Press.
moose populations and challenges to moose management Sæther, B.-E. and Engen, S. 2003. Routes to extinction. In:
in Fennoscandia. Alces, 39, 109–130. Blackburn, T. M. and Gaston, K. J. (eds.) Macroecology.
Lee, A. M., Sæther, B.-E., Markussen, S. S., and Engen, S. Concepts and Consequences, pp. 218–236. Oxford: Black-
2017. Modelling time to population extinction when indi- well.
vidual reproduction is autocorrelated. Ecology Letters, 20, Sæther, B.-E. and Engen, S. 2019. Towards a predictive con-
1385–1394. servation biology: the devil is in the behaviour. Philo-
Lewontin, R. C. and Cohen, D. 1969. On population growth sophical Transactions of the Royal Society B London 374
in a randomly varying environment. Proceedings of the (1781).
National Academy of Sciences, 62, 1056–1060. Sæther, B.-E. and Haagenrud, H. 1983. Life history of the
Lillegård, M., Engen, S., Sæther, B.-E., and Toresen, R. 2005. moose (Alces alces): fecundity rates in relation to age and
Harvesting strategies for Norwegian spring-spawning carcass weight. Journal of Mammalogy, 64, 226–232.
herring. Oikos, 110, 567–577. Sæther, B.-E. and Haagenrud, H. 1985. Life history of the
Lion, S. 2018. Class structure, demography, and selec- moose (Alces alces): relationship between growth and
tion: reproductive-value weighting in nonequilibrium, reproduction. Holarctic Ecology, 8, 100–106.
polymorphic populations. The American Naturalist, 191, Sæther, B.-E., Engen, S., Swenson, J. E., Bakke, Ø., and
620–637. Sandegren, F. 1998. Assessing the viability of Scandina-
Markussen, S. S., Loison, A., Herfindal, I., et al. 2018. Fit- vian brown bear, Ursus arctos, populations: the effects of
ness correlates of age at primiparity in a hunted moose uncertain parameter estimates. Oikos, 83, 403–416.
population. Oecologia, 186, 447–458. Sæther, B.-E., Engen, S., Filli, F., Aanes, R., Schröder, W., and
Markussen, S. S., Herfindal, I., Loison, A., et al. 2019. Deter- Andersen, R. 2002a. Stochastic population dynamics of
minants of age at first reproduction and lifetime breeding an introduced Swiss population of the ibex. Ecology, 83,
success revealed by full paternity assignment in a male 3457–3465.
ungulate. Oikos, 128, 328–337. Sæther, B.-E., Engen, S., Lande, R., Visser, M., and Both, C.
Mbeau-ache, C. and Franco, M. 2013. The time distribution 2002b. Density dependence and stochastic variation in a
of reproductive value measures the pace of life. Journal of newly established population of a small songbird. Oikos,
Ecology, 101, 1273–1280. 99, 331–337.
Moller, A. P. and De Lope, F. 1999. Senescence in a short- Sæther, B.-E., Solberg, E. J., and Heim, M. 2003. Effects of
lived migratory bird: age-dependent morphology, migra- altering sex ratio structure on the demography of an iso-
tion, reproduction and parasitism. Journal of Animal Ecolo- lated moose population. Journal of Wildlife Management, 67,
gy, 68, 163–171. 455–466.
Morris, W. F., Altmann, J., Brockman, D. K., et al. 2011. Sæther, B.-E., Solberg, E. J., Heim, M., Stacy, J. E., Jakobsen,
Low demographic variability in wild primate popula- K. S., and Olstad, R. 2004. Offspring sex ratio in moose
tions: fitness impacts of variation, covariation, and serial Alces alces in relation to paternal age: an experiment.
correlation in vital rates. The American Naturalist, 177, Wildlife Biology, 10, 51–57.
E14–E28. Sæther, B.-E., Engen, S., Persson, J., Brøseth, H., Landa, A.,
Newton, I. and Rothery, P. 1997. Senescence and reproduc- and Willebrand, T. 2005. Management strategies for the
tive value in sparrowhawks. Ecology, 78, 1000–1008. wolverine in Scandinavia. Journal of Wildlife Management,
69, 1001–1014.
Sæther, B.-E., Engen, S., Solberg, E. J., and Heim, M. 2007a. Toresen, R. and Østvedt, O. J. 2000. Variation in abun-
Estimating the growth of a newly established moose pop- dance of Norwegian spring-spawning herring (Clupea
ulation using reproductive value. Ecography, 30, 417–421. harengus, Clupeidae) throughout the 20th century and the
Sæther, B.-E., Lillegård, M., Grøtan, V., Filli, F., and influence of climatic fluctuations. Fish and Fisheries, 1,
Engen, S. 2007b. Predicting fluctuations of re-introduced 231–256.
ibex populations: the importance of density-dependence, Trask, A. E., Bignal, E. M., Mccracken, D. I., Piertney, S. B.,
environmental stochasticity and uncertain population and Reid, J. M. 2017. Estimating demographic contribu-
estimates. Journal of Animal Ecology, 76, 326–336. tions to effective population size in an age-structured wild
Sæther, B.-E., Grøtan, V., Engen, S., Noble, D. G., and Freck- population experiencing environmental and demograph-
leton, R. P. 2009. Critical paramters for predicting popula- ic stochasticity. Journal of Animal Ecology, 86, 1082–1093.
tions of some British passerines. Journal of Animal Ecology, Tufto, J., Sæther, B.-E., Engen, S., Swenson, J. E., and
78, 1063–1075. Sandegren, F. 1999. Harvesting strategies for conserving
Sæther, B.-E., Engen, S., Odden, J., Linnell, J. D. C., Grotan, minimum viable populations based on World Conser-
V., and Andren, H. 2010. Sustainable harvest strate- vation Union criteria: brown bears in Norway. Proceed-
gies for age-structured Eurasian lynx populations: the ings of the Royal Society London B, Biological Sciences, 266,
use of reproductive value. Biological Conservation, 143, 961–967.
1970–1979. Tuljapurkar, S. D. 1982. Population dynamics in variable
Sæther, B.-E., Coulson, T., Grøtan, V., et al. 2013. How environments. II. Correlated environments, sensitivity
life history influences population dynamics in fluctuating analysis and dynamics. Theorethical Population Biology, 21,
environments. The American Naturalist, 182, 743–759. 114–140.
Sæther, B.-E., Engen, S., Gamelon, M., and Grøtan, V. 2019. Tuljapurkar, S. and Haridas, C. V. 2006. Temporal autocor-
Predicting the effects of climate change on bird population relation and stochastic population growth. Ecology Letters,
dynamics. In: Dunn, P. O. and Møller, A. P. (eds.) Effects 9, 327–337.
of Climate Change on Birds, 2nd edition. Oxford: Oxford Vlam, M., Van der Sleen, P., Groenendijk, P., and Zuide-
University Press. ma, P. A. 2017. Tree age distributions reveal large-scale
Schaffer, W. M. 1974. Selection for optimal life histories: disturbance-recovery cycles in three tropical forests. Fron-
effects of age structure. Ecology, 55, 291–303. tiers in Plant Science, 7.
Silvertown, J., Franco, M., and Perez-ishiwara, R. 2001. Evo- Williams, G. C. 1966. Natural selection: costs of reproduc-
lution of senescence in iteroparous perennial plants. Evo- tion and a refinement of Lack’s principle. The American
lutionary Ecology Research, 3, 393–412. Naturalist, 100, 687–690.
Solberg, E. J., Garel, M., Heim, M., Grøtan, V., and Sæther, Zhang, H., Rebke, M., Becker, P. H., and Bouwhuis, S. 2015.
B.-E. 2008. Lack of compensatory body growth in a high Fitness prospects: effects of age, sex and recruitment age
performance moose Alces alces population. Oecologia, 158, on reproductive value in a long-lived seabird. Journal of
485–498. Animal Ecology, 84, 199–207.
CHAPTER 18
Applying comparative methods to

different databases: lessons from
demographic analyses across
mammal species
Jean-Michel Gaillard, Victor Ronget, Jean-François Lemaître,
Christophe Bonenfant, Guillaume Péron, Pol Capdevila,
Marlène Gamelon, and Roberto Salguero-Gómez
18.1 Introduction Gaillard et al. 2016 for a review). These life history
analyses generally included vital rates (e.g. survival
The diversity of life histories displayed by species
and fecundity) among a set of diverse life history
along the Tree of Life has fascinated researchers for mil-
traits describing the life cycle of organisms, and cov-
lennia. For several centuries, the Natural History of Ani-
ered a wide range of organisms (e.g. several groups:
mals written by Aristotle (350 bce) was the only work
Fenchel 1974; Blueweiss et al. 1978; insects: Blackburn
reporting a comprehensive evidence of this diversity.
1991; plants: Silvertown et al. 1993; homothermous
Much later, Georges-Louis Leclerc de Buffon proposed
vertebrates: Gaillard et al. 1989). Overall, these com-
his Histoire Naturelle (1749–1767), adding a tremen-
parative studies enabled the identification of covari-
dous amount of new observations on life histories and
ation among nondemographic life history traits (e.g.
unique synthesis, including the first human life tables
Russell 1982; Bennett and Harvey 1987), vital rates (e.g.
in contrasting environments (i.e. cities vs. country-
Gaillard et al. 1989; Silverton et al. 1993), or both types
side). This very early work was essentially descriptive,
(e.g. Promislow et al. 1990) across species. For a focal
and proper comparative demographic analyses across
taxonomic group (order, e.g. Swihart 1984, or class,
species searching for a structure and organisation of
e.g. Stearns 1983), species-specific life history traits
life histories appeared much later. For instance, one of
were collected from literature surveys. The associa-
the first comparative demographic analyses was pro-
tion between traits was investigated through bivariate
posed by Mitchell (1911), although it was not rooted in
correlations first, and then multivariate covariations
the grounds of life history evolution. The first question
among traits were examined. It became rapidly evident
formulated in an evolutionary framework was posed
that allometric rules (i.e. linear scaling of a life histo-
by Lamont Cole (1954) about the optimal timing of first
ry trait [e.g. life expectancy] as a function of size [e.g.
reproduction, which led him to launch the paradox of
body mass] on a log-log scale; Huxley 1932) strongly
the rarity of semelparous species in the Tree of Life
constrain life history variation among species (Peters
despite their evolutionary advantage over iteroparous
1983; Calder 1984) and should be accounted for to
species.
identify reliably covariation patterns among life his-
The modern and still current era of comparative
tory traits. Following Felsenstein’s work (1985), it was
analyses of life histories emerged with the seminal
also soon realised that phylogenetic relatedness—that
work by Stearns (1976, 1977), which generated a burst
is, the propensity of closely related species to share
of comparative studies in the 1980s and 1990s (see
Jean-Michel Gaillard et al., Applying comparative methods to different databases: lessons from demographic analyses across mammal species.
more similar trait values than distantly related ones, evolution (Eisenberg 1981), whether aquatic mammals
when everything else is kept constant—should also be display different demographic patterns compared to
accounted for in comparative analyses (Harvey and terrestrial mammals has remained little investigated,
Pagel 1991; Garamszegi 2014; Chapter 1). although habitat is a template (sensu Southwood 1977)
The comparative studies of life histories across that influences size, life history, and behaviour of
species published in the last decades of the twentieth organisms. We will thus test whether habitat types
century covered a broad range of taxonomic groups (aquatic vs. terrestrial) influence allometric relation-
(e.g. from insects to mammals among animals). Tak- ships of demographic metrics and/or the contribution
en together, these studies consistently showed that of dimensionless demographic metrics to major axes,
the first axis of life history variation across species which shape the variation in demographic trajectories.
corresponds to a time-scale ranking species along a In this chapter, we will illustrate the common
slow–fast continuum of life histories, opposing species approach currently used to identify covariation among
with a slow pace of life (i.e. with long biological times life history traits by performing a comparative demo-
[i.e. any life history trait measured in units of time graphic analysis across mammalian species. More
such as developmental time, age at first reproduction, specifically, we illustrate the principle and the power
or life span] and low frequencies (i.e. the inverse of of the comparative demographic approach by focus-
a biological time, such as heart beats per minute or ing on the shape of demographic trajectories (sensu
annual fecundity) to species with a fast pace of life (i.e. Ronget and Gaillard 2020) across mammalian species
with opposite characteristics). Offspring number (e.g. in relation to their occurrence in terrestrial vs. aquatic
litter size) and mass (e.g. birth mass), which are often habitats. The choice of this question was motivated by
included in comparative analyses of life histories (e.g. two reasons.
Promislow et al. 1990; Bielby et al. 2007), correspond First, while whether allometric constraints should
to dimensionless number and volume, respectively. be accounted for when performing comparative anal-
Therefore, while these traits can be linked to the slow– yses has been often discussed (e.g. Jeschke and
fast continuum of life histories, they cannot be inter- Kokko 2009), covariation among dimensionless met-
preted as defining this continuum. It is noteworthy rics, which solves this problem, has not been yet
that this continuum is retrieved both with and without investigated, although recent comparative studies of
accounting for allometric effects on traits, although, actuarial senescence included shape in their analyses
obviously, the rank of species along this continuum (see e.g. Baudisch 2011; Ronget and Gaillard 2020).
can differ markedly between analyses performed with Performing analysis of dimensionless metrics gives
and without correction for body size (e.g. Gaillard et al. us the opportunity to match the criteria of a dimen-
1989). Another continuum involving developmental sional analysis (sensu Stahl 1962). Dimensional anal-
(e.g. Stearns 1983) or reproductive (e.g. Gaillard et al. ysis, which tracks the dimensions (e.g. length, vol-
1989) traits is often reported as a second structuring ume, or time) and the measurement units when com-
axis of life history variation across species (see also paring among traits, is key to comparative analyses
Salguero-Gómez et al. 2016a for a study on plants). because life history traits do not share a common
While most studies performed to date have investi- dimension. For instance, life expectancy is measured
gated the structure of life history variation (reviewed in units of time, skeletal size is measured in units
in Gaillard et al. 2016), only a few have examined the of length, and body mass is measured in units of
ecological correlates (sensu Harvey and Clutton-Brock volume. As such, biologists should be careful not to
1985) of specific life history covariations. Firm evi- compare apples to oranges when performing compar-
dence that ecological factors or traits describing the ative analyses encompassing very diverse traits. For
lifestyle (in terms of, e.g., habitat, diet, mating sys- instance, mixing biological times and dimensionless
tem, or degree of sociality; sensu Dobson 2007) influ- numbers with a same analysis often leads to identify-
ence the covariation pattern among life history traits ing two independent axes of variation simply because
is so far limited to the role played by hibernation of differences in units (as in, e.g., Bielby et al. 2007,
(Turbill et al. 2011) and by the habitat, with aerial or where biological times structured the first axis and
subterranean species being slower than species liv- litter size the second one in most taxonomic-based
ing in other habitats (Healy et al. 2019) and species analyses of mammals reported in that work). Charnov
living in tropical environments being slower than (1993) popularised this type of analysis by encour-
species living in temperate areas (Wiersma et al. 2007). aging the use of dimensionless numbers that can be
However, while living in aquatic habitats has selected compared directly among species, without worrying
for obvious morphological changes in mammalian about allometric effects.
A P P LY I N G C O M PA R AT I V E M E T H O D S TO D I F F E R E N T D ATA B A S E S 301
Second, while we expect allometric constraints to dimensionless metrics. We also tested whether habitat
differ between terrestrial (with high constraints on types (i.e. terrestrial vs. aquatic) and the source of
locomotion) and aquatic (with less constraints on the demographic data (i.e. Malddaba vs. Comadre)
locomotion) habitats, this has been mostly overlooked influenced the relationship between body mass and
in previous life history analyses (but see Capdevila each demographic metric. We expected the relation-
et al. 2020 for a recent comparative analysis of differ- ships between demographic metrics and body mass to
ences between terrestrial and aquatic species of plants be stronger in Malddaba than in Comadre data sets due
and animals). With these issues at the forefront, we to the inclusion of full age-dependence in vital rates in
designed the current study with the following three Malddaba (see e.g. Festa-Bianchet et al. 2003). We also
key considerations, which highlight common pitfalls in expected stronger allometric relationships in species
comparative demographic research. living in terrestrial habitats because the biophysical
First, as the deposition of data in widely available constraints of a large body size in terms of locomotion,
repositories is rapidly becoming common in biological and thereby of resource acquisition, are much weaker
research, we have to identify the most suitable pri- in aquatic than in terrestrial environments (see Gaillard
mary data to address our question. To illustrate the et al. 1989 for a discussion).
potential influence of the source of species-specific Lastly, we performed phylogenetically controlled
data retrieved, we will compare the results of the statistical analyses to determine the covariation that
same analyses replicated on two different reference structured interspecific variation in the shape of
data sets. The first one was based on the detailed fully demographic trajectories. We thus used dimension-
age-specific lx and mx series (see Chapter 8) estimated less metrics that can be interpreted in terms of shape
for a single population (i.e. lx and mx series came from of demographic trajectories (see Ronget and Gaillard
the same population for a given species) (Malddaba 2020). Comparing the analysis performed using data
database;1 see Lemaître et al. 2020a, 2020b for recent sets extracted from Malddaba and Comadre allowed
applications using age-specific mortality and repro- us to assess the role of age dependence in vital rates for
ductive data, respectively). The second one was based shaping variation in demographic trajectories across
on st(age)-structured population matrix models (see species.
Chapter 9) that included most often an age for which
vital rates were assumed to be constant, independent
of age (i.e. in most cases adult survival and adult fecun-
dity is assumed constant over ages) for a given species 18.2 Methods
(Comadre database; see Salguero-Gómez et al. 2016b
for an overview of the database). The two databases
18.2.1 Building databases and calculating
have very different purposes. Malddaba only includes demographic metrics
age-specific traits (i.e. survival, reproduction, body We first looked for age-specific demographic data (i.e.
mass) of mammals and currently contains ca. 200 lx and mx series) for all mammalian species included in
species with at least partial age-specific demographic Malddaba. As generally done in comparative analyses
information. On the other hand, Comadre encom- of life history traits (but see Lemaître et al. 2020a), only
passes all major taxa and includes at least partial the mean trait value for each species was considered.
demographic information for thousands of species. For this analysis, we selected each species in which lx
We thus expected different outcomes for our analyses, series (i.e. percentage of individuals remaining alive at
depending on the database used. age x) and mx series (i.e. average number of female off-
Second, from both databases, we estimated biolog- spring born per female of age x) were reported for the
ical times and dimensionless metrics. We performed same population. We retrieved 66 species (54 species
an allometric analysis to test whether allometric with terrestrial habitats and 12 with aquatic habitats)
expectations were met. Indeed, we expected biologi- where this required information was available. When
cal times (i.e. metrics expressed in units of time) to be species-specific data were available for more than one
tightly related to body mass across species, contrary to population, we retained the population for which the
data were of highest quality (i.e. larger sample size,
longer monitoring, more reliable estimation methods
1
The database Malddaba (built and managed by JMG, VR, of vital rates). Next, for all these species, we extracted
and JFL) is currently under progress but will be released online
female adult body mass (in grams) from Malddaba,
and freely available for the scientific community in the com-
ing years. In the meantime, any query regarding the database and we calculated the five following demographic
should be sent to JMG, VR, or JFL. metrics using the lx and mx series:
Generation time (T): We calculated T as recommended with w the asymptotic distribution of the age class in
in Staerk et al. (2019) using the following formula: the population as weight. The metric ϕ measures the
P −x mean lifetime reproductive output and has a dimen-
xλ lxmx sion of time (years−1 ).
T = P −x
λ lxmx Number of genealogies (S): We calculated S as the
Demetrius’s entropy (Demetrius 1979) calculated using
which corresponds to Tb discounted for observed pop-
the following formula:
ulation growth (see Table 1 in Staerk et al. 2019, p. 3).
Assuming the population has reached a stable age dis- X
S=− λ−x lxmx * ln λ−x lxmx
tribution and overlapping generations, Tb corresponds
to the mean weighted age of mothers in a population.
This trait corresponds to the number of genealogies
The observed population growth (λ) was calculated
compatible with a given population growth rate and
from a population projection model (see Chapter 9)
can be interpreted as a measure of the diversity of
built from the full lx and mx schedules using the R
reproductive trajectories within the focal population.
package popbio (Stubben et al. 2007). T provides a mea-
The metric S is a dimensionless number.
sure of the ranking of a given species along the slow–
For the species retrieved from the Comadre
fast continuum of life histories (Gaillard et al. 2005). T
database, body mass and all the metrics described
has a dimension of time (in years).
above were directly obtained from a recent compara-
Age at first reproduction (AFR): AFR was defined as
tive analysis that derived them from matrix population
the earliest age at which at least 5% of females in a pop-
models (see R code and Supplementary Online Mate-
ulation give birth to newborn. AFR has a dimension of
rials available in Capdevila et al. 2020). In this case, the
time (in years).
information was available for 70 mammalian species
Mortality entropy from birth (Hbirth ): We calculated
(55 species with a terrestrial habitat and 15 with an
Hbirth as the Keyfitz entropy (Keyfitz 1985) using the
aquatic habitat).
following formula:
P
lx ln (lx)
Hbirth = − P 18.2.2 Statistical analyses accounting for
lx
phylogenetic inertia
Hbirth is often used to assess the shape of the mortali- We obtained two data sets (Malddaba and Comadre)
ty curve over the lifetime (Chapter 8). A value of Hbirth with species-specific information for five demographic
lower than 1 is typically interpreted as an increase of metrics, one measure of habitat type, and adult body
mortality with age, and a value of Hbirth higher than mass. We first analysed the allometric relationships
1 is typically interpreted as a decrease of mortality using phylogenetic regressions linking demographic
with increasing age. However, identical values of Hbirth metrics and body mass, as well as assessing the poten-
can correspond to rather different forms of mortali- tial influence of habitat type on these relationships.
ty curves (see e.g. Goldman and Lord 1986). We will Demographic traits and adult body mass were all
interpret more generally Hbirth as a metric measuring log-transformed (Pélabon et al. 2018). We fitted five
the lifetime lost due to age-specific mortality relative to different models: (1) a model including interactive
the life expectancy at birth (Goldman and Lord 1986). effects of adult body mass and habitat type on a
Hbirth is highly sensitive to differences between juve- given demographic trait only (i.e. different allometries
nile and adult mortality. It generally decreases with between habitat types); (2) a model including additive
survival increase at older ages (Fernandez and Beltran- effects of adult body mass and habitat type on a
Sanchez 2015) and can be interpreted as a measure given demographic trait (i.e. same allometry in both
of the diversity of ages at death. Hbirth is a positive habitat types with a constant difference in a given
nondimensional number. demographic trait at a given adult body mass); (3)
Mean number of recruits per lifetime (ϕ): We calculated a model only including the influence of adult body
ϕ as the weighted (with the proportional asymptotic mass on a given demographic trait (i.e. same allometry
representation of the age class in the population as and same trait value at a given adult body mass in
weight) mean number of recruits produced by females both habitat types); (4) a model only including the
over their life span using the following formula: influence of habitat type on a given demographic trait
X (i.e. no allometry and a constant difference in a given
φ= w mx
demographic trait between habitats); and (5) a model
including no influence of adult body mass or habitat are valid and generate nonspurious results when the
type on a given demographic trait (i.e. intercept-only resulting metrics are biological constructs (rather than
model corresponding to a same average value for a mathematical constructs) based on highly correlated
given demographic trait at any size in any habitat traits (Smith 1999), as proposed in the present analy-
type). Importantly, to avoid nonindependence of traits sis. We assessed the number of principal components
among species caused by the phylogenetic relatedness (i.e. defining the structuring axes) to retain based on
(see Chapter 1), we used phylogenetically corrected the broken-stick method (Jackson 1993) and tested the
analyses. We thus fitted phylogenetic generalised least potential influence of habitat type and of Log(T) (as
square (PGLS) models (Freckleton et al. 2002) and a measure of the ranking of a given species along the
used a consensus phylogenetic tree (Bininda-Emonds slow–fast continuum of life histories; Gaillard et al.
et al. 2007, 2008). Although more recent phylogenetic 2005) on this/these component(s). We fitted models
trees have been published (see e.g. Upham et al. 2019), including both additive and interactive effects of gen-
we keep using the most common tree considered in eration time (Log(T)) and habitat type and, again, used
comparative analyses of mammalian life histories (e.g. AIC for model selection.
Lemaître et al. 2020a). We estimated the strength of the The script to perform the statistical analyses in
phylogenetic signal for each model by using Pagel’s λ R is provided in the Appendix, together with the
(Pagel 1999) that varies from 0 (no phylogenetic signal data sets and the phylogenetic trees (please go to
detected) to 1 (perfect phylogenetic signal). Although www.oup.com/companion/SalgueroGamelonDM).
this signal is highly informative, it is important to keep
in mind that it requires some strong assumptions to be
18.3 Results
fulfilled, such as Brownian motion for trait evolution
and the same rate of trait change across the whole Both Malddaba and Comadre data sets encompassed
tree. It is noteworthy that Pagel’s λ should not be a wide range of taxonomic groups of mammals
confused with population growth (also called λ). We (Figure 18.1). The two databases shared 22 common
performed model selection based on AIC (Burnham species.
and Anderson 2002). We retained the model with the
lowest AIC, so long as the difference in AIC scores
18.3.1 Allometric analyses
was more than 2 points. If the difference in AIC
scores between candidate models was less than 2, we As expected, we detected allometric relationships for
retained the model with the least parameters based on both demographic metrics with a dimension of time
the criterion of parsimony. (i.e. generation time, T, and age at first reproduction,
We next identified the axes of variation that structure AFR) across mammalian species, from both Maldda-
the shape of demographic trajectories from the varia- ba (Table 18.1a) and Comadre data sets (Table 18.1b).
tion observed in dimensionless demographic metrics However, the strength of these relationships was con-
across species with a phylogenetic principal compo- sistently higher when using Malddaba data, with
nent analysis (PPCA) (Revell 2012). We entered four larger allometric coefficients and larger proportion of
dimensionless demographic metrics as PPCA loadings, variance explained (i.e. larger R2 ) than allometric rela-
including the two measures of entropy defined above tionships assessed from Comadre data (Table 18.2a
and available in both data sets (Keyfitz’s entropy Hbirth vs. Table 18.2b). As a result, because allometric rela-
and Demetrius’s entropy S, log-transformed) and two tionships are strongly influenced by phylogeny in
other measures corresponding to biological constructs mammals (e.g. Riek and Geiser 2013), the phyloge-
based on different demographic metrics defined ear- netic signal was stronger in analyses performed using
lier. We used the loge transformation in all cases. We Malddaba than using Comadre (Table 18.2), indicat-
defined a measure of dimensionless productivity (Prod) ing a higher explanatory power. For the third demo-
as the product between Log(T) (generation time in graphic trait with a dimension of time, the num-
years) and Log(ϕ) (number of recruits per lifetime in ber of recruits per lifetime ϕ, we found that its
years−1 ), and a measure of dimensionless timing of first response to mass changes differed between databas-
reproduction (TFR) as the ratio between Log(AFR) (age es. A quite strong positive relationship was found
at first reproduction in years) and Log(T) (generation when using the Malddaba data (Table 18.2a), where-
time in years). Although the use of ratios in life histo- as the number of recruits per lifetime ϕ was inde-
ry analyses has been criticised for leading to spurious pendent of adult body mass changes when using
or biased relationships (e.g. Atchley et al. 1976), they Comadre (Table 18.2b). The analyses of dimensionless
teralis
Tamiasciurus hudso
Cynomys ludovicianus
entris
Microtus
us
Sigmodo
lus la
marmota
uric
ibacin
s
Marmota flaviv
Tamias striatus
n eu
Dip
s da
Cas
ophi
Rattu
s
atu
Zapu
run
gi
od
ota ba
oecono
tor
n hispidu
hilu
ldin
rm
sperm
om
Marmota
ma
b
s fusc
Le
s prin sis
can
xi
a
Uro tellus
Cast
be
Ele
mop
ys
de
pu s ma ican s
Lo
Da
au
s
Marm
nicus
Uro tellu
i
dia
x
aden r
ph ta a inct s
spe
s
se
ds
sy
re
mus
Callo
a
od ove
llu
or fi ilis
ipes
ceps
Sper
go war
s
pu osu s vu
att
er
a
ci
uro imu
hu
Pr pit ecus
s
o
cta us
ite
a
ci
Tr
sv
Uro
sn
ich
nt
rc
d
Tr
be
ich uru
c
b
La opit ecu
rix us e
xa
ich
tr
mu
Pr pith
ae
x
po
fr
os
M
h
r
c
Le
hy rato
la
ru
go he
a
o
c
Pe cro
Pr
s c lpe
s
s
o
Ph pu le pe ollis
a
an cu
ta
th
as te im
u
co u se
in la
lar rus hy us ic
sc
ba
u
0.02 cto au ge
u
ac
Br guin us
fu
Mortality entropy from birth (Hbirth)

An a
E
CO ldda
s c str nii ell
DR
te
ch in a
er lis Sa guin s ap us
Did in ce
MA
Sa paju liva
ma
elp us a eus us
0.142 his gil
is Sa us o pucin
b
au e a
rit C sc i
a ebu abeli
0.264 C
ng
o gei
Po b erin
Arcto rilla lodytes
ceph Go
alus trog s
0.386 Arctoc trop Pan sa pien
epha icalis o
lus ga Hom atta
Arctocep zella a mul
halus Macac hinx
0.508 Arctoceph
australi
alus pus
s Mand
rillus sp
phalus
oce
illus Papio cyn
Phocarctos hooke mitis
0.63 ri Cercopithecus
Eumetopias jubatus Erythrocebus patas
Zalophus californianus Chlorocebus aethiops
0.752 us Eidolon helvu
Callorhinus ursin m
tris Saccoptery
angustiros x bilinea
0.874 Mirounga nina Phacoc ta
ga leo hoeru
Miroun llii Sus sc s aeth
iopicu
edde rofa s
tes w i Phys
ycho land
0.996 Lepton scha
uins na
et
Delp er cato
achu
s ituli hin do n
Mon oca v is Orc apte
P h lutr in
Glo us orc
rus
leu
1.118 En hy dra lutra
a is G.
bic
e a cas
Lutr dens ma phala
Da cro
a ea me
can rmin s Ru ma d rhyn las
1.24 nt r
a
la
e ele s Ce cerv ama ch
us
Lo te sm u
us Ce rvu us e
M ele ritim tos rv s e ld
M a rc us la ii
nis tra us
Ca es er
s m us a nip phu
la an
pr alc tar
u
lu ns
Alc ngif ame virgin

s s s
Ur Ur
nis ric
eo
po
Ra po
ca ulpa us
Ca me
lu s
n
pic us
p
Hip ocoil amelo
s
a
ca
Od ffa c ubali
e
u
us
pr
Gir alus
s
t
cyo ulpe
rs
Bu
eo
c
Ca
ide
U
a
lis
Bos
xc
lu
s
Biso
b
on
an
us
v
s
batu
ora
Aep
no
e
pe
hellii
Kobu
lu
us
s
Connoc
us
du
Aquatic species
fro
Oreamnos
s unicornis
pard
Rupicapra rupica
lpe
Capra
Capra ibex
Ovis aries
lo
sb
nb
tt
Ly
s
caball
ycer
ca
x ju
b
n li
nta
Vu
isu
s burc
pro
Ly
s lech
ison
hera
cyo
lcu
lis
ony
os m
Ovis canadensis
pa
ian
haetes
tes
s
Equus
rda
Uro
Equu
us
Pant
Acin
e
reu
elam
Rhinocero
americanu
lis
cte
taurin
pu
Ny
pra
s
us
s
Figure 18.1 Phylogenetic tree of mammalian species for which demographic information was extracted from Malddaba and COMADRE
databases to estimate the demographic metrics under study. The habitat type (terrestrial vs. aquatic) of each species is reported. As an illustration,
database-specific values of one demographic metric (Hbirth ) are displayed.
Table 18.1 Model selection of PGLS models fitted to assess allometric relationships of a given
demographic trait estimated for mammalian species from Malddaba (a) and Comadre (b) databases.
The table displays AIC of each model fitted, and the selected model occurs in bold.
(a)
Model np T Hbirth ϕ S AFR
Log(BM) × Hab 4 47.523 6.745 130.984 −74.887 78.930

Log(BM) + Hab 3 46.237 5.578 129.924 −76.407 77.145
Log(BM) 2 45.110 10.572 128.005 −77.219 75.412
Hab 2 76.931 3.585 143.878 −49.671 102.765
1 1 77.474 9.792 142.672 −48.528 102.330
(b)
Model np T Hbirth ϕ S AFR
Log(BM) × Hab 4 167.203 −8.522 −166.040 69.277 131.338

Log(BM) + Hab 3 165.770 −9.888 −165.714 67.373 129.718
Log(BM) 2 163.798 −11.590 −167.593 66.265 127.720
Hab 2 171.805 −7.565 −153.902 65.412 137.081
1 1 170.999 −8.141 −152.633 64.686 136.832
T: generation time (time); Hbirth (dimensionless): mortality entropy from birth; ϕ (time−1 ): mean number of recruits produced
per female during its lifetime; S (dimensionless): Demetrius’s entropy; AFR (time): age at first reproduction; np: number of
parameters of the model; BM: adult body mass; Hab: habitat type.
Table 18.2 Parameter estimates of the PGLS model retained to assess allometric relationships of a given demographic trait estimated for
mammalian species from Malddaba (a) and Comadre (b) databases. For each demographic trait, the table displays the selected model; the
parameters (intercept, allometric slope, and habitat effect), with standard error in parentheses; the estimate of the phylogenetic signal and its 95%
confidence interval; and the explanatory power of the selected model (measured by R2 , the proportion of variance explained).
(a)
Demographic Model Intercept Slope Effect of λ R2

Trait selected Hab
T Log(BM) 0.463 (0.414) 0.158 (0.023) NA 0.891 [0.733 – 0.970] 0.435

Hbirth Hab −0.309 (0.033) NA 0.028 (0.078) 0.000 [0.000 – 0.188] 0.117
ϕ Log(BM) 0.329 (0.689) −0.181 (0.040) NA 0.818 [0.582 – 0.947] 0.242
S Log(BM) 0.315 (0.163) 0.057 (0.009) NA 0.887 [0.738 – 0.962] 0.393
AFR Log(BM) −0.790 (0.441) 0.165 (0.026) NA 0.783 [0.550 – 0.921] 0.381
(b)
Demographic Model Intercept Slope Effect of λ R2

Trait selected Hab
T Log(BM) 1.448 (0.319) 0.096 (0.031) NA 0.000 [0.000 – 0.334] 0.123

