Principles of Soilscape and Landscape Evolution ( PDFDrive )

Principles of Soilscape and Landscape Evolution
Computational models are invaluable in understanding the

complex effects of physical processes and environmental factors
which interact to influence landform evolution of geologic time
scales. This book provides a holistic guide to the construction of
numerical models to explain the co-evolution of landforms, soil,
vegetation and tectonics, and describes how the geomorphology
observable today has been formed. It explains the science of
the physical processes and the mechanics of how to solve
them, providing a useful resource for graduates studying
geomorphology and sedimentary and erosion processes. It also
emphasises the methods for assessing the relative importance of
different factors at field sites, enabling researchers to select the
appropriate processes to model. Integrating a discussion of the
fundamental processes with mathematical formulations, it guides
the reader in understanding which processes are important and
why, and creates a framework through which to study the
interaction of soils, vegetation and landforms over time.
G A R R Y W I L L G O O S E is a professor in the Faculty of Engineering

and Built Environment at the University of Newcastle, Australia.
His research focuses on understanding the spatial and temporal
dynamics of drivers of hydrology and erosion including
landforms, soils, vegetation, soil moisture and fire. He developed
the internationally utilised SIBERIA landscape evolution model,
for which he was awarded the Lorenz Straub Award. He later
customised SIBERIA to examine the long-term sustainability
of rehabilitated mine sites, and developed further models for
co-evolving hydrology, vegetation and soil for these sites.
This book was worth the wait! What started as a description of a
pioneer modeling effort thirty years ago has ended up as a
comprehensive treatise on soil and landscape evolution enriched
by the experience of Willgoose. Hydrologists and geomorpholo-
gists interested in a quantitative understanding of what goes on in
the critical surface zone of the geosphere must read this book.
Rafael L. Bras, Georgia Institute of Technology
If it moves, model it! There is no better synthesis of all the

various elements in landscapes and soil than this lifetime compil-
ation. Willgoose examines the many mechanisms operating in the
landscape, at scales from continental tectonics down to the soil
profile, demonstrating how he and others have built them into
functional, mutually consistent and inter-connecting models. Its
greatest strengths lie in the incorporation of soil processes –
physical breakdown, mixing and weathering; and in how prin-
ciples and models have been applied to the management of
degrading spoil heaps.
Mike Kirkby, University of Leeds
An outstanding synthesis that thoroughly addresses both the
theoretical basis and practical application of landscape evolution
modelling — a benchmark of its kind.
Stuart Lane, University of Lausanne
Principles of
Soilscape and
Landscape
Evolution
Garry Willgoose
The University of Newcastle, Australia
University Printing House, Cambridge CB2 8BS, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India
79 Anson Road, #06–04/06, Singapore 079906
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit
of education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9780521858793
DOI: 10.1017/9781139029339
© Garry Willgoose 2018
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2018
Printed in the United Kingdom by TJ International Ltd. Padstow Cornwall
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Willgoose, Garry, author.
Title: Principles of soilscape and landscape evolution / Garry Willgoose, The University of Newcastle, Australia.
Description: Cambridge : Cambridge University Press, 2018. | Includes bibliographical references and index.
Identifiers: LCCN 2017035905 | ISBN 9780521858793 (hardback : alk. paper)
Subjects: LCSH: Soil structure. | Soil physics. | Soils. | Soil science. | Landscape changes.
Classification: LCC S593.2 .W55 2018 | DDC 631.4/3–dc23 LC record available at https://lccn.loc.gov/2017035905
ISBN 978-0-521-85879-3 Hardback
Additional resources for this publication at www.cambridge.org/willgoose.
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
Contents
Preface page ix
1 Introduction 1
1.1 Why Mathematically Model 1
1.2 Modelling Philosophy: What Is a Model? 2
1.3 Model Framework: The Model of Everything 5
1.4 Scope of the Book 9
1.5 Further Reading 10
2 Constructing a Landscape Evolution Model – Basic Concepts 11

2.1 Overview 11
2.2 The Governing Equations 11
2.3 Nondimensionalisation and Scaling Analysis 20
2.4 Perturbation Analysis for Analysis of Stochastic Processes 22
3 A Brief Hydrology and Geomorphology Primer 26

3.1 Hydroclimatology 26
3.2 Modelling Climate Temporal Variability 36
3.3 Geomorphology 38
3.4 Conclusions 46
4 Erosion and Other Water-Driven Processes 49

4.1 Rainsplash and Rainflow 49
4.2 Erosion and Deposition 52
4.3 Temporal Averaging of Erosion 68
4.4 Hillslope to Channel/Gully/Rill Transitions 72
4.5 Channel-Specific Issues 74
4.6 Some Synthesis of Fluvial Erosion 80
4.7 Numerical Issues 81
5 Soils: Constructing a Soilscape Evolution Model – Basic Concepts 87

5.1 Overview 87
5.2 The Interaction between the Soil Profile and Weathering 88
6 Soils: Soil Depth 90

6.1 Introduction 90
6.2 Soil Depth and Bedrock Conversion to Soil 90
v
vi Contents
7 Soils: Physical Weathering and Soil Particle Fragmentation 96

7.1 Physical Weathering and Soil Mineral Matter 96
7.2 The Evolution of the Soil Surface 96
7.3 The Evolution of the Full Soil Profile 97
7.4 Sediment Deposition 114
7.5 Differentiating between Mineral Components 115
7.6 The Evolution of Porosity 116
8 Soils: Chemical Weathering 119

8.1 Introduction 119
8.2 Chemical Weathering Principles 119
8.3 Full Profile Models 122
8.4 Multicomponent Chemical Weathering 139
8.5 Hillslope Modelling 142
8.6 Some Final Observations on Chemical Weathering 145
9 Soils: Slow Soil Flow and Creep 152

9.1 Creep and Soil Velocity 152
9.2 Creep as a Slow Flow Process 155
9.3 Creep as a Cyclic Expansion-Compaction Process 158
9.4 Creep as a Random Motion Process 159
9.5 Creep and Bioturbation 160
9.6 High Slopes 160
9.7 Conclusions 161
10 Soils: Colloids and Soil Organic Carbon 163

10.2 Colloid Migration and Illuviation 163
10.3 Single Soil Layer Soil Organic Carbon Models 164
10.4 Multilayer Soil Organic Carbon Models 168
10.5 Erosion 172
10.6 Soil Organic Matter-Hydrology Interactions 173
10.7 Catchment and River Network-Scale Carbon Cycling 175
10.8 Conclusions 177
11 Soils: Constructing a Soilscape Evolution Model – Details and Examples 178

11.1 Soilscape Modelling 178
11.2 Coupling Weathering Processes 178
12 Tectonics and Geology 190

12.2 Isostasy and Isostatic Compensation 191
12.3 Mountain Building 199
12.4 Extensional Settings 201
12.5 Renewal of Terrestrial Crust 202
Contents vii
13 High-Slope Gravity Processes 205

13.1 Trigger Criteria 205
13.2 Transport Principles, Momentum and Energy 213
13.3 Rockfall Modelling 219
13.4 Discrete Element Models 222
13.5 Escarpment Retreat 223
14 Vegetation and Wildfire 228

14.2 Landforms, Soils and Vegetation 228
14.3 Dynamic Vegetation Models 231
14.4 Soil Moisture and Water Limitation 238
14.5 Wildfire 244
14.6 Other Forms of Vegetation Disturbance 253
15 Constructing a Landscape Evolution Model – Details 255

15.1 How to Couple the Processes 255
15.2 Model Testing 257
15.3 Model Verification 258
15.4 Model Calibration 262
15.5 Model Prediction and Validation 266
16 Examples of Landscape Evolution Modelling and Final Thoughts 270