Hbirth Log(BM) 0.204 (0.091) 0.025 (0.009) NA 0.000 [0.000 – 0.488] 0.107
ϕ Log(BM) 0.205 (0.030) −0.013 (0.003) NA 0.000 [0.000 – 0.471] 0.221
S 1 0.413 (0.046) NA NA 0.000 [0.000 – 0.215] 0.000
AFR Log(BM) 0.804 (0.247) 0.082 (0.024) NA 0.000 [0.000 – 0.251] 0.147
λ: phylogenetic signal.
demographic traits (i.e. Demetrius’s entropy S and axis (PPC1) accounted for 42.94% of the total demo-
Keyfitz’s entropy Hbirth ) showed contrasting outcomes graphic variance among the four metrics (Table 18.3)
between the two databases. Keyfitz’s entropy Hbirth and was mostly structured by the trade-off between
did not depend on adult body mass in the analy- dimensionless timing of first reproduction and dimen-
ses using Malddaba (Table 18.2a), whereas a positive sionless productivity and, to a much lesser extent, by
but weak association between Keyfitz’s entropy Hbirth the trade-off between diversity of mortality trajectories
and adult body mass was identified in the analyses and diversity of reproductive trajectories (Figure 18.3).
using Comadre (Table 18.2b). Demetrius’s entropy S This structuring axis thus included a continuum
was positively associated with adult body mass for opposing species with a relatively late age at first
data from Malddaba, whereas S did not depend on reproduction (correlation between PPC1 and dimen-
adult body mass from Comadre analyses. From the two sionless timing of first reproduction TFR of 0.740),
databases we did not find any support for an influence a high diversity of reproductive trajectories (corre-
of living in water or on land on mammalian demo- lation between PPC1 and Demetrius’s entropy S of
graphic metrics either directly or through a differen- 0.577), a low relative number of recruits (correlation
tial response to adult body mass changes (Table 18.1). between PPC1 and dimensionless productivity Prod
Only Keyfitz’s entropy Hbirth displayed an average of −0.893), and to a lesser extent a low diversity of
difference between terrestrial and aquatic species ages at death (correlation between PPC1 and Key-
from Malddaba data, with aquatic mammals having fitz’s entropy Hbirth of −0.200) to species presenting
higher mortality entropy at birth than their terrestrial the opposite set of metrics. The second axis (PPC2)
counterparts. accounted for 31.34% of the total demographic vari-
ance among the four traits (Table 18.3) and was
mostly structured by a continuum of diversity of
18.3.2 Assessing axes of variation among
trajectories in terms of both mortality and reproduc-
dimensionless demographic metrics
tion. This structuring axis corresponded to a con-
From the PPCA performed using Malddaba we uncov- tinuum opposing species with a high diversity of
ered two major axes of variation among dimen- ages at death (correlation between PPC2 and Key-
sionless demographic metrics (Figure 18.2a). The first fitz’s entropy Hbirth of 0.855), a high diversity of
(a)
20
Order
Hbirth
Artiodactyla
S
10 Carnivora
Cetacea
PPC2 (31.3%)
Chiroptera
Diprodontia
Perissodactyla
0 Primates
Prod
Proboscidea
TFR Rodentia
–10 Habitat
Aquatic
Terrestrial
–20
–20 –10 0 10 20 30
PPC1 (42.9%)
(b)
10
TFR Order
Hbirth Artiodactyla
Carnivora
Cetacea Figure 18.2 Graphical displays of the
Chiroptera first factorial plan (PPC2 vs. PPC1)
Cingulata obtained from the multivariate analysis of
PPC2 (26.1%)
0 Dasyuromorphia
Didelphimorphia dimensionless demographic traits
S Diprotodontia performed on Malddaba (a) and on
Lagomorpha COMADRE (b) databases. Each data point
Prod Primates
Proboscidea corresponds to a mammalian species, and
Rodentia the nonindependence among species data
points was corrected for by using a
–10 Habitat phylogenetically controlled PCA. The
Aquatic
Terrestrial demographic metrics (lHbirth :
log-transformed mortality entropy from
birth; lS: log-transformed Demetrius’s
entropy; Prod: dimensionless productivity;
–10 0 10 TFR: dimensionless timing of first
PPC1 (46.8%) reproduction) are displayed in red.
reproductive trajectories (correlation between PPC2 demographic metrics were both closely related to
and Demetrius’s entropy S of 0.676), and to a less- generation time T, which measures the ranking of
er extent a relatively early age at first reproduction species along the slow–fast continuum (Figure 18.3).
(correlation between PPC2 and dimensionless timing Thus, PPC1 was strongly correlated positively with
of first reproduction TFR of −0.253) to species with generation time T, which explained most observed
opposite metrics. The dimensionless productivity con- variation in PPC1 (R2 of 0.958). Likewise, although
tributed very little to PPC2 (correlation between PPC2 mathematically independent of PPC1, PPC2 was also
and dimensionless productivity Prod of 0.036). The positively correlated with generation time T, which
two remaining axes accounted for only one-quarter explained almost half the variation observed in PPC2
of the demographic variance across species when (R2 of 0.427).
pooled and cannot be interpreted in biological terms We analysed the influence of habitat type on PPCs,
(Table 18.3). Interestingly, these two independent on generation time T, and on the covariation between
axes structuring the variation among dimensionless structuring axes and generation time T. On average,
Table 18.3 Proportion of variance explained by the four aquatic species had higher values of PPC1 (intercept
phylogenetically corrected principal components (PPC1–PPC4) difference of 6.731 ± 3.448, P = 0.055) and higher
obtained from PPCA performed on four dimensionless demographic values of PPC2 (intercept difference of 5.557 ± 1.614,
metrics measured from data extracted from Malddaba and Comadre
P = 0.001), likely because they displayed a slower pace
databases. Only the first two PPCs (in bold) are interpreted in terms of
structuring axes of demographic variation due to their relevance in of life on average (i.e. generation time T longer by
passing the broken-stick rule (Jackson 1993). 0.483 ± 0.203 [log-scale], P = 0.020) than terrestrial
species. We did not detect any evidence of an influence
Database PPC1 PPC2 PPC3 PPC4 for habitat type on the slope between PPCs and gener-
ation time T (a slope difference of −1.010 ± 2.149, P =
Malddaba 0.4294 0.3134 0.1700 0.0871
Comadre 0.4680 0.2614 0.1986 0.0720 0.640 and 4.540 ± 8.881, P = 0.611 with terrestrial habi-
tats as a reference for PPC1 and PPC2, respectively).
On the other hand, aquatic species had lower scores for
PPC1 (i.e. towards early and high reproduction with
Malddaba COMADRE a low diversity of reproductive trajectories) and high-
PPC1
- er scores for PPC2 (i.e. towards high diversity of both
High Hbirth Prod
Prod s High Hbirth s reproductive trajectories and ages at death) than ter-
Fast restrial species for a given generation time T (intercept
TFR TFR difference of −1.313 ± 0.750, P = 0.085 and 3.192 ± 1.340,
P = 0.020 with terrestrial habitat as a reference for PPC1
and PPC2 (Figure 18.4), respectively).
The PPCA using Comadre also yielded two major
Low Hbirth Prod s Low Hbirth
Prod s axes of variation among dimensionless demographic
Slow metrics (Figure 18.2b). The first axis (PPC1) accounted
TFR TFR for 46.80% of the total demographic variance among
+
the four traits (Table 18.3). This structuring axis cor-
Figure 18.3 Biological interpretation of the main structuring factor responded to a continuum opposing species with
shaping the variation among dimensionless demographic metrics in a relatively early age at first reproduction (correla-
the examined mammalian species according to the database used. tion between PPC1 and dimensionless timing of first
When analysing data from Malddaba, mammalian species are reproduction TFR of −0.417), a low diversity of repro-
distributed along a main axis opposing species with a relatively early ductive trajectories (correlation between PPC1 and
age at first reproduction (low TFR, brown trait), producing a relatively
Demetrius’s entropy S of −0.859), a low relative num-
large number of recruits (high Prod, open red circles) and displaying a
low diversity of reproductive trajectories for a given population growth ber of recruits (correlation between PPC1 and dimen-
rate (low S, green ellipses) and, to a much lesser extent, a high sionless productivity Prod of −0.828), and a low diver-
diversity of mortality trajectories from birth (Hbirth , in black) to species sity of ages at death (correlation between PPC1 and
with opposing characteristics. The size of the acronym indicates the Keyfitz’s entropy Hbirth of −0.525) to species with oppo-
contribution of the metric to the axis. This axis is closely correlated to site metrics. The second axis (PPC2) accounted for
the ranking of mammalian species along the continuum opposing 26.14% of the total demographic variance among the
species with a fast pace of life (Fast, in red) to species with a slow four metrics (Table 18.3). This structuring axis corre-
pace of life (Slow, in red), with generation time (T) as a metric. On the
sponded to a continuum opposing species with a high
other hand, when analysing data from Comadre, the main axis of
covariation among dimensionless demographic metrics opposes diversity of ages at death (correlation between PPC2
species with a relatively late age at first reproduction (high TFR, and Keyfitz’s entropy Hbirth of 0.553), a low diver-
brown trait), producing a relatively large number of recruits (high sity of reproductive trajectories (correlation between
Prod, open red circles) and displaying both a high diversity of PPC2 and Demetrius’s entropy S of −0.332), a relative-
reproductive trajectories for a given population growth rate (low S, ly late age at first reproduction (correlation between
green ellipses) and a high diversity of mortality trajectories from birth PPC2 and dimensionless timing of first reproduction
(Hbirth , in black) to species with opposing characteristics. This axis TFR of 0.706), and a low relative number of recruits
provides a ranking of mammalian species independent of their ranking
(correlation between PPC2 and dimensionless produc-
along the slow–fast continuum of life histories obtained when using
generation time (T) as a metric. Note that in both cases, the intervals tivity Prod of −0.362) to species with opposite met-
over which these covariations are assessed (flat blue lines) are rics. The two other axes also accounted for slightly
constant and dimensionless. These intervals can be interpreted more than one-quarter of the demographic variance
biologically as time-independent life cycles that are identical across all across species when pooled and cannot be interpreted
examined mammalian species. in biological terms (Table 18.3). While the partition of
on age-specific vital rates (Malddaba database) has

9 Terrestrial species
Aquatic species little effect on these patterns. Second, the strength
6
of allometric relationships and the demographic
3 covariation identified from the multivariate PPCAs
0 differ between Malddaba and Comadre. These dif-
PPC2
–3 ferent outcomes cannot be explained either by a

difference in explanatory power because the number
–6
of species was slightly larger in analyses performed
–9 on Comadre or by a different representation of mam-
–12 malian groups between data sets, which both equally
–15 covered the main mammalian orders (see Figure 18.1).
0.8 1.4 2.0 2.6 3.2 This finding supports the hypothesis that Maldda-
Generation time ba, which includes full age dependence in lx and
(years, log-transformed) mx schedules, provides an accurate description of the
age-based demographic trajectories across mammals.
Figure 18.4 Relationship between generation time (T, on a log Third, both axes of demographic variation assessed
scale) and PPC2 across mammalian species from Malddaba. For a from dimensionless traits are closely associated with
given generation time, aquatic mammalian species have higher PPC2 the slow–fast continuum, supporting the significance
scores, which correspond to higher diversity of reproductive and
of this time-scale to shape demographic tactics across
mortality trajectories and to a lesser extent an earlier age at first
mammals (Figure 18.3). Fourth, terrestrial and aquatic
reproduction.
mammals differ only slightly in terms of demograph-
ic covariation among their dimensionless demographic
metrics. However, in support of a recent compara-
demographic variance among axes of variation was tive analysis of the demographic influence of habitat
similar (Table 18.3), the covariation among demo- type performed across the Tree of Life (Capdevila et al.
graphic traits differed depending on which database 2020), aquatic species reproduce earlier and produce a
was analysed. Contrary to what we reported from anal- larger number of recruits than terrestrial species rela-
yses performed on Malddaba, these two independent tive to their position on the slow–fast continuum.
axes structuring the variation among dimensionless Our findings clearly support the view that demo-
demographic metrics were not related to the ranking graphic metrics with a dimension of time are hypo-
of species on the slow–fast continuum. Thus, neither allometrically (i.e. with an allometric coefficient less
PPC1 (slope of −0.434 ± 0.836, P = 0.605) nor PPC2 than 1) linked with body mass across species (Peters
(slope of −0.309 ± 0.592, P = 0.603) were associated with 1983; Calder 1984), independently of the database
generation time T, and the two negative trends only used. On the other hand, dimensionless metrics dis-
accounted for a negligible proportion of observed vari- played different patterns in relation to the database
ation in PPC1 and PPC2 (R2 of ~0.004 in both cases). used, which demonstrate that dimensionless metrics
We did not detect any evidence of an influence of habi- are not necessarily independent of adult body size
tat types on either of the two examined PPCs (intercept and can display a monotonous relationship with body
difference of −0.589 ± 1.650, P = 0.722 and 1.596 ± 1.153, mass. Using Malddaba, we found clear evidence that
P = 0.171 with terrestrial habitat as a reference for PPC1 S, Demetrius’s entropy, is positively associated with
and PPC2, respectively), and generation time T did adult body size across mammals. Demetrius (1979)
not differ between aquatic and terrestrial species either defined his dimensionless metric as the number of
(intercept difference of 0.256 ± 0.237, P = 0.283). genealogies in a population expected for a given
population growth. This high diversity of reproduc-
tive trajectories in slow-living species compared to
18.4 Conclusions fast-living ones might be the result of strong selec-
Our comparative demographic analyses provide four tion pressure against mortality patterns (illustrated
major findings. First, the demographic structures by the low diversity of Keyfitz’s entropy Hbirth in
including allometric relationships of biological times slow-living compared to fast-living species we report),
and the key axes of variation in demographic met- leading reproductive trajectories to respond to species-
rics are retrieved in both Malddaba and Comadre specific lifestyle and ecology. While there is only
databases. This suggests that detailed information
one way to have a slow pace of life (i.e. by avoid- adult stage to get an accurate picture of the demog-
ing risk to not jeopardise survivorship; Péron et al. raphy of a given population. More recently, Colchero
2019), the diversity of reproductive trajectories allows et al. (2019) showed how including age dependence
individuals in slow-living species to fine-tune state- in vital rates during the adult stage markedly mod-
dependent adjustments to the ecological context. We ulates the estimates of stochastic population growth,
can expect this number to increase with the duration which is crucial in the current context of climate
of the life cycle, which itself increases with adult body change.
size. When including detailed age dependence in vital
The most conflicting results between Malddaba rates, there was another key finding. The ranking of
and Comadre occurred for Hbirth , Keyfitz’s entropy of mammalian species along the main axis of variation
mortality from birth. This is not surprising because among dimensionless demographic metrics perfect-
Keyfitz’s entropy Hbirth is highly sensitive to age depen- ly matched the species ranking along the slow–fast
dence in mortality during the lifetime (Goldman and continuum. Thus, when standardising all mammalian
Lord 1986). For a substantial proportion of species, species to the same dimensionless life cycle duration
Keyfitz’s entropy Hbirth in Comadre was calculated starting at a value of 0 and ending at a value of 1,
without including potential actuarial senescence. On they ranked along an axis of demographic covaria-
the contrary, Keyfitz’s entropy Hbirth in Malddaba was tions among dimensionless metrics from early, high,
calculated from the full age dependence of mortal- and weakly diverse reproductive trajectories with (at a
ity reported, including the mortality increase with much lesser extent) a high diversity of ages at death to
age caused by actuarial senescence (i.e. the increase late, low, and highly diverse reproductive trajectories
in mortality rate with increasing age). As actuarial with (at a much lesser extent) a low diversity of ages at
senescence is the rule rather than the exception in death.
mammals (Nussey et al. 2013), we recommend includ- The identical nature of these two markedly differ-
ing age dependence in adult mortality to consider ent continua demonstrates that the speed of the life
Keyfitz’s entropy Hbirth in comparative demograph- cycle and the shape of the life cycle are strongly asso-
ic analyses focused on a given taxon (e.g. order or ciated, suggesting strong constraints in demographic
class). However, at the broad level of the Tree of variation across species. As pointed out by Péron et al.
Life, this problem is likely negligible because of the (2019), the constraint of adopting a risk-adverse life his-
huge diversity of mortality curves (see e.g. Jones et al. tory strategy to be long-lived and display a slow pace
2014), whereby mammals can be simply characterised of life strongly reduces variation in ages at death. Like-
by a decreased mortality between the juvenile and wise, the low demographic impact of a given change in
adult stage and a low mortality during the adult reproduction for slow-living species (Hamilton 1966)
stage. allows them to display a large array of reproductive
The multivariate analysis of covariations among trajectories, often shaped by environmental stochastici-
dimensionless demographic metrics identified the ty (Tuljapurkar et al. 2009). This similarity of biological
same structuring of the shape of demographic trajec- meaning between different continuums is parallel to
tories, with a main axis of variation accounting for the similarity of the nature of variation among demo-
more than 40% of the overall variation and the two graphic traits between analyses correcting or not for
first components accounting for about three-quarters allometric effects of body size (Stearns 1983; Gaillard
of the overall variation, irrespective of the database et al. 1989) and indicates that time is key to shape not
analysed. However, the nature of this axis of variation only the timing of demographic events but also the
differed between analyses performed using Maldda- shape of variation in the occurrence in these events
ba and Comadre, indicating that the occurrence, the along the life cycle.
strength, and the shape of age dependence in sur- Finally, our comparison between databases provides
vival and reproduction have a strong effect on the a kind of proof of concept. Some evolutionary ecolo-
shape of demographic trajectories, at least across mam- gists not used to dimensional analyses (see Stahl 1962)
malian species. This finding should not come as a and the appropriate use of ratios (see Smith 1999) might
surprise, as previous studies have pointed out the be tempted to see some circularity in our findings using
need to include full age dependence when analysing the Malddaba database. After all, two of our dimen-
demographic patterns. For instance, Eberhardt (1985) sionless traits included generation time explicitly in
was one of the first to highlight the importance of their definition, one in a product (Prod) and the oth-
considering age dependence in survival within the er in a ratio (TFR). At first glance, it may seem trivial
that a compound variable XxZ should be negatively general database including simplified descriptions of
correlated with a compound Y/Z and thus strong- species-specific life cycles for a large number of diverse
ly structure the first PCA axis when included with species such as Comadre allows suitable comparisons.
two other mathematically independent traits, and lead
the PCA axis to be strongly related to Z. However,
the replication of the analyses to the Comadre data References
set empirically proves that it is not the case, mostly
Atchley, W. R., Gaskins, C. T., & Anderson, D. (1976). Sta-
because, as well explained by Smith (1999) when X tistical properties of ratios. I. Empirical results. Systematic
and Y are each highly correlated with Z (which was Zoology, 25:137–148.
the case here because all metrics included in Prod and Baudisch, A. (2011). The pace and shape of ageing. Methods
TRF were biological times that are strongly associat- in Ecology and Evolution, 2:375–382.
ed across species; see e.g. Calder 1984), the correla- Bennett, P. M. & Harvey, P. H. (1987). Active and resting
tion between the ratios is not at all spurious. Indeed, metabolism in birds: allometry, phylogeny and ecology.
the first PCA axis from the Comadre analysis based Journal of Zoology, 213:327–344.
on the same metrics revealed a positive association Bielby, J., Mace, G. M., Bininda-Emonds, O. R., et al.
(2007). The fast-slow continuum in mammalian life his-
between Prod and TFR instead of the negative associa-
tory: an empirical reevaluation. The American Naturalist,
tion that a sceptical evolutionary ecologist would have
169:748–757.
expected. Bininda-Emonds, O. R., Cardillo, M., Jones, K. E., et al.
(2007). The delayed rise of present-day mammals. Nature,
446:507–512.
Bininda-Emonds, O. R., Cardillo, M., Jones, K. E., et al.
18.5 Prespectives (2008). Erratum: the delayed rise of present-day mam-
The inability to retrieve these demographic outcomes mals. Nature, 456:274-274.
when using Comadre data is highly informative but Blackburn, T. M. (1991). Evidence for a fast–slow continu-
does not preclude analyses from this database. As the um of life-history traits among parasitoid Hymenoptera.
Functional Ecology, 5:65–74.
devil is in the detail, reducing age dependence to an
Blueweiss, L., Fox, H., Kudzma, V., Nakashima, D., Peters,
opposition between juvenile and adult stages might
R., & Sams, S. (1978). Relationships between body
be more appropriate for broad-scale analyses at the size and some life history parameters. Oecologia, 37:
level of the Tree of Life, the clear objective of this 257–272.
database (see e.g. Paniw et al. 2018; Healy et al. 2019). Burnham, K. P. & Anderson, D. R. (2002). A practical
First, including detailed st(age)-specific demographic information-theoretic approach. Model selection and multimod-
information on species ranging from invertebrates to el inference, 2nd edition. Springer, New York, 2.
vertebrates will tremendously reduce the number of Calder, W. A. (1984). Size, function, and life history. Harvard
species analysed as our current knowledge of demo- University Press, Cambridge.
graphic traits across the Tree of Life is very scarce Capdevila, P., Beger, M., Blomberg, S. P., Hereu, B., Linares,
C., & Salguero-Gómez, R. (2020). Longevity, body dimen-
(Conde et al. 2019). Then, even if available, includ-
sion and reproductive mode drive differences in aquatic
ing detailed demographic information across a broad
versus terrestrial life-history strategies. Functional Ecology,
range of species would likely prevent us from identify- 34:1613–1625.
ing clear structures in demographic variation because Charnov, E. L. (1993). Life history invariants: some explorations
variation within taxon will likely blur the big pic- of symmetry in evolutionary ecology. Oxford University
ture of across-taxon variation. Therefore, instead of Press, Oxford.
opposing databases, our analysis should be interpret- Colchero, F., Jones, O. R., Conde, D. A., et al. (2019).
ed as illustrating the need to carefully choose the The diversity of population responses to environmental
database to analyse for answering a specific question change. Ecology Letters, 22:342–353.
about life history variation. This chapter illustrates the Cole, L. C. (1954). The population consequences of life
history phenomena. Quarterly Review of Biology, 29:
philosophy of this book: different types of questions
103–137.
can be tackled with different data sources and differ-
Conde, D. A., Staerk, J., Colchero, F., et al. (2019). Data
ent (and appropriate!) methodologies. Thus, to assess gaps and opportunities for comparative and conserva-
fine-scaled demographic variation, a specific database tion biology. Proceedings of the National Academy of Sciences,
including the key factor of variation (e.g. age in mam- 116:9658–9664.
mals) such as Malddaba is required. However, when Demetrius, L. (1979). Relations between demographic
looking for the big picture across the Tree of Life, a parameters. Demography, 16:329–338.
Dobson, F. S. (2007). A lifestyle view of life-history evolu- Jones, O. R., Scheuerlein, A., Salguero-Gómez, R., et al.
tion. Proceedings of the National Academy of Sciences, 104: (2014). Diversity of ageing across the tree of life. Nature,
17565–17566. 505:169–173.
Eberhardt, L. L. (1985). Assessing the dynamics of wild Keyfitz, N. (1985). Applied mathematical demography, 2nd
populations. Journal of Wildlife Management, 40:997–1012. edition. Springer, New York.
Eisenberg, J. F. (1981). The mammalian radiations: an analy- Lemaître, J. F., Ronget, V., Tidière, M., et al. (2020a). Sex
sis of trends in evolution, adaptation, and behaviour. Chicago differences in adult lifespan and aging rates of mortality
University Press, Chicago. across wild mammals. Proceedings of the National Academy
Felsenstein, J. (1985). Phylogenies and the comparative of Sciences, 117:8546–8553.
method. The American Naturalist, 125:1–15. Lemaître, J. F., Ronget, V., & Gaillard, J. M. (2020b). Female
Fenchel, T. (1974). Intrinsic rate of natural increase: the reproductive senescence across mammals: a high diversi-
relationship with body size. Oecologia, 14:317–326. ty of patterns modulated by life history and mating traits.
Fernandez, O. E. & Beltran-Sanchez, H. (2015). The entropy Mechanisms of Ageing and Development, 192:111377.
of the life table: a reappraisal. Theoretical Population Biolo- Mitchell, P. C. (1911). On longevity and relative viabili-
gy, 104:26–45. ty in mammals and birds; with a note on the theory of
Festa-Bianchet, M., Gaillard, J. M., & Côté, S. D. (2003). longevity. Proceedings of the Zoological Society of London,
Variable age structure and apparent density dependence 81:425–548.
in survival of adult ungulates. Journal of Animal Ecology, Nussey, D. H., Froy, H., Lemaitre, J. F., Gaillard, J. M., &
72:640–649. Austad, S. N. (2013). Senescence in natural populations
Freckleton, R. P., Harvey, P. H., & Pagel, M. (2002). Phyloge- of animals: widespread evidence and its implications for
netic analysis and comparative data: a test and review of bio-gerontology. Ageing Research Reviews, 12:214–225.
evidence. The American Naturalist, 160:712–726. Pagel, M. (1999). Inferring the historical patterns of biologi-
Gaillard, J. M., Yoccoz, N. G., Lebreton, J. D., et al. (2005). cal evolution. Nature, 401:877–884.
Generation time: a reliable metric to measure life-history Paniw, M., Ozgul, A., & Salguero-Gómez, R. (2018). Inter-
variation among mammalian populations. The American active life-history traits predict sensitivity of plants and
Naturalist, 166:119–123. animals to temporal autocorrelation. Ecology Letters,
Gaillard, J. M., Pontier, D., Allainé, D., et al. (1989). An anal- 21:275–286.
ysis of demographic tactics in birds and mammals. Oikos, Pélabon, C., Tidière, M., Lemaître, J. F., & Gaillard, J. M.
56:59–76. (2018). Modelling allometry: statistical and biological
Gaillard J. M, Lemaître, J. F., Berger, V., et al. (2016). Life his- considerations—a reply to Packard. Biological Journal of the
tories, axes of variation. In R. M. Kliman, ed. Encyclopedia Linnean Society, 125:664–671.
of evolutionary biology. vol. 2. Academic Press, Oxford. Pp. Péron, G., Lemaître, J. F., Ronget, V., Tidière, M., & Gaillard,
312–323. J. M. (2019). Variation in actuarial senescence does not
Garamszegi, L. Z. (ed.). (2014). Modern phylogenetic compar- reflect life span variation across mammals. PLOS Biology,
ative methods and their application in evolutionary biology: 17:e3000432.
concepts and practice. Springer, Berlin. Peters, R. H. (1983). The ecological implications of body size.
Goldman, N. & Lord, G. (1986). A new look at entropy and Cambridge University Press, Cambridge.
the life table. Demography, 23:275–282. Promislow, D. E. & Harvey, P. H. (1990). Living fast
Hamilton, W. D. (1966). The moulding of senescence by and dying young: a comparative analysis of life-history
natural selection. Journal of Theoretical Biology, 12:12–45. variation among mammals. Journal of Zoology, 220:
Harvey, P. H. & Clutton-Brock, T. H. (1985). Life history 417–437. “
variation in primates. Evolution, 39:559–581. Revell, L. J. (2012). phytools: an R package for phylogenetic
Harvey, P. H. & Pagel, M. D. (1991). The comparative method comparative biology (and other things). Methods in Ecology
in evolutionary biology. Oxford University Press, Oxford. and Evolution, 3:217–223.
Healy, K., Ezard, T. H., Jones, O. R., Salguero-Gómez, R., Riek, A. & Geiser, F. (2013). Allometry of thermal variables
& Buckley, Y. M. (2019). Animal life history is shaped in mammals: consequences of body size and phylogeny.
by the pace of life and the distribution of age-specific Biological Reviews, 88:564–572.
mortality and reproduction. Nature Ecology & Evolution, Ronget, V. & Gaillard, J. M. (2020). Assessing ageing pat-
3:1217–1224. terns for comparative analyses of mortality curves: going
Huxley, J. S. (1932). Problems of relative growth. Methuen and beyond the use of maximum longevity. Functional Ecology,
Co., London. 34:65–75.
Jackson, D. A. (1993). Stopping rules in principal compo- Russell, E. M. (1982). Patterns of parental care and parental
nents analysis: a comparison of heuristical and statistical investment in marsupials. Biological Reviews, 57:423–486.
approaches. Ecology, 74:2204–2214. Salguero-Gómez, R., Jones, O. R., Jongejans, E., et al. (2016a).
Jeschke, J. M. & Kokko, H. (2009). The roles of body size Fast–slow continuum and reproductive strategies struc-
and phylogeny in fast and slow life histories. Evolutionary ture plant life-history variation worldwide. Proceedings
Ecology, 23:867–878. of the National Academy of Sciences, 113:230–235.
Salguero-Gómez, R., Jones, O. R., Archer, C. R., et al. (2016b). Stearns, S. C. (1983). The influence of size and phylogeny
Comadre: a global data base of animal demography. Jour- on patterns of covariation among life-history traits in the
nal of Animal Ecology, 85:371–384. mammals. Oikos, 41:173–187.
Silvertown, J., Franco, M., Pisanty, I., & Mendoza, A. Stubben, C. & Milligan, B. G. (2007). Estimating and analyz-
N. A. (1993). Comparative plant demography—relative ing demographic models using the popbio package in R.
importance of life-cycle components to the finite rate of Journal of Statistical Software, 22:1–23.
increase in woody and herbaceous perennials. Journal of Swihart, R. K. (1984). Body size, breeding season length,
Ecology, 81:465–476. and life history tactics of lagomorphs. Oikos, 43:
Smith, R. J. (1999). Statistics of sexual size dimorphism. 282–290.
Journal of Human Evolution, 36:423–458. Tuljapurkar, S., Steiner, U. K., & Orzack, S. H. (2009).
Southwood, T. R. (1977). Habitat, the templet for ecological Dynamic heterogeneity in life histories. Ecology Letters, 12:
strategies? Journal of Animal Ecology, 46:337–365. 93–106.
Staerk, J., Conde, D. A., Ronget, V., Lemaitre, J. F., Gaillard, J. Turbill, C., Bieber, C., & Ruf, T. (2011). Hibernation is associ-
M., & Colchero, F. (2019). Performance of generation time ated with increased survival and the evolution of slow life
approximations for extinction risk assessments. Journal of histories among mammals. Proceedings of the Royal Society
Applied Ecology, 56:1436–1446. B: Biological Sciences, 278:3355–3363.
Stahl, W. R. (1962). Similarity and dimensional methods in Upham, N. S., Esselstyn, J. A., & Jetz, W. (2019). Inferring
biology. Science, 137:205–212. the mammal tree: species-level sets of phylogenies for
Stearns, S. C. (1976). Life-history tactics: a review of the questions in ecology, evolution, and conservation. PLOS
ideas. Quarterly Review of Biology, 51:3–47. Biology, 17:e3000494
Stearns, S. C. (1977). The evolution of life history traits: a cri- Wiersma, P., Muñoz-Garcia, A., Walker, A., & Williams, J. B.
tique of the theory and a review of the data. Annual Review (2007). Tropical birds have a slow pace of life. Proceedings
of Ecology and Systematics, 8:145–171. of the National Academy of Sciences, 104:9340–9345.
CHAPTER 19
Adaptive management: making

recurrent decisions in the face
of uncertainty
James D. Nichols
19.1 Introduction rates the transition probabilities that govern under-