16.1 Some Examples of Coupled Modelling 270
16.2 The ‘Future’ 286
References 292
Index 320
Preface
This book has been a long time coming. The original idea, model the key issues in the evolution of landscapes (land-
suggested by Peter Eagleson at my PhD defence in 1989, scapes being the combination of landform, soils, vegeta-
was to turn my PhD on landform evolution into a book tion and so on). The intellectual core of this book is the
co-authored with my PhD supervisors Rafael Bras and evolution of soilscapes. Soils are central because they
Ignacio Rodriguez-Iturbe. However, shortly thereafter constrain the evolution of (1) the landform through the
I became involved in applying my landform evolution soil’s interaction with the sediment transport processes
model, SIBERIA, to assess the long-term evolution and (2) the vegetation through vegetation’s dependence
(1,000–10,000 years in the future) of a rehabilitation on the water and nutrients stored in the soil. Moreover,
design for a uranium mine in Australia, Ranger Uranium soils are the single biggest store of carbon in the terrestrial
Mine (Willgoose and Riley, 1998a). It soon became clear environment and thus central to future trends in climate
to me that the characteristics of the eroding surface at change. The various storage processes within the soil
Ranger evolved much faster than the landform, and the profile determine the timing, and lags between cause and
major uncertainty on the safety assessment of the contain- effect, of almost all terrestrial processes. Soil’s importance
ment structure was the evolving erodibility of the land- is reflected in the Critical Zone research agenda where the
form surface, not the landform evolution per se. Only major component of the biogeochemical reactor that is the
millimetres of erosion significantly changed the erosion Critical Zone is the soil and its interface role in the water,
characteristics of landforms (which were many metres geology and atmospheric cycles on Earth (NAS, 2010).
high) long before any significant landform evolution was The focus of this book is on physically based model-
apparent. Parallel projects at other mine sites confirmed ling of processes, but I won’t shy away from empirical
that this behaviour was not unique to Ranger. Very and stochastic models if they are the best (current) fit. Yet
quickly the science focus of my work moved away from the book is not just about the mathematics of the processes
landform evolution into what we now call soilscape evo- because we first need to know what processes to model in
lution. The original 1989 book’s focus was too narrow. any particular situation. The book discusses the key con-
Unfortunately our progress in soilscape evolution model- cepts and the empirical evidence underpinning the math-
ling proved slow and computationally difficult, even more ematical representation of these processes, and studies of
so than the original development of SIBERIA, at least in when and where these processes may be dominant or
part, because there was very little existing quantitative otherwise. The book is not a comprehensive review of
work to build upon. The original book went on the back- the literature of landscape forming/evolving processes and
burner despite significant progress during short sabbatic- their implications. That would require a much, much
als at Lancaster University in 1995 and MIT in 2000. The bigger book. Hopefully, however, even those readers
final nail in the coffin was when I became seriously ill in whose primary interest is not modelling of landscape
2001 as a result of involvement in a flood-forecasting evolution may still find the discussions here worthwhile.
project in China, and the book was abandoned until Of course, where appropriate, I discuss, sometimes in
I recovered enough strength, some years later, to be able gruesome detail, the numerical methods and the
to revisit the project. In the meantime, landform evolution mathematics.
research had exploded worldwide. The range of processes that are potentially important in
The book you are holding is a very different book from a landscape evolution model is quite broad. To keep the
the original plan. It is a more holistic discussion of how to book treatment manageable, the book contains two types ix
x Preface
of chapters, process and synthesis. The process chapters the Mining Impacts Section) and Arthur Johnson (then
describe a specific set of processes in detail in a self- Institute Director) at the Environmental Research Institute
contained fashion independent, as much as is possible, of the Supervising Scientist (eriss) for supporting the
from the other process chapters. The variable length and original pioneering application of landform evolution
content of these process chapters reflect the variation in modelling to the assessment of the long-term 1,000-year
mathematical quantification heritage and potential, not waste containment safety at the Ranger Uranium Mine
any assessment on my part of their importance. The syn- back in 1990. This support is even more notable because
thesis chapters assemble those process chapters into what they assessed the likelihood of success at less than 1%,
we recognise as a landscape evolution model. These though fortunately they didn’t tell me that at the time. It’s
latter chapters focus on how the individual processes been a very productive 27-year collaboration, and
interact and how the processes are assembled into a counting. Among colleagues outside Newcastle there are
coherent model. a few that deserve particular thanks for their contribution
On the question of modelling philosophy my inclin- and collegiality. First among equals is Mike Kirkby who
ation is to favour simple models over complex ones. With has always had something interesting to say, a valued
all the nonlinear feedbacks in time and space even simple colleague when I was at Leeds University, and with whom
landscape evolution models can generate surprisingly I spent a very enjoyable week talking soils while writing
complex behaviour. Furthermore, complex models have this book. Others are Alex McBratney and Budi Minasny
the endemic problem of equifinality, where many of the at Sydney University who have been a mine of infor-
processes lead to the same or very similar behaviour, so it mation about mathematical soil science; the late Frank
is difficult to use the models alone to infer what processes Henderson (Henderson, 1966) who, when I was an under-
have caused the behaviour observed in the field. This graduate, impressed on me the power of mathematics to
complexity and equifinality tends to obscure the first order understand how processes work in the field; Keith Stol-
processes in the system. Some field workers have criti- zenbach at MIT who reinforced that same message during
cised my, and my modelling colleagues’, naivety in ignor- my PhD; and Bill Dietrich who after a very controversial
ing some process or other in our modelling, but I have AGU talk in 1987 (where I presented a mathematical
always been guided by this desire for simplicity over analogy between leaf veins and river networks) stressed
complexity. A colleague, Greg Tucker, once summed it the need to explain in plain English what the mathematics
up with a quote he had heard somewhere: ‘If we include in meant. The list of individual research sponsors is too long
our model every process that occurs in the field, then our to cite, but, aside from the ongoing collaboration with
model will be as difficult to understand as is the field’. eriss, special thanks go to the US National Weather Ser-
There are many people to thank. Rafael Bras and vice, which partly sponsored my SM and PhD at MIT; the
Ignacio Rodriguez-Iturbe started me on my landscape Australian Water Research Advisory Council (AWRAC),
evolution journey, and I know they were disappointed which sponsored my post-doc fellowship; the Australian
that the original book plan did not come to fruition. The Research Council, which has been an ongoing sponsor
soils chapters in this book were mostly written while on (notably with an Australian Professorial Fellowship in
sabbatical at the Community Surface Dynamics Model- 2006–2010); and the international mining and nuclear
ling Systems Facility (CSDMS) at Colorado University, waste industry.
Boulder, in 2015, and I thank James Syvitski for support- Some colleagues have been crazy enough to agree to
ing the quiet time I needed to write. I’d also like to thank review draft chapters, and the book is better for it. Thank you
Bob and Suzanne Anderson and Greg Tucker for interest- to (University of Newcastle unless otherwise noted) Pippa
ing conversations on soils and rivers during that time. Chapman (University of Leeds), Anna Giacomini, Eleanor
During my career there have been a number of PhD ‘Nellie’ Hobley (TU Berlin), George Kuczera, Budi Min-
students, post-docs and colleagues at Newcastle who have asny (University of Sydney), Gary Sheridan (University of
contributed to the landscape and soilscape research, Melbourne), In-Young Yeo and Omer Yetemen. Greg Han-
including Min Chen, Sagy Cohen, Ken Evans, Yeboah cock and Tom Vanwalleghem (University of Córdoba)
Gyasi-Agyei, Greg Hancock, Eleanor Hobley, George reviewed multiple chapters. Special thanks go to Natalie
Kuczera, Dene Moliere, Mariano Moreno de las Heras, Lockart (and Linden in utero, a budding earth system scien-
Hemantha Perera, Patricia Saco, Saniya Sharmeen, tist) for reviewing the entire text of the penultimate version.
Dimuth Welivitiya and Tony Wells. I particularly note Thanks also to all those colleagues who answered my ques-
the ongoing 25-year collaboration with Greg Hancock. tions about their published work. Of course, any remaining
I am especially indebted to Steven Riley (then Head of stupidity is my responsibility alone.
Preface xi
Finally I would like to thank everybody who person- then gave up. The reality is somewhat different. On a
ally supported me through the dark years after I fell ser- return visit to MIT soon after my PhD I was in Rafael
iously ill in 2001. I have thanked you all privately. I just Bras’s office with Ignacio and him. I was explaining my
want to put that thanks on the public record. And while latest work where I had derived an analytic solution relat-
she doesn’t feel she needs to be thanked publicly there has ing slope, area and elevation for landforms that were not
been my wife, Veronica Antcliff, who has put up with a in dynamic equilibrium (Willgoose, 1994a). Ignacio was
lot during my career and this book project. increasingly pacing up and down in the office, puffing on
Before we get to the serious stuff there is one story that his pipe faster and faster, the air getting thicker and
colleagues have insisted I recount. That is how my land- thicker, as he is wont to do when he is excited. After what
form evolution model, SIBERIA, got its name. Before seemed like 30 minutes of pacing and puffing (but was
I give the answer I’ll note that it’s not an acronym, and probably much less) he burst out in his thick Spanish
herein lies an amusing story involving researchers who accent ‘Garry, Garry, Garry . . . Garry, Garry, Garry . . .
should remain nameless to protect their reputation. These what on earth is this Siberia elevation stuff you are talking
researchers, on reading my early papers, were puzzled about?’ My Australian accent had defeated him, yet again.
why I had called it SIBERIA. Thinking that the name And I thought he was excited about the mathematical
was an acronym (and that perhaps the acronym had been result. I was deflated. That evening over a beer with Glenn
explained in a previous paper that they had not read) they Moglen (who was a PhD student at the time) I recounted
then spent several days over morning tea brainstorming this incident. Glenn burst out laughing and said that’s a
what SIBERIA was an acronym for. They got to E and great name for a computer code. And so it was.
1 Introduction
To look upon the landscape . . . without any recognition of the structures for containing mining and nuclear waste, there
labour expended in producing it, or of the extraordinary may not be any analogue to guide our management
adjustments of streams to structures and of waste to weather, decisions.
is like visiting Rome in the ignorant belief that the Romans of
today have had no ancestors.
(Davis, 1899, p. 496) 1.1 Why Mathematically Model
There can be no better, nor more succinct, description of In the absence of exact physical analogues or scaled
the motivation for, and content of, this book. For contin- model experiments (e.g. using nondimensionalised par-
ental and global environmental modelling applications we ameters), if we wish to quantify landform evolution, then
now have high-quality global databases of topography we are forced to rely on mathematical models to predict
(e.g. the Shuttle Radar Topography Mission [SRTM]), the future and guide decision-making. These models may
an emerging database for rates of change of topography range from empirical multiple regression correlations (e.g.
(e.g. from the GPS, GLONASS and Galileo global gridded soil data obtained by correlating soil chemistry
positioning systems, and the GRACE gravity mission), laboratory results with gridded landscape attributes such
and a global program to deliver high spatial resolution soil as topography; Holmes et al., 2004) through to graphical
functional and chemistry data (the GlobalSoilMap methods, or they may be computer models. Ideally the
program). For smaller site-specific applications LIDAR complexity of the model adopted for any particular study
provides high-resolution and accuracy topography is a function of the domain knowledge, the computational
(Tarolli, 2014), ground-penetrating radar and other tractability of the science and the project objectives (e.g.
proximal remote sensing technologies (e.g. gamma Larsen et al., 2016). Computer models can range from
spectroscopy) provide spatially distributed soils data, simple scripts in MATLAB or Python, to complex code
and sensors on drones can provide spatially distributed with thousands of lines in Fortran or C. This book is about
vegetation and soil chemistry data. While Davis could not these models.
foresee these datasets, his argument that for maximum Many existing computer codes in landscape evolution
benefit we need to know how the topography, soils and modelling tend to be complex. In some cases that com-
vegetation arose is as valid today as it was in 1899. plexity is essential, in others not so essential. In some of
By understanding how the landscape system changes the chapters in this book we will use quite short and
with time, then, we will be better able to diagnose why it simple computer models because these are best to display
is changing. This will allow us to better manage our the implications of the physical principles. These codes
landscapes, differentiate man-made impacts from natural can be downloaded from GitHub and other open source
evolution, and potentially predict (or at a minimum assess sites, or can be obtained from the author directly.
the risks of) future evolutionary scenarios and any anthro- These models are only as good as the process repre-
pogenic interventions. While paleo-analogue and current- sentations incorporated into them. Typically the generic
day analogue sites can indicate likely behaviour, their process representations are based on a synthesis of field,
usefulness depends on how well we can generalise from experimental and theoretical work. For quantitative appli-
their analogue behaviour to the site that we are interested cation at any specific site there will almost certainly be
in. That may be difficult. In some cases, such as the some need for site-specific customisation. This customi-
author’s engineering applications of designing man-made sation will involve the selection of what processes are 1
2 Introduction
dominant at that site and a calibration of the parameters of and soil flora and fauna. Much of this book is about the
those processes at that site. Most of the chapters in this evolution of the soils, and how they interact with the
book are about the generic process representations that are evolving landforms and the environment. However,
accepted in the respective research and applications com- the terms landscape evolution model, soilscape evolution
munities, and some characterisation of where and when model and dynamic vegetation model are a bit of a
these processes are important. As part of the presentation mouthful, so we will abbreviate references to them
of these processes there will be some discussion of pro- throughout the book:
cess rates and model parameters, and general information
about how they have been derived. We will discuss cali- • LEM: Landform evolution model, modelling only the
evolution of the landform/topography and nothing else.
bration and validation briefly in a moment, but the topic of
This is currently the common usage.
site-specific process calibration (i.e. experimental design,
parameter calibration and model uncertainty analysis) is • SEM: Soilscape evolution model, modelling the evolu-
tion of the soils across the landscape and nothing else
quite important for defensible quantitative landscape evo-
(i.e. assuming the landform is fixed). Soilscape is the
lution applications and will be addressed in detail in
term used in the soil science community to refer to a soil
Chapter 15.
model (typically without any time-varying evolutionary
One problem with complex models is that it can be
component) applicable across a landscape (e.g. a grid-
hard to understand how they work and identify the cause-
ded digital soil map), and the term soilscape distin-
and-effect linkages. For researchers this makes them dif-
guishes these types of models from soil profile/pedon
ficult to work with because there is a significant learning
models that model the vertical profile of the soil at a
curve to climb before we can be completely confident that
given point in isolation. A soil catena is a subset of
we can apply these codes to our (novel) sites. Moreover,
soilscape modelling, typically applying to individual
in very complex models a single person cannot hope to be
hillslopes rather than an entire catchment.
completely familiar with every aspect of the model, par-
ticularly when a model has been extended numerous times • SLEM: A coupled soilscape and landform evolution
model, a model that links an LEM and SEM.
by a series of PhD students and post-docs who have
subsequently moved on to other things. • BioSEM, BioSLEM: One of an SEM or SLEM that has
a coupled biophysical model. A BioSEM may have an
In conclusion, if possible, simple models are best.
active biological weathering component to the soil
They allow simple transparent explanations. That is the
profile development (e.g. organic acids from the break-
guiding principle throughout this book.
down of soil organic matter, fungal breakdown of rock
This book is about much more than landforms alone.
fragments). A BioSLEM might feature the coupling of a
A central rationale is to support the emerging field of
vegetation model (typically something like the dynamic
soilscape evolution modelling. Many authors over many
vegetation models [DVMs] that are typically coupled
years have recognised the link between soils and the
with ecohydrology and climate models) with an SLEM.
landforms over which they drape (e.g. Jenny, 1941,
1961; McBratney et al., 2000) and that soils evolve in When I think of a landscape evolution model I think of
response to the climate, geology and biosphere (e.g. something like a BioSLEM. This book will touch on the
Birkeland, 1990; McFadden and Knuepfer, 1990; construction of all four types of models, with a heavy
Schaetzl, 2014). It seems only natural that these two views emphasis on SEM.
should be merged into the ongoing work with landform
evolution models to create coupled models of soilscape
and landform evolution. Many gaps remain to be filled 1.2 Modelling Philosophy: What Is a Model?
(particularly with regard to quantification), but the field is
now sufficiently mature that we can see a way forward Before we get down to the nuts and bolts of model
and conceptualise what a coupled model should look like. making, it is worthwhile to reflect on the applications for
This book pulls together the different threads of work in a (landscape evolution) model. These applications will
soilscape and landform evolution to provide this intellec- influence the way we construct and use our models. It is
tual framework. often said, ‘All models are wrong, it’s just that some are
The abbreviation LEM has been used interchangeably useful.’ While this is a gross overgeneralisation, this does
in recent years to refer to either landform evolution capture the idea that models are approximations of the
models or landscape evolution models. Landforms are just field, but that they capture to first order the characteristic(s)
one component of landscapes, along with soils, vegetation of interest. They do not replicate everything in the field
1.2 Modelling Philosophy: What Is a Model? 3
because of the assumptions and approximations that are complex models, while having the kitchen sink of all
either implicit or explicit in the construction of the model. processes built into them, tend to be less well suited to
Models have two main uses: (1) numerical experimen- this problem because it can be hard to be confident that
tation tools to improve our understanding of natural you fully understand all the assumptions built into the
systems and/or (2) predictive tools to predict behaviour model. Also, large complex models more commonly have
when the system is modified in some way. problems with equifinality (e.g. Hancock et al., 2016).
Equifinality (Beven, 1996) is where (1) two processes or
1.2.1 Numerical Experimentation Mode (2) multiple combinations of parameters lead to similar
responses so it is difficult to discern which process/
In this mode models are used in the same way as field and parameters actually cause the observed behaviour.
laboratory experiments. Experiments are done where
components of the physics are changed to explore 1.2.2 Prediction Mode
changes in behaviour. These changes may be changes in
rate parameters or parameters that change the interactions In this mode models are used for making predictions of
and feedbacks, or may involve complete substitution of behaviour (1) at other sites or (2) for future situations.
one set of physics for another. In this way we can quanti- These predictions may then be compared with some
tatively discern the impact of changes in process proper- societal, environmental or regulatory requirements to see
ties and how they change system response. The main if the modified system’s behaviour is satisfactory. In
problem with this is that, just as with field and laboratory principle our numerical experimentation models can be
experiments, it can be difficult to determine whether the used for this problem, but there are a number of subtle
observed changes in system response occur only for the differences. The most important difference is that the
parameters adopted for those experiments (or a particular model must be capable of predicting system behaviour
site for field experiments). The main advantage that for the criteria that are used for assessment and ideally
numerical experiments have is that it is possible to com- provide a measure of how reliable that prediction is. In
pletely control the system. The user has the ability to hold some cases this acceptance criteria may be relative to
some parameters steady while changing others, something unmodified behaviour (e.g. how much better will the
that can be difficult to do for field catchments and even system be after modification, expressed as, say, a percent-
sometimes in laboratory experiments. By doing this it age or absolute change relative to no modification).
may be possible to ascertain how to design an experiment It is common to observe that the exact deterministic
that can distinguish and test competing hypotheses for results from a landform evolution model are sensitive to
field behaviour. For instance, if changing a process many model inputs. While small changes may change the
(parameters, mathematical formulation and so on) does average behaviour of the model only a small amount (e.g.
not change some characteristic we can observe and meas- average erosion rate across a catchment), these same
ure in the field, then, at least for this process, there is no changes may dramatically change the location of specific
value in performing such an experiment because the field features (e.g. the two simulations may have a completely
experiment will also probably not show any significant different location for a gully). Thus if the acceptance
differences. In this sense using models for numerical criterion is a limit on the maximum erosion at a specific
experimental laboratories is no different from the way location, it may be difficult to demonstrate that the modi-
science was done before computers, it is just that we can fied behaviour at a specific location is better, even though
now do preliminary tests of what to expect in the field and the average for the entire landform is, in fact, better.
hopefully design more robust experiments. This does not Moreover, if small changes in model inputs have a dra-
negate the value of purely curiosity-driven experiments to matic effect on the location of a gully, it may be impos-
‘see what happens’. However, it does impose rigour into sible to convince regulators that the model predictions are
experiments where quantitative conclusions might be useful indicators of the performance of proposed field
possible about cause and effect. Without the model it is modifications.
not possible to know if the proposed processes, when Finally, if project proponents (e.g. a low-level nuclear
quantified, actually lead to the quantitative effect that is waste repository that needs to be shown to be safe for
observed. Quantification is key. 10,000 years – current requirements in the United States
The type of model that is well suited to numerical and Australia for radioactive mining waste and low-level
experimentation is one where it is easy to change the nuclear waste) are to spend large amounts of money, they
construction and/or parameters of the model. Large need to know that the quantitative predictions of the
4 Introduction
model are accurate, or at least justifiable to regulators and something that is not modelled in the agricultural erosion
defensible in court. Regulators and courts, for instance, models, so differences become apparent only over the
will often want to see short-term predictions of behaviour longer term.
that can be tested quantitatively against observed behav- One unique aspect of landform evolution models (and
iour to assess whether the model is ‘correct’ or not. it is not yet clear whether this is also true of soilscape
evolution models; Phillips, 1993) is they show sensitivity
1.2.3 Model Calibration, Validation and to initial conditions and external inputs so that the deter-
Uncertainty Analysis ministic output (e.g. the exact values of elevation at every
node in the model) can vary dramatically with only small
Finally, we need to be able to test our models. This is changes in inputs or parameters (e.g. Willgoose and
challenging because we are dealing with systems that Gyasi-Agyei, 1995; Willgoose et al., 2003). Channel and
evolve over thousands to millions of years, and because valley locations can move significant distances laterally
it is unusual to have situations where we can do independ- for only small changes in model inputs. Thus it is not
ent replicates of landform evolution. Thus we are left with possible to make a direct comparison of the exact values
the difficult situation that our models will not match at each location of the model and the field data. Rather it is
observed data perfectly (since they are approximations necessary to identify statistics of the landforms that will
of the field) and we don’t know how much difference be the same if the physics is modelled correctly and then
between the models and the field is acceptable. Moreover, compare these statistics. One might then ask, What is the
we may be able to calibrate our models to a specific field set of essential statistics that is required to completely
case (i.e. change the physics and its parameters until we describe a landscape? This set of essential statistics can
get a satisfactory fit), but this demonstrates only that the then form the focus of any model testing.
model is feasible (i.e. with appropriate parameters the Key outputs needed for model testing are confidence
model can fit the data), not that the model is correct. limits on model predictions or confidence limits on field
Indeed, the hypothesis-testing literature (and a computer measurements. For instance, the t-test is commonly used
model is simply a hypothesis written in a computer lan- for testing the differences between models and experi-
guage) indicates that we can never prove a model correct, ments. The t-statistic is the difference between the two
we can only prove it incorrect. It may fit all the current experiments divided by the variability of the experiments,
data, but the data to disprove it may be just around the and this variability can be determined from the confidence
corner. This ability to reject a model is called falsifiability limits. As we have mentioned, field replication of land-
and is a key component of model testing (Popper, 1959). scapes is difficult, so it is rarely possible to calculate
A related aspect of this is that a good model makes confidence limits on field measurements. An alternative
predictions that can potentially be falsified. If the model is to calculate confidence limits on model output. In
cannot make predictions that can be falsified, then it is not principle this is simple if we have accuracy estimates on
possible to test it (Pfister and Kirchner, 2017). model parameters and exogenic inputs. We can use Monte
Nor can we blindly adopt similar physics from other Carlo simulation with the model. This process is to repeat-
types of models that have been tested for different appli- edly run the model with different parameters, where the
cations. For instance, there are numerous well-validated different parameters are determined by sampling the
and tested agricultural erosion models (e.g. RUSLE, parameter values from the probability distribution of the
CREAMS, WEPP). They are based on one of the broad model parameters. In this way we can generate a number
range of shear stress–driven detachment and sediment of output simulations, and we can derive the probability
transport models (see Chapter 4). They have all been distribution of the outputs for the model and thus deter-
empirically fitted to field data so that they all satisfactorily mine confidence limits on the predictions (e.g. Willgoose
predict erosion over relatively short time scales of a few et al., 2003). Likewise we can explore the effect of
years for paddock-scale problems. However, when unknown climate by running simulations with randomly
applied over long time scales of millennia or more, the generated climate data to explore the effect of climate
specific parameters used in the different transport models variability. The problem with this process is that it can
yield qualitatively different landforms, even though over be extremely computer intensive. Depending on how
agricultural time scales of a few years the differences many parameters need to be varied, it is common for
between the models are small (e.g. Willgoose and thousands of simulations to be necessary.
Gyasi-Agyei, 1995). The reason for this is the long-term This problem of excessive computer time is even
feedbacks between changes in topography and erosion, worse for satisfying many regulatory requirements in
engineering applications because it is common that will be presented in three stages starting at the largest
acceptance criteria are expressed as a small probability length scale, the continental scale, zooming down to the
of some criterion or criteria being exceeded (e.g. one in catchment/river network scale, and then zooming in until
one million chance of failure in any one year is common we reach the smallest length scale, the hillslope scale.
for high-consequence industrial applications). Thus not
only is it necessary to do Monte Carlo simulation to obtain
the probability distribution of the output, but the small 1.3.1 Continental Scale
probability required for acceptance means that the results
Figure 1.1 overviews the important processes at the con-
we are trying to determine are on the tail of the probability
tinental scale. At this length scale (10–100 km) the weight
distribution. The further out we are on the tail of the
of the land on the crust has significant interactions with
probability distribution, the greater is the number of simu-
the tectonics of mountain building. The loss of mass by
lations that are required to adequately sample enough rare
erosion lightens the load on the crust and the crust rises in
events to define the tail of the probability distribution.
response by isostacy (Section 12.2). If mountains are
The key conclusion from this is that to be able to
being created by convergence of two plates, then this
provide confidence limits on model simulations it is cru-
unloading by erosion also has significant feedbacks on
cial to have fast models. Thus it is not simply enough to
the rate and nature of the interaction between the two
be able to simulate some processes, but it is also necessary
plates underneath the emerging mountain range (Section
to be able to simulate these processes efficiently so that
12.3). An important interaction is that the topography of
confidence limits can be generated in a reasonable com-
the mountain range changes the spatial distribution of
pute time. The book will return to this issue of algorithmic
rainfall (Section 3.1.1) with higher rainfall on the wind-
efficiency repeatedly.
ward side of mountains than the leeward side. However,
not only does this increased rainfall increase runoff (and
1.3 Model Framework: The Model of everything else being equal increases erosion), but it also
Everything increases the vegetation density (increased density
decreases erosion), and it is the competing effects of
The title of this section is a tongue-in-cheek description increased runoff and increased vegetation that determines
by a colleague of the vision of this book and what consti- whether the erosion rate is higher on the rainy side or the
tutes a ‘complete’ landscape evolution model. Given the drier side (Section 16.1.2).
complexity of the processes and interactions, it is worth The rivers (and the hillslopes linked to them) are the
taking a step back for a moment to present a big picture transport mechanism to remove sediment from the moun-
overview of the landform evolution system: What are the tain ranges to the ocean (Chapter 4); these river network
components of the modelling framework and how do and hillslopes will be discussed below.
these components interact? This framework informs the While this book will not cover marine processes, the
organisation of the chapters that follow. This overview fate of the eroded sediment is shown in Figure 1.1. Most of
FIGURE 1.1: A schematic of the processes that occur at the continental scale including mountain building, isostatic rebound, river
erosion, spatially variable rainfall and deep marine deposition. The heavy grey arrows are process fluxes and their predominant direction
of action. The two circular arrows represent the feedbacks between (left) erosional reduction of elevation and rising in the crust, and (right)
deep-sea deposition and depression of the crust.
6 Introduction
the particulate and dissolved load is deposited either on the Broadly speaking the river network consists of two
continental shelf (either in the delta at the river mouth or types of channels (Sections 4.4 and 4.5): (1) bedrock
farther offshore) or in basins in the deep ocean beyond the channels where the river flows across bare exposed bed-
shelf. The extra load from this deposited material results in rock and (2) alluvial channels where the river flows
the crust being depressed downward; that is, not only does through deposited sediment. These two channel types
erosion on the land interact with the crust and mantle so are end members in a transition from bedrock channels
the crust rises, but the deposited sediment in the ocean also to alluvial channels as the sediment load in the river
interacts with the crust and mantle, so the crust falls. increases, but this distinction suffices for the moment.
The continental shelf is an important transition region Alluvial channels occur in regions of deposited sediment
between terrestrial processes and marine processes. The called floodplains and typically meander from side to side
depth below current sea level of the ocean edge of contin- within this floodplain. Bedrock channels are typically
ental shelves worldwide is of the order of 140 m, which constrained laterally by the erosion-resistant bedrock
coincides with the level of the ocean during the ice ages of channel sides and so tend to be straight rather than
the last million years. Ice core data indicate that about meandering. The floodplain is a result of net deposition
80% of the last million years has been ice age, while the of sediment on the floodplain by the alluvial channel
current, interglacial, conditions (since about 10,000 years (typically at high flows), and these sediments are remobi-
ago) have prevailed for less than 20% of this period. Thus lised as the river meanders from side to side (typically at
the topography of the continental shelf regions around lower flows). The residence time of the sediment within
many continents will largely reflect terrestrial processes, the floodplain is measured in many thousands of years.
with about 10,000 years of recent marine processes over- Recent work has also indicated that floodplains are a
printed on them. major storage site for organic carbon (Section 10.7).
Some final points to note about Figure 1.1 are the Thus the process of erosion of the mountains is one of
following: delivery of sediments from the hillslopes to the channel.
This sediment (both mineral and organic matter) then
1. The arrows for uplift of the crust under the mountain passes through a mixture of bedrock and alluvial channels
range are not vertical, because mountain building on its way to the ocean. In the alluvial channels there is an
involves convergence of the crust as well as vertical exchange of sediment between the channel and the
uplift. Moreover there is the possibility of relative floodplain deposits with some of the channel sediment
lateral motion horizontally between the converging being deposited in the floodplain, while some of the flood-
plates (e.g. the San Andreas fault in California). plain materials are remobilised from the floodplain
2. The rainfall is spatially and temporally variable to deposits (Sections 4.5 and 10.7). Typically the finest,
reflect the effect of the changing topography on rain- lightest and most mobile sediments will pass through the
fall. Note that this spatial distribution of rainfall may floodplain without being exchanged with the floodplain
vary seasonally, and year to year, as a result of changes sediments (Section 4.2.1.3). This leads to what is com-
in the prevailing wind directions. Some signature monly referred to as a ‘strings and beads’ plan of the river
research sites have broadly consistent prevailing wind network where the strings are channel reaches without
directions (e.g. Taiwan, New Zealand, Chile), while adjacent floodplains and the beads are river reaches with
for other sites wind directions even today vary in their adjacent floodplains.
direction seasonally as a result of seasonal monsoons Figure 1.2b shows a schematic of a river long profile
(e.g. the Himalayas). showing the transition from a bedrock channel reach to an
alluvial channel/floodplain reach and back again. For low
1.3.2 Catchment and River Network Scale sediment loads the bottom of the bedrock channel will be
clear of sediment so bedrock incision will erode the chan-
Figure 1.2 overviews the important processes at the catch- nel. For high sediment loads the bottom of the channel in
ment and river network scale. At the continental scale the the bedrock reach will be periodically covered by sedi-
main transport process for sediment is the channel net- ment waves and pulses that move down the reach. Thus at
work. Most of the sediment that is transported by the river high sediment loads the bottom of the bedrock channel
network comes from the hillslopes, either by (1) mass will be partially protected by sediment. Incision can occur
movement (e.g. landslides) in parts of the river network only when the bedrock channel base is not covered by
that are steep or (2) by fluvial erosion and soil creep in sediment. Thus the bedrock channel will incise into the
flatter areas (Figure 1.2a). bedrock periodically, with the percentage of time in the
FIGURE 1.2: A schematic of the

processes and interactions in
channel networks at the
catchment scale: (a) a plan of the
river network showing the ‘string
and beads’ arrangement of
channels and floodplains, (b) a long
section along a river showing the
periodic change in channel type
from bedrock to alluvial and back
again, (c) a typical alluvial channel/
floodplain cross section at low
flow showing the lateral
movement of the channel (left to
right in this figure) within the
floodplain as a result of
meandering and (d) a typical
alluvial channel/floodplain cross
section at high flow showing the
excavation of the channel cross
section down to and exposing the
underlying bedrock, and the
deposition of particulate matter on
the floodplain.
bedrock incision mode being a function of the percentage are deposited on the floodplain (Figure 1.2d). The resi-
of time that the bedrock bed is not covered by sediment. dence time in the floodplain deposits is then a function of
Likewise for the alluvial channel there will localised the rate of floodplain excavation by meandering relative to
stretches where the bedrock underneath the floodplain is the width of the floodplain. The residence time is a key
exposed. Moreover, during high flows the channel is parameter for organic matter sequestration in the flood-
scoured and can be excavated down to bedrock plain because it determines how much of the deposited
(Figure 1.2d). Thus periodically an alluvial channel can organic matter decomposes in situ versus being remobi-
be incising bedrock (and thus the bedrock under the flood- lised by bank erosion at some time after its original
plain is lowered), though for most of the time and for most deposition.
locations the channel is largely in balance with the
meandering channel excavating the floodplain sediment 1.3.3 Hillslope Scale
on the outside of the meander bend and depositing inside
the bend, so that there is an exchange of sediment and Figure 1.3 overviews the important processes at the
organic matter between the channel and the floodplain hillslope scale. The four panels overview different com-
(Figure 1.2c). At high flow, sediment and organic matter ponents of the hillslope processes: (a) mineral transport,
8 Introduction
(b) soil organic matter and carbon transport, (c) water