lying dynamics. Finally, studies of disease dynamics
Many, if not most, decisions in conservation biology
may focus on the fraction of a population infected
and wildlife management require a focus on system
by the disease, with vital rates the individual tran-
dynamics. One or more system state variables is iden-
sition probabilities among uninfected, infected, and
tified as being relevant to programme objectives, and
post-infection states, as well as state-specific mor-
management then focuses on effects of potential man-
tality rates. Some of these vital rates may not be
agement actions on these state variables. These effects
immediately recognisable as demographic parame-
of management actions generally influence system
ters, but I argue (see section 19.6) that they are
state via their effects on system vital rates, for example,
all functions of survival, reproduction, and move-
the demographic parameters that are the focus of this
ment and hence relevant to the central topic of
book. Thus, estimation of vital rates and the func-
this book.
tional relationships between management actions and
Decision processes for all of these situations require
vital rates is a key information need for conserva-
estimates of system state, the vital rates that drive state
tion programmes. These functional relationships form
dynamics, and the relationships between management
the basis for projecting effects of management on sys-
actions and these vital rates. Estimates of these quan-
tem state and hence for selecting optimal management
tities must typically deal with nondetection (counts of
actions.
entities in a spatial sample are typically undercounts,
This focus of decision-making on ecological state
as entities are missed) and sometimes misclassifica-
variables and the vital rates that drive their dynamics
tion (e.g. an individual classified as not infected may
is very general, extending across taxa. Single-species
actually carry the infection). The estimates of system
conservation problems typically focus on population
state are used to make state-dependent decisions, to
size as the focal state variable, and rates of survival,
assess progress towards meeting objectives, and to
reproductive recruitment, and movement as the vital
learn about system dynamics (see section 19.2). Esti-
rates that drive dynamics. Larger-scale single-species
mates of vital rates and their relationships with man-
problems sometimes focus on the fraction of a land-
agement actions are used for developing projection
scape occupied by a species, with vital rates iden-
models in order to make management decisions.
tified as local probabilities of extinction and coloni-
In this chapter, I first lay out the basic principles
sation. For community studies, species richness may
for structured decision-making. I then focus on the
be the focal state variable, with vital rates being
class of decisions that are recurrent and characterised
local species-level rates of extinction and colonisation.
by uncertainty, and the adaptive management (AM)
In some land management programmes, the num-
approach developed to deal with such decisions. I
ber of sites in each of a number of potential vegeta-
then move to three examples that use AM, outlining
tion types may be the focal state variable, with vital
James D. Nichols, Adaptive management: making recurrent decisions in the face of uncertainty.
the basic approach and emphasising the role of esti- perhaps maximisation of species harvest, conditional
mating vital rates in this approach. I conclude with on population size not dropping below a specified
a discussion of integrated demographic parameters threshold level.
and a plea for more methodological sharing among Managers must be able to identify a set of poten-
disciplines. tial actions (including ‘do nothing’) that can move
the system in a direction consistent with management
objectives. At each decision point, the decision problem
19.2 Structured decision-making is to select which action is most consistent with pro-
Humans are faced with numerous decisions every day, gramme objectives. Each decision thus requires models
some of which have important consequences for our that predict system responses to the potential actions
lives. In fact, an individual’s life can be viewed as a such that utility of each action can be judged relative to
decision process, as Gandalph reminded Frodo: ‘All objectives. Some managers are reluctant to use models,
we have to decide is what to do with the time that viewing them as academic exercises with little ground-
is given us’ (Tolkien 1954). Given this central role of ing in reality. However, models are simply structures
decisions, it is surprising that most of us spend so that permit predictions about system responses, and
little time thinking about useful processes for mak- without such predictions there is no basis for selection
ing them. ‘Management’ and ‘conservation’ both entail of an appropriate action; hence informed management
actions designed to achieve specified objectives, yet it is is not possible. Demographic vital rates are key compo-
possible to obtain graduate degrees in wildlife manage- nents of management models, and indeed the key rela-
ment and conservation biology without taking a single tionships in these models predict the effects of potential
course in decision-making! actions on vital rates.
Structured decision-making, for the purposes of this Monitoring of system state variables (e.g. popula-
chapter, is simply an attempt to formalise a common- tion size) and other variables relevant to objectives
sense approach to making decisions. The approach is important for multiple reasons. One is simply to
decomposes a decision into component parts, first assess the degree to which management is successful.
focusing on each separately and later combining Another principal reason is that decisions are typi-
them into a complete decision process (e.g. Clemen cally state-dependent. That is, we would expect to
and Reilly 2001; Martin et al. 2009). An important make very different decisions depending on whether
advantage of this approach is reduction of confusing a population size is above or below some target val-
discussions among stakeholders who disagree and ue based on objectives. A third reason is for learning,
tend to shift arguments from one component to anoth- as comparison of estimated state variables with model-
er. The approach also blunts the natural tendency of based predictions provides a basis for discriminating
stakeholders to focus directly on preferred actions, among competing models of system response to man-
sometimes without adequately considering true agement (see section 19.3.3). Finally, monitoring data
objectives. can be used to obtain updated estimates of model
The components of structured decision-making are parameters.
objectives, potential actions, models of system resp- The final component of structured decision-making
onse to actions, monitoring, and a decision algorithm. is a decision algorithm. The algorithm is used to select
Objectives are simply clear statements about what the action most consistent with objectives, based on the
stakeholders would like to achieve through manage- current state of the system (estimated via monitoring)
ment of the focal system. Even when all stakeholders and the projected consequences of the different poten-
are in general agreement, it can be surprisingly diffi- tial actions (based on models). Frequently, the decision
cult to develop specific objectives. In the more com- algorithm resides in the head of the decision-maker,
mon case, when different groups of stakeholders have but use of optimisation methods is more objective,
competing objectives, the development of compromise transparent, and defensible.
objectives can be very difficult. However, objectives are
not optional, as the concept of management depends
on them. Indeed, objectives drive the entire process
and are an important consideration in development
19.3 Adaptive management (AM)
of all other decision components. Management objec- Adaptive management (e.g. Walters and Hilborn 1976,
tives might include maintenance of population size of 1978; Walters 1986; Williams et al. 2007) is a spe-
an endangered species above some threshold value, cial case of structured decision-making developed to
attainment of a threshold value of species diversity, or deal with recurrent decision processes characterised
A D A P T I V E M A N A G E M E N T: M A K I N G R E C U R R E N T D E C I S I O N S I N T H E FA C E O F U N C E RTA I N T Y 315
by uncertainty. The basic idea underlying AM is disagreements about the likely effectiveness of poten-
that repeated management decisions can be used tial actions, with a tendency in wildlife management
not only to manage the focal system, but also to and conservation biology to overestimate the effects
reduce uncertainty about system responses to man- of many potential actions (Nichols 2012). Adaptive
agement. Each management decision and resultant management provides the opportunity to eventual-
action is based on model predictions of system ly resolve such disagreements via learning. Decisions
responses, and these predictions can be compared about the set of potential management actions are
against estimates of actual response, as obtained via sometimes as contentious as those involving objectives.
monitoring, to learn about the predictive abilities In summary, stakeholder meetings can be difficult to
of the model(s). The process of AM includes three direct and navigate but are an integral part of the
main phases: deliberative, iterative, and double-loop deliberative phase of AM for most decision problems.
learning. The purpose of such meetings is to emerge with clear
statements of decision-process objectives and a set of
available actions.
Models of system response to management actions,
19.3.1 Deliberative phase and monitoring designed to estimate system state
The deliberative, or set-up, phase focuses on develop- variables and other variables relevant to objectives,
ing the components of AM. The first two components, must also be developed. These deliberative phase
objectives and potential actions, typically require input tasks usually involve a smaller group than the
from all relevant stakeholder groups. Many decisions stakeholders who develop objectives and actions.
in conservation and natural resource management Development of models and monitoring is typical-
are about systems that are valued in different ways ly handled by professional managers and associated
by different stakeholder groups. For example, deci- scientists. A common approach is to establish work-
sions commonly involve systems that some groups ing groups that include some scientists and man-
want to protect from any potential deleterious action, agers who are experts for the focal systems and oth-
whereas others want to exploit at least some sys- ers who are experts in the quantitative aspects of
tem components. The deliberative phase of AM fre- modelling and monitoring. Once developed, mod-
quently includes stakeholder meetings such that all els and monitoring plans may be shared with the
relevant groups can have input into development larger stakeholder groups, and even modified in
of programme objectives and actions. Such meetings response to stakeholder concerns, as a means of insur-
can become contentious when different stakeholders ing transparency and stakeholder engagement in the
have competing interests, and efforts must be made process.
to develop compromise objectives to which all stake- Use of AM is predicated on a desire to reduce
holders agree. The objectives should ultimately be uncertainty about system response to management
incorporated into an objective function, an explicit actions. This uncertainty is typically expressed in sys-
(usually mathematical) statement of objectives. In the tem models in one of two ways. The approach with
case of multiple objectives, an objective function may which I have most experience actually identifies mul-
include different ‘weights’ for the different compo- tiple models that incorporate the range of plausible
nents of the function. Another approach to dealing system behaviours, and hopefully the range of stake-
with multiple objectives is to focus on one objec- holder beliefs. For example, in the case of resource
tive component, subject to a constraint or thresh- exploitation, stakeholders with protective conserva-
old associated with another component (e.g. Mar- tion perspectives frequently believe in larger effects of
tin et al. 2009). For example, an objective function exploitation on the focal system, whereas those favour-
might entail maximising harvest of a system coming exploitation often believe in minimal effects. One
ponent, subject to the constraint of that component advantage of AM is the ability to incorporate these
exceeding some threshold value (e.g. Johnson et al. competing views of system responses. If competing
1997). models are weighted equally in the deliberative phase,
Development of a set of potential actions usual- as is often wise in the case with competing stakeholder
ly requires stakeholder input and may be difficult as views, then no single model has undue influence in the
well. For example, some actions (e.g. predator con- initial management decision. As learning proceeds in
trol) that are acceptable to some stakeholders may be the iterative phase, models with greater predictive abil-
unacceptable to others for ethical or other reasons asso- ities exert greater influence in the decision process. If
ciated with individual values. There are frequently the model set includes at least one model that is a good
approximation to reality, then management decisions that satisfies all relevant stakeholders and keeps them
should improve. engaged and supportive of the AM process.
The other way to incorporate uncertainty about sys-
tem response entails use of a single general model
that includes one (or perhaps more) key parameter(s), 19.3.2 Iterative phase
different values of which can produce very different Armed with the described components of struc-
system responses to actions. In the deliberative phase, tured decision processes, as developed during the
it may be decided to begin the process with a value deliberative phase of AM, the iterative phase proceeds
that is intermediate with respect to the different effect as outlined in Figure 19.1. At each decision point,
hypotheses under consideration. The iterative phase denoted as time t, the decision algorithm is used to
then leads to increasingly better estimates for use in select the optimal action based on the other process
management decisions. components, that is, objective function, set of poten-
Finally, a decision algorithm must be selected. Recur- tial actions, system model(s), and estimated state of the
rent decision processes for state-dependent decisions system (based on monitoring). That action is taken (1)
typically cannot simply consider returns based on the driving the system to a new state and (2) providing
upcoming time-step and hence require dynamic opti- returns (desired changes in variables specified in objec-
misation algorithms. For example, consider a simple tives). The new state of the system is then identified via
objective of maximising harvest of a focal species over monitoring and is compared against predictions based
a 10-year time horizon. Optimisation of harvest for a on each model, yielding changes in model weights (see
one-step problem would be to harvest as many individ- below). At the next decision point, t+1, a new deci-
uals as possible. However, in a recurrent problem, such sion is made using the new model weights and new
an action may leave few or no individuals to harvest in estimate of system state. The process proceeds in this
the next time step. The decision at time t must thus con- manner through time.
sider not only returns between t and t+1, but also all In this case of multiple discrete models, learning
time steps for the remainder of the management pro- occurs via the evolution of model weights. Specifically,
gramme time horizon. Such decisions require dynamic we define the ‘information state’ as a vector of model
optimisation algorithms such as stochastic dynamic weights, πt (model i) for model i at time t, that reflect the
programming (Bellman 1957; Puterman 1994; Williams relative predictive abilities of models in the model set.
et al. 2002). In the case of M models in the set:
In summary, the purpose of the deliberative phase
is to develop all of the components required for the X
M
iterative phase in which AM is actually implemented. πt (model i) = 1. (19.1)
Hopefully, this development is conducted in a manner i=1
Management of Dynamic Resources
management management management

action action action
... resource resource resource ...

system system system
Figure 19.1 Diagrammatic
representation of a recurrent decision
problem. At each decision point an action
is taken based on the objectives, potential
return return return
actions, models, and current system state.
The action induces a return and drives the
system to a new state that becomes the
basis for the next decision. (Source:
time
t–1 t t+1 Nichols and Williams 2013.)
We are more confident in models with higher weights, where Pr(data t+1 | model i) is the probability that the
viewing them as more likely to represent reasonable new observations at time t+1 would have arisen, given
abstractions of the modelled natural processes. The that model i is a good representation of the actual pro-
weights of the models at a decision point determine cess that generated them. Equation (19.2) is based on
their relative influence on the optimal decision. Initial Bayes’ theorem or rule (Bayes and Price 1763). The
weights (at the beginning of the management pro- updating of model weights is based on the relative
gramme) can be based on historic information (e.g. one confidence in the model that has accumulated through
might use AIC weights from a single analysis; Burn- time t, πt (model i), and the consistency of the new
ham and Anderson 2002) or intuition, or simply set observations with that model, Pr(data t+1 | model i). If
equal (1/M) for each model. the model set includes a good approximating model,
Subsequent model weights evolve with new obser- then the weight for that model should evolve to
vations according to: approach 1, whereas the weights of models that pre-
dict more poorly should eventually approach 0 (see
(πt+1 (model i) | datat+1 )
Figure 19.2).
πt (model i) Pr (datat+1 |model i) The iterative phase thus entails making decisions
= PM , (19.2)
i πt (model i) Pr (datat+1 |model i)
that are optimal with respect to specified objectives,
Observed
Model SaRw
10 Model ScRw
Model SaRs
Model ScRs
Population in millions
9 Model average
0.6
Model weights
0.4
0.2
0.0
1995 2000 2005 2010

Year
Figure 19.2 Top: population estimates of midcontinent mallards compared to predictions of each model (Sa = additive mortality; Sc =
compensatory mortality; Rw = weakly density-dependent reproduction; Rs = strongly density-dependent reproduction) from 1996 to 2013. Grey
shading represents 95% confidence intervals for population estimates. Arrows begin at the population estimate of year t and point to the weighted
mean annual prediction for year t+1. Bottom: changes in model weights from 1995 (equal weights) to 2013. (Source: Nichols et al. 2015.)
while simultaneously reducing uncertainty and learn- vital rates and state variables (e.g. Seber 1982; Williams
ing which model(s) represents the best approximation et al. 2002) play an important role in AM. Similarly,
to the processes governing system responses to man- projection models for predicting the consequences of
agement actions. This combination of wise decision- management actions for system dynamics and other
making and simultaneous learning distinguishes adap- variables identified in objectives are essential for AM
tive management from other forms of management. as well, forming the basis for optimal decisions. In
sections 19.4.1–19.4.4, I briefly discuss inference for
four common state variables and their vital rates.
19.3.3 Double-loop learning phase
Learning that occurs during the iterative phase of AM 19.4.1 Population size/abundance
may lead managers to revisit decisions made during
Frequently, the state variable of single-species man-
the deliberative phase. For example, objectives may
agement is population size, and the associated vital
evolve, resulting in a need for revision (Johnson et al.
rates are survival, reproduction, and movement in
2015). Similarly, if experience with the set of identi-
and out of the focal population. Ecologists have devel-
fied actions indicates that none of them is particularly
oped a variety of methods for estimating population
effective, development of new actions will be useful.
size in the face of unknown detection probabilities.
In addition, model weights may fluctuate nonmono-
These methods include closed and open capture–
tonically, and/or none of the models in the set may
recapture models, removal methods, catch-effort
be predicting well. In such cases, it may be useful to
methods, change-in-ratio methods, distance sampling,
revisit the model set, perhaps adding models and even
double-observer models, time-at-detection models,
deleting some. Finally, the monitoring may not be pro-
and N-mixture models, as well as approaches that
viding estimates of sufficient precision to permit wise
combine two or more of these methods (e.g. see
decisions and rapid learning, leading to a need to revise
Chapters 5, 7, and 14 and reviews in Seber 1982;
the current monitoring programme.
Williams et al. 2002; Lancia et al. 2005; Nichols
Thus, a perception of programme organisers that
et al. 2009).
any of the process components should be revised may
Methods for inference about survival rates include
lead to a double-loop phase in which one or more
open capture–recapture models and band recovery
elements of the deliberative phase is revisited. These
models, both of which deal with unknown detection
double-loop learning phase steps typically occur on a
probabilities, and methods based on known deaths for
larger time-scale than the decision points of the iter-
some radio-telemetry studies (e.g. see Chapter 13 and
ative phase. So if the iterative phase entails a 1-year
Seber 1982; Lebreton et al. 1992, 2009; Williams et al.
time-step, for example, the double-loop steps might
2002). Inference about reproductive rates may focus
occur at irregular 5- or 10-year intervals, as needed.
on an overall reproductive rate (e.g. young per breed-
ing female) or the various components of reproductive
rate (e.g. the proportion of reproductive-age animals
19.4 Demographic methods that breed, nest success, brood survival) (e.g. Seber,
The demographic methods that form the subject mat- 1982; Williams et al. 2002; Skalski et al. 2010). Rates of
ter of this book are integral to the application of AM. movement in and out of populations are often estimat-
Management typically focuses on key state variables ed using capture–recapture or band recovery methods
of managed systems, as well as other variables asso- (e.g. Williams et al. 2002; Lebreton et al. 2009). Virtual-
ciated with management objectives. Dynamics of sys- ly all of the methods developed for estimation of vital
tem state variables are determined by system vital rates can also be used to draw inferences about the
rates, and management actions are usually designed effects of management actions and environmental and
to influence dynamics by affecting these vital rates. other variables on these rates.
AM typically entails estimation of system state vari- Population projection models used to predict con-
ables, the vital rates that drive them, and the rela- sequences of management actions on population size
tionships between vital rates and both management are also well developed. The most commonly used
actions and influential environmental variables. These approaches are based on discrete age or stage class-
estimates are then used to populate models of system es (Chapters 9 and 11 and reviews by Caswell 2001;
dynamics in order to project consequences of different Tuljapurkar and Caswell 2012), but integral models
management actions. Thus, models for estimation of using continuous variation among individuals in traits
influencing vital rates are also available (Chapter 10 and colonisation. When richness is estimated using
and Rees and Ellner 2009). For both of these classes capture–recapture thinking, these rate parameters can
of model, dynamics usually follow first-order Markov be estimated using a community-level version of
processes, providing access to the variety of decision Pollock’s (1982) robust design (Nichols et al. 1998;
algorithms developed for such problems (e.g. Bellman Hines et al. 1999). Rate of local colonisation can be
1957; Puterman 1994). defined in multiple ways, depending on whether the
members of the species pool of a sample location are
known a priori. If so, then a reasonable approach is to
19.4.2 Occupancy define colonisation as the probability that a member
of the species pool not present at the focal location at
Another state variable used in single-species manage-
time t is present at time t+1 (see Nichols et al. 1998;
ment is occupancy, or the proportion of area occu-
Boulinier et al. 2001). When occupancy approaches are
pied by a species. Large areas are often subdivided
used, the most general models permit estimation of
into patches the size of individual animal territories
species-specific rates of local extinction and colonisa-
or home ranges, but sample unit size can be more
tion using multiseason occupancy models (MacKenzie
arbitrary. The state variable is then defined as the pro-
et al. 2018).
portion of those designated units that is occupied by
A natural way to model dynamics of species rich-
the focal species, recognising that species may occupy
ness at a focal site, when the species pool associated
a site yet go undetected. This state variable is thus not
with the site is known, is to consider the number of
equivalent to abundance, although the two are related.
pool species in each of two states, present and not
Inference about occupancy in the face of nonde-
present at the focal site at time t. Rates of local colonisa-
tection is possible using a basic approach that has
tion and extinction are then used to project the system
been extended to a variety of useful applications
state through time. These vital rates can be modelled
(MacKenzie et al. 2018). Dynamics are driven by the
as functions of management actions, habitat character-
vital rates of unit-level extinction and colonisation.
istics, and weather variables in order to produce time-
Local extinction and colonisation, and the relationship
specific projections. The first-order Markovian nature
between these rates and management actions and envi-
of the system dynamics model (Boulinier et al. 2001)
ronmental variables, can be estimated using multisea-
is again preadapted for use with decision algorithms.
son occupancy models developed for these purposes.
This same basic approach can also be applied to sys-
These same multiseason models provide the frame-
tems in which the species pool is not known but must
work for projecting occupancy dynamics through time
be estimated, for example using the data augmen-
(e.g. Yackulic et al. 2014). These dynamic occupancy
tation approach of Royle et al. (2007; see Kery and
models are also first-order Markov processes and thus
Royle 2008).
amenable to use with many decision algorithms.
19.4.4 Landcover vegetative states

19.4.3 Species richness
Demographic performance and population growth of
Some community-level studies focus on species rich- many species are dependent on habitat (e.g. Breininger
ness for specific taxonomic groups. Richness is defined et al. 2009). When habitat is dynamic, projection of
as simply the number of species inhabiting an area. focal species dynamics often requires joint models that
Because species are not always detected at a loca- include the managed species and associated habitat
tion when present, inference methods must deal with (e.g. Johnson et al. 2011; Miller et al. 2012). In such sit-
unknown detection probabilities. Richness estimation uations, habitat can be categorised into discrete states,
is often based on capture–recapture thinking, with a such that the state variable for a managed area is the
variety of specific estimators now available (Williams fraction of patches or habitat units in each of the dif-
et al. 2002; Chao et al. 2014). Occupancy modelling can ferent habitat classes. In some situations where ground
also be used in either of two forms, one dealing with assessments are made, the state of each habitat unit can
species data from multiple sites and the other deal- be made with no error or need to deal with nonde-
ing with data from only one site (Zipkin et al. 2009; tection or misclassification. When state assignment is
MacKenzie et al. 2018). based on aerial or satellite imagery, it may be necessary
Vital rates associated with dynamics of species to deal with misclassification, which can be accom-
richness are local rates of species-level extinction plished using methods similar to those developed in
multistate capture–recapture modelling (Pradel 2005; harvested in the United States. In the 1960s and ear-
Veran et al. 2012). ly 1970s, divergent views among stakeholders about
Habitat state dynamics can be modelled using mat- the effects of hunting on duck populations led to
rices of state transition probabilities (e.g. Breininger annual arguments about appropriate hunting regula-
et al. 2010; Johnson et al. 2011). Transition proba- tions (Nichols 2000). In an effort to reduce this uncer-
bilities can be estimated directly using multinomial tainty and associated contentious debate, FWS scientist
modelling in the absence of state uncertainty. When F. A. Johnson led an effort to develop an AM pro-
state assignment is based on imagery characterised by gramme for the setting of annual hunting regulations
uncertainty, capture–recapture thinking can be used for mid-continent mallard ducks. This programme was
to estimate transition probabilities using data from formally adopted by FWS in 1995 and has been in use
multiple seasons (Veran et al. 2012). since that time (see Johnson et al. 1993, 1997, 2015;
The state variable is a vector consisting of the Nichols et al. 1995a, 2007, 2015; Williams et al. 1996; US
number of patches in each habitat state, for all k Fish and Wildlife Service 2018).
habitat classes. The associated transition matrices are The established objective of harvest management
of dimension k × k and are used to project the state for mid-continent mallards is to maximise cumulative
variable through time. Transition probabilities may harvest over the long term, requiring persistence of
be modelled as functions of management actions and a viable population into the future. This objective is
environmental conditions, or different matrices may constrained to avoid regulations expected to result in
be constructed for each potential management action. a population size below a specified numerical goal
This kind of modelling is Markovian and amenable to of 8,500,000 birds. Based on this constraint, the value
use with decision algorithms. of harvest decreases proportionally as the difference
between the goal and expected population size increas-
es, for expected population sizes less than the goal.
19.5 Case studies
Management actions include four discrete options
A large part of the motivation for this book is to defined by length of the hunting season and daily bag
demonstrate the importance of demographic methods limits: closed season, restrictive, moderate, and lib-
across a variety of taxa and types of questions. The AM eral regulations; with season length and daily limits
approach described in this chapter is similarly general increasing from restrictive to liberal packages (US Fish
and applicable to a variety of taxa and kinds of man- and Wildlife Service 2018).
agement problems. As stated, AM is especially applica- The relevance of demographic analyses is clear in
ble for recurrent management problems characterised the modelling and monitoring components of this AM
by uncertainty, especially uncertainty associated with programme. The basic model projecting the mallard
system responses to management. AM is thus applica- population through time is:
ble to a very large class of ecological and resource prob-
lems. This section includes overviews of three different
problems conducive to use of AM, involving verte-
brates, plants, and disease organisms. The vertebrate Nt+1 = Nt mSt,AM + 1 − m St,AF + Rt
example is by far the most detailed, describing the
longest-running formal AM programme in the world. St, JM ϕsum
F
St, JF + (19.3)
An overview is provided for the vegetation manage- ϕsum
M
ment example, which corresponds to a relatively new
AM programme. The disease example is hypothetical
and does not correspond to any real-world example where
that I know of. I simply provide a sketch of what Nt = breeding population size in spring of year t,
an AM programme for disease management should m = proportion of males in the breeding population,
look like. St,AM , St,AF , St,JF , St,JM = annual survival rates (year t)
for adult males, adult females, juvenile females,
and juvenile males, respectively,
19.5.1 Mallard (Anas platyrhynchos) duck
Rt = reproductive rate for year t, defined as autumn age
harvest in North America ratio (juveniles per adult) of females,
The US Fish and Wildlife Service (FWS) is charged with ϕsum
F /ϕ sum = ratio of female to male summer survival
M
setting annual hunting regulations for migratory birds rate.
If we choose to focus on adult females and males, Reproductive rate is then modelled as either strong-
separately, we can write eqn. (19.1) as a matrix projec- ly or weakly density dependent, representing two
tion model: extremes in the ability of a population to respond to
hunting via increased reproduction. The general model
Nt+1,AM SAM RSJM ϕsumF /ϕM
sum Nt,AM defining reproductive rate is:
=
Nt+1,AF 0 SAF + RSJF Nt,AF
(19.4) RH H H H
t = β0 + β1 Pt − β2 Nt , (19.7)
The stakeholder controversy over the effects of hunt- where