balance and (d) vegetation and fire. The hillslopes deliver
the material that is then transported to the coast by the
rivers. Most of the material transported by the rivers has
been sourced in some fashion from the hillslopes, so the
characteristics of the material eroded from the hillslope
determine the material in the river and the rate of river
transport.
Starting with the mineral matter (Figure 1.3a) the main
transport mechanisms are fluvial erosion (Chapter 4), soil
creep (Chapter 9) and mass movement (Chapter 13). The
fluvial erosion is largely a function of runoff generation
from rainfall (Chapter 3). The main modifier of the fluvial
erosion rate is the groundcover provided by vegetation and
its leaf litter (Chapter 14) though the canopy also provides
some protection, but it is less important than groundcover.
The dynamics of soil production and weathering influence
the characteristics of the sediment on the surface of the
soil. Soil production is the process of conversion of the
underlying saprolite into soil, and its rate is a function of
the thickness of the soil (Chapter 6). Within the soil profile
a range of weathering processes are active that are (1)
breaking the soil fragments into smaller and smaller frag-
ments (Chapter 7) and (2) chemically transforming the soil
rock fragments into secondary minerals (mostly clays) and
leachate (that is transported out of the hillslope by water as
dissolved load) (Chapter 8). The soil is vertically mixed by
biological processes called bioturbation (Sections 7.3.5,
8.3.4). The roots of vegetation provide much of the
strength of the hillslope to resist landsliding and other
forms of mass movement (Chapters 13 and 14). Finally
at mid-latitudes the aspect of the hillslopes can make a
significant difference to vegetation cover, so many of the
transport processes above vary depending on whether the
hillslope is polar-facing or equatorial-facing.
Figure 1.3b shows the organic matter cycle on the
hillslope. The main processes are vegetation providing
organic matter, in the form of leaf litter, which is incorpor-
ated into the soil profile (Chapters 10 and 14). Much
organic matter is generated within the profile from plants
roots (the rhizosphere), mycorrhizae hyphae and micro-
biology. Mycorrhizae hyphae (i.e. fungi roots) are import-
ant in developing soil structure and increasing soil
infiltration rates, and by mass are the most significant
CAPTION FOR FIGURE 1.3: (cont.) and (d) vegetation and fire.
FIGURE 1.3: A schematic showing two hillslopes draining down The heavy grey arrows show process fluxes and their predominant
to a central channel. The four panels each show the processes direction of action. The aspect label indicates those processes and
operating for the four main components of the hillslope system: properties that are different between polar- and equatorial-facing
(a) soil mineral matter, (b) soil organic matter, (c) the water balance hillslopes for mid-latitude catchments.
1.4 Scope of the Book 9
component of soil organic matter (SOM). In the same way impacts of vegetation. Fire removes groundcover vegeta-
that bioturbation mixes the mineral matter from top to tion and leaf litter layer and, if intense enough, the canopy
bottom in the soil profile, so bioturbation mixes the vegetation as well. The loss of ground protection by the
SOM. Within the profile the SOM decomposes over time, removal of groundcover and leaf litter results in fluvial
and the decomposition rate is initially fast for fresh SOM erosion increasing dramatically until the vegetation
as the most reactive components decompose. The decom- recovers in a year or two. A further fire impact is that
position rate slows over time since the remaining SOM the SOM in the top of the soil is volatised so that the soil
contains a greater percentage of components more resist- structure created by SOM (mostly mycorrhizae hyphae
ant to decomposition (Chapter 10). SOM is preferentially and polysaccharides) is destroyed. This results in reduced
eroded (i.e. sediment is enriched with SOM) at the surface infiltration, increased runoff and increased erosion. For
relative to the mineral matter because it is less dense than those trees killed by fire their root strength is reduced so
the mineral matter. Finally the decomposition of the SOM that the resistance to mass movement and creep provided
(as well as respiration from vegetation roots; see below) by plant roots is reduced, and mass movement rates
generates carbon dioxide within the soil profile. Carbon increase for a decade or so after fire. The charcoal that is
dioxide levels within the soil profile can be very high created by fire is incorporated into the soil, and since it is
(near 100%), and when this dissolves in water it generates decomposition resistant, it is effectively sequestered for-
acid that drives chemical weathering of the soil mineral ever within the soil. The main impact of aspect on vege-
matter (Chapter 8). The main impacts of aspect on the tation is that the differences in vegetation and soil
organic matter cycle are in changing vegetation density moisture lead to different fire recurrence rates and inten-
(and thus the generation rate for SOM source material) sity. The denser the vegetation the higher the fire inten-
and soil temperature (the SOM decomposition rate sity, while the lower the soil moisture the dryer the fuel
increases strongly with soil temperature). (e.g. leaf litter), leading to higher fire intensity.
The water cycle is illustrated in Figure 1.3c. For most What should be clear from the complexity of Figure 1.3
of the processes in this book the main characteristic of is that many factors on the hillslope interact and may
importance is the soil moisture because it influences (1) result in either positive or negative feedbacks. Thus mod-
the growth rate for plants, (2) the amount of soil organic elling the evolution of only the mineral matter processes
matter stored in the soil (wetter soils biodegrade faster), (as was common for the early landform evolution models)
(3) the rate of chemical weathering in the soils and (4) the independently of the soil organic matter, vegetation, fire
initial wetness of the soil when rainfall occurs so that and water is modelling only one part of a highly con-
wetter soil generates more runoff. The main sources that nected story. The challenge is that some parts of the
increase the soil moisture in the water cycle are rainfall processes in Figures 1.1–1.3 are quite difficult to quantify,
and infiltration. The main sinks that decrease soil moisture so modelling them is difficult. This difficulty is no reason
are (1) bare soil evaporation, (2) plant transpiration to ignore them, however, and the aim of this book is to
(a function of plant growth rate, Chapter 14), (3) surface provide an overarching framework within which we can
water runoff (a driver of fluvial erosion, Chapter 4) and quantify these processes and test their impacts on land-
(4) groundwater flow and leachate transport (Chapter 8). scape evolution.
Higher soil moisture is associated with higher plant
growth rate, surface runoff, creep and landsliding rates 1.4 Scope of the Book
and lower fire occurrence. The main impacts of aspect are
(1) the infiltration rate is changed by changes in soil This book focuses on modelling principles, not modelling
structure due to SOM and soil grading and (2) higher results. Of course, we are guided by what questions we
transpiration from vegetation on equatorial-facing slopes. wish to answer, but a book including extensive discus-
Finally, the vegetation and fire cycles are summarised sions of modelling results (and by necessity an extensive
in Figure 1.3d. The main feedbacks from vegetation are in discussion of, and comparison with, field behaviour) is
(1) plant roots, which increase the strength of the soil beyond the scope of this one. Likewise the book is not
making it more resistant to landsliding, (2) groundcover primarily about the underlying physics that we are mod-
and leaf litter, which provide soil surface protection elling other than to focus on what questions we wish to
against fluvial erosion, and (3) root respiration, which answer, and this will determine what physical principles
generates carbon dioxide within the soil profile, a source are important. Finally, all models are approximations,
of acid active in chemical weathering. The main impacts and we need to understand both our problem and its
of fire on landscape evolution are by modifying the dominant physics to understand which approximations
10 Introduction
are appropriate. While we would like our models to be as an evolutionary framework and exemplifies some of the
realistic as possible, sometimes that is not feasible or key work in the area using these models.
practical. We may not fully understand the physics, in Chapters 12 through 14 examine the nonsoil landscape
which case we may use the model as a numerical experi- evolution processes. They discuss the various components
ment as discussed in the previous sections. Even if we of a landscape evolution model that characterise how
fully understand the physics we may not have the landforms evolve. Some physical processes such as creep,
required computer power to fully solve the problem, so typically discussed as parts of landform evolution models,
we are forced to simplify the model formulation to gen- appear in the soilscape section because they are truly soil
erate any solution whatsoever. processes rather than landform processes. Accordingly the
Accordingly, the focus here is to provide the reader with observant reader will find that organisation of some topics
a broad holistic background of the processes that might within chapters purposely departs from the traditional
need to be modelled in any particular situation, and how organisation of discussions about landform evolution
modellers currently solve these processes. The aim is to models.
provide the reader with enough intuition to understand why Chapter 15 discusses how the landform evolution
modellers solve problems with the methods they do, and components fit together. This chapter covers the important
with enough background to launch into the gruesome topics of calibration and validation of soilscape and land-
details in the scientific literature when required. scape evolution models, and uncertainty analysis of their
The focus of this book is on temperate terrestrial pro- output.
cesses. Unfortunately, that has meant leaving out specific Chapter 16 concludes with some examples of coupled
discussion on marine and associated processes (e.g. sub- landscape evolution models. The linkage between land-
aerial, submarine and estuarine processes), the cryosphere form, soilscape and vegetation evolution models is exem-
(e.g. snow, glaciers and ice sheets), extra-terrestrial terrains plified here. It ends with a little future-gazing at what I see
(e.g. Mars, Pluto), aeolian landforms (e.g. dunes) and as new and challenging applications, and some of the
anthropogenic effects (e.g. farming, land degradation, for- unresolved science challenges.
estry, waste repositories). Furthermore, other than high- Readers can download the codes used in this book to
lighting links between vegetation and soil organic carbon, exemplify the evolutionary processes discussed at Github
soil microbiology and its impacts on soil chemistry is also (search for 'Willgoose soilscape') or from the Cambridge
not discussed. This was a pragmatic decision to keep the University Press webpage for this book (www.cambridge
scope of the book manageable, even though it is recognised .org/willgoose).
that these processes have shaped some terrestrial locations.
The book is organised into a number of logical sections.
Chapter 2 discusses in general terms the principles 1.5 Further Reading
underpinning a landscape evolution model, and how the
mathematical components fit together in a mass and For anybody interested in the history of landform evolu-
energy balance framework. tion modelling there are two key references: Leopold et al.
Chapter 3 presents a general climate, hydrology and (1964) and Carson and Kirby (1972). While showing
geomorphology background that will be repeatedly used their age, these two books, through their focus on funda-
throughout the book. mental processes, still provide good overviews and
Chapter 4 discusses fluvial erosion and deposition. It is insight. A more current overview, albeit not modelling
discussed early in the book because it is central to both specific, is provided by Anderson and Anderson (2010).
soilscape and landform evolution. An excellent overview of current challenges in modelling
Chapters 5 through 11 discuss soilscape modelling. soils and processes within the soils is provided by Ver-
They discuss the various components of a soilscape evo- eecken et al. (2016).
lution modelling framework and how it involves aspects For the areas not covered in this book, suggested
of pedogenesis and soil transport. Each of these chapters readings are anthropogenic interactions (Wainwright and
is relatively self-contained, and the totality of the chapters Millington, 2010), engineering applications (Willgoose
describes how to model soilscape evolution on a landform and Hancock, 2010), glaciers and cryosphere (LeB
that doesn’t change significantly with time. Chapter 11 Hooke, 2005) and submarine processes (Peakall and
focuses on how the soilscape components fit together in Sumner, 2015).
2 Constructing a Landscape Evolution
Model – Basic Concepts
2.1 Overview for erosion, the sign convention defined to be consistent

with z). The s subscript indicates that it is a flux of
In this book we will be discussing the components sediment, the x, y subscripts indicate the flux in the x
involved in the construction of a landscape evolution and y directions, respectively, the a subscript indicates that
model, and the mathematical and numerical representation it is the actual sediment transport to distinguish it from
of those components. This landscape evolution model will potential sediment transport to be used later in the book.
simulate landforms, soils and vegetation. The details of The sediment transport is a vector quantity pointed in the
the components will follow in Chapters 4 and beyond. direction of sediment transport and is in units of volume
This chapter aims to provide the overarching framework of sediment per unit width per unit time. The sediment
within which these components will fit, to provide a transport in volume units can be converted to sediment
structure for the following chapters. transport in mass units by multiplying by the bulk density
of sediment (i.e. q ðmass=unit width=timeÞ ¼ ρb q
sam sav
ðvolume=unit width=timeÞ, where the subscripts m and v
2.2 The Governing Equations
are used to differentiate the units). Note that Equation
2.2.1 A Simple Landform and Soil Evolution (2.1) is simply a statement of volume balance since the
Model Formulation divergence term r q is the difference between the
sa
sediment transport inflow and the sediment transport
We start the discussion by overviewing how a landform outflow at a point. Equation (2.1) can be formulated as a
evolution model is traditionally constructed. The landform mass balance equation
is discretised in space (either on a regular grid of nodes in
∂z ∂qsam,x ∂qsam,y
x and y, or on an irregularly spaced set of nodes with a ρ b ¼ ρs U þ (2.2)
∂t ∂x ∂y
Triangular Irregular Network [TIN]). Given some initial
condition for this landform, the landform is subjected to a where ρs is the density of the saprolite/bedrock underlying
suite of environmental processes and evolved through the surface. The subscript m indicates that the sediment
time. This evolution through time requires a discretisation flux is mass units rather than the volume units used in
of the processes with time. Equation (2.1). Note that the difference between Equations
A commonly used model of landform evolution on its (2.1) and (2.2) is that the bulk density of the sediment ρb
own, with no soils, is is typically less than the density of the underlying rock
ρs because sediment has voids between the sediment