ing focused on the demographic rates of survival and RHt = reproductive rate under the hypothesis of weak
reproduction. The basic model structure of eqns. (19.3) (H = W) or strong (H = S) density dependence,
and (19.4) was thus used to create four different mod- Pt = number of ponds (in millions) in prairie Canada
els defined by two different survival functions and two breeding grounds in May of year t,
different reproductive rate functions (Johnson et al. β0H = intercept parameter under hypothesis H,
1997; US Fish and Wildlife Service 2018). The two β1H = slope parameter for the relationship between
survival functions are termed the compensatory and reproductive rate and ponds (in millions) under
additive mortality hypotheses (Anderson and Burn- hypothesis H,
ham 1976; Cooch et al. 2014). Survival rate under the β2H = slope parameter for the relationship between
additive mortality hypothesis is written as: reproductive rate and mallard breeding popula-
tion size (in millions) under hypothesis H.
St,sex,age = SA
0,sex,age 1 − Kt,sex,age , (19.5) Reproductive rate for the mallard population is esti-
whereas under the compensatory mortality hypothe- mated via combined use of banding and recovery
sis, survival is: data and data from harvest surveys based on hunter
( ) questionnaire and parts collection surveys (Martin and
SC0,sex,age if Kt,sex,age ≤ 1 − SC0,sex,age Carney 1977). Specifically, reproductive rate (autumn
St,sex,age = , ,
1 − Kt,sex,age if Kt,sex,age > 1 − SC0,sex,age age ratio of females) is estimated as:
(19.6)
Ĥt,J /Ĥt,A
R̂t = , (19.8)
where ĥt,J /ĥt,A
SH
0,sex, age = annual sex- and age-specific survival rate
in the absence of hunting under hypothesis H where
(additive, A, or compensatory, C), Ĥt,age = estimated total harvest for year t and age =
Kt,sex, age = sex- and age-specific hunting mortality rate J (juvenile) or A (adult), based on the FWS harvest
for year t. survey (Martin and Carney 1977),
ĥt,age = estimated harvest rate for year t and age = J
Hunting mortality always decreases overall sur- (juvenile) or A (adult), based on banding and
vival under the additive mortality hypothesis. How- recovery data.
ever, under the compensatory mortality hypothesis,
changes in hunting mortality rates over a specified Equation (19.8) can be viewed as a ratio of Lincoln–
range cause no such changes in overall survival. Petersen estimates (e.g. see Lincoln 1930; Seber 1982)
Estimation of annual survival rates, survival rates in for age-specific autumn abundance. For example, the
the absence of hunting, and hunting mortality rates is number of adults present in the autumn of year t can be
based on large-scale banding of ducks on the breed- estimated as the ratio of number of adults in the harvest
ing grounds just prior to the hunting season and to the probability that an autumn bird became part of
reported recoveries of bands during the subsequent that harvest, that is, Ĥt,A /ĥt,A .
hunting seasons (e.g. see Anderson 1975; Brownie Because pond numbers are such important determi-
et al. 1985). These models allow for the nonreport- nants of mallard reproductive rate, they are treated as
ing of ducks that are shot and retrieved by hunters. a state variable as well. Specifically, ponds are mod-
The specified relationships between annual survival elled as:
and both hunting mortality rate and survival in the Pt+1 = 2.2127 + 03420Pt + εt , (19.9)
absence of hunting are estimated directly using either
likelihood or Markov Chain Monte Carlo (MCMC) where εt is distributed as normal (0, 1.2567). Using the
approaches. estimates of reproductive rate and pond numbers, the
coefficients of the relationship of eqn. (19.7) were esti- and model weights are modified, increasing for models
mated via maximum likelihood. The parameters of the that predict well and decreasing for those predicting
two density-dependent hypotheses were then based on poorly. The next season’s regulations are then select-
these estimates and their associated variances. ed with the new model weights and estimated system
The two different hypotheses about survival state, and the process proceeds.
rates, and the two about reproductive rates, were The AM process developed by FWS for mid-
combined in pairs to yield four different models continent mallards has been a major success, produc-
intended to capture the uncertainty about mallard ing optimal decisions that are defensible, objective,
population responses to hunting. Let Sa denote the and transparent. Learning has progressed, with model
additive mortality hypothesis (eqn. 19.5), Sc denote SaRw achieving twice the weight of ScRw and weights
the compensatory mortality hypothesis (eqn. 19.6), Rs of the other two models approaching 0 (Figure 19.2).
denote strongly density-dependent reproductive rate These changes in model weights have produced
(eqn. 19.7), and Rw indicate weakly density-dependent corresponding changes in the optimal state-specific
reproductive rate. The resulting four population- regulations (Johnson 2011).
dynamic models are thus SaRs (additive mortality, The central roles of demographic methods in this
strongly density-dependent reproduction), SaRw process are clear. Estimates of survival rates and repro-
(additive mortality, weakly density-dependent reproductive rates are key components of the underlying
duction), ScRs (compensatory mortality, strongly population projection models on which optimal deci-
density-dependent reproduction), and ScRw (com- sions are based. Relationships between these rates and
pensatory mortality, weakly density-dependent both management actions and environmental variables
reproduction). This AM programme of harvest man- are key sources of uncertainty that are reduced dur-
agement was formally adopted for mid-continent ing the management process. Estimates of the key state
mallards in 1995 and has been ongoing ever since. In variable, population size, are essential for informing
1995, all four models were given equal weights of 0.25. state-dependent decisions and for learning about pop-
Multiple FWS monitoring programmes support this ulation responses to management.
harvest management programme for mid-continent
mallards. Breeding population size and pond numbers
19.5.2 Management of vegetation states for
are the two system state variables and are estimated via
an extensive aerial survey programme (Smith 1995). A
Florida scrub jays (Aphelocoma coerulescens)
large-scale banding programme is used to estimate sur- The Florida scrub jay is a threatened species under
vival, harvest, and hunting mortality rates, based on the US Endangered Species Act and generally viewed
band recovery models (Brownie et al. 1985) and band as a flagship species for scrub and flatwoods habitat
reporting rates (Nichols et al. 1995b). Finally, total age- found along central Florida’s Atlantic coast. Florida
and sex-specific harvest is estimated using the harvest scrub jays prefer a transitional stage of vegetative suc-
survey (Martin and Carney 1977). These various mon- cession in these systems. However, human influences
itoring programmes are summarised by Martin et al. have fragmented preferred habitat and reduced fire fre-
(1979) and Blohm (1989). quency, resulting in more habitat in older stages of
The iterative phase of mallard AM entails selecting succession that are not as productive for jays. Tradition-
a regulations package each year in the summer, just al habitat management for Florida scrub jays has used
after aerial survey monitoring results are available and prescribed burning to set back succession and provide
shortly before the hunting season begins. This selection more habitat suitable for jays.
uses stochastic dynamic programming (e.g. Williams Florida scrub jay is often categorised into four dif-
1996; Lubow 1997) and is based on programme objec- ferent habitat states: short (Sh: <1.2 m tall; open sand
tives, the four regulations packages, the models and between patches of oak scrub; burned within the last
their respective weights, and the estimated state of the few years), optimal (Op: 1.2–1.7 m tall; open sand
system (from monitoring). The optimal regulations are between patches of medium-height oak scrub), tall mix
implemented during the subsequent autumn–winter (Tm: short- or medium-height oaks with large patches
hunting season, driving the system to a new state of tall oak), and tall (Ta: no short–medium oaks, all tall
(estimated the following spring via monitoring) and with little open sand) (Breininger et al. 2009). Florida
yielding returns (harvest, also estimated via monitor- scrub jay survival and reproductive rates vary across
ing). The predictions of each of the four models are these habitat states such that birds with territories in
compared against the new estimate of population size, the optimal habitat state have higher survival and
reproductive rates than birds in the other three habi- probability ψtTmSh , for a patch going from state tall mix
tat types (Breininger et al. 1996, 2009; Breinenger and to short. Under the ‘do nothing’ management action
Carter 2003; Breinenger and Oddy 2004). Johnson et al. (action 1), this probability is very small, approaching
(2011) used these vital rates to compute habitat-specific 0. Under the burning action (action 2), it is larger, and
population growth rates, λH , with only λOp > 1. under mechanical cutting (action 3), this probability is
Two AM programmes have been proposed for larger still (i.e. ψtTmSh (1) < ψtTmSh (2) < ψtTmSh (3), where
Florida scrub jay populations in central Florida. The management action is denoted in parentheses).
first programme was developed by F. A. Johnson Estimation of transition probabilities is based on
and D. R. Breininger for Merritt Island National observed patch-level state transitions under the dif-
Wildlife Refuge. Johnson et al. (2011) carried out the ferent management actions. Estimation is accom-
deliberative phase of AM, focusing on maximising the plished via any software that provides maximum
amount of habitat in each management unit that is likelihood estimates for multinomial model parame-
in the optimal state. D. R. Breininger, M. J. Eaton, ters. Breinenger et al. (2009, 2010) and Johnson et al.
and F. A. Johnson have brought together a group of (2011) used multistate capture–recapture models (e.g.
state and private landowners and managers to initi- Lebreton et al. 2009) for which detection probabilities
ate another AM programme for the species in Brevard were set equal to 1 (because all habitat patches were
County, FL (Eaton et al. 2021). The programmes have observed and classified correctly to habitat type each
somewhat different approaches, but habitat manage- year).
ment is central to each programme. Here, I summarise Uncertainty about system response to management
the approach of Johnson et al. (2011), which focuses on actions could be acknowledged by developing multi-
maximising the amount of optimal habitat in a man- ple discrete models reflecting different transition prob-
′
agement unit over a long time horizon. This approach abilities, ψtHH , for at least some habitat pairs, HH’, and
is consistent with the purpose of this section to focus actions. The approach to learning (reducing uncertain-
on vegetation dynamics. ty) taken in the AM programme developed by Eaton
The objective of maximising the number of 10-ha et al. (2021) for Brevard County is to view this as an
patches (the approximate territory size of Florida scrub estimation problem, focusing on transition probability
jays) of optimal habitat in a management unit requires estimates, with more observations of state transitions
a set of actions designed to alter vegetation structure. leading to better estimates.
One possible set of three potential actions is: to do Monitoring in support of this AM problem entails
nothing (action 1), to implement prescribed burning of visits to each patch every year in order to assess habitat
a management unit (action 2), and to use mechanical type. These assessments are conducted on the ground
cutting of vegetation, followed by burning (action 3). (as opposed to use of remote imagery), hence assumed
The decision at each time-step (e.g. each year) for each to be made without error, although approaches to deal-
management unit is then selection of which action to ing with misclassification error are available (Veran
implement. et al. 2012). Monitoring of Florida scrub jay patch occu-
The basic model for habitat transitions among the pancy also occurs using modified occupancy models
four habitat states can be written as: (MacKenzie et al. 2018), but as the focus here is on veg-
   OpSh    etation dynamics, I will not describe these inference
nSh
t+1
ψtShSh ψt ψtTmSh ψtTaSh nSh
t
 Op   ShOp OpOp TmOp TaOp   Op  methods.
nt+1  ψt ψ ψ ψ   nt 
 Tm  =  ShTm tOpTm tTmTm tTaTm   Tm , Equation (19.10) represents a first-order Markov
nt+1  ψt ψt ψt ψt   nt 
OpTa
process, so optimal decisions can be obtained using
nTa ψtShTa ψt ψtTmTa ψtTaTa nTa
t+1 t such approaches as stochastic dynamic programming
(19.10) (Bellman 1957; Puterman 1994). Johnson et al. (2011)
provide examples of optimal decisions for different
where
action-specific matrices of transition probabilities. The
nH
t = number of patches in habitat state H at time t,
′ AM programme for Brevard County is in the early
ψtHH = probability that a patch in state H at time t is
stages of implementation and appears to be working
in state H’ at time t+1.
well. One important message emerging from the AM
The key to using eqn. (19.10) in AM is to recognise programmes for Florida scrub jays is the degree to
that each of the management actions leads to a different which we are constrained by available management
set of transition probabilities, that is, a different tran- actions. The nature of succession in scrub oak–flatlands
sition matrix. As an example, consider the transition habitat in central Florida, and the effects of the
identified management actions on the successional population models in which stages or states are defined
process, are such that maintaining substantial amounts with respect to presence of disease organisms.
of transitional (in our case, optimal) habitat is extreme- Two classes of demographic parameters drive the
ly difficult, leading to the need to at least consider other underlying dynamics of such multistate models,
possible management actions for influencing Florida state-specific transition probabilities, and state-specific
scrub jay populations. mortality rates. Estimation of both the state vari-
ables and the underlying rate parameters can be
accomplished using multiple list data at different
19.5.3 Disease control via AM points in time. For example, one list (data source)
Although AM has been recommended as a potentially might be individuals diagnosed while seeking treat-
useful approach to epidemiological programmes (e.g. ment in hospital emergency rooms. Another might
Bogich et al. 2013; Shea et al. 2014; Nichols 2019), be diagnoses of individuals tested when entering
I am not aware of formal adoption of AM for disease drug rehabilitation programmes. Such lists are char-
control. The argument in favour of AM is a simple acterised by detection probabilities that are <1 and
one. Agencies responsible for disease monitoring and that vary among lists, leading to the suggestion to use
control efforts tend to devote substantial efforts to dis- multistate capture–recapture models (Arnason 1973;
ease modelling and monitoring, often using rigorous Brownie et al. 1993; Schwarz et al. 1993; Lebreton
mathematical approaches. Yet the final decisions about et al. 2009) for inference about numbers of individuals
control actions such as vaccination strategies are often in each state and associated transition and mortali-
made by groups of experts studying estimates and ty rates (see Cooch et al. 2012; Nichols et al. 2017;
model results, integrating them in their heads, and Nichols 2019).
then making a decision. The methodological asym- Substantial uncertainty is likely to surround the
metry between modelling and estimation on the one effectiveness of potential management actions (e.g.
hand and decision-making on the other is striking. clean needles, law enforcement), leading to multiple
The recommendation is thus to confront uncertainty models that include large and small effects for each
in a rigorous manner, using the same rigour to make action. Specifically, actions would be expected to affect
decisions as is currently used in modelling and esti- the probability that an individual in state S at time
mation. Here I simply provide sketches of potential t will be in state I at time t+1, denoted as ψtSI . This
AM programmes at two different geographic scales transition probability or infection rate is a key driv-
in order to emphasise their applicability and discuss er of SIR models (Anderson and May 1991; Cooch
the kinds of demographic parameters useful to such et al. 2012), an important source of uncertainty, and a
endeavours. natural target of management actions. Compartmental
First consider an effort within a large city to reduce models are usually written as first-order Markov pro-
the number of people infected by a disease such as cesses and thus amenable to use of associated decision
hepatitis C that is spread through intravenous drug algorithms.
use with dirty or shared needles. A formal statement Now consider the epidemiological problem of man-
of objectives might specify a reduction below some aging disease problems such as measles at a country
threshold number of infections. Potential management level. The management objective might be to minimise
actions might include such divergent efforts as public the number of locations (e.g. counties) with infected
programmes to supply drug users with clean needles individuals, and actions might include various com-
or increased law enforcement targeting intravenous ponents of a nationwide vaccination strategy (specific
drug users and suppliers. disease strain(s) targeted, types of efforts to increase
Because objectives focus on numbers of persons number of vaccinations [mandatory for school atten-
infected with the disease, compartmental models (e.g. dance, kinds and intensities of advertising], possible
SIR models; Kermack and McKendrick 1927; Bailey geographic variation in these components to target
1975; Anderson and May 1991; Cooch et al. 2012) are predicted hotspots, etc.).
a natural choice. Such models categorise individuals An AM programme at this country-level scale might
of a focal population into disease states (e.g. suscep- use as a state variable the number or proportion of
tible [S], infected [I], and recovered or removed [R], counties at which the infection is present, suggesting
depending on the disease). Numbers of individuals the use of occupancy modelling (MacKenzie et al.
in each state are thus the state variables of this mod- 2018). The occupancy parameter estimates the pro-
elling. These models can be viewed as stage-structured portion of sites with at least one infected individual,
and the vital rates are probabilities of location-specific The fundamental equation of demography focuses
colonisation (disease moving into a county) or extinc- on population size and the vital rates that influence
tion (loss of disease from a county that contained it: rates of mortality, reproduction, and movement in
infected individuals during the prior time-step). The and out of the population. I have argued that for some
fundamental equation governing occupancy dynam- kinds of questions and management problems, differ-
ics is: ent state variables and vital rates should be considered.
Here, I claim that these different kinds of parame-
θt+1 = θt (1 − εt ) + (1 − θt ) γt , (19.11)
terisations can simply be viewed as different ways
where of integrating the fundamental demographic variables
θt = proportion of sites occupied (disease present) at and parameters.
time t, Consider occupancy, for example, defined as the
εt = probability that a site occupied at time t is not proportion of sites occupied by a focal species; that
occupied at time t+1, is, the proportion of sites at which abundance is >0.
γt = probability that a site not occupied at time t is The vital rates of local extinction and colonisation
occupied at time t+1. can be written as functions of the fundamental demo-
graphic variables and parameters. For example, local
County-specific vaccination actions and effective- extinction, εt , for a site must entail death or move-
ness might be expected to influence colonisation prob- ment out of the site by all individuals present at t
ability, and perhaps local extinction as well. Both and no movement into the site by individuals from
colonisation and extinction may also be affected by elsewhere. Dynamics of species richness are similar-
occupancy (disease presence) in surrounding counties, ly determined by demographic variables and param-
suggesting the use of autologistic occupancy models eters, this time of multiple species rather than just
(Yackulic et al. 2012; Eaton et al. 2014; MacKenzie one. The state variables of occupancy and richness
et al. 2018). These effects of management actions and thus focus on one tail of the abundance distribution
neighbourhood disease status on colonisation (dis- (abundance = 0 or >0), and associated vital rates are
ease spread) and extinction are key sources of uncer- direct functions of rates of death, reproduction, and
tainty that would be foci for an AM programme. movement of individual organisms. State dynamics of
Equation (19.11) is a first-order Markov process and vegetation can also be viewed as integrating demo-
thus amenable to the various decision algorithms graphic variables and parameters. In this case, the state
developed for such processes. variables are defined by the abundances of different
species. Transition probabilities for moving from one
state to another are functions of species-specific rates
19.6 Conclusions of death, reproduction, and movement (often at the
Development of demographic models and estimation seed stage). In the case of compartmental disease mod-
of model parameters are not inherently useful activi- els, we are modelling disease dynamics by focusing on
ties. Instead, they attain their value as parts of larger dynamics of host organisms classified by their disease
endeavours, typically either science or management. state. Transition probabilities of hosts from one dis-
The focus of this chapter has been on a form of manage- ease state to another reflect the survival, reproduction,
ment developed to deal especially with the common and movement of disease organisms within the host
problem of uncertainty, particularly uncertainty asso- population.
ciated with system responses to management actions. All of the examples in this chapter are thus based on
One theme of this chapter is that a natural approach demography. The state variables of occupancy, species
to modelling the effects of management is to focus on richness, and vegetative state can all be written in
vital rates. It would be possible to develop models terms of species abundances. Similarly, the vital rates
in which implementation of a management action at of local extinction, colonisation, and state transition can
time t would yield a direct change in the state vari- all be rewritten as functions of species-specific rates
able(s) at time t+1. I have argued that a more natural of death, reproduction, and movement. As is always
and mechanistic approach is to model the effects of the case for modelling, the choices of state variables
management actions on the vital rates underlying sys- and associated vital rates are very much determined
tem dynamics. This perspective is consistent with the by the objectives of the modelling effort (Nichols 2001).
theme of this book and its focus on demographic rates The use of integrated variables and parameters is sim-
and associated modelling. ply an approach to increasing efficiency and reducing
costs when modelling objectives can be met with inte- Anderson, R.M. and May, R.M. (1991). Infectious Diseases of
grated parameters. This general approach is consistent Humans: Dynamics and Control. Oxford University Press,
with Levins’ (1966) discussion of sufficient parameters, Oxford.
where ‘sufficient’ is defined relative to model use (see Arnason, A.N. (1973). The estimation of population size,
migration rates, and survival in a stratified population.
Nichols 2001; Williams et al. 2002).
Researches on Population Ecology, 15, 1–8.
Despite the importance of demography and associ-
Bailey, N. T. J. (1975). The Mathematical Theory of Infectious
ated methods to so many kinds of questions, classes of
Diseases, 2nd edition. Macmillan, London.
decision problems, and classes of organisms, substan- Bayes, T. and R. Price. (1763). An essay towards solving a
tial methodological variation exists among the differ- problem in the doctrine of chances. Philosophical Transac-
ent disciplines that investigate demography. Some of tions (1683–1775), 53, 370–418.
this variation is justified. Human demographers, for Bellman, R. (1957). Dynamic Programming. Princeton Univer-
example, have developed research methods that rely sity Press, Princeton, NJ.
on direct counts of individuals experiencing different Blohm, R.J. (1989). Introduction to harvest: understanding
fates over time. Scientists studying animal popula- surveys and season setting. Proceedings of the International
tions, however, have developed methods that account Waterfowl Symposium, 6, 118–133.
Bogich, T.L., Anthony, S.J., and Nichols, J.D. (2013).
for undercounts (usually), recognising that detection
Surveillance theory applied to virus detection: a case for
probabilities associated with virtually all animal sam-
targeted discovery. Future Virology, 8, 1201–1206.
pling methods are <1. Similarly, classification mistakes Boulinier, T., Nichols, J.D., Hines, J.E. Sauer, J.R., Flather,
are sometimes made, with species being misidentified C.H., and Pollock, K.H. (2001). Forest fragmentation and
or misclassified to various state variables (sex, age, bird community dynamics: inference at regional scales.
reproductive status). I believe that many of these meth- Ecology, 82, 1159–1169.
ods developed to deal with nondetection may be useful Breininger, D.R. and Carter, G.M. (2003). Territory quality
to human investigations in epidemiology and the social transitions and source-sink dynamics in a Florida scrub-
sciences, especially when dealing with ‘hidden’ human jay population. Ecological Applications, 13, 516–529.
populations (e.g., homeless, criminals) and new data Breininger, D. R. and Oddy, D.M. (2004). Do habitat pot-
ential, population density, and fires influence Florida
sources (e.g. health web site queries with symptom
scrub-jay source-sink dynamics? Ecological Applications,
lists indicative of specific diseases) (Nichols 2019). I
14, 1079–1089.
believe the emphasis of this book on the commonalities Breininger, D.R., Larson, V.L., Oddy, D.M., Smith, R.B., and
of demographic processes across taxa is a useful step to Barkaszi, M.J. (1996). Florida scrub-jay demography in
the integration of associated methods. different landscapes. Auk, 112, 617–625.
Breininger, D.R., Nichols, J.D., Carter, G.M., and Oddy,
D.M. (2009). Habitat-specific breeder survival of Florida
Acknowledgements
scrub-jays: inferences from multistate models. Ecology, 90,
Thanks to Marlene Gamelon and Roberto Salguero- 3180–3189.
Gomez for developing the concept of this book and Breininger, D.R., Nichols, J.D., Duncan, B.W., et al. (2010).
inviting me to contribute to it. Thanks to Mitch Eaton Multistate modeling of habitat dynamics: factors affect-
ing Florida scrub transition probabilities. Ecology, 91,
and an anonymous reviewer for providing useful com-
3354–3364.
ments on the initial draft manuscript. My ideas on this
Brownie, C., Anderson, D.R., Burnham, K.P., and Robson,
topic have been influenced by far more people than I
D.S. (1985). Statistical Inference from Band Recovery Data: A
can list here, but some major influences include Scott Handbook, 2nd edition. USFWS Resource Publication No.
Boomer, Mike Conroy, Evan Cooch, Paul Fackler, Fred 156. US Fish and Wildlife Service, Washington, DC.
Johnson, Bill Kendall, Mike Runge, and Ken Williams. Brownie, C., Hines, J. E., Nichols, J.D., Pollock, K.H., and
Hestbeck, J.B. (1993). Capture–recapture studies for mul-
tiple strata including non-Markovian transition probabil-
References ities. Biometrics, 49, 1173–1187.
Anderson, D.R. (1975). Population Ecology of the Mallard: V. Burnham, K.P. and Anderson, D.R. (2002). Model Selection
Temporal and Geographic Estimates of Survival, Recovery and and Multimodel Inference, 2nd edition. Springer, New York.
Harvest Rates. USFWS Resource Publication No. 125. US Caswell, H. (2001). Matrix Population Models. Sinauer Asso-
Fish and Wildlife Service, Washington, DC. ciates, Sunderland, MA.
Anderson, D.R. and Burnham, K.P. (1976). Population Ecol- Chao, A., Gotelli, N.J., Hsieh, T.C., et al. (2014). Rarefac-
ogy of the Mallard: VI. The Effect of Exploitation on Sur- tion and extrapolation with Hill numbers: a framework
vival. USFWS Resource Publication No. 128. US Fish and for sampling and estimation in species diversity studies.
Wildlife Service, Washington, DC. Ecological Monographs, 84, 45–67.
Clemen, R.T. and Reilly, T. (2001). Making Hard Decisions with Lebreton, J.-D., Nichols, J.D., Barker, R., Pradel, R., and
Decision Tools. Duxbury Press, Pacific Grove, CA. Spendelow, J. (2009). Modeling individual animal
Cooch, E.G., Conn, P.B., Ellner, S.P., Dobson, A.P., and histories with multistate capture–recapture models.
Pollock, K.H. (2012). Disease dynamics in wild popu- Advances in Ecological Research, 41, 87–173.
lations: modeling and estimation: a review. Journal of Levins, R. (1966). The strategy of model building in popula-
Ornithology, 152, 485–509. tion biology. American Scientist, 54, 421–431.
Cooch, E.G., Guillemain, M., Boomer, G.S., Lebreton, J.-D., Lincoln, F.C. (1930). Calculating Waterfowl Abundance on the
and Nichols, J.D. (2014). The effects of harvest on water- Basis of Banding Returns. Circular Number 118. US Depart-
fowl populations. Wildfowl Special Issue, 4, 220–276. ment of Agriculture, Washington, DC.
Eaton, M.J., Hughes, P.T., Hines, J.E., and Nichols, J.D. Lubow, B. (1997). Adaptive Stochastic Dynamic Programming
(2014). Testing metapopulation concepts: effects of patch (ASDP) Supplement to SDP User’s Guide. Version 2.0. Col-
characteristics and neighborhood on occupancy dynamics orado Cooperative Fish and Wildlife Research Unit, Col-
of an endangered lagomorph. Oikos, 123, 662–676. orado State University, Fort Collins, CO.
Eaton, M.J., D.R. Breininger, J.D. Nichols, et al. (2021). MacKenzie, D.I., Nichols, J.D., Royle, J.A., Pollock, K.H., Bai-
Integrated hierarchical models to inform management of ley, L.A. and Hines, J.E. (2018). Occupancy Modeling and
transitional habitat and the recovery of a habitat specialist. Estimation, 2nd edition. Academic Press, San Diego, CA.
Ecosphere, 12(1), e03306. Martin, E.M. and Carney, S.M. (1977). Population Ecology
Hines, J.E., Boulinier, T., Nichols, J.D., Sauer, J.R., and of the Mallard: IV. A Review of Duck Hunting Regulations,
Pollock, K.H. (1999). COMDYN: software to study Activity and Success, with Special Reference to the Mal-
the dynamics of animal communities using a capture– lard. USFWS Resource Publication No. 130. US Fish and
recapture approach. Bird Study, 46 (Suppl.), S209–S217. Wildlife Service, Washington, DC.
Johnson, F.A. (2011). Learning and adaptation in the man- Martin, F.W., Pospahala, R.S., and Nichols, J.D. (1979).
agement of waterfowl harvests. Journal of Environmental Assessment and population management of North Amer-
Management, 92, 1385–1394. ican migratory birds. In J. Cairns, G.P. Patil, and W.E.
Johnson, F.A., Williams, B.K., Nichols, J.D., et al. (1993). Waters (eds.) Environmental Biomonitoring, Assessment, Pre-
Developing an adaptive management strategy for har- diction, and Management. Certain Case Studies and Relat-
vesting waterfowl in North America. Transactions of the ed Quantitative Issues. Statistical Ecology Vol. Sll, pp.
North American Wildlife and Natural Resources Conference, 187–239. International Co-operative Publishing House,
58, 565–583. Fairland, MD.
Johnson, F.A., Moore, C.T., Kendall, W.L., et al. (1997) Uncer- Martin J., Runge, M.C., Nichols, J.D., Lubow, B.C. and
tainty and the management of mallard harvests. Journal of Kendall, W.L. (2009). Structured decision making as a
Wildlife Management, 61, 202–216. conceptual framework to identify thresholds for con-
Johnson, F.A., Breininger, D.R., Duncan, B.W., Nichols, servation and management. Ecological Applications, 19,
J.D., Runge, M.R., and Williams, B.K. (2011). A Markov 1079–1090.
decision process for managing habitat for Florida scrub- Miller, D.A.W., Brehme, C.S., Hines, J.E., Nichols, J.D. and
jays. Journal of Fisheries and Wildlife Management, 2, Fisher, R.N. (2012). Joint estimation of habitat dynam-
234–246. ics and species interactions; disturbance reduces co-
Johnson, F.A., Boomer, G.S., Williams, B.K. Nichols, J.D., and occurrence of non-native predators with an endangered
Case, D.J. (2015). Multi-level learning in the adaptive man- toad. Journal of Animal Ecology, 81, 1288–1297.
agement of waterfowl harvests: 20 years and counting. Nichols, J.D. (2000). Evolution of harvest management
Wildlife Society Bulletin, 39, 9–19. for North American waterfowl: selective pressures and
Kermack, W.O. and McKendrick, A.G. (1927). A contribution preadaptations for adaptive harvest management. Trans-
to the mathematical theory of epidemics. Proceedings of the actions of the North American Wildlife and Natural Resources
Royal Society of London, Series A, 115, 700–721. Conference, 65, 65–77.
Kery, M. and Royle, J.A. (2008). Hierarchical Bayes esti- Nichols, J.D. (2001). Using models in the conduct of science
mation of species richness and occupancy in spatial- and management of natural resources. In T.M. Shenk and
ly replicated surveys. Journal of Applied Ecology, 45, A.B. Franklin (eds). Modeling in Natural Resource Manage-
589–598. ment: Development, Interpretation and Application, pp. 11–34.
Lancia, R.A., Kendall, W.L., Pollock, K.H., and Nichols, J.D. Island Press, Washington, DC.
(2005). Estimating the number of animals in wildlife pop- Nichols, J.D. (2012). Evidence, models, conservation pro-
ulations. In C.E. Braun (ed.) Research and Management grams and limits to management. Animal Conservation, 15,
Techniques for Wildlife and Habitats, pp. 106–153. The 331–333.
Wildlife Society, Bethesda, MD. Nichols, J.D. (2019). Confronting uncertainty: contributions
Lebreton, J.-D., Burnham, K.P., Clobert, J., and Anderson, of the wildlife profession to the broader scientific commu-
D.R. (1992). Modeling survival and testing biological nity. Journal of Wildlife Management, 83, 519–533.
hypotheses using marked animals: a unified approach Nichols, J.D. and Williams, B.K. (2013). Adaptive manage-
with case studies. Ecological Monographs, 62, 67–118. ment. In A.H El-Shaarawi and W.W. Piegorsch (eds).
Encyclopedia of Environmetrics, 2nd edition, p.28. John of information: learning via intervention in epidemiology.
Wiley, New York. PLOS Biology, 12(10), e1001970.
Nichols, J.D., Boulinier, T., Hines, J.E., Pollock, K.H., and Skalski, J.R., Ryding, K.E., and Millspaugh, J.J. (2010).
Sauer, J.R. (1998). Estimating rates of local extinction, col- Wildlife Demography: Analysis of Sex, Age and Count Data.
onization and turnover in animal communities. Ecological Elsevier/Academic Press, New York.
Applications, 8, 1213–1225. Smith, G.W. (1995). A Critical Review of the Aerial and
Nichols, J.D., Johnson, F.A., and Williams, B.K. (1995a). Ground Surveys of Breeding Waterfowl in North America.
Managing North American waterfowl in the face of Biological Scientific Report 5. US Department of the Inte-
uncertainty. Annual Review of Ecology and Systematics, 26, rior, Washington, DC.
177–199. Tolkien, J.R.R. (1954). The Fellowship of the Ring. George Allen
Nichols, J.D., Reynolds, R.E. Blohm, R.J., Trost, R.E., Hines, and Unwin, London.
J.E., and Bladen, J.P. (1995b). Geographic variation in band Tuljapurkar, S. and Caswell, H. (2012). Structured-Population
reporting rates of mallards based on reward banding. Models in Marine, Terrestrial, and Freshwater Systems. Chap-
Journal of Wildlife Management, 59, 697–708 man and Hall, New York.
Nichols, J.D., Runge, M.C., Johnson, F.A., and Williams, US Fish and Wildlife Service. (2018). Adaptive Harvest Man-
B.K. (2007). Adaptive harvest management of North agement: 2019 Hunting Season. US Department of the Inte-
American waterfowl populations: a brief history and rior, Washington, DC.
future prospects. Journal of Ornithology, 148 (Suppl. 2), Veran, S., Kleiner, K.J., Choquet, R., Collazo, J., and Nichols,
S343–S349. J.D. (2012). Modeling habitat dynamics accounting for
Nichols, J.D., Thomas, L., and Conn, P.B. (2009). Infer- possible misclassification. Landscape Ecology, 27, 943–956.
ences about landbird abundance from count data: recent Walters, C.J. (1986). Adaptive Management of Renewable
advances and future directions. In D.L. Thomson, E.G. Resources. MacMillan, New York.
Cooch, and M.J. Conroy (eds.) Modeling Demographic Walters, C.J. and Hilborn, R. (1976). Adaptive control of fish-
Processes in Marked Populations, pp. 201–235. Springer, ing systems. Canadian Journal of the Fisheries Research Board
New York. of Canada, 33, 145–159.
Nichols, J.D., Johnson, F.A., Williams, B.K., and Boomer, Walters, C.J. and Hilborn, R. (1978). Ecological optimization
G.S. (2015). On formally integrating science and policy: and adaptive management. Annual Review of Ecology and
walking the walk. Journal of Applied Ecology, 52, 539–543. Systematics, 9, 157–188.
Nichols, J.D., Hollman, T., and Grand, J.B. (2017). Mon- Williams, B.K. (1996). Adaptive optimization and the har-
itoring for the management of disease risk in animal vest of biological populations. Mathematical Biosciences,
translocation programmes. EcoHealth, 14, S156–S166. 136, 1–20.
Pollock, K.H. (1982). A capture–recapture design robust to Williams, B.K., Johnson, F.A., and Wilkins, K. (1996). Uncer-
unequal probability of capture. Journal of Wildlife Manage- tainty and the adaptive management of waterfowl har-
ment, 46, 752–757. vests. Journal of Wildlife Management, 60, 223–232.
Pradel, R. (2005). Multievent: an extension of multistate Williams, B.K., Nichols, J.D., and Conroy, M.J. (2002). Analy-
capture–recapture models to uncertain states. Biometrics, sis and Management of Animal Populations. Academic Press,
61, 442–447. San Diego, CA.
Puterman, M.L. (1994). Markov Decision Processes: Discrete Williams, B.K., Szaro, R.C., and Shapiro, C.D. (2007). Adap-
Stochastic Dynamic Programming. Wiley, New York. tive Management: US Department of the Interior Techni-
Rees, M. and Ellner, S.P. (2009). Integral projection mod- cal Guide. Adaptive Management Working Group. US
els for populations in temporally varying environments. Department of the Interior, Washington, DC.
Ecological Monographs, 79, 575–594. Yackulic, C.B., Reid, J., Davis, R., Hines, J.E., Nichols, J.D.,
Royle, J.A., Dorazio, R.M., and Link, W.A. (2007). Analysis of and Forsman, E. (2012). Neighborhood and habitat effects
multinomial models with unknown index using data aug- on vital rates: expansion of the barred owl in the Oregon
mentation. Journal of Computational and Graphical Statistics, Coast Ranges. Ecology, 93, 1953–1966.
16, 67–85. Yackulic, C.B., Reid, J., Nichols, J.D. Hines, J.E., Davis, R.,
Schwarz, C.J., Schweigert, J.F., and Arnason, A.N. (1993). and Forsman, E. (2014). The roles of competition and
Estimating migration rates using tag-recovery data. Bio- habitat in the dynamics of populations and species distri-
metrics, 49, 177–193. butions. Ecology, 95, 265–279.
Seber, G. A. F. (1982). The Estimation of Animal Abundance and Zipkin, E.F., DeWan, A., and Royle, J.A. (2009). Impacts
Related Parameters. MacMillan, New York. of forest fragmentation on species richness: hierarchical
Shea, K., Tildesley, M.J., Runge, M.C., Fonnesbeck, C.J., and approach to community modelling. Ecological Applications,
Ferrari, M.J. (2014). Adaptive management and the value 46, 815–822.
CHAPTER 20
Heritability, polymorphism, and

population dynamics: individual-based
eco-evolutionary simulations
Anna Kuparinen
20.1 Introduction denoted as evolutionary ecology. It was characteristic

of evolutionary ecology to study how ecological differ-
Life history traits, such as rate of somatic growth
ences in the environment caused selection and drove
and metrics of reproductive output, vary across
evolution of species. Conversely, adaptive differences
individual phenotypes. Consequently, the demo-
among populations could cause differences in demo-
graphic rates expressed at the population level depend
graphic properties of different populations of the same
on the phenotypic composition of the populations.
species. Most commonly, the focus was on the one-
For example, if the somatic growth rates of young
way feedbacks from ecology to evolution or vice versa
individuals are higher than those of older individuals,
(Hendry 2016).
a population dominated by young individuals shows
More recently, increased computational power and
higher biomass growth per unit time. On the other
numeric simulation approaches have enabled a broad-
hand, reproductive output of a population might
ening of evolutionary ecology from the study of
be promoted by the presence of large, experienced
one-way feedbacks to iterative dynamics and the
individuals that produce viable juveniles. Thus, the
equilibrium conditions in which the eco-evolutionary
trade-off between somatic growth and reproduction at
dynamics settle into a stable state or fluctuate about it.
the level of individuals manifests as differences in pop-
Of course, entirely stable states rarely exist in nature
ulation demographic rates, mediated by differences in
due to continuously changing environmental condi-
population age structure.
tions. Eco-evolutionary dynamics should therefore be
More generally, trade-offs are manifested at the lev-
seen as an ongoing continuum of feedbacks between
el of individuals, among phenotypic traits, such as life
ecological and evolutionary processes orchestrated by
history invariants (Charnov 1993), but they may also
selection arising from ecological conditions and modi-
translate to population-level differences in basic demo-
fying population evolution within genetic, physiologi-
graphic traits. The key is to understand that both the
cal, and spatiotemporal boundaries (Hendry 2016).
individual- and population-level interfaces of a biolog-
ical system affect each other and to identify their inter-
action pathways, such as age structure. Interactions
20.2 Heritability and selection
among individual traits and population dynamics are
not limited to ecological dynamics, as phenotypes can in the context of population dynamics
evolve on a decadal time-scale or across few genera- Most phenotypic traits such as age and size at repro-
tions (Reznick 1997). Such evolution within contempo- duction and growth rate are to some extent herita-
rary time-scales is likely to alter demographic rates and ble (Mousseau and Roff 1987), meaning that traits
the dynamics of populations (Hendry 2016). are partially coded by genes. The trait frequencies
Research that integrated the fundamentals of pop- in a population can, therefore, be modified through
ulation ecology and evolutionary biology was long selection on the coding genes. This may further lead
Anna Kuparinen, Heritability, polymorphism, and population dynamics: individual-based eco-evolutionary simulations.
to co-evolution of correlated traits even if they were trade-offs and correlations among life history traits.
not subjected to selection directly. The extent to Moreover, no restricting assumptions need to be made
which phenotypic variation is associated with genetic as would often be the case in analytically solvable
variation is typically quantified through heritability, models. Evolutionary changes in the population sim-
the ratio of genetic variance, and phenotypic variance ply depend on which individuals survive to maturity
(Roff 1992). and with whom they mate. The juveniles will to some
In theory, one can estimate evolutionary responses extent reflect the properties of their parents, and the
to selection by using the Breeder’s equation R = h2 S, new cohort will reflect the ways in which selection
where R is the response to selection, h2 is the heritabil- filtered the parental cohorts. The annual number of
ity, and S is the selection differential (i.e. the change survivors, individuals that died, and new juveniles
in the cohort trait value from birth to reproduction). will reflect demographic rates of survival, mortality,
This equation will quantify the expected response to and reproduction. Box 20.1 and Boxed Figure 20.1 pro-
selection in the presence of nonoverlapping genera- vide a generic illustration of how to construct such a
tions but also makes several restricting assumptions simulation approach in practice.
(Morrissey et al. 2010). For example, coupled with
traditional life table analyses, one can then approx- 20.3.1 Parental trait values coupled
imate how an evolutionary change in age at matu- with known heritability
rity alters a population’s net reproductive rate and
In the case of quantitative traits, the trait of an off-
the intrinsic rate of population growth. However, this
spring can be approximated based on the trait values
approach gives at its best only very rough approxi-
of its parents coupled with some random variability
mations. Firstly, in most populations generations do
typically drawn from a normal distribution with a
overlap. Secondly, the selection differential integrates
mean of zero and the standard deviation calibrated to
all the selection on the given trait from birth up to
yield the desired level of phenotypic versus genetic
reproduction, and, specifically, it cannot be estimat-
variability, that is, the desired level of heritability.
ed based on a single selective event such as elevat-
This approach has been used particularly in so-called
ed mortality at a certain age or stage. Thirdly, the
ecogenetic simulation models applied to predict how
selection differential will likely be affected by many
fishing-induced evolution shapes fish phenotypes
correlated traits: a trait generally favoured by selec-
and the dynamics of the population, particularly its
tion at one stage of life may be disadvantageous at
reproductive capacity and recovery ability (Dunlop
other life stages. The selection differential will joint-
2009; Enberg 2009). These modelling approaches gave
ly integrate the outcome of multiple selective events
important insights into broad-scale eco-evolutionary
across the lifetime of an individual (Roff 1992). Captur-
dynamics of natural populations and influenced
ing the complexity of eco-evolutionary dynamics thus
recommendations regarding fisheries management
requires extensive long-term experiments or modelling
practices (Jørgensen et al. 2007).
approaches with empirically based parameterisations.
However, from an eco-evolutionary theory perspec-
This is the case even if eco-evolutionary dynamics
tive, the above approach suffers from some limitations.
were studied only at the level of one population or
Assuming a constant heritability might overesti-
species. Involving community or ecosystem aspects
mate the true adaptive capacity of the population—
obviously makes things even more complicated. In
namely, in the absence of, for example, mutations
the following, this chapter focuses on an individual-
and immigration—heritability should decline in the
based eco-evolutionary simulation approach as a tool
presence of strong, directional selection, as selec-
to understand trait evolution in the context of popula-
tion removes less-fit phenotypes/genotypes from the
tion dynamics.
population. As heritability declines, trait evolution
slows down, and one expects that the population finds
20.3 Generic ways to model the a new adaptive optimum at which its resilience and
evolution of demographic rates persistence are maximised, given the prevailing condi-
tions (Chevin et al. 2010).
A straightforward and intuitively appealing way to
integrate phenotypic evolution into ecological dynam-
20.3.2 Dummy genes coding the traits
ics is to use individual-based simulation models
(Box 20.1; see also Chapter 12). In such models, growth,
in a classic Mendelian manner
survival, mating, and reproduction are described at the An alternative to fixing the heritability was presented
level of individuals, allowing for full consideration of by Savolainen et al. (2004), and this approach is based
H E R I TA B I L I T Y, P O LY M O R P H I S M , P O P U L AT I O N DY N A M I C S 331
Box 20.1 The basic concept of individual-based eco-evolutionary simulations.
The simulation tracks through time, from one time-step to juveniles. The individual can have one or several mates
the next one (e.g. day or year), and within each time–step a within the time step.
set of demographic and evolutionary processes are modelled (5) Juvenile genotype: The genotype of each individual juve-
at the level of each individual, going through all individuals nile depends on its parents. Genotype can be determined
in the population one by one. These processes are shown in based on the average of the parental trait values coupled
Boxed Figure 20.1 and involve the following: with some stochasticity or drawn from parental alleles.
(6) Juvenile phenotype: Once juvenile genotype is known, the
(1) Mortality: Based on a random number, an individual
juvenile is assigned with some environmental variation in
either survives or dies, and in the case of the latter, the
the phenotypic expression.
individual is simply removed from the population.
(7) Updated population: New juveniles will be added to the
(2) Growth: The surviving individual grows, depending on
population.
the growth model used. It can progress along a growth
curve or, in a discrete stage-based setting, shift from After going through steps 1–7 for each individual and
one life history stage to another. Again, a random num- their juveniles, the population will be updated with new
ber can be involved to make the process stochastic, and juveniles and time shifts to the next time step. The above
density-dependent restrictions can be applied. processes will be repeated for each time-step. Each of the
(3) Maturity: At some point of its life, an individual starts to ecological and evolutionary processes can be based on ran-
reproduce. It may exceed a simple size or age threshold, dom numbers drawn from appropriate distributions, given
or its possibility to reproduce may depend on the popula- the average rate and the desired level of stochasticity. The
tion or age class/stage density. The following steps apply choice of distributions, the scale of time, forms of densi-
to mature individuals only. ty dependence, models for growth, maturity, and fecundity,
(4) Number of juveniles: Depending on the individual’s as well as parameter values are species- and study-system
body size, age, or condition, it produces a number of specific.
The recipe for individual-based eco-evolutionary simulations
Ecological dynamics Evolutionary dynamics