∂z ∂qsa,x ∂qsa,y particles, while the bedrock/saprolite has less voids
¼U rq ¼U þ ¼U þE (2.1) (ρb ¼ ρs ð1 ΔnÞ where Δn is the change in porosity
∂t sa ∂x ∂y
resulting from the conversion of saprolite into soil). For
where z is elevation (positive vertically upwards in igneous and metamorphic bedrocks there will be virtually
metres), t is time, U is tectonic uplift (in metres/year) no voids, while for sedimentary bedrocks the porosity
and qsa is the sediment transport vector expressed as a may be very similar to the sediment.
volume flux (cubic metres of sediment/unit width/time) That Equation (2.2) is an equation for mass balance
and E is the erosion (positive for deposition and negative can be demonstrated if we look at the mass balance on a
11
12 Constructing a Landscape Evolution Model – Basic Concepts
FIGURE 2.1: Schematic of the

llslope planar hillslope with unit width for
of hi
top surface erosion.
qsa
¶qsa
qsa + Dx
¶x
Dx
one-dimensional hillslope that is unit width (Figure 2.1) dissolution there is a corresponding volume increase.
for a small section of the hillslope Δx long and for a short The simplest way to extend the equations is to define the
time period Δt elevation of the saprolite-soil boundary as S, the bedrock
conversion rate, P, as the depth of saprolite converted to
∂z
ρb ΔxΔt ¼ ρs UΔxΔt þ ðinflow outflowÞΔt soil per unit time, and the increase in porosity in the
∂t
∂q conversion from saprolite to soil is Δn so that
¼ ρs UΔxΔt þ qsam,x qsam,x þ sam,x Δx Δt
∂x ∂z ∂S ∂D PΔn E
(2.3) ¼ þ ¼Uþ þ
∂t ∂t ∂t 1 Δn ρb
where ‘inflow’ and ‘outflow’ are the sediment mass fluxes ∂D P E
¼ þ
into the upslope boundary and out of the downslope ∂t 1 Δn ρb
boundary of the control volume, respectively. This sim- ∂S
plifies to ¼UP
∂t

∂z ∂q ∂qsam,x ∂qsam,y
ρb ΔxΔt ¼ ρs UΔxΔt sam,x ΔxΔt (2.4) E¼ þ (2.7)
∂t ∂x ∂x ∂y
and dividing through by ΔxΔt we obtain We note that Equation (2.7) is consistent with Equation
∂z ∂q (2.2) when uplift equals soil production (U ¼ P) so the
ρb ¼ ρs U sam,x (2.5) soil depth is constant, and this can be seen by reformu-
∂t ∂x
lating Equation (2.2) and the elevation equation in Equa-
which is the one-dimensional form of Equation (2.2). tion (2.7), respectively
Ahnert (1976) proposed a simple extension of Equa-
tion (2.1) to simultaneously model soil thickness, D: ∂z ∂z ρ E
ρb ¼ ρs U þ E ) ¼ U þ s 1 U þ
∂t ∂t ρb ρb
∂z ∂qsa,x ∂qsa,y
¼U þ ¼UE ∂z PΔn E ρs E
∂t ∂x ∂y ¼Uþ þ ¼Uþ 1 Pþ (2.8)
∂t 1 Δn ρb ρb ρb
∂D
¼PE (2.6)
∂t recalling that ρb ¼ ρs ð1 ΔnÞ: This formulation makes
several points explicit:
where E is the erosion rate (in metres/year) and P is the
rate of conversion of saprolite to soil at the bottom of the • The surface elevation is a result of the combination of
soil profile (also in metres/year). Current coupled land- the elevation of the saprolite boundary and the depth of
form and soilscape evolution models typically use some soil so that rather than the elevation of the surface being
variant of this formulation. Ahnert implicitly used a the fundamental property to be modelled, it is the
volume formulation for flux since he did not incorporate saprolite-soil interface that is the fundamental elevation
a sediment bulk density term. property to be modelled.
An extension to Equation (2.6) recognises that in the • The elevation of the surface is a function of the conver-
conversion from saprolite to soil the density and porosity sion rate from saprolite to soil and the increase in the
changes. In the absence of mass loss due to chemical porosity that results from it. Consider a simple thought
experiment where uplift and erosion are zero but the soil downslope (e.g. from the top of the catchment divide),
∂qdm,x
is thickening due to bedrock weathering. The elevation then the creep transport rate also increases (i.e. ∂x >0
of the surface increases as the lower porosity saprolite is if x is positive in the downslope direction), and from
converted to higher porosity soil and the soil layer Equation (2.9) the soil thins and the elevations decreases
thickens. by the same amount as the soil thins. Another soil
• This formulation is written in terms of porosity change transport process is shallow landsliding where the soil
in the absence of mass loss due to chemical dissolution. layer fails and flows/slides rapidly downslope
When the change in porosity is solely due to chemical (Chapter 13). While it is clear that after the landslide
dissolution it is possible for porosity to increase without event the soil has thinned, the more interesting question
a consequent increase in the volume of the soil so the is whether over the long term and averaged over many
depth of soil does not change. landslides (some of which occur upslope and deliver soil
to the location, while others remove sediment from the
However, this formulation for soil depth is still defi-
location and deliver it downslope) the net effect of
cient because it only allows the soil thickness to vary as a
landslides is to thin the soils. It seems intuitively reason-
result of vertical fluxes: (1) removal of soil at the surface
able to suggest that at the tops of hillslopes soils will
due to erosion and (2) the soil production at the soil
thin (more soil is removed than delivered), and at the
profile base. It does not allow the soil column to move
bottom of the hillslopes soils will thicken and form
down the hillslope. In a similar way to sediment transport
colluvial deposits.
at the surface, if more soil flows out downslope than
The formulations above are referred to as Eulerian
flows in from upslope, then the soil will thin, so if we
equations because they solve the mass balance equations
define a soil transport flux, qdm , where the subscript d
in time at a fixed location. In a following section (Section
indicates that this is the flux of the soil, and m indicates
2.2.3) we will introduce an alternative approach to model
that it is the mass flux (Figure 2.2), then Equation (2.7)
formulation that has some conceptual advantages over
becomes
this approach, called a Lagrangian formulation. But, first,

∂z E PΔn 1 ∂qdm,x ∂qdm,y we will talk about numerically solving the Eulerian
¼Uþ þ þ equations above.
∂t ρb 1 Δn ρb ∂x ∂y

∂D P E 1 ∂qdm,x ∂qdm,y
¼ þ þ
∂t 1 Δn ρb ρb ∂x ∂y 2.2.2 Time Discretisation and Operator
∂S Splitting
¼UP (2.9)
∂t
It is rare that we can solve the equations in the previous
where we note that not only does the soil mass flux term
section analytically, so we need to discretise these equa-
potentially thin the soil but it also reduces the surface
tions so that we can solve them numerically. We now
elevation. A number of processes cause soil to move
discuss how the equations are solved with time. If we
downslope, and they will be discussed in detail in
discretise Equation (2.9) in time, we can write a simple
Chapters 9 and 13. One soil transport process is soil
explicit Euler timestepping discretisation to this equation
creep (Chapter 9), which is akin to a viscous flow of the
(the Euler timestepping algorithm should not be confused
soil downslope. The creep transport rate increases with
with Eulerian equation formulation; Euler was a man of
increasing slope. So as the slope increases as you move
many talents):
soil
surf
ace qsa ztþ1 zt E PΔn 1 ∂qdm,x ∂qdm,y
¼ Uþ þ þ
¶qsa D Δt ρb 1 Δn ρb ∂x ∂y
qsa +
t¼t
soil ¶x x
-sap
rolit qd Dtþ1 Dt P E 1 ∂qdm,x ∂qdm,y
e inte ¼ þ þ
rfac Δt 1 Δn ρb ρb ∂x ∂y
e ¶qd D t¼t
qd +
¶x x
Stþ1 St
Dx ¼ ðU PÞjt¼t (2.10)
Δt
FIGURE 2.2: Schematic of the planar hillslope with unit width for where the subscripts in t and t þ 1 are the times at which
surface erosion and downslope movement of the soil profile. the values are defined, and the ð Þjt¼t indicates that the
term inside the parentheses is evaluated at time t. After Step 1: Erosion modelling
reorganisation, the equations solved numerically are
1 ∂qsam,x ∂qsam,y
zþ tþ1 ¼ zt þ U þ Δt
ρb ∂x ∂y
E PΔn 1 ∂qdm,x ∂qdm,y t¼t
ztþ1 ¼ zt þ U þ þ þ Δt
ρb 1 Δn ρb ∂x ∂y 1 ∂q ∂q
t¼t
Dþ tþ1 ¼ Dt þ sam,x
þ
sam,y Δt
ρb ∂x ∂y
P E 1 ∂qdm,x ∂q t¼t
Dtþ1 ¼ Dt þ þ þ
dm,y Δt
1 Δn ρb ρb ∂x ∂y Sþ tþ1 ¼ St þ ðU Þjt¼t Δt (2.12)
t¼t
Stþ1 ¼ St þ ðU PÞjt¼t Δt (2.11)