Population at the Mortality Juvenile genotype Population at the end
beginning of the time- • Rate of natural mortality of the time-step
step • External sources • Parental genotypes
• Inheritance • Survivors
• Abundance • New juveniles
• Properties of the
individuals Growth Juvenile phenotype
• Along a growth curve or • Genotype
from one stage to • Environmental
another variation in gene
• Density dependence expression
Maturity
• Random or deterministic
• Based on a rule or
current stage
Reproduction
• Number of juveniles
• Random mating or mate
choise
Time
• Scale of the time step
Boxed Figure 20.1 Key processes affecting individual-based eco-evolutionary dynamics of a population. Examples of factors
that may affect each process are detailed in the boxes. In principle, simulations track trough time, and each process is simulated
for each individual, thus updating the population from the previous to the next time point.
on the idea that most quantitative traits are determined influences of the genes can be easily varied, and the
by small, additive effects of a large number of loci (Roff number of coding genes can be low or high (i.e. the
1992). model allows different genetic architectures of traits).
In this approach, a trait of interest is determined This will be demonstrated in the third case study
by a set number of loci, the alleles of which are cod- (section 20.4.3). Similarly, one can set a rate of muta-
ed by 0 and 1. Now, the sum across the allele values tion for all or some loci, if this seems relevant for the
can be translated into a realistic range of trait values, time-frame of the study or the system in question.
such that sums close to 0 are associated with the low-
est trait value and the highest sums with the highest 20.4 Case studies
trait value. Naturally, the direct trait value sum must
The objective of the present book is to present demo-
be coupled with some random variability to allow for
graphic methods that can be applied across different
environmental variation in phenotypes. This variabili-
types of study organisms. In the context of eco-
ty can be initially tuned to yield realistic heritabilities in
evolutionary dynamics, this applies not only to ani-
pristine or stable conditions. When disturbances such
mals and plants but also to traits that are coded in
as harvesting or changes in environmental drivers are
very different ways and, thus, can evolve very differ-
then induced to the simulated eco-evolutionary popu-
ently, with associated effects of co-evolving traits and
lation dynamics, the heritability can change similarly
ecological dynamics. To this end, all the chosen case
to the other population-level parameters.
studies utilise the approach detailed in section 20.3.2
It is notable that the number of loci considered in
and demonstrate its functioning for different types of
such simulations does not need to reflect the actu-
organisms and genetic architectures.
al number of genes/loci coding the trait. Instead, the
modelled loci are simply a representation of the genet-
20.4.1 Boreal forest adaptation to warming
ic basis of the trait in question. Therefore, the num-
ber of modelled loci is mainly just a computational
climate
question: There needs to be enough loci such that the Global warming has increased temperatures in bore-
trait can evolve fairly smoothly but not too many to al areas, and this warming is likely to continue in the
slow down the computation speed. Namely, in this future. In practice, forest trees living close to or at their
approach, genes of an offspring depend on the genes species ranges will experience improved growth condi-
of the parents and are passed via the principles of clas- tions not only due to increased temperature during the
sic Mendelian heritance. This involves going through growing season but also as a result of the prolonged
each locus one by one and randomly drawing the growth season. Namely, transplantation experiments
juvenile genes from the two alleles of both the par- have demonstrated that in the northernmost forests a
ents (Figure 20.1). While this may sound laborious, the tree’s individual fitness will increase as those trees are
obvious advantage is that the inherent stochasticity reared at slightly more southern latitudes than where
associated with passing genes to offspring is accounted their parental population originated from (Savolainen
for in the model. et al. 2007). This observation suggests that, as opposed
The flexibility that this approach offers is that the to the core of the species range, where individuals are
trait in question does not necessarily need to be coded typically adapted to their local conditions and suffer
by many loci or by their additive effects. The relative from transplantation northwards or southwards, at the
Each individual has X loci with two alleles in each, coded as 0 or 1
Empirical data on
0 1 the range of the trait value
→ Sum S of the allelic values
1 1
across the X loci codes
Trait value
1 0
the mean phenotypic trait
1 1
. . Figure 20.1 Schematic illustration of the
. . → Phenotypic plasticity ensured model, where a phenotypic trait is coded by
by adding random variation S + ε S+ε X loci and two alleles in each.
northern edge of the species range local adaptation simulations looked into how the populations adapted
does not occur, or it breaks down. This can be explained to the predicted change in the growth season duration
by gene flow from more southern populations that is across 100 years: from 170 to 220 days in the south end
not balanced by gene flow from the north, as the species of the cline and from 120 to 170 days in the north end
does not exist beyond its northern range boundary. of the cline (Savolainen et al. 2004). In between, the
Thus, northernmost forests are likely to experience change was linearly related to the location along the
some level of fitness benefit from more ‘southern-type’ cline, and the shift in the growth season duration also
conditions (Savolainen et al. 2007). occurred linearly across the 100 years.
One obstacle in this reasoning is that in boreal for- Genetic variability and gene flow from southern
est trees the growth cessation in the autumn is to populations (in the forms of pollen and seeds) as well
some extent genetically controlled (Savolainen et al. as phenotypic variability served as sources for such
2004). As the first frosts occur, a tree that has not seedlings, whose phenotypic growth season length
ceased its growth and developed winter hardening was longer than the local adaptive average. These
will be harmed, and the younger the tree, the higher seedlings were able to outcompete seedlings with
the potential for death. On the other hand, too ear- shorter growth season, but if the phenotypic growth
ly growth cessation is a disadvantage in competition season of a seedling exceeded that at the location at
among better growing conspecifics. Purely ecological the given year, it was considered killed by frost. Old
forest growth models do not take into account this established trees were assumed more cold tolerant than
evolutionary aspect of forest growth but assume that seedlings. Therefore, selection acting at the seedling
forests will be able to utilise their full growth capacity stage favoured longer growth season but had also a
given the prevailing conditions. knife-edge threshold. The final step in the evolution
This mismatch makes it interesting to explore how process was the establishment of a seedling into an
much genetic control of growth cessation might affect adult reproducing tree; this required that an estab-
tree population dynamics, compared to an assump- lished tree died first, which occurred at the annual
tion that growth cessation is able to track the changing probability of 2–4%. Once a previously established
climate without any delay. Answering this question adult died, ten seedling-stage individuals were ran-
requires coupling the phenotypic trait evolution to for- domly drawn from the population. Among these, the
est tree population dynamics. seedling with the latest growth cessation possible, giv-
In northern Fennoscandia, Scots pine (Pinus en the prevailing climate at the current time point, was
sylvestris) and silver birch (Betula pendula) are the dom- allowed to grow into an established adult stage.
inant forest trees. Following the approach detailed in The simulated population adaptation to increasing
section 20.3.2, Kuparinen et al. (2010) assumed that growth season length was much slower than the envi-
the timing of growth cessation is coded by ten loci ronmental change; that is, evolution lagged behind
with additive effects. The loci sum was coupled with the changing environment (Kuparinen et al. 2010)
phenotypic variability, to yield heritabilities similar to (Figure 20.2). Thus, forests were not able to utilise right
those estimated in Scots pine common garden experi- away their improved growth conditions, and forest
ments (Notivol et al. 2007). For birch the phenotypic growth predictions that omitted this evolutionary lim-
variability was assumed the same, in the absence of itation would therefore be vastly overestimated. The
similar estimates as for Scots pine. slow adaptation was not due to the lack of genetic vari-
Scots pine and silver birch differ substantially in ability but the slow rate at which the seedlings could
their maturation times and pollen and seed disper- establish themselves. As the second author, Professor
sal abilities. These ecological differences between the Outi Savolainen, interpreted the results: ‘The pattern
species affect their population renewal rates and, thus, is similar to academic life: young scientists can’t get
potentially modify the rate at which evolution towards positions because those are held by old established
later growth cessation could take place (Kuparinen professors until they retire’ (O. Savolainen, personal
et al. 2010. To investigate this, we constructed an eco- communication).
evolutionary simulation model following the princi-
ples described in Box 20.1. The model had two distinct
20.4.2 Co-evolution of correlated traits
developmental life history stages: seedling pool stage
and the established adult stage.
in Atlantic cod
The eco-evolutionary dynamics of ten popula- The previous example was simple in the sense that
tions along an 800-km cline were simulated. These it only looked at one evolving trait and how that
40 (a) by simply adding the number of loci and having certain

loci to code each trait. However, this has major points
of consideration: (1) computation time can increase
2%
substantially; and (2) one needs to make assumptions
30 3% about the correlations among the traits (or lack of cor-
relation). If strong well-established correlations exist,
4%
those can be utilised without the need to increase the
number of loci. For example, let’s assume we have
20
three traits, A, B, and C, of interest and we know how
they are correlated with each other. We can then just
simulate the loci coding trait A and then determine the
value of B and C based on the value of A. Thus, the
Lag behind the thermal growth period length (days)
10
traits co-evolve according to their correlation structure,
but only the loci coding trait A is simulated.
An alternative way, of course, would be to allow
0
overlap among the sets of loci coding each trait, but
0 20 40 60 80 100 this requires tuning the model such that the emerging
correlations match with the empirically detected ones.
Computation time would also increase as the num-
40 (b)
ber of simulated loci increases. Nonetheless, such an
approach is interesting and worth testing in the future.
In ectotherms, many key life history traits are cor-
30 related, and these correlations are often known as life
history invariants (Charnov 1993). In fishes, these cor-
relations are closely involved with the so-called von
Bertalanffy growth model (Figure 20.3), which tells at
20 a given age t the length of the fish L(t). The model is
based on three parameters: the asymptotic body length
10
Von Bertalanffy growth curve
0 L(t)=L∞ – (L∞ – L0)e–tk
0 20 40 60 80 100
L∞
Time (year)
Figure 20.2 Evolutionary lag behind the changing growth period

length for (a) Scots pine and (b) silver birch. The lag between the
L(t)
mean genotypic growth period length and the climatic optimum 66% of L∞
Length at maturity
(thermal growth period length) is shown by the y-axis. The lag is given
as a function of time during which the environment changes gradually
(x-axis). Results for different rates of adult tree mortality are indicated
by different line types. (Reproduced from Kuparinen et al. (2010) by
permission of Elsevier.)
affected the projected population dynamics. However,

t (age)
in reality, fitness is often the product of a few
important life history traits. The method presented Figure 20.3 Von Bertalanffy growth model and its life history
in section 20.3.2 and utilised throughout the three invariants (von Bertalanffy 1938; Charnov 1993). L(t): length at age t;
presented applications (sections 20.4.1–20.4.3) can be L∞ : asymptotic body length; L0 : length at age t=0; k: the rate at
straightforwardly extended to several evolving traits which an individual reaches its L∞ .
L∞ , the rate k at which L(t) approaches L∞ , and the

average length at t = 0. The two key invariants related
to L∞ are as follows: (1) k and L∞ are strongly and neg-
atively correlated; and (2) length at maturity is about
66% of L∞ (Charnov 1993; Jensen 1997).
By utilising these correlations, the evolution of fish
growth trajectory and fish maturation can be modelled
by simulating the loci that code L∞ and then predict-
ing the age and size at maturation and the growth
parameter k based on L∞ . While the length at matu- Figure 20.5 Atlantic cod (Gadus morhua L.). (Source: Wikipedia.)
rity (and thus the age at maturity) can be derived
straightforwardly from the value of L∞ , predicting k mortality due to fishing is likely to select for earli-
requires knowledge about the empirical relationship er maturity at smaller body size. These life history
between L∞ and k. In fishes this information can be changes, in turn, can affect fish population dynamics,
obtained by reading lengths-at-age of the fish from more specifically population productivity, resilience
scales or otoliths. Von Bertalanffy growth curves can to disturbances, and the ability to recover from low
then be fitted to individual growth trajectories and k abundances.
modelled as a function of L∞ . If the relationship is Atlantic cod is in many ways an iconic species to
particularly skewed, this can be fixed by taking a log- study fishing impacts, owing to several population col-
transformation of k, as shown in Figure 20.4. Now the lapses in the 1980s and 1990s in the Atlantic Ocean and
linear model can be fitted to describe the association of the Baltic Sea, followed by cod fishing closures in many
L∞ and k. For each value of L∞ , the value of k can be management units. Despite dramatic reductions in cod
predicted using the fitted model and coupled with ran- fishing pressure, in many areas such as the coasts of
dom variation, the magnitude of which should match Newfoundland and the Gulf of St Lawrence in Cana-
the residual variation in the model fit. da, cod has shown very little or no signs of recovery.
The above described modelling approach was While many reasons such as environmental conditions,
utilised to predict eco-evolutionary responses of ecosystem regime shifts, and changes in predator–prey
Atlantic cod (Gadus morhua L.) (Figure 20.5) to selec- dynamics can hinder recovery, the dramatic reduc-
tive fishing (Kuparinen and Hutchings 2012; Kupari- tions in cod size and age at maturity suggest that some
nen et al. 2014). As theory, empirical observations, and life history evolution might have occurred, and this
experiments demonstrate (e.g. Law 2007), increased may interact with the ecological dynamics of cod pop-
ulations. Namely, based on demographic population
parameters alone, fisheries scientists have for long pre-
dicted using their stock-assessment models that cod
140 should recovery rapidly after fishing pressure relaxes.
Indeed, simulated eco-evolutionary dynamics of cod
120 populations exposed to fishing suggest that cod L∞
might evolve to substantially smaller values within
just a few decades and under moderate fishing pres-
L∞ (cm)
100
sure (Kuparinen and Hutchings 2012). Not only the
average age and size of the reproducing individuals
80
is reduced but also the net reproductive rate of the
population R0 and per-capita population growth rate
60 r are affected, particularly at low population abun-
dances (Kuparinen et al. 2014). Similarly, after the
40 life history changes the populations could not rebuild
−2.5 −2.0 −1.5 −1.0 −0.5 their biomasses back to prefishing levels but only to
log(k) biomasses that were 10–20% lower. Thus, life histo-
ry evolution can also change the population carrying
Figure 20.4 Association between L∞ and k parameters of the von capacity, a fundamental ecological parameter regulat-
Bertalanffy growth models fitted to lengths-at-age read from 258 ing population dynamics (Kuparinen and Hutchings
Atlantic cod otoliths collected by Hardie (2007). 2012).
20.4.3 Evolution of a sexually polymorphic age females tend to mature late, similarly to the homozy-
at maturity in Atlantic salmon gotes for the one allele, but heterozygous males are
more similar to the early maturing homozygotes for
The previous two examples were based on the assump-
the other allele (Table 20.1). This genetic architecture
tion that quantitative traits, such as the length of the
makes the age at maturity in Atlantic salmon a poly-
growth season in trees (section 20.4.1) and asymptotic
morphic trait, with phenotypes being either early or
body length (L∞ ) in fishes (section 20.4.2), are coded
late reproducers. However, this polymorphism is part-
by a large number of loci with small, additive effects
ly masked by sex and can therefore be difficult to
(Roff 1992). While this might be the case for many
distinguish by eye from raw data. In fact, more specifi-
organisms and traits, more recent genomic discover-
cally one would say that the trait is sexually dimorphic
ies have demonstrated that the genetic architecture of
among heterozygotes.
a phenotypic trait can be more diverse.
Flexibility of the modelling approach described in
Atlantic salmon (Salmo salar L.) (Figure 20.6) is
section 20.3.2 becomes particularly convenient when
an anadromous fish species inhabiting rivers in the
modelling a specific genetic trait architecture such as
northern hemisphere. Many Atlantic salmon popula-
the one discovered for Atlantic salmon age at maturi-
tions have declined to vulnerable levels or even gone
ty. Instead of an arbitrary number of loci, salmon age at
extinct (WWF 2001). Due to their critical state, suc-
maturity can now be described by a single locus with
cessful reproduction is extremely important for the
two alleles. The trait expression is then sex-dependent.
future of many populations. One key trait involved
But in individual-based simulations, tracking a trait
in salmon reproduction is the age at which the indi-
like sex is simple. And, as usual, there is some environ-
vidual returns to its natal river to spawn. The longer
mental variance about the alleles coding the trait. In the
the salmon postpones maturity, the lower its odds of
case of Atlantic salmon maturation, gene expression is
reproducing before it dies. On the other hand, larg-
defined in a probabilistic manner: different genotypes
er older spawners produce much more and possibly
are associated with the probabilities for postponing
better quality juveniles, so postponing might also be
maturity or, inversely, becoming mature at the given
beneficial. This largely depends on prereproductive
age. Thus, for an individual at a given age t with a giv-
mortality and uncertainty about it. Consequently, the
en genotype and sex, one can simply draw a binomial
question of what triggers maturity in Atlantic salmon
random number to determine whether the individual
has puzzled scientists for decades.
postpones maturity or becomes mature at age t. Sim-
Recently, genomic analyses have discovered that
ulated in this manner, there will be some variability
salmon age at maturity is strongly controlled by one
about the expression of the same genotypes across a
single locus with sexually dimorphic expression; this
number of individuals.
one locus explains about 40% of variation in salmon
Compared to the traditional assumption that quanti-
age at maturity (Barson et al. 2015). In practice,
tative traits are coded by a large number of genes with
homozygotes for one allele tend to mature young and
small additive effect, the genetic coding of Atlantic
homozygotes for the other allele tend to mature old,
salmon age at maturity is quite different. Firstly, there
irrespective of the sex. But the age at maturity of
are much fewer genotypes; and, secondly, the sexual
the heterozygotes depends on the sex: heterozygote
dimorphism of the heterozygotes is likely to promote
genetic diversity in a population (Barson et al. 2015).
This is because even if selection strongly favoured
early or late maturity, heterozygote males or females
carry alleles promoting the opposite maturity sched-
ule. Therefore, it becomes very interesting to inves-
tigate how evolution driven by directional selection
differs between the two scenarios for trait control: (1)
one locus with sexually dimorphic expression; and (2)
multiple loci coding the trait additively without sex-
dependence.
The above question was addressed by systematic
simulations of the evolution of Atlantic salmon age
at maturity in the presence of fishing targeting old
(large) individuals (Kuparinen and Hutchings 2017).
Figure 20.6 Atlantic salmon (Salmo salar L.). (Source: Wikipedia.) These simulations revealed several interesting aspects
Table 20.1 Probabilities of postponing maturity beyond threshold ages for Vgll3TOP genotype × sex, as estimated by
Barson et al. (2015). These probabilities were utilised in salmon fishing simulations, in the model where the age at maturity
was coded by one locus having two alleles (0 and 1). (The table is reproduced from Kuparinen and Hutchings (2017).)
Sex Homozygote (11) Heterozygote (10 or 01) Homozygote (00)
2 SW → 3 SW 0.754 0.949 0.983

Female
1 SW → 2 SW 0.101 0.404 0.665
2 SW → 3 SW 0.266 0.277 0.835
Male
1 SW → 2 SW 0.058 0.061 0.467
about the interaction of the genetic trait architecture, This is opposite to the multi-locus scenario without
the nature of trait evolution, and its feedbacks to pop- sexual dimorphism, where all evolutionary trajectories
ulation dynamics. are very similar and, thus, indicative of the general
Among the simulated populations, single-locus con- outcome.
trol did not lead to large changes in average age at The second important aspect is that Atlantic salmon
maturity but vastly increased the variance among the populations that seem to respond differently to the
simulated populations (Figure 20.7). The pattern was same fishing pressure and selectivity need not differ
opposite to multi-locus control, as in this scenario fish- in any of their other genetic or ecological aspects. The
ing led to an expected decrease in age at maturity simulated populations in Figure 20.7 are identical, but
but also reduced variability among replicated popula- the single-locus sexually dimorphic control of age at
tions. In brief, multi-locus control led to the expected maturity drives them completely randomly in differ-
and predictable pattern, whereas single-locus control ent directions when exposed to fishing. This increased
made the evolutionary response to selection highly uncertainty within and among populations in terms
uncertain. Uncertainty was further manifested within of evolutionary dynamics also reflects their ecological
populations, as temporal variability in age at maturity dynamics: Both in the absence and presence of selective
was much higher in the presence of single-locus con- fishing, the per-capita population growth rate r varies
trol compared to multi-locus control (Kuparinen and more if age at maturity is under single-locus sexually
Hutchings 2017). Thus, one can conclude that sexual- dimorphic control. This metric (r) is the major correlate
ly dimorphic single-locus control of age at maturity of population recovery ability and, inversely, the risk
causes trait evolution to be disruptive and divergent. of extinction (Dulvy et al. 2004; Mace et al. 2008). Thus,
The former of these terms means increased variability instability in r may increase the uncertainty of popu-
among identical populations, while the latter describes lation persistence and recovery ability, which can be
increased temporal variability within each population. particularly problematic if the population is exposed
Further insights into the relative roles of single-locus to external environmental stochasticity in addition to
control and sexual dimorphism were recently provided its internal inherent stochasticity.
by Oomen al. (2020), as the same simulation design was In summary, the case of Atlantic salmon age at
repeated in the absence of dimorphism: the increased maturity suggests that population dynamics and
drift that underlies divergent and disruptive elements demography can be intimately linked with the genetic
of the trait evolution was virtually completely subject architecture of key life history traits and that this
to the single-locus control, as compared to multi-locus linkage and its ecological consequences are far from
control. intuitive.
From practical conservation and fisheries manage-
ment perspectives, the nature of this trait evolu-
tion has several consequences. Firstly, an Atlantic
20.5 Conclusions and future directions
salmon population might experience fishing-induced The above three examples (sections 20.4.1–20.4.3) illus-
evolution that cannot be inferred from phenotypic trate why it is important to consider evolutionary
data. As shown in Figure 20.7, fishing can either slight- processes as a part of demographic analyses. This
ly increase or decrease the age at maturity or leave is because evolution can change key demographic
it relatively unchanged. This results from the inher- parameters and population dynamics, and these, in
ent stochasticity among the evolutionary trajectories. turn, can feed back to selection driving evolution.
One simply cannot predict how a trajectory develops. In brief, ecology and evolution are just two sides of
Multi-locus Single-locus
2 SW
2 SW
F = 0.35
1 SW
1 SW
50 100 150 200 250 300 50 100 150 200 250 300
Age at maturity (sea winter)
2 SW
2 SW
F = 0.40
1 SW
1 SW
50 100 150 200 250 300 50 100 150 200 250 300
2 SW
2 SW
F = 0.45
1 SW
1 SW
50 100 150 200 250 300 50 100 150 200 250 300
Time (year)
Figure 20.7 Development of age at maturity in Atlantic salmon under two scenarios for the genetic architecture of the trait (multi- and
single-locus). Alternative fishing mortalities (F) are indicated on the right of each pair of graphs; the beginning and end of the fishing period are
indicated by dashed vertical lines. Each individual line represents one replicated simulation line (n = 50). (Reproduced from Kuparinen and
Hutchings (2017). Reproduced with permission from Royal Society Publishing.)
the same coin—the eco-evolutionary dynamics (see population dynamics and further demonstrated why
Chapter 16). this is important from the perspectives of ecological
The main objective of this book is to present demo- dynamics, conservation, and the management of
graphic methods that are applicable across the Tree of natural populations.
Life, and the present chapter did this in the context of As described in section 20.2, the method applied
eco-evolutionary trait dynamics. More specifically, the here is by no means the only way to couple evolu-
present chapter provided a recipe (Boxed Figure 20.1) tion with individual-based ecological dynamics, but it
to integrate the evolution of a phenotypic trait into has some advantages, being both flexible and allowing
for heritability to change in response to selection. Like- already for decades, practical applications to conser-
wise, the method has been applied to plants, animals, vation, management, and eco-evolutionary predictions
and different modes of inheritance. Even though the remain rare.
method has here been applied to the evolution of The range of demographic methods available to
fitness-related traits, it can also be used to describe neu- analyse and predict species dynamics is already wide
tral genetic diversity and drift, as well as how those (Chapter 5), but there are two clear directions for
are affected by ecological dynamics and directional future development: (1) Efficient integration of genom-
selection (Kuparinen et al. 2016). ic knowledge is necessary to better understand the
Rigorous simulation-based analyses of eco-evolu- nature and boundaries of phenotypic evolution as well
tionary population dynamics are still a very new as eco-evolutionary feedbacks to population dynam-
field in biological research. Nonetheless, the need for ics. (2) The consideration of species dynamics as a
such analyses becomes increasingly clear as new case ‘dimension’ of multispecies dynamics in an ecosys-
studies and empirical observations demonstrate the tem is likely to illuminate mechanisms underlying
interplay between contemporary evolution and ecolog- fundamental demographic parameters and population
ical dynamics. At the same time, the need for biological dynamical properties. Capturing complex biological
forecasts is increasing: projections for how human dynamics from genes to communities and ecosystems
impacts and climate change affect natural populations, is undoubtedly challenging, but the potential gains
particularly their resilience, recovery potential, and are also tremendous: discoveries of the mechanisms
macroecological dynamics, require mechanistic eco- behind the patterns, identification of the key pathways
evolutionary modelling tools (Chapter 19). Although that regulate the eco-evolutionary system dynamics,
exploitation, climate change, and species range shifts and improved ability to predict the destiny species
are anticipated to drive rapid evolutionary changes in face with environmental changes and anthropogenic
life histories, evidence of such change remains provi- disturbances.
sional (e.g. Gienapp et al. 2008; Audzijonyte et al. 2016).
Thus, evolution seems to be an integral component of
current and future ecological dynamics of species.
As illustrated by the Atlantic salmon case study References
(section 20.4.3), our understanding of the drivers Audzijonyte, A., Fulton, B., Haddon, M., et al. (2016) Trends
of eco-evolutionary population dynamics will like- and management implications of human-influenced life-
ly drastically change as new genomic discoveries history changes in marine ectotherms. Fish and Fisheries,
accumulate (e.g. Barson et al. 2015; Kuparinen and 17, 1005–1028.
Hutchings 2017). To date, we know little about the Barson, N.J., Aykanat, T., Hindar, K., et al. (2015) Sex-
genomic structures coding multiple correlated traits. dependent dominance at a single locus maintains varia-
Nonetheless, as those are discovered, we will be in a tion in age at maturity in Atlantic salmon. Nature, 528,
much better position to predict the boundaries of phe- 405–408.
Charnov, E.L. (1993) Life-history invariants. Oxford Univer-
notypic adaptation and how phenotypes might evolve
sity Press, Oxford.
under human-induced selection (see Chapter 1). In
Chevin, L.-M., Lande, R., and Mace, G.M. (2010) Adapta-
practice, much tighter collaboration among genomi-
tion, plasticity, and extinction in a changing environment:
cists, evolutionary biologists, and ecologists will be towards a predictive theory. PLOS Biology, 8, e1000357.
needed to synthesise and integrate research findings Dulvy, N.K., Ellis, J.R., Goodwin, N.B., Grant, A., Reynolds,
into an eco-evolutionary framework. J.D., and Jennings, S. (2004) Methods of assessing extinc-
On a final point, one aspect often lacking from tion risk in marine fishes. Fish and Fisheries, 5, 255–276.
these analyses is an ecosystem perspective. Basic Dunlop, E.S., Heino, M., and Dieckmann, U. (2009) Eco-
demographic parameters such as natural mortality are genetic modeling of contemporary life-history evolution.
often a product of predator–prey interactions. Like- Ecological Applications, 19, 1815–1834.
wise, fluctuations in species abundances often relate to Enberg, K., Jørgensen, C., Dunlop, E.S., Heino, M., and
Dieckmann, U. (2009) Implications of fisheries-induced
fluctuations in prey availability (Olsen et al. 2004) and
evolution for stock rebuilding and recovery. Evolutionary
environmental stochasticity mediated by the food web
Applications, 2, 394–414.
(Kuparinen et al. 2019), or are generated by phenotypic Gienapp, P., Teplitsky, C., Alho, J., Mills, J.A., and Mer-
changes in the focal or other species in the commu- ilä, J. (2008) Climate change and evolution: disentangling
nity (Wood et al. 2018). While ecological networks environmental and genetic responses. Molecular Ecology,
and their functioning have been studied theoretically 17, 167–178.
Hardie, D.C. (2007) The evolutionary ecology of Atlantic Morrissey, M.B., Kruuk, L.E.B., and Wilson, A.J. (2010) The
cod (Gadus morhua) in Canadian Arctic lakes. PhD thesis, danger of applying the Breeder’s equation in observation-
Dalhousie University, Halifax. al studies of natural populations. Journal of Evolutionary
Hendry A.P. (2016) Eco-evolutionary Dynamics, Princeton Biology, 23, 2277–2288.
University Press, Princeton. Mousseau, T.A. and Roff, D.A. (1987) Natural selection
Jensen., A.L. (1997) Origin of the relation between K and Linf and the heritability of fitness components. Heredity,
and synthesis of relations among life history parameters. 59,181–197.
Canadian Journal of Fisheries and Aquatic Sciences, 54, Notivol, E., Garcia-Gil, M.R., Alia, R., and Savolainen, O.
987–989. (2007) Genetic variation of growth rhythm traits in the lim-
Jørgensen, C., Enberg, K., Dunlop, E.S., et al. (2007) Manag- its of a latitudinal cline in Scots pine. Canadian Journal of
ing evolving fish stocks. Science, 318, 1247–1248. Forest Research, 37, 540–551.
Kuparinen, A. and Hutchings, J.A. (2012) Consequences of Olsen, E.M., Heino, M. Lilly, G.R., et al. (2004) Maturation
fisheries-induced evolution for population productivity trends indicative of rapid evolution preceded the collapse
and recovery potential. Proceedings of the Royal Society of northern cod. Nature, 428, 932–935.
B, 279, 2571–2579. Oomen, R.A., Kuparinen, A., and Hutchings, J.A. (2020)
Kuparinen, A. and Hutchings, J.A. (2017) Genetic architec- Consequences of single-locus and tightly linked genomic
ture of age at maturity can generate divergent and dis- architectures for evolutionary responses to environmental
ruptive harvest-induced evolution. Philosophical Trans- change. bioRxiv, BIORXIV/2020/928770.
actions of the Royal Society B, 372, 20160035. Reznick, D.N., Shaw, F.H., Rodd, F.H., and Shaw, R.G. (1997)
Kuparinen, A., Savolainen, O., and Schurr, F.M. (2010) Evaluation of the rate of evolution in natural populations
Increased mortality can promote evolutionary adaptation of guppies (Poecilia reticulata). Science, 275, 1934–1937.
of forest trees to climate change. Forest Ecology and Man- Roff, D.A. (1992) Life history evolution. Sinauer Associates,
agement, 259, 1003–1008. Sunderland, MA.
Kuparinen, A., Stenseth, N.C., and Hutchings, J.A. (2014) Savolainen, O., Bokma, F., Garcia-Gil, R., Komulainen, P.,
Fundamental population–productivity relationships and Repo, T. (2004) Genetic variation in cessation of
can be modified through density-dependent feedbacks growth and frost hardiness and consequences for adapta-
of life-history evolution. Evolutionary Applications, 7, tion of Pinus sylvestris to climatic changes. Forest Ecology
1218–1225. and Management, 197, 79–89.
Kuparinen, A., Hutchings, J.A., and Waples, R. (2016) Savolainen, O., Pyhäjärvi, T., and Knürr, T. (2007) Gene
Harvest-induced evolution and effective population size. flow and local adaptation in trees. Annals of Reviews in
Evolutionary Applications, 9, 658–672 Ecolology, Evolution and Systematic, 38, 595–619.
Kuparinen, A., Perälä, T., Martinez, N.D.M., and Valdovinos, Von Bertalanffy, L. (1938) A quantitative theory of organic
F.S. (2019) Environmentally-induced noise dampens and growth (inquiries on growth laws II). Human Biology, 10,
reddens with increasing trophic level in a complex food 181–213.
web. Oikos, 128, 608–620. Wood, Z.T., Palkovacs E.P., and Kinnison, M.T. (2018) Eco-
Law, R. (2007) Fisheries-induced evolution: present status evolutionary feedbacks from non-target species influence
and future directions. Marine Ecology Progress Series, harvest yield and sustainability. Scientific Reports, 8,
335, 271–277. 6389.
Mace, G.M., Collar, N.J., Gaston, K.J., et al. (2008) Quan- World Wildlife Fund (2001) The status of wild Atlantic
tification of extinction risk: IUCN’s system for classifying salmon: a river by river assessment. [http://d2ouvy59p0-
threatened species. Conservation Biology, 22, 1424–1442. dg6k.cloudfront.net/downloads/salmon2.pdf; accessed
19 August 2019]
CHAPTER 21
Demographic processes in socially

structured populations
Maria Paniw, Gabriele Cozzi, Stefan Sommer, and Arpat Ozgul
21.1 Introduction depending on their social status, with dominants typi-

cally benefitting from larger groups (Ozgul et al. 2014).
In animals, a staggering diversity of social organi-
In addition, group cohesion is highest at intermediate
sations has evolved, ranging from solitary living (Kap-
group sizes. At such sizes, the benefits for helpers
peler and van Schaik 2002) to group decision-making in
of staying at home despite reproductive suppression
eusocial species (Wilson and Hölldobler 2005). Among
outweigh the costs of dispersing and forming a new
these life history strategies, sociality, where individ-
group, where the chances of reproducing are higher
uals form cohesive groups that are typically struc-
(Packer et al. 2005; Bonte et al. 2012; Hoogland 2013;
tured by kinship, has evolved in numerous clades
Markham et al. 2015). These examples illustrate that
(Alexander 1974; Clutton-Brock 2002). Such group
social structure plays an important role in regulating
living confers direct and indirect fitness benefits to
vital rates. To accurately infer group- and population-
group members (Hamilton 1964; Clutton-Brock 2002;
level processes in social species, social structure
Silk 2007). Moreover, population ecologists increasing-
therefore needs to be included in the respective demo-
ly recognise that accounting for social complexity in
graphic models (Courchamp et al. 1999; Bateman
population models can improve population viability
et al. 2013); this is particularly true when addressing
estimates (Walters et al. 2002) and inference from pop-
environmental change questions, as changes in group
ulation genetics (Parreira and Chikhi 2015) and there-
structure and traits of group members may precede
fore lead to better conservation management decisions
population collapse (Paniw et al. 2019).
(Courchamp and Macdonald 2001; Berec et al. 2007).
One straightforward way to account for direct
In socially structured groups, vital rates of indi-
demographic effects of social structure on population
viduals are affected by social rank, group size, and
dynamics (e.g. differential responses in survival and
group structure (Courchamp and Macdonald 2001;
reproduction depending on social state) is to parame-
Clutton-Brock 2016; Angulo et al. 2018). Depending on
terise matrix population models (MPMs) (Chapter 9).
the degree of sociality, reproduction within a group
MPMs have been used in numerous systems to
may be monopolised by a dominant pair (Weldon
study differences in demography among social stages
McNutt and Silk 2008; Armitage 2014; Clutton-Brock
(Figure 21.1). For instance, Brault and Caswell (1993)
and Manser 2016), and their offspring are then
investigated population dynamics of killer whales
co-reared by nonreproducing group members (Cour-
(Orcinus orca) accounting for the structure of differ-
champ et al. 1999; Griesser et al. 2017), which gain
ent pods. MPMs that include reproductive dominant
indirect fitness benefits by raising close kin (Hamilton
and subordinate stages have been developed for the
1964). As a consequence of such cooperative breeding,
group-living Siberian jay (Perisoreus infaustus) (Layton-
reproductive success increases with group size (Cour-
Matthews et al. 2018); and they have been used to
champ and Macdonald 2001; Clutton-Brock 2002).
derive optimal group size in the obligate cooperatively
Group size also regulates the survival of individuals
Maria Paniw et al., Demographic processes in socially structured populations. In: Demographic Methods Across the Tree of Life.
(a) (b) (c)
Figure 21.1 Three examples of social animals for which structured population models are available: (a) the yellow-bellied marmot (Marmota
flaviventris), a facultative cooperative breeder; (b) the Kalahari meerkat (Suricata suricatta), an obligate cooperative breeder; and (c) the Siberian
jay (Perisoreus infaustus), in which breeding within a group is monopolised by a dominant pair but helpers do not occur. (credits: Arpat Ozgul
(a, b) and Michael Griesser (c).)
breeding meerkat (Suricata suricatta) (Bateman et al. requires individual-based models (IBMs) of group and
2018). population dynamics. Such models have been devel-
In addition to direct demographic effects of social oped for meerkats (Ozgul et al. 2014) and red-cockated
structure, individual phenotypic traits can mediate woodpeckers (Zeigler and Walters 2014).
these effects. For instance, body size can play an Despite increasing calls for an individual-based ecol-
important role in acquiring and maintaining social sta- ogy (Stillman et al. 2015), the scarcity of mechanistic
tus, where the largest individuals gain reproductive modelling frameworks for socially structured popu-
dominance (Hodge et al. 2008). Integral projection lations partly results from the type of information
models (IPMs) (Chapter 10) can account for age- or required to parameterise IBMs, that is, long-term,
stage-specific dynamics of phenotypic traits such as individual-based data spanning several family groups.
body size (Ellner et al. 2016). In social species, IPMs Furthermore, these models tend to require increased
have been developed for marmots (Ozgul et al. 2010; computational capacity and lack analytical tractabil-
Maldonado-Chaparro et al. 2018) and meerkats (Paniw ity because population-level metrics can be obtained
et al. 2019). In both species, the IPMs have shown through simulations only. Thus, in many cases, ecol-
the critical role of body mass in mitigating effects of ogists prefer simplified structured population mod-
environmental factors on vital rates. els such as MPMs. The crucial question is then: To
Despite their widespread use, both MPMs and IPMs what extent does simplification of mechanistic process-
may be a poor modelling choice for socially struc- es governing social structure affect key inferences of
tured populations (Zeigler and Walters 2014). This group and population dynamics?
is because these two modelling frameworks cannot In this chapter, we try to provide an answer by test-
easily account for demographic stochasticity inher- ing population models of increasing complexity. To
ent to small groups and complex behavioural mech- this end, we test population models that can be applied
anisms generating group structure. For instance, the to socially structured, individual-based data sets and
acquisition of a high social status such as breeding quantify consequences of model simplification.
dominance, is typically reserved for a specific indi- In the following sections, we implement several
vidual within a group, such as the oldest or strongest approaches, increasing in complexity, to incorporate
female (Armitage 2014; Ozgul et al. 2014) and there- social structure into population models and assess
fore constitutes a deterministic process. MPMs and how much model complexity is required to adequately
IPMs, however, define transitions among social stages capture the dynamics of a socially structured popu-
as probabilistic processes, assigning a likelihood to lation. To parameterise all models, we use simulated
gain reproductive dominance based on, for example, individual-based data of a hypothetical population of a
age, mass, group size, or environmental covariates (e.g. cooperative breeder, for example the African wild dog,
Paniw et al. 2019). It is possible to incorporate demo- Lycaon pictus, or Kalahari meerkat, Suricata suricatta
graphic stochasticity into MPMs and IPMs, but this (Figure 21.2). From these data, we first develop a sim-
can jeopardise the analytical tractability of these mod- ple stage-structured MPM, in which stages correspond
els (Ovaskainen and Meerson 2010). Modelling deter- to social status, whereby group dynamics, such as the
ministic transitions among social stages, in contrast, processes determining the extinction and formation of
D E M O G R A P H I C P R O C E S S E S I N S O C I A L LY S T R U C T U R E D P O P U L AT I O N S 343
(a) S(MH) Rep(MH) Rec(MH) Rec mass(H)
S(MH) Rep(MH) Rec(MH) Rec mass(H)
Subadult S(MH) G(MH) Adult (1- E(MH))