Step 2: Soil Production
A number of problems exist with using this explicit Euler
PΔn
method to solve this set of equations in practice (mostly its zþþ tþ1 ¼ zþ tþ1 þ Δt
1 Δn t¼t
poor numerical stability requiring small timesteps), but it
P
will suffice for the current discussion, as it is simple and
Dþþ tþ1 ¼ Dþ tþ1 þ Δt
easy to explain. To solve Equation (2.11) we need to 1 Δn t¼t
evaluate all the properties on the right-hand side of the Sþþ tþ1 ¼ Sþ tþ1 þ ðPÞjt¼t Δt (2.13)
equation at time t to determine the new value on the left-
hand side at time t þ 1. Equation (2.11) is in the form of a
Step 3: Transport Downslope
state-space equation. The terms z, D and S are referred to

as states, and their values at time t define everything we 1 ∂qdm,x ∂qdm,y
ztþ1 ¼ zþþ tþ1 þ þ Δt
need to know about the system. If we know only z, D and ρb ∂x ∂y
t¼t
S at time t, then we completely know the system at time t;
1 ∂q ∂q
this is the definition of a state. In state-space systems the Dtþ1 ¼ Dþþ tþ1 þ dm,x
þ
dm,y Δt
ρb ∂x ∂y
behaviour of the system at time t (i.e. its rate of change) is t¼t
completely defined by the ‘states’ (and properties that can Stþ1 ¼ Sþþ tþ1 (2.14)
be derived from the states). The flux terms (qsa and qd )
change depending on the states, and they are determined where the + and ++ superscript notation is used to indicate
by their ‘constitutive equations’. For instance, in the pre- the intermediate states produced by the operator splitting.
vious section we talked about soil creep, qd. The form of It can be seen that by substituting Equations (2.12) and
the qd dependence on the states is determined by the (2.13) into (2.14), we retrieve the original Equation
constitutive equation for qd (e.g. how it changes with (2.11). For this fairly trivial example there seems little
slope, which itself is a function of the surface elevation advantage to this reformulation, but operator splitting is
state z). The constitutive equation in turn will contain a useful in controlling the complexity of the equations that
series of ‘parameters’ that determine the rates of the pro- we need to solve. One significant advantage is that it
cesses, and how they change with changes in the states. allows us to formulate optimal numerical solvers that are
The form of the constitutive equations and their par- different for each of the different components of the
ameters is the subject of the subsequent chapters in this physics. We will discuss this later. However, as an
book. However, one aspect of Equation (2.11) is worth example, Equation (2.12) can be a very difficult equation
discussing here, because it conceptually underlies the to solve because of numerical instability and nonlineari-
construction of the overarching landscape evolution ties, and it typically requires very small timesteps. By
framework. Equation (2.11) consists of three processes: operator splitting we can solve Equation (2.12) at very
(1) the evolution of the landform and soil due to erosion, small timesteps but have large timesteps for (2.13) and
(2) the soil production at the base of the soil profile and (2.14) reducing the computations. Experience also shows
(3) the soil transport downslope. While it is convenient to that we can sometimes improve the performance per
consider them all together in one set of equations for the timestep if the ðÞjt¼t terms are recalculated for the new +
timestepping, it is not necessary to calculate them simul- and ++ intermediate states rather than for the original
taneously as in Equation (2.11). Equation (2.11) can be states at t ¼ t as written in Equations (2.12)–(2.14),
split into three sets of equations that are solved sequen- but that may be offset by the extra effort that is required
tially at each timestep. This technique is called ‘operator to recalculate the terms inside the parentheses at the new
splitting’ (Celia and Gray, 1991). Equation (2.11) can be intermediate states rather than doing all the calculations
rewritten as follows: for the original states at t ¼ t. The operator splitting in
Equations (2.12)–(2.14) has used a very simple explicit the landslide erosion problem. This has been used as the
Eulerian timestepping algorithm. This example was used basis of landform evolution models (Chase, 1992; Crave
because of the simplicity of the explanation above; oper- and Davy, 2001). The conceptual appeal of this approach
ator splitting can be used for any timestepping algorithms. is obvious. The problem is, however, that we need to be
There are other conceptual advantages to operator split- able to characterise the three components above from field
ting, but it is premature to discuss them now, and they will data. The first two can and have been characterised from
be discussed later. remote sensing studies of landslide-prone areas (e.g.
Hovius et al., 1997; Stark and Hovius, 2001). The third
is more difficult because it relies on matching a specific
2.2.3 The Lagrangian Conceptualisation and landslide scar with a specific deposition area. Moreover
Process Locality the distance the package travels is a function of the exact
details of the path that is taken and the energy dissipation
The formulation for transport continuity in the previous
along that path. We discuss this in more detail in
sections is based on applying a mass balance at a given
Chapter 13.
location (e.g. a grid node point) with time. Because these
Here we will show how the two model formulations,
equations track mass balance by balancing fluxes into and
Eulerian and Lagrangian, are consistent, and by doing so
out of a fixed location, these equations are called Eulerian
highlight aspects of both approaches. Consider the
equations. An alternative approach tracks a unit of sedi-
landslide example again and assume a unit width slope.
ment as it moves downslope from its starting location to
Assume that every landslide has the same probability per
the end location where it stops moving. This specified
unit area per unit time of being triggered and this prob-
volume or mass of soil/sediment has been given a number
ability is the same at every location in the landform p. We
of names in the literature including package, cohort and
will also assume the mass of the package of sediment
precipiton; we will use the term ‘package’. A method of
transported by each landslide M, and the distance travelled
tracking the dynamics of a package of sediment from
by this package L are always the same. The probability p is
source to sink is called a Lagrangian method.
defined as the probability that a landslide will be triggered
How a Lagrangian formulation works is best shown
in unit time within a unit area, so the probability of
with an example. Consider the example of a landslide
triggering in a region of length Δx on the unit width
(Figure 2.3). There are three components to a single
hillslope in a time period Δt is pΔxΔt: We start from the
landslide: (1) the probability that any given location will
top of the catchment divide and consider the package
have a landslide at any given point in time (i.e. a package
passing a location A (Figure 2.3) that is distance L from
will start to move), (2) how much sediment will be moved
the catchment divide. For every trigger location upslope of
by that landslide when it occurs (i.e. how big is the
A the package of sediment triggered will pass A. If we
package) and (3) how far that landslide will travel once
consider location B that is a location L þ ΔL from the
triggered (i.e. what is the path followed by the package).
divide, then it has the same mass of sediment pass it as
Given those three pieces of information you can track the
location A. For location B those landslides that are trig-
sediment package as it travels downslope and determine
gered at a distance between 0 and ΔL from the divide
the trajectory of that sediment package. If we did this for
deposit their sediment between A and B and do not pass
many landslide events over some period and summed up
location B. The key concept is that at any location only
the net effect of all the package trajectories on the evolu-
landslides within a distance L upslope of that location
tion of the landform, then this is a Lagrangian solution to
contribute to the sediment flux past that location. Accord-
ingly we can calculate the average number of landslides in
the region distance L upstream for the unit width of hill-
slope as N ¼ pLΔt, and if each of these landslides contains
mass M of sediment, then the flux past location A is
qs ¼ pLM (2.15)
This is then the Eulerian equivalent of the Lagrangian
representation of the landsliding process. The Eulerian
flux equation in Equation (2.15) is equivalent to the aver-
FIGURE 2.3: Schematic of the Lagrangian representation of a age of the Lagrangian description. If we insert Equation
landslide moving downslope. (2.15) into Equation (2.1), then this example results in
zero erosion anywhere since ∂q ∂x ¼ 0 with x positive in the