Dominant
helper helper S(MH) G(MH)
I(H)
(1- E(MH)) S(MH) G(MH) S(MH) G(MH) Figure 21.2 Three-stage life cycle (a) of a socially structured
population simulated (b) and analysed in this chapter. Life cycle
(b)
Group (social) stages are represented by circles. Transitions among
Number of helpers (group size)
30 1 4 7 10 stages and reproductive output are depicted by black solid and

2 5 8 11 dashed arrows, respectively; they depend on stage-specific
3 6 9 12 vital rates: that is, survival (S), emigration (E), reproduction
(Rep), and recruitment (Rec) (blue); and trait transitions, that
20
is, growth (G) and recruit mass (Rec mass) (red). Vital rates
and trait transitions (i.e. demographic processes), in turn, are
functions of mass (M) and number of helpers (H). Adult
10 helpers can also immigrate (I) into the population when the
number of helpers falls below a threshold. Using the life cycle,
individual stage transitions and fates are simulated for
0 50 years, resulting in a data set (b) consisting of 12 groups of
0 10 20 30 40 50 different sizes (average group size is plotted in black).
Year
new groups, are ignored. We then expand this basic in Ozgul et al. (2014) (see also Paniw et al. 2019).
MPM into an IPM by including individual traits. In Our adaptation (Figure 21.2a) is representative of
doing so, we first ignore and then incorporate group life cycle dynamics in numerous other cooperative
dynamics. Lastly, we attempt to reconstruct group breeders such as African wild dogs (Courchamp and
dynamics with a flexible IBM approach. We compare Macdonald 2001) and Arabian babblers (Ridley 2007).
the fit of each modelling framework to the observed Our simulated population consists of females in differ-
data using the following metrics: instantaneous popu- ent groups. In each group, one dominant monopolises
lation growth rate, relative abundances of social stages, most reproduction (although this reproductive dom-
stage-specific mean masses, and average group sizes. inance is relaxed when groups are small). The other
These metrics are typically used to assess goodness-of- group members then help co-rear the offspring. To gain
fit of individual-based or structured population mod- reproductive opportunities, adult helpers can either
els of social animals (e.g. Ozgul et al. 2014; Paniw et al. ‘queue’ for dominance (i.e. wait for the death of the
2020). dominant to take
We perform all analyses in R and provide the corre- her position) or emigrate from their natal group
sponding scripts (in the Appendix; (please go to www. and form new groups. Our simulation represents these
oup.com/companion/SalgueroGamelonDM)). These social dynamics and is detailed in the R script gen-
scripts represent tools for researchers working with erate_social_data.R. The script is designed to
socially structured populations to both simulate data simulate fates and body masses of individuals, while
that best describe their study system and assess the lev- taking into account the effects of individual (life histo-
el of model complexity needed to capture the dynamics ry stage) and social factors (group size) on vital rates
of the system. (survival, reproduction, recruitment, and emigration).
The simulation follows individual females in dis-
crete time-steps of 1 year over 50 years. It comprises
21.2 A simulated, socially structured
three hierarchical levels: individual, social group, and
population population. All females are characterised by the follow-
The simulated, individual-based data set used to ing state variables: age; mass (i.e. log of body mass,
develop socially structured population models loosely in grams); group identity and associated group size
follows the life cycle structure of meerkats presented (i.e. number of subadult and adult helpers); social stage
(i.e. dominant, subadult helper, and adult helper); We consider two levels of variation in demographic
reproductive or pregnancy stage (pregnant or nonpreg- processes. First, within groups, the simulation imple-
nant); and recruitment success (number of offspring ments demographic stochasticity by treating survival,
that survive to become subadults). Only adults can emigration, and pregnancy as binomial probability
reproduce, and adult subordinate helpers can emigrate events, and recruitment as a Poisson count event.
from a given group (e.g. McNutt 1996; Maag et al. Second, to account for environmental stochasticity,
2018). The vital rates of survival, emigration, reproduc- parameters that define the demographic-process mod-
tion, and recruitment determine the fate of individuals, els are sampled from 10-year-specific values, akin to
while growth and recruit-mass functions determine random-year effects on demographic-process averages
changes in body mass (Figure 21.2a). We base the (Paniw et al. 2019). These year-specific values assume
parameterisation of the vital rates and trait transitions a spatially homogeneous environment and affect all
(collectively referred to as demographic processes hence- groups simultaneously, that is, a specific random-effect
forth) on published models of meerkat demographic structure is chosen at the beginning of each year in the
processes (Bateman et al. 2013; Ozgul et al. 2014; Paniw simulation and is incorporated into all demographic-
et al. 2019). In these models, all demographic process- process models for all groups simultaneously. The
es increase with mass. Helper survival and dominant simulated data result from one reproducible iteration
recruitment also increase with group size, but emi- of demographic and environmental stochasticity, that
gration is higher and adult helper reproduction and is, by setting a fixed random seed in R (see gen-
recruitment are lower when groups are larger. There- erate_social_data.R). This data set follows 12
fore, an intermediate group size is favoured in long- groups across the 50 years, with average group sizes of
term projections of group dynamics (Figure 21.2b). 14 helpers (Figure 21.2b). We later implement several
The simulation determines group extinction at the stochastic iterations in the IBM in order to reconstruct
beginning of each time-step. A group is considered group dynamics in the simulated data set (see 21.6).
extinct if it consists of maximally one dominant female We initialise the simulation of the individual-based
and one adult helper. If a group goes extinct, the data with four groups of different sizes; these groups
available pool of individuals that emigrate at a giv- constitute the starting population (Figure 21.2b). We
en time-step form a new group. More specifically, the assign masses to individuals within these initial groups
largest number of emigrants from one of the four resi- based on their social stage. Demographic process-
dent groups settles as a new group. All other emigrants es are then calculated for all groups simultaneously,
disperse and leave the population at the end of the upon which the simulation advances by one time-step.
time-step. If no local emigrants are available at a given Demographic processes in existing and new groups are
time-step, immigrants from outside the focal popula- equivalent, with the exception that individuals do not
tion form a new group. These immigrants consist of emigrate from a new group in the year of founding.
a dominant (i.e. founding) female and a random sam-
ple of six to eight adult helpers (Maag et al., pers.
21.3 A socially structured matrix
observation).
Individuals that survive from one time-step to the population model
next change mass. Adult helpers can emigrate or stay To assess how well the dynamics of a socially
in the group, and those that stay can become dominant structured population can be captured by a sim-
or stay helpers. Surviving adult helpers and domi- ple structured population model, we parameterise
nants can get pregnant, and those that do wean ≥0 an MPM using the simulated data set described
subadult helpers with a certain body mass. Body mass in 21.2. Details on the parameterisation can be
and number of helpers in the group differentially affect found in the R script social_MPM.R. The MPM
each of these processes at each time-step. Due to the consists of three stages—subadult helpers, adult
alloparental care of offspring, their body mass is not helpers, and dominants—with transitions among
affected by the mass of their mother in the simula- stages, reproduction, and recruitment determined by
tion. When all individuals have completed the demo- stage-specific demographic processes (Figure 21.2a).
graphic processes, their age is updated, and subadult We parameterise the demographic processes using
helpers automatically become adult helpers; surviving generalised linear mixed models (GLMMs). We ignore
dominants and adult helpers remain in their perspec- group dynamics in the MPM but assume that popula-
tive stages unless a dominant dies, in which case the tion size can potentially affect demographic processes
heaviest adult helper >2 years of age takes its place (Bateman et al. 2013; Paniw et al. 2019). We therefore
(Hodge et al. 2008). model each demographic process as a function of
population size, a fixed effect. We also include year indirectly consider group dynamics by incorporat-
as a random effect on the mean. We use Akaike’s ing population size into demographic-process mod-
information criterion (AIC) (Burnham et al. 2011) to els and projecting population size through time.
assess whether including population size provides a However, while in our simulated data, the interac-
better model fit compared to the null model including tion between mass and group size (i.e. number of
only the random-year effect. We build an MPM for subadult and adult helpers) determines adult helper
each of the 49 yearly transitions available from the survival, which is a critical demographic process
data. Starting with the population vector in year 1, (Paniw et al. 2019), modelling adult helper survival
we project population dynamics until year 50 by as a function of population size and mass does not
multiplying an MPM, which describes demographic capture this interactive effect (see the GLMMs in
transitions from time t to t+1, by a population vector at social_IPM_popSize.R). We therefore attempt to
time t to obtain a vector at t+1. We incorporate immi- account for group dynamics more directly by refit-
gration by adding the number of observed immigrant ting demographic-process GLMMs using group size
adult helpers and dominants at time t to the modelled instead of population size as the predictor variable.
population vector at time t. From the projections, we We then model average group size in the population
obtain yearly instantaneous population growth rates as a linear function of population size and parame-
and relative abundances of different social stages in terise and project IPMs inferring average group size
the population. from population size (i.e. the sum of the population-
vector elements) at each time-step of the projec-
tion. Details on the parameterisation can be found in
21.4 A socially structured integral the R script social_IPM_groupSize.R. As before,
projection model we obtain yearly instantaneous population growth
rates, relative abundances of different social stages in
Next, we expand the MPM approach by explicit-
the population, and the distribution of stage-specific
ly considering the effect of a continuous phenotypic
masses.
trait, that is, body mass, on demographic process-
Parameterising IPMs with average group sizes also
es and population dynamics. For each of the three
allows for determining optimal group size analyti-
social stages that characterise our population, we test
cally. At the optimal group size, average population
the effects of mass, population size, and their inter-
fitness, that is, the long-term asymptotic population
action on all demographic processes using GLMMs.
growth rate λ, is the highest. To obtain an estimate
We then parameterise a stage-mass classified IPM for
of optimal group size in our population, we calcu-
each of the 49 annual demographic transitions. For
late λ by performing an additional analysis (optimal_
each stage a, the IPM tracks the number of individu-
groupSize.R), in which we parameterise IPMs with
als (na ) in 100 discretised mass classes at time t. Details
group sizes ranging from 2 to 30 individuals, using
on the parameterisation can be found in the R script
averaged demographic processes across years.
social_IPM_popSize.R. The IPM approach allows
to project trait and population dynamics simultane-
ously (Ellner et al. 2016). These projections follow the 21.6 An individual-based model
methods described in the MPM section (21.3). In the as a mechanistic approach to
yearly IPMs, immigration is accounted for by adding modelling sociality
the observed number of immigrant adult helpers and
The most mechanistic approach to model the dynamics
dominants in a given mass class at time t to the pro-
of a socially structured population is to parameterise
jected population vector at time t. In addition to yearly
an IBM, because IBMs account for various levels of
instantaneous population growth rates and relative
stochasticity and can impose deterministic transitions
abundances of different social stages in the population,
among social stages (here, gaining dominance in a
we obtain the distribution of stage-specific masses.
group). We implement an IBM for our simulated data,
which essentially replicates the protocol of simulating
21.5 Explicitly accounting for group life history data outlined in 21.2 (see social_IBM.R).
However, unlike our simulated data set, which was
sizes in a population model generated using one random iteration of demograph-
As population size in our simulated data is correlat- ic and environmental stochasticity, we use 100 random
ed with average group size (see generate_social_ iterations of the IBM to fully account for stochasticity.
data.R), the socially structured MPM and IPM As in the structured models described in 21.3 and
21.4, we use parameters of the most parsimonious of new group sizes when a group goes extinct in the
demographic-process models to parameterise the IBM. population.
The structured modelling frameworks (here the
IPMs), in contrast to the IBM, allow for determining
21.7 Does increased model complexity optimal group size analytically (Figure 21.5). Aver-
improve inferences of population age population fitness (i.e. the asymptotic popula-
tion growth rate) is highest at a group size of nine
dynamics in a socially structured
helpers and declines sharply as group size increases
population? (Figure 21.5). This sharp decline is in line with previ-
For most metrics, the three modelling frameworks that ous investigations of optimal group size in meerkats
we use to parameterise and project the dynamics of (Bateman et al. 2018) and other cooperative breeders
a socially structured population generate similar esti- (e.g. Markham et al. 2015). The analytically obtained
mates (Figure 21.3). In particular, increasing model optimal group size is lower than the average group
complexity does not reduce relative differences in pro- size (13 individuals) in the observed data. One rea-
jected and observed instantaneous population growth son for this discrepancy in optimal group size is that
rates (Figure 21.3a). Model complexity also does little the observed data are based on short-term simulations
to reduce such differences in number of adult helpers of 50 years and ignore asymptotic long-term dynam-
and their mass. Here, the mechanistic model with the ics. Another reason is that such asymptotic dynamics
highest complexity (i.e. the IBM) only performs best will likely never be reached due to the dynamic nature
(averaged across the 100 random iterations) for the first of changes in group size (Figure 21.2b). That is, calcu-
years of the projection (Figure 21.3b,c). In subsequent lating λ at different group sizes assumes that a given
years, as projections diverge from observations, dif- group size remains stable through time, which does not
ferences between the modelling frameworks decrease. occur in the system.
With respect to average masses, the IPM based on
population size slightly outperforms the IPM based 21.8 Implications for life history models
on group size in projecting masses of adult helpers,
but differences between IPM parameterisations do not
of socially structured populations
translate into differences in projected average mass- Perhaps due to perceived similarities to human soci-
es of subadult helpers and dominants (Figure 21.3c). eties, the behaviour and life histories of social ani-
The MPM performs worse than the IPMs in projecting mals have been studied extensively (Emery et al. 2007;
the numbers of adult helpers (with the IPM parame- Clutton-Brock 2016). Accounting for sociality, how-
terised with average group size outperforming the IPM ever, introduces additional levels of complexity into
parameterised with population size) (Figure 21.3b). population models, questioning the relative useful-
However, for most years, the MPM actually does a ness of different modelling frameworks for investi-
better job than the IPMs in projecting the number of gating social species. Here we show that structured
subadult helpers and dominants (Figure 21.3b). One population models (MPMs and IPMs) can be a rea-
metric in which the IBM clearly outperforms all other sonable choice for projecting the dynamics of socially
models is in its accuracy to project the number of domi- structured populations even though they don’t con-
nants in the population, which, due to the deterministic sider deterministic changes in social stages and demo-
nature of parameterising transitions to the dominant graphic stochasticity. Most importantly, the MPM did
stage, is always equivalent to the number of groups (i.e. not perform worse than the IPMs in quantifying
four). most metrics that inform on the goodness-of-fit of
It is important to emphasise that, although the aver- population models, suggesting that relatively simpler
age results of the IBM do not differ substantially from structured models do not necessarily compromise the
the results of the MPM and IPMs, the IBM captures accuracy of results.
the demographic and environmental stochasticity We recommend using an IBM if assessing total
embedded in the demographic data (Figure 21.4). group number in a population is crucial for an
That is, observed dynamics are within the confidence analysis. The most useful attribute of highly mech-
interval of the 100 random iterations of the IBM anistic models, such as IBMs, in projecting social-
(Figure 21.4; see also plots in social_IBM.R). Most ly structured populations is their ability to quanti-
variation among the iterations is seen in the number fy deterministic transitions among social stages in a
of adult helpers, which reflects the random sampling population (Ozgul et al. 2014). In our simulations,
(a)
0.2 MPM IPMpop IPMgroup IBM
Relative difference to observed
0.1
dynamics
0.0
–0.1
–0.2
0 10 20 30 40 50
Year
(b)
2
Relative difference to observed dynamics
Subadult
helper
1
0
0.8
Adult helper
0.4
0.0
–0.4
Dominant
0.5
0.0
0 10 20 30 40 50
Year
(c)
0.050
Subadult
helper
Year Relative difference to observed
0.025
0.000
–0.025
Adult helper
0.0050 Figure 21.3 The three modelling frameworks used to

dynamics
0.0025 describe the dynamics of a socially structured population:

0.0000 MPM; IPM accounting for population size (pop) or
–0.0025 average group size (group); and IBM. The three models
–0.0050 perform similarly in quantifying (a) instantaneous
0.01 population growth rates, (b) relative abundance, and
Dominant
(c) average mass in each of the three social stages

0.00 modelled. For the IBM, results averaged across 100
–0.01 random iterations of the model are shown. Note that the
relative difference in the number of dominants is 0 for
0 10 20 30 40 50 the IBM.
Year
150
Average projected
Subadult helper
100 Observed
50
0
150
Number of individuals
Adult helper
100
50
0
150
Figure 21.4 The IBM accurately
simulates the stochastic dynamics of a
Dominant
100 socially structured population projected
for 50 years. The grey lines show
50 projections under each of the 100 random
iterations of the IBM; the blue line
0 indicates the average across the 100
iterations; numbers of projected and
0 10 20 30 40 50
observed dominant individuals coincide.
Year
IBM can account for various levels of stochasticity in

1.00 a straightforward manner. This is important in sys-
Asymptotic population growth rate
tems in which demographic stochasticity may strongly

affect modelling outcomes (Zeigler and Walters 2014).
0.95 Despite such advantages, mechanistic models require
detailed information on the processes that determine
social interactions among individuals and group for-
0.90 mation (e.g. when immigration occurs in our example).
Due to these challenges, simpler models are often pre-
ferred (Ellner et al. 2002).
0.85 Among the structured population models that we
use to parameterise and project the dynamics of a
socially structured population, the MPM, which is
0.80 the simplest in parameterisation, outperforms the IPM
0 10 20 30
(and even the average results of the IBM) in the
Group size
assessment of instantaneous population growth rates.
Figure 21.5 Asymptotic population growth rate as a function of This occurs because the MPM estimates the number
group size. Population growth (i.e. fitness) peaks at a group size of of dominants and subadult helpers in the popula-
nine individuals. tion more accurately in most years. Because simulated
numbers of subadults and dominants are lower than
for example, the IBM is the only model that accu- simulated numbers of helpers, demographic stochas-
rately projects the number of dominants and, as a ticity affects the former two life stages more strongly
consequence of model specifications, the number of and, as a consequence, increases the uncertainty in pre-
social groups in the population. This accuracy is dictions of demographic processes (Petchey et al. 2015).
unsurprising as we parameterised the IBM to reflect Moreover, models that project mass change in addi-
the observed data in which maximally four social tion to stage change propagate additional uncertainty,
groups co-exist in the population at any given time. resulting from assessments of mass change, to esti-
In addition to allowing deterministic transitions, an mates of population dynamics. Thus, simple MPMs
may prove useful for studying socially structured pop- Bateman, A.W. et al., 2018. Matrix models of hierarchi-
ulations, if projecting mass change is not a critical cal demography: linking group- and population-level
component of the study. dynamics in cooperative breeders. The American Natural-
If projecting mass change is a critical component ist, 192, 188–203.
Bateman, A.W. et al., 2013. Social structure mediates envi-
of the study, IPMs can be a valuable alternative to
ronmental effects on group size in an obligate cooperative
IBMs. However, although we obtain similar results
breeder, Suricata suricatta. Ecology, 94, 587–597.
from IPMs parameterised with either population size
Berec, L., Angulo, E., and Courchamp, F., 2007. Multiple allee
or average group size, we caution against using these effects and population management. Trends in Ecology and
two metrics interchangeably. Even under a strong cor- Evolution, 22, 185–191.
relation of group and population size, demographic Bonte, D. et al., 2012. Costs of dispersal. Biological Reviews of
processes do not necessarily respond equally to these the Cambridge Philosophical Society, 87, 290–312.
two metrics. In our example, while the probability to Brault, S. and Caswell, H., 1993. Pod-specific demography
reproduce of heavy dominants is negatively affected of killer whales (Orcinus orca). Ecology, 74, 1444–1454.
by population size, this interaction effect is not signif- Burnham, K.P., Anderson, D.R., and Huyvaert, K.P., 2011.
icant when reproduction is modelled as a function of AIC model selection and multimodel inference in behav-
group size. In addition, when simulating the observed ioral ecology: some background, observations, and com-
parisons. Behavioral Ecology and Sociobiology, 65, 23–35.
data, we consider the effect of group size on demo-
Clutton-Brock, T., 2002. Breeding together: kin selection
graphic processes but not the effects of population size.
and mutualism in cooperative vertebrates. Science, 296,
For instance, these two metrics have opposite effects 69–72.
on meerkat emigration (Bateman et al. 2013), since Clutton-Brock, T., 2016. Mammal Societies, John Wiley &
females emigrate more readily at large group sizes, Sons.
when within-group competition is high, but not at high Clutton-Brock, T.H. and Manser, M., 2016. Meerkats: coop-
population densities, when between-group competi- erative breeding in the Kalahari. In: W.D. Koenig and J.L.
tion is high (Paniw et al. 2019). Dickinson (eds) Cooperative Breeding in Vertebrates: Stud-
The three modelling frameworks that we describe ies of Ecology, Evolution, and Behavior, 294–317, Cambridge
in this chapter represent commonly used approaches University Press.
for socially structured populations. This is, howev- Courchamp, F. and Macdonald, D.W., 2001. Crucial impor-
tance of pack size in the African wild dog Lycaon pictus.
er, no exhaustive list, and alternative models that can
Animal Conservation, 4, 169–174.
address some of the challenges associated with mod-
Courchamp, F., Grenfell, B., and Clutton-Brock, T., 1999.
elling sociality have recently been developed (Plard Population dynamics of obligate cooperators. Proceed-
et al. 2019; Shizuka and Johnson 2019). For instance, ings of the Royal Society of London B: Biological Sciences,
integrated population models (IPM2), particularly if 266, 557–563.
fitted in a Bayesian framework, can incorporate demo- Ellner, S.P. et al., 2002. Precision of population viability
graphic stochasticity and account for error propagation analysis. Conservation Biology, 16, 258–261.
due to parameter uncertainty in population projections Ellner, S.P., Childs, D.Z., and Rees, M., 2016. Data-Driven
(Kery and Schaub 2011; Plard et al. 2019). Furthermore, Modelling of Structured Populations: A Practical Guide to the
individual-based models do not need to be constructed Integral Projection Model, Springer.
‘from scratch’ but can be implemented in various Emery, N.J. et al., 2007. Cognitive adaptations of social bond-
programmes (e.g. Vortex) (Zeigler and Walters 2014), ing in birds. Philosophical Transactions of the Royal Society of
London. Series B, Biological Sciences, 362, 489–505.
which can facilitate their inference and perturbations.
Griesser, M. et al., 2017. Family living sets the stage for coop-
Therefore, we encourage readers to use the R code
erative breeding and ecological resilience in birds. PLOS
that accompanies this chapter as a possible foundation Biology, 15, e2000483.
for analyses to be expanded and integrated into novel Hamilton, W.D., 1964. The genetical evolution of social
applications. behaviour. I. Journal of Theoretical Biology, 7, 1–16.
Hodge, S.J. et al., 2008. Determinants of reproductive success
References in dominant female meerkats. Journal of Animal Ecology, 77,
92–102.
Alexander, R.D., 1974. The evolution of social behavior. Hoogland, J.L., 2013. Prairie dogs disperse when all close kin
Annual Review of Ecology and Systematics, 5, 325–383. have disappeared. Science, 339, 1205–1207.
Angulo, E. et al., 2018. Allee effects in social species. Journal Kappeler, P.M. and Van Schaik, C.P., 2002. Evolution of pri-
of Animal Ecology, 87, 47–58. mate social systems. International Journal of Primatology, 23,
Armitage, K.B., 2014. Marmot Biology: Sociality, Individual 707–740.
Fitness, and Population Dynamics, Cambridge University
Press.
Kery, M. and Schaub, M., 2011. Bayesian Population Analy- Petchey, O.L. et al., 2015. The ecological forecast horizon, and
sis Using WinBUGS: A Hierarchical Perspective, Academic examples of its uses and determinants. Ecology Letters, 18,
Press. 597–611.
Layton-Matthews, K., Ozgul, A., and Griesser, M., 2018. The Plard, F. et al., 2019. IPM 2: toward better understand-
interacting effects of forestry and climate change on the ing and forecasting of population dynamics. Ecological
demography of a group-living bird population. Oecologia, Monographs, 89, e01364.
186, 907–918. Ridley, A.R. 2007. Factors affecting offspring survival and
Maag, N. et al., 2018. Density-dependent dispersal strategies development in a cooperative bird: social, maternal and
in a cooperative breeder. Ecology, 99, 1932–1941. environmental effects. Journal of Animal Ecology, 76,
Maldonado-Chaparro, A.A. et al., 2018. Transient LTRE 750–760.
analysis reveals the demographic and trait-mediated pro- Shizuka, D. and Johnson, A.E., 2019. How demographic pro-
cesses that buffer population growth. Ecology Letters, 23, cesses shape animal social networks. Behavioral Ecology.
1353. http://dx.doi.org/10.1093/beheco/arz083.
Markham, A.C. et al., 2015. Optimal group size in a high- Silk, J.B., 2007. The adaptive value of sociality in mam-
ly social mammal. Proceedings of the National Acade- malian groups. Philosophical Transactions of the Roy-
my of Sciences of the United States of America, 112, al Society of London. Series B, Biological Sciences, 362,
14882–14887. 539–559.
McNutt, J.W. 1996. Sex-biased dispersal in African wild Stillman, R.A. et al., 2015. Making predictions in a changing
dogs, Lycaon pictus. Animal Behaviour, 52, 1067–1077. world: the benefits of individual-based ecology. Bioscience,
Ovaskainen, O. and Meerson, B., 2010. Stochastic models of 65, 140–150.
population extinction. Trends in Ecology and Evolution, 25, Walters, J.R., Crowder, L.B., and Priddy, J.A., 2002. Pop-
643–652. ulation viability analysis for red-cockaded woodpeckers
Ozgul, A. et al., 2010. Coupled dynamics of body mass and using an individual-based model. Ecological Applications,
population growth in response to environmental change. 12, 249–260.
Nature, 466, 482–485. Weldon Mcnutt, J. and Silk, J.B., 2008. Pup produc-
Ozgul, A. et al., 2014. Linking body mass and group dynam- tion, sex ratios, and survivorship in African wild dogs,
ics in an obligate cooperative breeder. Journal of Animal Lycaon pictus. Behavioral Ecology and Sociobiology, 62,
Ecology, 83, 1357–1366. 1061–1067.
Packer, C. et al., 2005. Ecological change, group territoriality, Wilson, E.O. and Hölldobler, B., 2005. Eusociality: origin
and population dynamics in Serengeti lions. Science, 307, and consequences. Proceedings of the National Academy
390–393. of Sciences of the United States of America, 102, 13367–
Paniw, M. et al., 2019. Life history responses of meerkats to 13371.
seasonal changes in extreme environments. Science, 363, Zeigler, S.L. and Walters, J.R., 2014. Population models
631–635. for social species: lessons learned from models of red-
Paniw, M. et al., 2020. Assessing seasonal demographic cockaded woodpeckers (Picoides borealis). Ecological Appli-
covariation to understand environmental-change impacts cations, 24, 2144–2154.
on a hibernating mammal, Ecology Letters, 23, 588–597.
Parreira, B.R. and Chikhi, L., 2015. On some genetic conse-
quences of social structure, mating systems, dispersal, and
sampling. Proceedings of the National Academy of Sciences of
the United States of America, 112, E3318–E3326.
CHAPTER 22
Demographic methods
in epidemiology
Petra Klepac and C. Jessica E. Metcalf
22.1 Why demography is important infection can only spread via contact between a suscep-
to epidemic dynamics tible and infected individual, unless new susceptible
individuals are introduced into the population, the
Bernoulli’s work on smallpox in the 1760s perhaps pathogen will go extinct. For immunising infections,
represents the earliest infectious disease model (Dietz in many settings, the main source of new susceptible
and Heesterbeek 2002; Bernoulli and Blower 2004). individuals will be via a core demographic process,
Bernoulli estimated the gain in life expectancy that that is, births (infants may be transiently protected by
would emerge from smallpox elimination using life maternal immunity, but this generally wanes rapidly).
tables constructed by Edmond Halley (who also dis- Birth rates thus crucially determine pathogen dynam-
covered Halley’s comet). This early work reveals one ics. As an example, in England and Wales, during the
half of the inextricable link between demography and baby boom that followed World War II, measles out-
infectious disease epidemiology: Infectious disease can breaks occurred every year. As the birth rate fell and
shape host demography directly by changing host sur- susceptibles were replenished more slowly, dynamics
vival or fertility. For example, smallpox mortality was switched to outbreaks occurring every other year (Earn
estimated to be as high as 30% (Kotar and Gessler 2013), et al. 2000; Bjørnstad et al. 2002). The magnitude of such
plant pathogens like the Citrus tristeza virus may be effects diminishes as the duration of immunity declines
invariably fatal (Dawson et al. 2013), pathogens like (Morris et al. 2015), but signatures of host demograph-
brucellosis cause spontaneous abortion in many ungu- ic features such as sudden increases in the birth rate
lates (Moreno 2014), and fungal pathogens transmit- on pathogen dynamics are still expected across a broad
ted by pollinators can be sterilising for many species array of pathogen life histories. In species like plants
of plants (Antonovics 2005). However, just as host that lack adaptive immunity, pathogen-linked mortal-
demography may be shaped by infectious disease
ity and spatial heterogeneity might also result in sus-
dynamics, host demography may in turn shape the
ceptible depletion sufficient to limit pathogen spread.
spread of infections.
Another important feature of pathogen and host life
Many features of infectious pathogen and host life
history that shapes dependence of pathogen dynamics
history shape this reverse dependence. In vertebrates,
on host demography is the fact that both transmission
infection can leave hosts protected from subsequent
(or transfer of infection between infected and suscep-
infection by the same pathogen as a result of adaptive
tible individuals) and recovery can differ by demo-
immunity. The time span of this protection can range
graphic features. For example, in human populations,
from a few years (for antigenically variable pathogens,
like influenza) to lifelong (for childhood infections such younger individuals may have higher transmission
as measles). Focusing on the latter case (lifelong immu- rates than older individuals because they have greater
nity), following a large outbreak, most individuals will rates of contact with other individuals (Mossong et al.
have been infected, and thus immunised, and are con- 2008), and transmission rates depend on both rates of
sequently no longer susceptible to infection. Because contact between a susceptible and infected individual
Petra Klepac and C. Jessica E. Metcalf, Demographic methods in epidemiology. In: Demographic Methods Across the Tree of Life.
and probabilities of transfer of the infectious agent. In et al. 2009). However, whatever the scale of modula-
this situation, populations with higher birth rates, and tion, this additional dependency of pathogen dynamics
thus a higher density of younger individuals, might on host demography needs to be considered when
experience faster pathogen spread. Susceptibility to modelling infectious disease dynamics.
infection, recovery, and/or duration of shedding can To conclude, host demography both depends on
also be age dependent. For example, juvenile plants and shapes pathogen dynamics. Host fertility shapes
in a species of alpine carnations appear far more sus- the introduction of new susceptible individuals into
ceptible to fungal pathogens than adults (Bruns et al. the population (Earn et al. 2000); host demographic
2017). In humans, duration of shedding of respirato- structure may be crucial to allowing pathogen spread,
ry syncytial virus declines with age (Munywoki et al. as a result of the form of transmission (density or
2015), which might directly reflect age effects or reflect frequency dependent; McCallum et al. 2001), age-
reduced shedding following a previous exposure to the (or stage-) specific patterns of contact (Mossong et al.
virus. 2008; Klepac et al. 2009), or differences in susceptibility
Age is not the only demographic feature that can (Bruns et al. 2017) or recovery (Munywoki et al. 2015)
contribute to variation in transmission and recovery across age (or stage). Further, beyond direct effects of
(or indeed growth, survival, or fertility). An individu- pathogens on host survival or fertility, the burden of
al’s ontogenetic trajectory can be measured by aspects infection may vary as a function of age (e.g. rubella
of size, such as height or weight, or discrete life his- infection is of greatest concern in women of child-
tory phases, such as pupal stage. These aspects may bearing age (Metcalf et al. 2012a); SARS-CoV-2 shows
not map perfectly to an individual’s age (Caswell striking increases in severity with age). These depen-
2001 see Chapter 9), yet they may importantly affect dencies require modelling strategies that account
both demography and pathogen-relevant processes. for population structure in both host and pathogen
For example, risk of infection can be shaped by size and how they affect each other. Furthermore, model
of floral display in plants as a result of insect foraging structures have to be able to encompass the fact that
behaviour (Ferrari et al. 2006b); and nematode-linked the relevant time-scales for host and pathogen may
sheep mortality is higher in lower-weight individuals be very different. For example, human demography
(Gulland 1992). Where such effects occur, the linked unfolds on a scale of years, while the generation time
demographic and epidemiological dynamics may be (or average interlude from a primary infection to a sec-
better captured by tracking ‘stage’ rather than (or as ondary infection) of a pathogen like measles is about
well as) ‘age’ within a model. All of these scenarios rep- 2 weeks. In this chapter, following an overview of core
resent the effect of demography on disease dynamics. concepts in infectious disease modelling, including a
Another example of an important link between brief overview of the parallel between the concepts of
infectious disease dynamics and demography is cap- R0 in demography and infectious disease dynamics,
tured by the distinction made between ‘density-’ we introduce a methodological approach for formally
and ‘frequency-’dependent infectious disease trans- addressing the combined interaction between demog-
mission. In the former case (density-dependent trans- raphy and infectious disease dynamics and illustrate
mission), as the population grows, so does the number this approach using an example. We conclude by
of individuals infected by one infectious individual, pointing to the broad spectrum of options available for
resulting in a direct dependence of pathogen dynam- modelling the intersection between host demography
ics on host population dynamics. This might broadly and infectious disease dynamics.
capture the dynamics of directly transmitted infections
like tuberculosis (Ramsey et al. 2002). In the latter
22.2 Modelling the intersection between
case (frequency-dependent transmission), the number
of individuals infected by one infectious individual is epidemiology and demography
approximately fixed. This is often thought of in the con- In a basic unstructured infectious disease model, ini-
text of sexually transmitted infections (where the num- tially ignoring demography, individuals are classi-
ber of partners an individual has is not dependent on fied only according to their epidemiological state, for
population size) or the dynamics of vector-transmitted example, ‘Susceptible’ (S), ‘Infected’ (I), or ‘Recovered’
infections (where the number of hosts visited also does (R)—noting that further states are possible, depen-
not scale with population size but saturates at a certain dent on the biology of the focal host–pathogen system
point). It is increasingly recognised that this dichotomy (e.g. ‘Maternally immune’, ‘Exposed’, ‘Vaccinated’).
is a simplification, and dynamics may be somewhere Individuals’ movement between these epidemiologi-
between the two extremes (Wonham et al. 2006; Smith cal states is tracked using tools such as differential
DEMOGRAPHIC METHODS IN EPIDEMIOLOGY 353
births
b
births
b
births b
βΙ γ
S I R
infection recovery
μ μΙ μ
mortality mortality mortality
Figure 22.1 A susceptible–infected–recovered model showing transitions governed by host demography (text underlined) and infection
processes (not underlined); parameters as in section 22.2. Colours reflect the fact that both infection and its aftermath (individuals in the
R compartment) might also modify host demography, illustrated by the changed mortality for infected individuals μI .
equations. Susceptible individuals may become infect- This framing indicates that how the infectious
ed at a rate defined by the product of a transmission pathogen affects host demography will depend
rate (often denoted β) and the density (or frequency; on both the pathogen transmission rate (β) and
see above) of infected individuals; infected individuals pathogen-related mortality ( μI ) because together
recover (at a rate often denoted γ such that the average these determine the excess mortality within the
duration of the infectious period is equal to 1/γ); and host population (Figure 22.1). However, the framing
recovered individuals may lose immunity and return also reveals that host demography can modulate
to the susceptible state. This may be represented as: the pathogen’s spread, which may complicate this
dS relationship.
= −βSI To describe pathogen spread, we can derive an
dt
dI expression for R0 , the expected number of new infec-
= βSI − γI tions per infectious individual in a completely suscep-
dt
dR tible population, R0 = β/ (γ + μI ), which reveals that
= γI, pathogen population growth rate will decline as infect-
dt
ed host mortality μI increases. The conditions for ‘herd
which is the classic Kermack and McKendrick (1927)
immunity’, or the fraction of the population that must
compartmental model with constant rates. It is sim-
be immunised pc such that one infected individual
ple to extend this so-called closed population framing
results in less than one new infection (i.e. the pathogen
to encompass core demographic processes, such that
is doomed to extinction), require that (1 − pc ) R0 is less
individuals may enter or leave the population. All
than one (Cobey 2020), that is, pc = 1 − 1/R0 . The
epidemiological classes (S, I, and R) are subject to mor-
underlying logic is as follows: the proportion of the
tality (e.g. at rate μ), which may be higher for infected
population immunised by vaccination must be such
individuals (e.g. μI ). Births generally enter the suscep-
that the effective reproduction number, RE , or num-
tible class (or may briefly enter a maternally immune
ber of new infections per one infectious individual in
class before this maternal immunity wanes and they
one generation time of the disease in a population that
become susceptible). This could be dependent on the
is only partially susceptible, RE = R0 S, must be less
population size:1 for example, at a rate b × the popula-
than 1, where S is the proportion of the population that
tion size (shown below), or a constant rate, depending
is susceptible (here assuming a frequency-dependent
on assumptions about fertility:
transmission formulation, so that the proportion, S, is
dS the relevant quantity). From this definition of pc , we can
= b (S + I + R) − βSI − μS
dt see that, all else being equal, higher pathogen-related
dI mortality can reduce the threshold for pathogen elim-
= βSI − γI − μI I
dt ination by inflating the denominator of R0 . Moving
dR beyond pathogen-related mortality, how the thresh-
= γI − μR
dt old for elimination is approached will also depend
on fertility and mortality more broadly. Assuming
1
Discrete time analogues of capturing demography of for simplicity that μ = μI and the host population
births arguably date back to Leonardo Bonacci (or Leonardo
of Pisa), now known as Fibonacci, in 1202, where number of
is at equilibrium, that is, b = μ, the equilibrium
births at time Bt is governed by the population size at two pathogen incidence I∗ = μ (R0 − 1) / (R0 (γ + μ)), so
previous time-steps, Bt = Bt−1 + Bt−2 . increasing the rate of birth or mortality can increase
Average age of infection (A*)

Equilibrium incidence (I*)
0.4 40
birth per
0.3 30 1000 Figure 22.2 Impact of birth rate and
10 vaccination coverage on equilibrium
0.2 20 20 incidence and average age of infection.
30
40 Dotted line shows the critical elimination
0.1 10 50 threshold, or the proportion of the
population we need to vaccinate in order
0.0 0 to interrupt disease transmission in a
0% 25% 50% 75% 100% 0% 25% 50% 75% 100% population, pc = 1 − 1/R0 . Parameters:
Vaccination coverage Vaccination coverage R0 = 12; infectious period is 14 days.
the equilibrium incidence of the pathogen (Figure 22.2, (ODEs) (e.g. see Tudor (1985) for a measles example).
left). Numerical solutions are generally obtained using finite
Further relevant demographic information is difference methods, essentially discretising the sys-
embedded in this simple structure. For example, at tem. Since most formulations of these models are
equilibrium, the rate at which susceptible individuals equivalent to assuming exponential waiting times for
are infected is (broadly) βI∗ , and this will define transitions between states, it is important to veri-
the average age of infection, A∗ , that is, A∗ = 1/βI∗ , fy that model outcomes do not reflect unrealistic
so that infections with higher transmission rates β patterns of ageing (i.e. some individuals may age
should have lower ages of infection, and, given the very quickly into older age groups, and individu-
dependence of I∗ on birth rates, birth rates can have als may even be lost if model age groups are too
a similar effect (Figure 22.2, right). This illustration large).
is essentially a version of the Kermack–McKendrick ‘Cohort ageing’ is one possible alternative approach
model, with births and constant rates studied in 1932 to modelling ageing as a continuous processes—under
(Kermack and McKendrick 1932), which notes the ‘cohort ageing’ all individuals are aged at a single
link between increasing births and increasing rate of time point in the year. This may be appropriate for
disease and corresponding endemic level. Related hosts with very sharply seasonal life cycles (Peel et al.
expressions that explicitly encompass demography 2014) or human infections circulating in school set-
can also be obtained (McLean and Anderson 1988). tings (Schenzle 1984). Effectively, cohort ageing corre-
However, these basic models do not allow for dif- sponds to using what is referred to as a ‘birth pulse
ferent rates (infection, mortality, waning of immunity) model’, where, for example, all individuals are born
for individuals of different demographic ages or stages. on 1 January and age at the end of the year.
The clear next step is to further classify individuals Clear dynamical consequences (e.g. birth pulses shape
according to their demographic stage or age, as well pathogen persistence [Peel et al. 2014]) indicate that
as their epidemiological state. In this scenario, patterns care must be taken in defining the form of ageing
of ageing or ontogeny must be modelled alongside used by these models. Similar considerations around
epidemiological processes. the framing of ageing apply to discrete-time matrix-
One approach is to use partial differential equations based model frameworks (Klepac and Caswell 2011)
(PDEs), where age is modelled as a continuous vari- which represent an alternative approach to captur-
able. This approach dates again to McKendrick (1925) ing both demographic and epidemiological dynam-
(though it is commonly known as the McKendrick– ics. The basic principles developed can be expand-
von Foerster equation) and can be linked to Lot- ed to consider further structure within populations
ka’s integral equation for the number of births in (e.g. households [Hilton and Keeling 2019]). Finally,
the populations by the method of characteristics to as computational power increases, individual-based or
solve PDEs (Keyfitz and Keyfitz 1997). Technical chal- agent-based models represent an increasingly tractable
lenges assocated with such formulations (ranging from option to represent infectious disease transmission
issues with convergence in simulations, to challenges in demographically heterogeneous populations (e.g.
in aligning models with data, and sheer computational Iozzi et al. 2010).
intensiveness) mean that in practice PDE framings To conclude, many model frameworks can capture
are often reduced to ordinary differential equations the intersection between demography and infectious
disease dynamics. All have in common explicit expres- long before epidemic R0 took its prominent place in
sion of epidemiological and demographic rates (trans- infectious disease control. In population biology, R0 is
mission, mortality, etc.) and may additionally include defined as the expected number of offspring that an
cross-classification of individuals by epidemiological individual will produce during its lifetime, or the pop-
and demographic state (age or stage); nuances in terms ulation growth rate from one generation to the next
of how ageing or ontogeny is modelled must generally (Caswell 2001). The link with epidemic R0 is evident
be considered. Before providing the mathematical even from the survival function. In demography,
details underpinning one particular approach (a dis- Z ∞
crete time-matrix-based framing [Klepac and Caswell R0 = l (a) m (a) da
0
2011]), to provide further intuition on the links between
demography and infectious disease dynamics, we where l (a) is survival probabily from birth to age a,
expand further on how classic demography modelling and m (a) is reproduction rate at age a. However, if
relates to an essential parameter in infectious disease we let l (a) instead be the probability that a newly
dynamics, R0 , briefly introduced above. infected individual remains infected at least for time
a, and m (a) be the rate of infectiousness of an indi-
vidual infected a units of time, the same expression
gives us the epidemic R0 (Heesterbeek and Roberts
22.3 Demographic underpinnings 1995). Even in a model without explicit age structure,
of a core parameter, R0 epidemic R0 is influenced by demography and can
In infectious disease dynamics, the basic reproduction be expressed by R0 ≃ L/A (derived in Dietz 1975),
number, R0 , is defined as the expected number of sec- where L is the life expectancy (1/μ in our case) and
ondary cases produced by a single (typical) infection A is the average age of infection. While framing R0 is
in a completely susceptible population. As a result, R0 straightforward in a population structured only by epi-
acts as a threshold of whether or not disease will spread demiological states (e.g. in the basic ‘closed population’
in a population and is a central quantity in infectious SIR defined in section 22.2 R0 = β/γ), if individuals
disease dynamics. If R0 > 1 one person will on aver- are also structured by age (or stage), the issue of defin-
age infect more than one other person and the infection ing a ‘typical’ infection is more complex, and infections
will spread within a population, while when R0 < 1 at different ages (or stages) may have different conse-
the infection dies out. This has important implications quences, depending on both their characteristics (e.g.
for control, as R0 < 1 indicates the conditions for suc- how much they transmit) and the population context
cessful containment and eventual elimination of an (e.g. concentrations of susceptibles into age or stage
infection. classes that a focal infected age or stage may have more
Even though the concept of a critical threshold arose or less contact with). The challenge of estimating R0
from analysis of mathematical models for vector-borne in these settings was resolved via the development of
diseases (Ross 1911, 1916; MacDonald 1952) and direct- next-generation-matrix approaches (Dietz 1975, 1993;
ly transmitted diseases (Kermack and McKendrick Hethcote 1975; Diekmann et al. 1990; Van Den Driess-
1927, 1932) at the beginning of the last century, the che and Watmough 2002), which essentially charac-
potential use of R0 in epidemiology to inform control terises the duration of infectiousness for each age or
of infectious diseases was only fully recognised in stage class and the number of new infected individuals
the 1970s (Dietz 1975; Hethcote 1975). Kermack and that will result from each.
McKendrick noted in 1927 that disease won’t spread
if the population density is below the ratio of recovery 22.4 A discrete-time multistate matrix
rate and contact transmission rate (here designated approach
γ/β or 1/R0 ). This is directly linked to the critical
elimination threshold designating the proportion of the We can classify individuals according to their epi-
population we need to vaccinate in order to interrupt demiological state (which could include, for example,
disease transmission in a population, pc = 1 − 1/R0 maternally immune, susceptible, infected, recovered,
(Dietz 1975; Anderson and May 1991). vaccinated); the epidemiological rates between those
The analogous concept of ‘net reproductive rate’ states determine the epidemiological dynamics. We
(denoted by R0 by Dublin and Lotka (1925)) had focus here on discrete time formulations. It is straight-
widespread applications in demography and ecolo- forward to take the expressions for the SIR mod-
gy (Sharpe and Lotka 1911; Dublin and Lotka 1925) els and frame them in discrete time using a matrix
formation (Klepac and Caswell 2011). Taking the first Note that the Leslie matrix approach can be inter-
example illustrated (SIR with no demography) and set- preted as an example of an approximation of the PDE
ting the discrete time-step to the generation time of approach. However, epidemiological and demograph-
the infection, such that all individuals recover in one ic transitions are not independent. Infection can affect
time-step, we discretise the probability of infection as survival and reproduction rates, while rates of trans-
ϕt = 1 − exp (−βIt ), additionally scaling β to the appro- mission and recovery from infection can differ by
priate time-step. We can then express the matrix T age. Capturing this interaction between epidemiolog-
that can be used to project the population vector nt ical and demographic processes requires a combined
(of length 3, corresponding to numbers of individuals model structure. In a combined model, each individual
within the S, I, and R epidemiological states respective- is classified by both their infection status and demo-
ly) to nt+1 (likewise) according to: graphic stage. If there are s demographic states and
     c epidemic states, the population at time t can be
S (1 − ϕt ) 0 0 S described by an s × c array.
 I  =  ϕt 0 0  I   
R t+1 0 1 0 R t n11 · · · n1c
 . 
n
nt+1
n
T
n
nt N (t) = 
 .. ... 

ns1 · · · nsc
Many operations used for matrix population models
can be deployed (sensitivity analysis, etc. [Klepac et al. The population vector can be organised either into
2009]). Other important demographic processes could epidemic categories by stacking columns on top of one
be added, for example, by multiplying each of the rates another (grouping demographic stages within epidem-
in T by the probability of survival over the focal time- ic categories) or into demographic stages by stacking
step, and adding a birth rate b to each cell of the the first transposed rows of N(t) using the vec operator (a func-
row of the matrix to capture the fact that all individuals tion that stacks the columns of the matrix) (Hunter
produce offspring that are all born susceptible. and Caswell 2005; Klepac and Caswell 2011). The vec-
Going beyond this basic framework to add more permutation matrix approach allows modular struc-
detailed demography requires accounting for the ture of the model, where demographic and epidemic
fact that individuals also differ in age, maturity, and processes are separately captured with block-diagonal
reproductive status, and might die at different rates as matrices (see Klepac and Caswell 2011). In this way,
a function of age. Demographic models capture move- additional processes could be conveniently captured
ment of individuals as they age, grow, and reproduce with additional block-diagonal matrices. A challenge
and the resulting population dynamics. One exemplar here is to ensure the right orientation of the population
is the Leslie matrix approach [∧2], a discrete-time vector.
model focusing on the survival and fertility of each
age class, which we take inspiration from below. The
number of births here is determined by the fertility, 22.5 A specific example
Fs , of each age class (s) and their year-to-year survival To provide a concrete example, we illustrate the devel-
probabilities, Ps . Then a number of individuals of opment of a model structure for completely immu-
each age s, ns , at time t (or the age distribution at time nising childhood infections like measles, mumps, and
t) is easy to project to time t + 1 by multiplying the rubella. Focusing first on the epidemiological transi-
population vector ns with the matrix of fertilities and tions, we start by setting the time-step to be equal to
survival probabilities known as the Leslie matrix, L the average infectious period (i.e. γ = 1 discrete time-
(Leslie 1945, 1948) [∧2]. step), such that all infections recover at the end of each
     time-step. This duration is typically between a week
n0 F0 F1 L Fs n0 and a month for many childhood infections (Metcalf
 n  P  n 
 1   0 0 L 0  1  et al. 2009). We also expand the set of epidemiologi-
    
 n2  =  0 P1 L 0   n2  cal states considered (M), susceptible (S), infected (I),
    
 M  M  M  recovered (R), and vaccinated (V) state. For our focal
ns t+1 0 L Ps−1 0 ns t pathogens, a key driver of dynamics is highly variable
n n n
nt+1 L nt patterns of contact and thus opportunities for transmis-
sion between individuals of different ages (Mossong et
nt+1 = L nt al. 2008). The epidemic transitions in the model can be
described by the following matrix: age class (i + 1) at the end of the time-step, so ui = 1
  (e.g. for cohort ageing). If the age class i is longer than
1 − dm 0 0 0 0
 d the duration of the time-step, ui is set so that the aver-
 m (1 − Φ (n)) (1 − v) 0 0 0 

  age duration in that age class is equal to the duration
A (n) =  0 Φ (n) (1 − v) 0 0 0 
  of the age class. There is no growth out of the last age
 0 0 1 1 0 
class s, so the last entry in the matrix only has mor-
0 v 0 0 1
tality (or survival), Dss = ss . Here we use the vector
that groups the infection categories within demograph-
where dm is loss of maternal immunity, v is the propor- ic stages (by stacking transposed rows of N (t)). The
tion of susceptible individuals that are vaccinated, the resulting population vector n (t) is:
infection probability is given by
n (t) = vec N⊤
Φ (n) = 1 − exp(−Λ (n (t)) = (M1 , S1 , I1 , R1 , V1 , . . . , Ms , Ss , Is , Rs , Vs ) (t)⊤ .
and the force of infection for stage i, Λi is given by The full transition matrix includes ageing, mortality,
and infection dynamics:
P α
j Cij Ij (t)
Λi (n (t)) = β Pn . A (n (t))
(t)  
s1 (1 − u1 ) A1 0 0 L 0
 
 s1 u 1 A 1 s2 (1 − u2 ) A2 0 L 0 
= 
 M 
Here, β scales overall transmission which is set to 0 0 L ss−1 us−1 As−1 ss A s
be frequency dependent (the population size appears
in the denominator of the expression), Cij is the age- where Ai is the transmission matrix for demographic
specific mixing matrix, capturing relative encounter stage i.
and transmission probability between, and α captures We can define a corresponding fertility matrix, focus-
heterogeneities in mixing not directly modelled and the ing first within an age group. Here we illustrate a
effects of discretisation (α = 0.97 for measles [Bjørnstad scenario where fertility does not depend on epidemi-
et al. 2002]). Since the time-step is taken to reflect the ological state but only on age. However, recovered or
infectious period, and assuming that rate of mortality vaccinated mothers give birth to maternally immune
is likely to be negligible at this time-scale, the over- offspring; all other mothers give birth to susceptible
all magnitude of transmission should approximate R0 . offspring:
For this to be appropriately reflected in the expression  
0 0 0 b1 b1
for the force of infection, given that infection is com-  b b b
 1 1 1 0 0  
pletely immunising, the product of the contact matrix  
Fi (n) =  0 0 0 0 0 
C and susceptible population is equivalent to the next-  
 0 0 0 1 0 
generation matrix because the duration of infection is
0 0 0 0 1
set to one time-step in this scenario. We therefore scale
Cn so that its dominant eigenvalue is equal to 1; the Here, bi is the probability that individuals of age i pro-
magnitude of β should then be close to that of R0 . duce one offspring over the duration of the time-step,
Ignoring births and epidemic transitions for the and these are born susceptible (second row of matrix)
moment, the process of ageing (growth) and survival unless mothers were recovered or vaccinated, in which
can be modelled with the following transition matrix: cases they are born maternally immune (first row of
matrix). The full transition matrix is:
 
(1 − u1 ) s1 0 0 ··· 0  
F1 F2 F3 L Fs
 (1 − u2 ) s2 
 u1 s1 0 0   0 0 0 L 0 
D=
 .. 
 F=  M M M L M 

 . 
0 ··· ss 0 0 0 L 0
The full dynamics can be captured according to:
Here, si is the probability of surviving age i for the dura-
nt+1 = (A (n (t)) + F) nt
tion of the time-step, and ui is the rate of ageing out of
that class. If the age class i is equal to the time step, which will project the population forwards, encom-
all of the individuals in that class move on to the next passing transitions between demographic stages
(either stage or age) and epidemiological stages model structure. Measles is a directly transmitted, com-
(maternally immune, susceptible, infected, etc.). pletely immunising infection, with a generation time
of ∼2 weeks (Bjørnstad et al 2002). The existence of a
safe, inexpensive, and completely immunising vaccine,
22.6 Case study in conjunction with the high case fatality rate of this
infection, means that every World Health Organiza-
A key contribution that infectious disease models have tion region currently has a measles elimination goal.
brought to public policy is the concept of ‘herd immu- Estimates of R0 ∼ 15 for measles indicate that pc = 0.93
nity’ (also referred to as ‘community immunity’) and is the simplest (unstructured) analysis. How does this
the related insight that not everyone need be immu- change if we add detail to the demographic context?
nised in order to eliminate an infection, as some indi- For directly transmitted infections like measles, pat-
viduals will be indirectly protected by the immunity terns of contact over age have been suggested to
of the ‘herd’ (Cobey 2020). In the simplest case, the play an important role in shaping spread (Edmunds
critical threshold for population immunity achieved et al. 2000). Patterns of contact over age have been
by vaccination is pc = 1 − 1/R0 , but this expression inferred from diary studies, where individuals are
ignores age (and other forms of) structure in the pop- asked to record the age and number of people they
ulation. Heterogeneity in individual contacts can lead interact with over the course of some time period,
to herd immunity being achieved at a lower thresh- for example a week (Mossong et al. 2008). We start
old than classical 1 − 1/R0 (Gomes et al. 2020), while by assuming that such measurements provide a rea-
reduced variation in contacts can lead to a herd immu- sonable reflection of relative patterns of contact over
nity threshold higher than 1 − 1/R0 (Ferrari et al. 2006a). age of relevance for disease transmission (while not-
Here we focus on the scenario where contacts and/ ing that this does not necessarily follow) and use
or magnitude of transmission between individuals results from an existing study of age contact patterns
vary across ages, and we work to identify how the (Mossong et al. 2008) as a starting point (Figure 22.3,
vaccination coverage required to achieve pathogen left). We focus our investigation on a country whose
elimination is altered. The structure of the proportion demography is assumed to still be experiencing popu-
of the population that is susceptible, rather than the lation growth, with a population pyramid that narrows
absolute proportion of the population that is suscep- as age increases, and one with a much flatter popu-
tible, is what will shape the number of new infections lation profile (determined arbitrarily, for illustration;
following introduction of one (average) infected indi- Figure 22.3, right).
vidual and, conversely, the pattern of immunity (and Taking 2 weeks as our discrete time-step, as this
thus vaccination coverage) that is required for elimina- reflects the life history of measles, we scale the pattern
tion. of contacts reported from the diary studies, Cij , and
To illustrate this, we take the example of vaccination
coverage required for measles elimination using the
70 50000 Figure 22.3 (Left) Age contact data,

where colours indicate intensity of
60 recorded contacts between different age
classes (x and y axes), ranging from dark
50 (low) to white (high) contacts, and
population size
Age of contact
30000 showing that individuals tend to mostly

40 have contacts with individuals of their
own age (diagonal white coloured) or
30 their own parents/offspring (off-diagonals
light-coloured white). (Right) Population
20 structures explored to evaluate the
10000
required vaccination profile: solid
10 line—exponential population structure,
0 typical for developing world; dashed
0 line—rectangular population structure,
0 10 30 50 70 0 10 30 50 70 commonly observed in developed
Age focal individual Age countries.
1–4 1–4
10.0 10.0 1–9
1–14
1–9 1–19
1–24
1–29
5.0 1–14
5.0 1–34
1–39
1–44
RE structured
1–49
2.0 2.0
1–19 1–54
1–59
1.0 1.0
1–24
0.5 0.5 1–64
1–69
0.5 1.0 2.0 5.0 0.5 1.0 2.0 5.0 10.0

RE unstructured RE unstructured
Figure 22.4 Comparison of estimated RE under different age windows of vaccination (text on the figure indicates the age range of vaccination
corresponding to each point) for unstructured (x axis) and structured (y axis) population models in either a growing (left, corresponding to the solid
line in Figure 22.3) or static population (right, corresponding to the dashed line in Figure 22.3)
the transmission term β to obtain R0 = 15, using the across a range of upper ages targeted in the vaccina-
first population (Figure 22.3, right, solid line) as our tion campaign, comparing results for structured and
reference. To illustrate the importance of age profiles unstructured populations across the two population
of vaccination, we assume for simplicity that measles structures (Figure 22.4).
does not circulate in this population (i.e. that there In the first example, the upper age class that must be
is no natural immunity) and that there has been no targeted to achieve herd immunity is the same (up to
vaccination to date (there is also no vaccine-derived age 24) for both the structured and the unstructured
immunity; both are unrealistic). We then define v as case; but one can also see very different patterns in
a vector reflecting individuals immunised by vacci- each. In particular, the structured model shows much
nation across age. We assume that every individual larger expected numbers of new cases for correspond-
within a targeted age group is reached by vaccina- ing scales of vaccination of younger individuals (right-
tion, but, in line with estimates for the measles vaccine, hand size of x = 1) and smaller expected numbers of
we assume that vaccination efficacy is 97%, and we new cases if older individuals are targeted (lefthand
consider delivery of only a single dose. The public pol- size of x = 1). In the second example, very wide range
icy decision might then be what the upper age of the age campaigns are required, and the threshold for
vaccination campaign should be to achieve measles immunity differs between unstructured and structured
elimination—are we targeting children up to age 5 models.
years? Up to 10 years? Even higher?
In the unstructured model, where we include no dif-
ferences in patterns of contact over age, the effective 22.7 Conclusions
reproduction number RE is defined by RE = R0 s = Developing models that tractably engage with the
P
R0 vn/ n (t), where n is the number of individu- varying time-scales and dynamics of infectious dis-
als in each age class. For the structured population eases and host biology is a challenge. However, the
model, RE is the dominant eigenvalue of the prod- intricate interdependence of demography and epi-
uct of R0 (1 − v) nC. We can plot the impact on RE demiology makes engaging this challenge essential to
appropriately characterising a range of biological sce- Reproduction Ratio R0 in Models for Infectious Diseases
narios, from comparing different management stragies in Heterogeneous Populations. Journal of Mathematical
of infected ungulates (Jennelle et al. 2014), to quan- Biology 28 (4): 365–382.
tifying how changing demographic patterns such as Dietz, K. 1975. Transmission and Control of Arbovirus Dis-
eases. SIMS 1974 Utah Conference Proceedings, January
declining birth rates might alter the burden of a human
1974: 104–121.
infection (Metcalf et al. 2012a), to identifying life stages
Dietz, K. 1993. The Estimation of the Basic Reproduction
critical to pathogen transmission in partially observed
Number for Infectious Diseases. Statistical Methods in Med-
outbreaks in wildlife (Klepac et al. 2009), to anticipat- ical Research 2 (1): 23–41.
ing perennial grassland community invasion by annual Dietz, K. and J. A. P. Heesterbeek. 2002. Daniel Bernoulli’s
plants with and without a pathogenic virus (Borer Epidemiological Model Revisited. Mathematical Biosciences
et al. 2007). In this chapter, we detail one potential 180 (1–2): 1–21.
framework rooted in discrete time approaches (not- Dublin, L. I. and A. J. Lotka. 1925. On the True Rate of
ing that there are many other possibilities, as briefly Natural Increase: As Exemplified by the Population of
described). The balance of availability of analytic meth- the United States, 1920. Journal of the American Statistical
ods, such as sensitivity (see Klepac and Caswell 2011; Association 20 (151): 305–339.
Earn, D. J. D., P. Rohani, B. M. Bolker, and B. T. Grenfell.
Metcalf et al. 2012a) and tractable inference from data
2000. A Simple Model for Complex Dynamical Transitions
makes such discrete-time a powerful tool in address-
in Epidemics. Nature 287: 667–670.
ing applied questions from a range of fields. Exciting Edmunds,W. J., N. J. Gay, M. Kretzschmar, R. G. Pebody, and
directions for future research include extension of such H. Wachmann. 2000. The Pre-Vaccination Epidemiology
tools and relevant statistical approaches for a range of of Measles, Mumps and Rubella in Europe: Implications
data-streams. for Modelling Studies. Epidemiology & Infection 125 (3):
635–650.
Ferrari, M. J., S. Bansal, L. A. Meyers, and O. N. Bjørnstad.
2006a. Network Frailty and the Geometry of Herd Immu-
References nity. Proceedings of the Royal Society B: Biological Sciences 273
Anderson, R. M. and R. M. May. 1991. Infectious Diseases of (1602): 2743–2748.
Humans: Dynamics and Control. Oxford: Oxford University Ferrari, M. J., O. N. Bjørnstad, J. L. Partain, and J. Antonovics.
Press. 2006b. A Gravity Model for the Spread of a Pollinator-
Antonovics, J. 2005. Plant Venereal Diseases: Insights from a Borne Plant Pathogen. The American Naturalist 168 (3):
Messy Metaphor. New Phytologist 165 (1): 71–80. 294–303.
Bernoulli, D. and S. Blower. 2004. An Attempt at a New Gomes, M. G. M., R. Aguas, R. M. Corder, et al. 2020.
Analysis of the Mortality Caused by Smallpox and of the Individual Variation in Susceptibility or Exposure to Sars-
Advantages of Inoculation to Prevent It. Reviews in Medical Cov-2 Lowers the Herd Immunity Threshold. medRxiv.
Virology 14 (5): 275–288. https://www.medrxiv.org/content/10.1101/2020.04.27.
Bjørnstad, O. N., B. F. Finkenstädt, and B. T. Grenfell. 2002. 20081893v3.
Dynamics of Measles Epidemics: Estimating Scaling of Gulland, F. M. D. 1992. The Role of Nematode Parasites in
Transmission Rates Using a Time Series SIR Model. Eco- Soay Sheep (Ovis aries L.) Mortality During a Population
logical Monographs 72 (2): 169–184. Crash. Parasitology 105 (3): 493–503.
Borer, E. T., P. R Hosseini, E. W. Seabloom, and A. P. Dobson. Heesterbeek, J. A. P. and M. G. Roberts. 1995. Mathemat-
2007. Pathogen-Induced Reversal of Native Dominance ical Models for Microparasites of Wildlife. In Ecology of
in a Grassland Community. Proceedings of the National Infectious Diseases in Natural Populations, edited by B. T.
Academy of Sciences 104 (13): 5473–5478. Grenfell and A. P. Dobson, 90–122. Cambridge: Cam-
Bruns, E. L., J. Antonovics, V. Carasso, and M. Hood. 2017. bridge University Press.
Transmission and Temporal Dynamics of Anther-Smut Hethcote, H. W. 1975. Mathematical models for the spread
Disease (Microbotryum) on Alpine Carnation (Dianthus of infectious diseases. In Epidemiology, edited by D. Lud-
pavonius). Journal of Ecology 105 (5): 1413–1424. wig & K. L. Cooke, 122–131. Philadelphia: Society for
Caswell, H. 2001. Matrix Population Models: Construction, Industrial and Applied Mathematics.
Analysis, and Interpretation. Sunderland, MA: Sinauer Hilton, J. and M. J. Keeling. 2019. Incorporating Household
Associates. Structure and Demography into Models of Endemic
Cobey, S. 2020. Modeling Infectious Disease Dynamics. Sci- Disease. Journal of the Royal Society Interface 16 (157):
ence 368 (6492): 713–714. 20190317.
Dawson, W. O., S. M. Garnsey, S. Tatineni, S. Y. Folimonova, Hunter, C. M. and H. Caswell. 2005. The Use of the
S. J. Harper, and S. Gowda. 2013. Citrus Tristeza Virus– Vec-Permutation Matrix in Spatial Matrix Population
Host Interactions. Frontiers in Microbiology 4: 88. Models. Ecological Modelling 188 (1): 15–21.
Diekmann, O., J. A. P. Heesterbeek, and J. A. J. Metz. Iozzi, F., F. Trusiano, M. Chinazzi, et al. 2010. Little Italy:
1990. On the Definition and the Computation of the Basic An Agent-Based Approach to the Estimation of Contact
Patterns—Fitting Predicted Matrices to Serological Data. Infectious Disease: Inference with Discrete Data. Theoreti-
PLOS Computational Biology 6 (12): e1001021. cal Population Biology 82 (4): 275–282.
Jennelle, C. S., C. Henaux, G. Wasserberg, B. Thiagarajan, Moreno, E. 2014. Retrospective and Prospective Perspectives
R. E. Rolley, and M. D. Samuel. 2014. Transmission of on Zoonotic Brucellosis. Frontiers in Microbiology 5: 213.
Chronic Wasting Disease in Wisconsin White-Tailed Deer: Morris, S. E., V. E. Pitzer, C. Viboud, C. J. E. Metcalf, O. N.
Implications for Disease Spread and Management. PLOS Bjørnstad, and B. T. Grenfell. 2015. Demographic Buffer-
One 9 (3): e91043. ing: Titrating the Effects of Birth Rate and Imperfect
Kermack, W. O. and A. G. McKendrick. 1927. A contribution Immunity on Epidemic Dynamics. Journal of the Royal
to the mathematical theory of epidemics. Proceedings of the Society Interface 12 (104): 20141245.
Royal Society of London. Series A 115 (772): 700–721. Mossong, J., N. Hens, M. Jit, et al. 2008. Social Contacts
Kermack, W. O. and A. G. McKendrick. 1932. Mathemati- and Mixing Patterns Relevant to the Spread of Infectious
cal Theory of Epidemics. II. The Problem of Endemicity. Diseases. PLOS Medicine 5 (3): e74.
Proceedings of the Royal Society of London. Series A 138 (834): Munywoki, P. K., D. C. Koech, C. N. Agoti, et al.
55–73. 2015. Influence of Age, Severity of Infection, and
Keyfitz, B. L. and N. Keyfitz. 1997. The McKendrick Partial Co-Infection on the Duration of Respiratory Syncytial
Differential Equation and its Uses in Epidemiology and Virus (RSV) Shedding. Epidemiology & Infection 143 (4):
Population Study. Mathematical and Computer Modelling 804–812.
26 (6): 1–9. Peel, A. J., J. R. C. Pulliam, A. D. Luis, et al. 2014. The Effect
Klepac, P. and H. Caswell. 2011. The Stage-Structured Epi- of Seasonal Birth Pulses on Pathogen Persistence in Wild
demic: Linking Disease and Demography with a Multi- Mammal Populations. Proceedings of the Royal Society B:
State Matrix Approach. Theoretical Ecology 4:301–319. Biological Sciences 281 (1786): 20132962.
Klepac, P., L. W. Pomeroy, O. N. Bjørnstad, T. Kuiken, A. Ramsey, D., N. Spencer, P. Caley, et al. 2002. The Effects of
Osterhaus, and J. M. Rijks. 2009. Stage Structured Trans- Reducing Population Density on Contact Rates Between
mission of Phocine Distemper Virus in the Dutch 2002 Brushtail Possums: Implications for Transmission of
Outbreak. Proceedings of the Royal Society of London. Bovine Tuberculosis. Journal of Applied Ecology 39 (5):
Series B 276: 2469–2476. 806–818.
Kotar, S. L. and J. E. Gessler. 2013. Smallpox: A History. Ross, R. 1911. Some Quantitative Studies in Epidemiology.
Jefferson, NC: McFarland. Nature 87 (2188): 466–467.
Leslie, P. H. 1945. On the Use of Matrices in Certain Popula- Ross, R. 1916. An Application of the Theory of Probabilities
tion Mathematics. Biometrika 33 (3): 183–212. to the Study of a priori Pathometry. Part I. Proceedings of
Leslie, P. H.. 1948. Some Further Notes on the Use of the Royal Society of London: Series A 92 (638): 204–230.
Matrices in Population Mathematics. Biometrika 35 (3/4): Schenzle, D. 1984. An Age-Structured Model of Pre-
213–245. and Post-Vaccination Measles Transmission. Mathemat-
MacDonald, G. 1952. The Analysis of Equilibrium in Malar- ical Medicine and Biology: A Journal of the IMA 1 (2):
ia. Tropical Diseases Bulletin 49 (9): 813–829. 169–191.
McCallum, H., N. Barlow, and J. Hone. 2001. How Should Sharpe, F. R. and A. J. Lotka. 1911. A Problem in
Pathogen Transmission Be Modelled? Trends in Ecology & Age-distribution. The London, Edinburgh, and Dublin
Evolution 16 (6): 295–300. Philosophical Magazine and Journal of Science 21:
McKendrick, A. G. 1925. Applications of Mathematics to 435–438.
Medical Problems. Proceedings of the Edinburgh Math- Smith, M J., S. Telfer, E. R. Kallio, et al. 2009. Host–Pathogen
ematical Society 44: 98–130. Time Series Data in Wildlife Support a Transmission
McLean, A. R. and R. M. Anderson. 1988. Measles in Devel- Function Between Density and Frequency Dependence.
oping Countries. Part II. The Predicted Impact of Mass Proceedings of the National Academy of Sciences 106 (19):
Vaccination. Epidemiology & Infection 100 (3): 419–442. 7905–7909.
Metcalf, C. J. E., O. N. Bjørnstad, B. T. Grenfell, and V. Tudor, D. W. 1985. An Age-Dependent Epidemic Model with
Andreasen. 2009. Seasonality and Comparative Dynamics Application to Measles. Mathematical Biosciences 73 (1):
of Six Childhood Infections in Pre-Vaccination Copen- 131–147.
hagen. Proceedings of the Royal Society B: Biological Sciences Van Den Driessche, P. and J. Watmough. 2002. Reproduc-
276 (1676): 4111–4118. tion Numbers and Sub-Threshold Endemic Equilibria for
Metcalf, C. J. E., J. Lessler, P. Klepac, F. T. Cutts, and Compartmental Models of Disease Transmission. Mathe-
B. T. Grenfell. 2012a. Impact of Birth Rate, Seasonality matical Biosciences 180 (1–2): 29–48.
and Transmission Rate on Minimum Levels of Coverage Wonham, M. J., M. A. Lewis, J. Rencławowicz, and P. Van
Needed for Rubella Vaccination. Epidemiology and Infection Den Driessche. 2006. Transmission Assumptions Generate
16: 1–12. Conflicting Predictions in Host–Vector Disease Models:
Metcalf, C. J. E., J. Lessler, P. Klepac, A. Morice, B. T. Gren- A Case Study in West Nile Virus. Ecology Letters 9 (6):
fell, and O. N. Bjornstad. 2012b. Structured Models of 706–725.
Subject Index
Tables, figures, and boxes are indicated by an italic t, f, and b following the page number.
abiotic drivers 115 allelic information 23

ABMs see Individual-Based Models (IBMs) allometric analysis 303, 304f, 305
abscisic acid (ABA) 42 allometric constraints 300–301
abundance (N) 133, 134–149 AM see adaptive management (AM)
abundance analysis 135–145 Amniote database 4t
dynamic abundance model 136–143 AnAge database 4t,38t
state–space model 143–145 antioxidant deficiencies 42–43
accelerated failure time (AFT) model 231 AR model see autoregressive (AR) model
accelerometers 45 artificial intelligence (AI) 66b
acoustic transmitters 57 ascorbate (vitamin C) 41
acquired immune response 44 Asian elephant (Elephas maximus) 169, 171,
adaptive habitat selection 215 171f
adaptive management (AM) 313–328 Asian tiger mosquito (Aedes albopictus) 7t
Adélie penguin (Pygoscelis adeliae) 56f assortativity, social behaviour 67
adrenocorticotropic hormone (ACTH) 41 asymptotic analysis 199t
Advanced Research and Global Observation Atlantic cod (Gadus morhua) 332–333, 334f,
Satellite (ARGOS) 56 335f
African wild dog (Lycaon pictus) 342–343 Atlantic salmon (Salmo salar) 336–337, 336f,
AFT see accelerated failure time (AFT) model 337t, 338f, 339
Agassiz’s desert tortoise (Gopherus automated tracking systems 102
agassizii) 59t autoregressive (AR) model
age at first reproduction (AFR) 302 dynamic abundance model 141–142
ageing 43, 158, 279 spatial demography 260t, 261–262
agent-based model (ABMs) see auxins 42
Individual-Based Models (IBMs) avuncular links 19
age-structured population 285–297 axes of variation 305–308, 305t, 307t, 308f
demographic data 301 azure damselfly (Coenagrion puella) 139, 140f,
initial stage structure 202 141
life table 164–165, 164t
survival probability 155 B10 survey data 4t
temporal autocovariance in environmental bacterial killing assays 44
noise 289 badger (Meles meles) 60t, 61t, 62t, 103
agricultural pests 122 Balearic wall lizard (Podarcis lifordi) 235
Akaike information criterion (AICs) 191, 303, Balt myna (Leucopsar rothchildi) 156, 157t
345 barn swallow (Hirundo rustica erthrogaster) 59t
alignment, genetic phylogeny 26 basal metabolic rate (BMR) 37–38
364 SUBJECT INDEX
bats (Chiroptera) 64, 103 classical swine fever (CSF) 220–221, 222f, 223f
between-individual variation 37 CLC Main Workbench 23t
betweenness, social behaviour 67 climate range 119
BIDDABA database 4t clonality 106
BIDE model 232 Clustal2 23t
biochemical trait 40–43 Clustal Omega 23t
Biofuel Ecophysiological Traits and Yields CMR model see Capture–Mark–Recapture
database (BETYdb) 38t (CMR) model
biogenic time series data 77–95, 88f coded nanotags 57
biologging 45, 55–57 CodonCode Aligner 23t
biomarkers 36–37 cohort life tables 152–153, 152f, 153–154, 153t,
biopsies, genetic samples 21t 154f
biotic drivers 115 colorednoise 7t
black-headed gulls (Chrolocephalus COMADRE database 4t, 177
ridibundus) 173, 174f, 175, 176f see also comparative methods
blood common cuckoo (Cuculus canorus) 136, 137f
collection 37 common ragweed (Ambrosia
genetic samples 21t artemisifolia) 251–253, 252f, 253f
blue tit (Cyanistes caeruleus) 102, 247–251, 248f, common rough woodlouse (Parcello
249f, 250f scaber) 171–173, 172f, 173f, 173t, 174f
body mass 276, 308–309 comparative methods 299–312
bootstrapFP 7t conservation biology 313
brassinosteroids 42 conStruct 7t
Breeder’s equation 330 consumer-resource interactions 281
Brownian motion 20 continuous drivers 117–118, 119
brown trout (Salmo trutta) 59t Convolutional Neural Network
brushtail possum (Trichosurus vulpecula) 61t (CNN) 65–66b
bssm 7t cooperative breeding 341, 343
BTSPAS 7t Coral Trait Database 4t, 38t
butterfly monitoring programme 134 cowbird (Molothrus) 59t
Cox’s proportional hazards 231
caloric restriction 152 critical elimination threshold 355
capm 7t cross-dating 80
Capture–Mark–Recapture (CMR) cryptic life stages 106
model 232–234 cyclooxygenase 1 (COX1) 20
Cormack–Jolly–Seber (CJS) model 232, 233 Cyclops 7t
multi-event model 233–234 cytochrome-b 20
multi-state model 233
single-state model 232–233 DATLife database 4t, 153
survival analysis 229, 230, 237 death rates 155
capwire 7t deep learning 65–66b, 65f
carbohydrate biomarkers 40–41 demogR 7t
CARE1 7t dendRoAnalyst 7t
carotenoids 41 dendrochronology 77, 81–82
cattle (Bos taurus) 60t, 61t dendrometeR 7t
census time 165–166 dendroTools 7t
chelating anticoagulants 37 density-dependent models 215, 218–219,
chlorophyll 39, 41 280–281
chloroplast markers 20 detrending 79f, 80
SUBJECT INDEX 365
developmental traits 300 filtering, data preparation 23

deviance information criterion (DIC) 235 FinePop 7t
diameter at breast height (DBH) 89 fire 120–121
diffusion equation 260t, 265 fish
disease control 324–325 age, otolith growth rings 78
dispersal 122, 123, 217, 260t, 264–266 biologging web survey 59t
distance-matrix methods 27t Fisherian stable age distribution 286, 288f
distance sampling 101 Florida scrub (Aphelocoma
DNA amplification/extraction/ coerulescens) 322–324
sequencing 22–23 fossil record 21
DPWeibull 7t FRANz 24
dragonhead (Dracocephalum austriacum) 124 free radical 42
drought-induced tree mortality 85, 88 FREQ 7t
dynamic abundance model 136–143, 140f freshwater mussel (Margaritifera) 82
dynamichazrd 7t fruit fly (Drosophila melanogaster) 151
dynsurv 7t functional trait 35b
eco-evolutionary dynamics 215–245 Galapagos sea lion (Zalophus wollebaecki) 63t

eigenvector 67 gauseR 7t
elasticity analysis 203 Genemapper 23
elephant seal (Mirouga leonina) 56f Genemarker 23
elk (Cervus candensis) 62t generalised linear mixed model (GLMM) 236,
Elkhorn coral (Acropora palmata) 121 345
El Niño-Southern Oscillation (ENSO) 87f generalised linear model (GLM) 185–186, 236
Encyclopedia of Life TraitBank 38t generation overlap 330
epidemic dynamics 351–352 generation time 156–302
epidemiology 351–361 genetic data 17–18, 217
epigenetics 45 genetic markers 20
epigenomics 45 genetic pedigrees 19
Epoppler 9t genetic phylogenies 19–21
E-SURGE 235 genetic sample collection 21–22, 21t
ethylene 42 gibberellins 42
Euclidian distance 259 GloTherm 38t
Euler–Lotka equation 152 glucocorticoids 41
EURING database 4t Gompertz model 142–143
European starling (Sturnus vulgaris) 154–155, goodness-of-fit tests, Capture–Mark–
155t Recapture model 235
European wild rabbit (Oryctolagus GPS technology 2, 56
cuniculus) 62t grandparent assignment 19
European woodcock (Scolopax rusticola) 262 great tit (Parus major) 102, 263
Europop database 4t grey reef shark (Carcharhinus
evolutionary demography 273–284 amblyrhynchos) 59t
experimental approaches, driver group sizes 341
quantification 117–118 growth rings 81–82
extinction rates 123
hair samples 21t
faecal samples 21t haptoglobin 44
feathers, genetic samples 21t hepatitis C 324
fertility, life table 156–157, 157t herbchronology 81–82
366 SUBJECT INDEX
herbivory 119 integrated metapopulation model 260t,

heritability 329–330 264–265
heterogeneity see individual heterogeneity Integrated Population Model (IPM) 245–255
heterophily, social behaviour 67 interactions see social data
hierarchical Bayesian framework 89 Intergovernmental Panel on Climate Change
homophily, social behaviour 67 (IPCC) 121
HormoneBase 38t International Database of Longevity 4t
hormones 41–42 International Transcribed Spacer (ITS) 20
cytokines 44 International Tree-Ring Data Bank (ITRDB) 78
cytokinins 42 intraspecific density 125
horn, sclerochronology 82 intraspecific dynamics 281
house sparrow (Passer domesticus) 25, 25t, 26f intrinsic population growth rate 259
Human Cause-of-Death Database 4t intrinsic rate of increase 280
Human Fertility Database 4t intrinsic water use efficiency 41
Human Life-Table Database 4t inverse modelling, missing life cycle
Human Mortality Database 4t, 239 parts 184–185b, 184f
humans, biochemical traits 40 IPMpack 7t
hurricanes, episodic drivers 121 IPMR 8t
island fox (Urocyon littoralis) 62t
IBMPopSim 7t
ICARUS project 56–57 jacpop 8t
iDiv Data 38t JAGS 236
ID system, sessile species 106 jasmonate 42
iEcology 107 joint likelihood 249–250, 249f
Illumina GenCall 23 Jolly–Seber (JS) model 232, 236
Imanishi, Kinji 54
immune markers 44–45 Kalign 23t
immunity 358 Kalman filter 143
immunoglobins (Igs) 44 Kaplan–Meier estimates
imperfect detection, known fate data vs. 230 life table 158
indirect effects, drivers 125 survival analysis in laboratory 230–231,
Individual-Based Models (IBMs) 213–227 231f
individual heterogeneity 170b, 234 kernel 187
individual likelihoods, Integrated Population Keyfitz’s entropy 309
Models 246 killer whale (Orcinus orca) 341, 342f
individual reproductive values 287, 287b, Kimura parameter model 26
288f, 289 known fate data, imperfect detection vs. 230
individual size, Integral Projection Model koala (Phascolarctos cinereus) 63t
(IPM) 182–183 K-selection 280–281
individual social interactions 57–58, 64
individual state (i-state) 163 lady orchid (Orchis purpurea) 198
infected (I) individuals, epidemiology 352, landscape genetics 265
356 leaf stomatal conductance 39
infectious diseases, dynamics & LEDA database 4t
demography 352 Lee–Carter model, human survival
Infra-Red Gas Analyser (IRGA) 39 analysis 239
innate immune response 44 lefko3 8t
instantaneous sampling 55 lemon shark (Negaprio breviostris) 59t
Integral Projection Model (IPM) 88, 181–195 Leslie matrix element 356
SUBJECT INDEX 367
Lexis diagram 152f mallard duck (Anas platyrhynchos) 320–322

LexisplotR 8t mammal
lhmixr 8t biologging web survey 59–63t
life expectancy, matrix population model 169 individual-level long-term
LifeHist 8t monitoring 102–103
life history 3, 4–5t, 4t, 7–10t, 11, 17, 19–20, mandrill (Mandrillus sphinx) 63t
35, 38t, 41–42, 78, 82, 84–89, 90, 101, 116, mangrove jack (Lutjanus argentimaculatus) 83
123, 134, 141, 151–152, 154, 157–158, 164, marbled murrelet (Brachyramphus
167–169, 170b, 171, 173, 175, 177, 193, 197, marmoratus) 263
202, 205, 210, 224, 229, 230, 237, 239, 246, marine ecosystem 78, 104
248, 273–281, 286, 288, 290, 294, 299–303, MatLab scripts 6
309–310, 329–330, 331b, 333–335, 334f, 337, Matrix Population Model (MPM) 3, 4–5t, 6,
339, 341, 343, 345, 346, 348–349, 351–352, 7–10t, 11, 118, 121, 152, 157, 158, 163–179,
358 181, 182, 183, 184b, 186, 187, 246, 262, 341,
Life Table Response Experiments 342–343, 346, 348, 351, 356
(LTREs) 205–206, 207 age-structured 164–166
life tables 151–161 socially structured 344–345
lifetime reproductive success 169 stage-structured 166–167
light detection and ranging (LIDAR) transient analysis 197–212
techniques 99f, 104 maximum likelihood (ML) method 27–28t,
linear models 99f, 135, 142, 176, 185–186, 189, 142–143, 322, 323
237–238, 239, 260t, 273, 335, 358 Mead’s milkwood (Asclepias meadii) 237
lipids, plants 40–41 meerkat (Suricata suricata) 107, 341, 342–343,
lipophilic antioxidants 43 342f, 346
Living Planet Index 5t
metabolomics 45
lmf 8t
metafolio 8t
local interactions, individual-based
metasim 9t
model 215–216
metric calculation, comparative
locomotion 281
methods 301–302
loggerhead sea turtle (Caretta caretta) 83
microbial interactions, continuous drivers 120
log-rank test, survival analysis in
micro-nutrients 41
laboratory 230–231
longevity records 5t microPop 8t
longitudinal demographic data microsatellite loci 22
collection 97–113 migration 4–5t, 7–10t, 55, 78, 84, 85, 89, 98,
see also long-term field studies 99f, 103–104, 108, 116, 122, 134, 151, 164,
longitudinal human studies, biomarkers 40 182, 184b, 217, 218, 229–230, 232–233, 236,
long-term field studies 97–113 245, 247, 249, 262, 264–265, 320–322, 330,
Lotka–Volterra equation 271–281 343–345, 349
LTER database 5t minor allele frequency (MAF) 24
LTREs see Life Table Response Experiments missDeaths 8t
(LTREs) missing data, survival analysis 230
LYRCforests 8t missing life cycle parts, inverse
modelling 184–185b, 184f
macro-nutrients, plant biomarkers 41 mitochondrial DNA 20, 21
MAFFT 23t mobile species long-term studies 98–99,
magnetic-inductive tracking techniques 103 101–103
MALDABBA database 5t modularity, social networks 67, 68–69f
see also comparative methods moose (Alces alces) 289–290, 290f
368 SUBJECT INDEX
Moran’s index 259 North American small yellow lady

morphometric variables 105 (Cypripedium parviflorum) 237
morse 8t North Pacific High (NPH) winter 86, 87f
MortalityLaws 8t npsurv 8t
MortalityTables 8t
mountain brushtail possum (Trichosurus OBMbpkg 8t
cunninghami) 61t observation error model 136, 138–139, 140f
mountain date palm (Phoenix loureirin) 124 occupancy 325
movement 4–5t, 45, 78, 82, 84, 87, 101, 103, oestrogen 41
124, 176, 217–218, 217f, 223, 224, 233, 259, omics data 45
260, 260t, 263–264, 265–267, 268–269, 313, open-access databases 3, 4–5t
318, 325, 352, 356 openCR 8t
MPM see matrix population model (MPM) Ordinary Differential Equation (ODE) 354
mptools 8t oxidative stress 42–43
MRsurv 8t oxygen consumption measurement 38
multi-event model, Capture–Mark–Recapture
Pacific geoduck (Panopea generosa) 82
model 233–234
PADRINO database 5t
multisite capture–recapture model 260t,
Pantheria database 5t
264–265
parameter uncertainty 216
multistate model, Capture–Mark–Recapture
parentage analysis 19, 25, 25t, 26f
model 233
parental traits 330
MUSCLE 23t parents & children, genetic pedigrees 19
mutation–selection nonlinear model 280 Partial Differential Equations (PDEs) 265
mutations, genetic pedigree construction 24 Passive Integrated Transponders (PIT tags) 57
mutualism 119–120 individual-level long-term monitoring 102
pathogen 67, 117, 119, 121, 220–221, 222f, 223f,
N see abundance (N) 229–230, 351–354, 356–357, 358, 360
ncappc 8t pattern-oriented modelling 219, 220f
NEff 8t PCAClone 9t
net reproductive rate 355 pec 8t
networks 4–5t, 6, 38t, 44, 53, 54–56, 54f, 57, penguin (Spheniscidae) 56f, 57, 64, 100f
59–63t, 64, 65–66b, 66, 68–69f, 69–70, 85–86, per capita growth rate 280
90, 123–124, 176, 223, 339 period life table 152f, 154–156, 155t
New Caledonian crow (Corvus perturbation analysis 187–189
moneduloides) 59t pest control management, matrix population
new infections per infected individual model 175
(R0 ) 353–354, 354f, 355 PHYLACINE 29
next-generation individual-based model 215 phylogenetic data, statistical analysis 302–303
next generation sequencing (NGS) 21 phylogenetic generalised least squares (PGLS)
nodes 68–69f, 69t model 303
characterisation 66 phylogenetic principal component analysis
social behaviour analysis 66–67 (PPCA) 303, 307–308
nonclimatic abiotic factors 119 phylogenetic signal 20
nonlinear feedback, individual-based phylogenetic supertree 21
model 215–216 phylogenetic tree, rooted vs. unrooted 26
nonlinear growth model, tree growth ring 81 physiological traits 37–40
nonlinearity, perturbation analysis 189, 190f animal metabolism 37–39
nonlinear population dynamics 124 databases 38t
SUBJECT INDEX 369
immune markers 44–45 pre-reproductive censuses 165–166

plant metabolism 39 principal component analysis (PCA) 45–46
semen physiology 39–40 prospective perturbation analysis 199t
telomeres 43–44 PROSPER 9t
pigment, plant biomarkers 41 proteomics 45
pitcher plant (Sarracenia purpurea) 124 proximity biologger 58, 64
PIT tags see Passive Integrated Transponders P-spline 239
(PIT tags) PSPManalysis 9t
Planteome 38t pyramid 9t
pletherm 8t
polar bear (Ursus maritimus) 198, 200 quadrats, sessile species long-term studies
polyamines 42 106
polymerase chain reaction (PCR) 22 quantitative fluorescence in situ hybridisation
polymorphic gene markers 24 (Q-FISH) 43
quantitative genetic model 279–280
polymorphic species markers 19
pop 8t
racoon (Procyon lotor) 60t, 61t
popbio 8t radio-frequency identification (RFID)
popdemo 8t stations 57, 58f, 107
POPdemog 8t individual-level long-term monitoring 102
PopED 9t radio receiver 56
popEpi 9t radiotelemetry 2
popkin 9t radio-tracking 55–56, 56f
poppr 9t Rage 9t
population growth rate 106, 117–122, 124, random forest algorithms 260–261, 260t
133, 135, 142, 154, 156–157, 158, 165f, 166f, random varying environments 276–279, 278f
167–169, 172f, 173, 183, 187, 199t, 200, raneMapper 9t
203–205, 204f, 205f, 206f, 207–209, 208b, rare events, episodic drivers 121–122
219, 220f, 246, 250, 259, 262–263, 265, 267, RCOMPADRE 9t
277, 289, 291, 294, 302, 307f, 323, 335, 337, Reactive Species (RSs) 42
343, 345, 346, 347f, 348, 348f, 353, 355 real-time quantitative polymerase chain
population inertia 201f, 202 reaction (qPCR) 43
population-level data collection 98–99, 99f, Recovered (R) individuals 352, 356
100f, 101 recovery, epidemiology 352
population momentum 202 red-billed chough (Pyrrhocorax
population prediction interval (PPI) 290 pyrrhocorax) 263, 289
population sink 259 red deer (Cervus elaphus) 103
population size (N) 97 red kite (Milvus milvus) 235–236
population source 259 redox state marker 42
positive interaction, continuous regional curve standardisation (RCS) 80
drivers 119–120 Registry of Research Data Repositories 38t
positive matrix, Markovian sequence 120 relatedness-based methods, genetic
post-disturbance patterns, episodic pedigree 24
drivers 121 relative estimate, abundance 135
post-reproductive censuses 165–166 relative water content 40
post-reproductive life 279–280 remote tracking device 235–236
predation 41, 64, 103, 119, 121–122, 198, 215, reproductive value 285–297
217f, 218–220, 219f, 220f, 224f, 281, 290, 315, individual reproductive value 287, 287b,
335, 339, 381 288f, 289
370 SUBJECT INDEX
reproductive value (Continued) sheep (Ovis aries) 98, 184–185b, 184f, 191
Matrix Population Model (MPM) 167–168 shrub, dendrochronology 81–82
see also age-structured populations siamang (Symphalangus syndactylus) 120
reptiles, biologging web survey 59t Siberian jay (Perisoreus infaustus) 341, 342f
resource selection 266–267 sibship clustering 19
respirometry 38 silver birch (Betula pendula) 333
resting metabolic rate 38 Simultaneous Auto Regressive Model
restricted maximum likelihood (REML) (SAR) 262
methods 142–143 Single Nucleotide Polymorphisms (SNPs) 19,
retrospective perturbation analysis 209 22
matrix population model 199t Single Telomere Length Analysis (STELA) 43
RFID see radio-frequency identification SIR model 355–356
(RFID) stations SMITIDstruct 9t
ribd 9t snakeroot (Eryngium cuneifolium) 120–121
risk of infection 352 Soay sheep see sheep (Ovis aries)
river otter (Lontra canadensis) 61t social behaviour analysis 64–67
RMark package 9t social data 53–76
Capture–Mark–Recapture model 235 collection 53
roe deer (Capreolus capreolus) 267 field-based sampling 54–55
root voles (Microtus oeconomus) 122–123, see also biologging
218–220, 219f, 220f social hierarchy position 67, 68–69f
R packages 7–10t socially structured populations 341–350
Rramas 9t Social Network Analysis (SNA) 64, 66, 69
r-selection 280–281
social rank 341
rSHAPE 9t
Sonoran desert tortoise (Gopherus
salicylates 42 morafkat) 262
SARS-CoV-2 352 sparrow-hawk (Accipter nisus) 289
satellite data collection 101 spatial associations, biologging 57–58, 64
SCALES database 5t spatial capture–recapture 263
Scandinavian lynx (Lynx lynx) 290–293, 290f, spatial demography 259–272
291–292b, 293f spatially explicit matrix population model
sclerochronology 80–81, 82 262
see also growth rings spatially-referenced population 101
Scots pine (Pinus sylvestris) 333 species richness 319
sea otter (Enhydra lutris) 266 spectrophotometric assay 43
seed bank data 251 SPI-Birds database 5t
selection differential 330 spline, spatial demography 260t, 261
semiparametric models, human survival sspe 9t
analysis 239 SSsimple 9t
sensitivity stagePop 9t
analysis 168–169, 173, 175, 216, 356 standard size-classified model 171
evolutionary demography 274–275, 275f State-Space Modelling (SSM) 235
matrix 168 state transition, Capture–Mark–Recapture
sequence alignment software 23t model 233
Serengeti: survey of age structure in ungulates stochastic analysis 199t
and ostrich database 5t stochastic population model 141, 142–143
sessile species long-term studies 103–106 strigolactone 42
sex differences, mortality 152 structured decision-making 314
SUBJECT INDEX 371
submodel, individual-based model 214 see also dendrochronology; growth rings

survival analysis 229–243 treestructure 10t
in laboratory 230–232 truncation, life table 158
see also Capture–Mark–Recapture (CMR) TRY Plant Trait Database 38t
model TSSS 10t
survivalmodels 9t turtle exclusion device (TED) 121
survParamSim 9t turtle scute, sclerochronology 82
Susceptible (S) 352, 356
swab, genetic samples 21t Unified Modelling Language (UML) 214
synchrony 123 unmanned aerial vehicle (UAV) 104, 105f
taproot, dendrochronology 81–82 vaccinated (V) individuals 356

Tasmanian devil (Sarcophilius harrisii) 62t validation, biomarkers 37
teeth, sclerochronology 82 vertebrate bones, biogenic time series data 89
telomere 43–44 vervet monkey (Chlorocebus pygerythrus) 56f
temporal autocovariance 289 Virginia opossum (Didelphis virginiana) 60t
terminal fragment (TRF) analysis 43 vital rate 1, 3, 4–5t, 11, 35, 37, 78, 84, 88–90, 98,
testosterone 41 99f, 105f, 115, 116–117, 119, 122, 124–125,
Thesaurus Of Plant Characteristics for Ecology 133, 158, 168–169, 172, 174f, 177, 181, 183,
and Evolution (TOP) 38t 185–186, 188–189, 190, 191, 193, 197–198,
thistle, state–space abundance analysis 199t, 200, 202, 203–208, 208b, 210, 215,
model 144–145, 146f 217, 217f, 227, 245–246, 249, 251, 260t,
tiger flathead (Platycephalus richardsoni) 80–81 262–263, 274, 276–279, 280, 299, 301, 308,
tiger salamander (Ambystoma californiense) 123 309, 313–314, 318–319, 323, 325, 341–342,
TIMELY model 224f 343–344, 343f
timereg 9t vitamin C (ascorbate) 41
vole (Microtus pensylvanicus) 264
time-varying environment, matrix population
von Bertalanffy growth model 334–335, 334f,
model 197, 204–207, 205f, 206f, 207–209,
335f
208b
time-varying model, matrix population
Weddell seal (Leptonychotes weddellii) 57
model 199t
white-tailed deer (Odocoileus virginialis) 60t,
timing of first reproduction (TFR) 303, 305 63t
tocopherol 41 wild boar (Sus scrofa) 220–221, 222f, 223f
total antioxidant capacity (TAC) 43 wiqid 10t
trackdem 10t wire-tailed manakin (Pipra filicauda) 59t
TrailNet 38t within-individual variation, data 37
Traits database 5t, 38t within-species variation, genetic
transcriptomics 45 phylogenies 20
transients, Capture–Mark–Recapture wolves (Canis lupus) 266
model 234 wood thrush (Hylocichla mustelina) 262
transition matrix 171
transmission rate, epidemics 352 xylem anatomical variation 84–85
transplantation experiment 118
tree growth ring 78–81, 79f, 81f year assignment, tree growth rings 80

Demographic Methods Across The Tree of Life

Uploaded by

Copyright:

Available Formats

Demographic Methods Across The Tree of Life

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Demographic Methods Across The Tree of Life

Uploaded by

Copyright:

Available Formats

Demographic Methods across the Tree of Life

Introduction: From lions, to lion’s manes, to dandelions: how different

Part 1 Demographic Data Collection: From Genes to Environment 15

1 Genetic data collection, pedigrees, and phylogenies 17

2 Biochemical and physiological data collection 35

3 Social data collection and analyses 53

4.3 Growth rings in other organisms 81

5 Longitudinal demographic data collection 97

6 Drivers of demography: past challenges and a promise for a changed

Part 2 Data- and Research Question-Driven Methods 131

7 Abundance-based approaches 133

7.3 Analytic approaches 135

8 Life tables: construction and interpretation 151

9 Introduction to matrix population models 163

10 Integral projection models 181

10.4 Combining the vital functions in the IPM 186

11 Transient analyses of population dynamics using matrix projection

12 Individual-based models 213

13 Survival analyses 229

14 Efficient use of demographic data: integrated population

Part 3 Applications 257

15 Spatial demography 259

16 Evolutionary Demography 273

17 Reproductive value and analyses of population dynamics of

17.6 Conclusions 293

18 Applying comparative methods to different databases: lessons from

19 Adaptive management: making recurrent decisions in the face of

20 Heritability, polymorphism, and population dynamics: individual-based

20.3 Generic ways to model the evolution of demographic rates 330

21 Demographic processes in socially structured populations 341

22 Demographic methods in epidemiology 351

From lions, to lion’s manes,

understand the biology of three different species of 4 From data acquisition to

Data source Description Source

Brief description Source

albopictus Age-structured population dynamics models https://cran.r-project.org/web/packages/albopictus/index.html

Brief description Source

IPMR Analysis and interpretation of simple and complex IPMs https://github.com/levisc8/ipmr

Brief description Source

trackdem Particle tracking and demography https://cran.r-project.org/web/packages/trackdem/index.html

Demographic Data Collection:

Genetic data collection, pedigrees,

Sampling technique Suitable taxa Suitable analyses Limitations Sample storage

Name Description Data set size Availability URL

Population Relationship link/running time 50 SNPs 605 SNPs

Aldra Maternal links 12 12 12 11 12 12 12 12

(a) distinguish homoplasy from homology and to estimate

30 and Cantor 1969), which assumes equal probability of

Name of approach: Optimisation principle:

Description Key benefits Key drawbacks Example programs

Name of approach: Optimisation principle:

Description Key benefits Key drawbacks Example programs

Name of approach: Optimisation principle:

Description Key benefits Key drawbacks Example programs

Name of approach: Optimisation principle:

Description Key benefits Key drawbacks Example programs

Biochemical and physiological

2.1 Introduction in functional ecology (Salguero-Gómez et al. 2018) or

Database name Description Link