s
the node where we are calculating erosion/deposition).
downslope direction. For net erosion to occur one or more A local process can be written as an Eulerian differential
of p, L and M must increase downslope. Of course, this is equation at that location. A process is ‘nonlocal’ if it
a very simple example. and all three terms will vary in requires information from other parts of the landform to
space. Moreover when the three terms change in space an solve the equation at the node. Thus a process that requires
equation equivalent to Equation (2.15) will not be as easy only the local slope (e.g. diffusion, Chapter 9) to deter-
to derive. We will discuss these issues in Chapter 13 when mine sediment flux (and thus elevation and soil depth
landsliding is discussed. change at that point) is local. A process that requires
The reason for going through the derivation leading to knowledge of what the catchment looks like upstream is
Equation (2.15) is that in later chapters we will discuss nonlocal. A simple example to consider is erosion. Erosion
several processes that are more naturally understood and (Chapter 4) requires knowledge of discharge at the node,
mathematically formulated in a Lagrangian framework. which requires knowledge of the catchment area (and
In some cases it will easier to numerically solve them by perhaps upstream soil, vegetation and so on), so erosion
converting them to a Eulerian form, while in other cases is strictly speaking nonlocal. However, we can convert
it may be easier to numerically solve the Lagrangian erosion to a local process if we separately calculate the
form using cellular automata or particle-tracking catchment area and discharge, and then use this in the
approaches. The reader needs to remember that though erosion equations. This is a somewhat trivial example,
the conceptualisation is different, Eulerian and Lagran- but, on the other hand, considerable debate surrounds
gian representations are simply different ways of looking whether landsliding, with its strong dependence on energy
at the same problem and the reader should view them dissipation along the landslide pathway, can be modelled
as fundamentally consistent methods. This will be par- as a local process, and we will discuss this further in
ticularly useful when we consider fluvial erosion in Chapter 13. The reason for discussing locality here is that
Chapter 4, where it will be convenient to switch between the derivation resulting in Equation (2.15) is for a very
the formulations as each approach highlights different simple geometry. Consider now the case where the
aspects of the physics. hillslope gradient varies downslope and where these
The numerical methods for directly solving Lagran- variations are significant within the distance L. This means
gian methods are generally lumped under the headings of that the conditions that apply when the sediment package
particle tracking, cellular automata or agent-based starts moving are different from the condition distance L
methods. To demonstrate the principle for particle downslope. Specifically L, the distance the package
tracking we can think of the landslide problem again. At travels, may change as it moves downslope. Thus the flux
every timestep it is determined whether a landslide has at location A will depend on the way the hillslope interacts
been triggered at a node. If it is triggered, then the land- with the package as it moves from the point from which it
slide package size is calculated (i.e. the mass). This pack- started moving to the location A. This makes the flux
age then moves downslope based on whatever dynamics process at location A nonlocal. However, it may be
for the transport has being adopted, thus determining the possible to reformulate the process as local by modelling
trajectory. For instance, Chase (1992) moved the mass in not only the mass transport but also the momentum and/or
the steepest downslope direction until it reached a location energy. In this reformulation the energy dissipation rate
where the slope gradient was below a threshold. Another would be calculated locally at A based on the energy
way would be to model the kinetic energy of mass (e.g. dissipation processes at A.
gain of kinetic energy as it moves to lower elevation, loss
of energy through frictional losses) until the kinetic 2.2.4 Lagrangian Simulation Approaches
energy is zero, at which location the mass stops moving
and is deposited. Each landslide is modelled separately The discussion in the previous section leads to an obvious
and the landform evolution is the net result of the many conceptualisation of how to model landslides (and per-
landslides that occur over the time modelled in the haps other processes) within LEMs based on the explicit
simulation. tracking of material as it moves downslope.
Finally one thing that the Lagrangian formulation high- Essentially the idea is to track the movement of a
lights is a property of transport processes called ‘locality’. particle that represents a property of interest based on
A process is referred to as ‘local’ if the process is com- some forcings. A simple example is the movement of a
pletely determined by the properties of the process at the sediment particle as a result of a flow field. This example
location at which the equations are being solved (e.g. at falls under the heading of ‘particle tracking’ in the
literature and is a technique that has been widely used for modelling. The first reason is that it is hard to see how
pollutant transport modelling for many decades as an many geophysical processes can be recast as agents, even
alternative to modelling pollutant transport with the Euler- though its relatively easy to see how the human, fauna and
ian equations for advection and diffusion (for applications flora processes in Chapter 14 can be cast into this frame-
dating back to the 1970s, see Williams, 2006). Typically work. The second reason is that the technique has
those particles are passive (simply responding to the sur- developed as a spinoff of simulation modelling systems
rounding environment) and the particle properties do not (LOGO, various frameworks implemented in Java and so
change with time, so the modelling simply involves on) that have not been widely used in the traditional
tracking the movement of thousands to millions of par- simulation literature, and the systems that have been used
ticles (parameterised only by their location) and examin- have developed a reputation for being computationally
ing the aggregate behaviour. slow. It is not clear whether this slowness is inherent in
A simple extension of this particle tracking is to allow the agent-modelling approach or if it simply reflects the
the particles to have properties that may change in slowness of the underlying modelling frameworks and
response to the surrounding environment. For example, their implementation languages. Finally the technique
the particles may be radioactive (e.g. naturally decaying has been poorly explained in the literature, and many
radioactive elements where radioactivity is passively papers simply refer back to underlying simulation systems
decaying, or cosmogenic nuclides where there is also a and give sparse information about how their agents work,
potential source term based on depth below the soil making it difficult to replicate findings independently
surface) so that over time the particle itself is changing, (Grimm et al., 2006).
but the movement of the particle is still passively
responding to the surrounding environment. There are 2.2.5 Boundary and Initial Conditions
other conceptually similar extensions where the particle
properties may respond to the other temporally and Since we are solving a set of partial differential equations
spatially varying properties of the environment. For with time, there is a need for both boundary conditions
instance, some chemical property of the particle may and initial conditions.
change over time. An example is sorption where the Determining the initial conditions for landform evolu-
chemical on a sediment particle (typically a cation) is tion models is difficult. For geologic problems we gener-
either attracted (sorbed) or released (desorbed) from the ally have only a vague idea of what the initial conditions
particle surface depending on the balance of the chemical were. Yet we know that landform evolution models will
concentration on the particle surface and in the surround- yield (markedly) different planar channel networks
ing solution. While it is not presented in the same way as depending on the initial conditions. Thus if we don’t
Section 2.2.2 this is suitable for operator splitting where know the initial conditions, it is very unlikely that we will
the three operators are (1) the particle tracking, (2) the be able to replicate known observed channel networks
Eulerian solution for the solution concentration and (3) unless there is some ongoing constraint on landform
the sorption-desorption calculation. Conceptually this development (e.g. patterns of high and low erodibility
approach is similar to the ‘particle-in-cell’ and ‘marker- rocks). We can, however, replicate statistics of the land-
in-cell’ solvers. form. Experiments have shown us that the statistics of the
A conceptual advance on this tracking technique that network can be matched if the model is correct but the
has emerged in the last decade is agent-based modelling exact spatial pattern is rarely reproduced (e.g. Hancock
(e.g. Perez and Dragicevic, 2012). In the particle-tracking and Willgoose, 2002). Essentially what this means is that
approaches above the particle motions respond passively there is no unique best solution for a channel network or
to their surrounding flow field. An ‘agent’ is a particle that landform evolution, but instead there are a suite of equiva-
has some ability to modify its motion and other behaviour lent network solutions that all solve the governing equa-
based on its surrounding environment (e.g. a fish tions equally well. Thus the idea that there is a single
swimming upstream), or the agent may interact with other global optimal network (e.g. the optimal channel network;
agents. Thus an agent simulating, say, a cow, may interact see Chapter 3) is probably misguided.
with other cow agents to generate the aggregate cow herd The question of the potential impact of initial condi-
behaviour, or an insect agent may change its flying based tions on soilscape and vegetation models is still poorly
on wind direction and speed. understood. For instance, we need to understand whether,
There are three potential reasons why agent modelling given a known landform, we can deterministically repli-
has not been adopted widely for landform evolution cate the soilscape distribution or only replicate statistics of
the soilscape (e.g. depth of soil versus distance from the 2.2.6 Separation of Response Time Scales and
divide). Likewise given a known landform and soilscape Effective Process Representations
we need to understand whether we can deterministically
replicate the vegetation’s spatial distribution. The final overarching issue to discuss is that of response
The boundary conditions required for a simulation are time scales. In a coupled model containing many different
by comparison relatively straightforward. The boundaries components with different dynamics, the time to adjust to
of the domain are largely determined by the landform and some perturbation of the system will be different for the
should be where the surface water flux at right angles to different components of the system. Some parts of the
the boundary is zero. For most processes these will be the system will respond quickly and equilibrate faster, while
catchment divides, though the catchment divides them- other parts of the system will respond more slowly and
selves may move during simulations as a result of pro- equilibrate more slowly. In a strict sense the system will
cesses that can cross the divide (e.g. landslides may not be perfectly equilibrated until the slowest part of the
remove part of the upstream reaches of an adjacent catch- system has equilibrated to the perturbation. In a pragmatic
ment). I will note in passing that some authors errone- sense this makes the entire system rather difficult to model
ously state that discharge is zero at the catchment divide. numerically because typically the size of timesteps is set
The discharge at right angles to the catchment divide is by the fastest responding part of the system, even while
zero, and the discharge parallel to the catchment divide is the duration of the simulation is determined by the slowest
undefined (i.e. it can be nonzero). responding part of the system. This is a characteristic of
There is a need to specify a known elevation for at some of the most difficult systems to solve numerically,
least one point, typically the downstream boundary of the ‘stiff equations’.
catchment (e.g. the river outlet at the bottom of the catch- If it is the slower parts of the system that are of interest
ment). There is one subtlety with this, and that is the case and not the faster ones, then a common solution is to
where the downstream known elevation varies with time develop effective parameterisations of the fastest respond-
(e.g. variable sea or lake levels; Figure 2.4). In this case ing parts of the system that reflect the average effect of the
the elevation doesn’t simply move up and down at the faster parts of the system on the slower parts. A common,
same location, but if the elevation increases, then this though not necessarily correct, solution is to take just take
elevation floods the lower reaches of the catchment and the average of the faster responding part and use that as
the location of the known elevation moves upstream. input into the slower part. There are subtle differences
More difficult is if the elevation drops, in which case the between the average of the process and the geomorpholo-
location of the known elevation moves downstream and gically effective representation of that same process, which
potentially outside the domain being modeled. For revolve around nonlinearities and how variability propa-
instance, in the last ice age the sea level was about gates through a nonlinear relationship, which will be dis-
140 m lower than current sea level and in many cases cussed in detail in Section 2.4 and Chapter 4. Many times
was at the edge of the continental shelf. Even today in we are forced to approximate the fast process because either
areas where there is isostatic rebound from the last ice age one of the spatial or temporal discretisation is too coarse to
(see Chapter 12) and the land is rising relative to current capture the fine scale detail. We thus need to develop ‘sub-
sea level, then the outlets of coastal catchments are grid effective parameterisations’ that capture the variability
moving out to sea. in the process at finer time or space resolution.
FIGURE 2.4: Schematic showing how the downstream boundary moves when the downstream boundary is an ocean boundary, where
the elevation of the downstream boundary is changing relative to the land surface through either sea level changes or land surface elevation
changes due to tectonics.
crushed rock that can break down rapidly and faster

than the landform changes). SCORPAN, CLORPT
and soil catena fall into this category.
• Coevolving soils: These are soils that respond with
similar response times as the landform so they change
at roughly equal rates. In this case both the soil and
landforms reflect the history of both the soils and the
landforms, and they cannot be decoupled.
The reason for making this classification is to highlight
that it may be possible to identify three regions of model
behaviour separated by the vertical dotted lines A and B in
Figure 2.5.
Fast soils can be modelled by a soilscape model that
uses a fixed landform, because they equilibrate quickly to
the slowly changing landform. Thus there is value in dis-
FIGURE 2.5: Schematic showing the relative time scales of cussing a soilscape evolution model that can potentially be
adjustment for the landform versus the soils, and highlighting the
used decoupled from a landform evolution model. This is
three cases where (1) soils may adjust more slowly than landforms
(Slow soils), (2) landforms may adjust more slowly than soils (Fast the rationale for the organisation of the book, where all the
soils) and (3) the soils and landforms adjust at rates that overlap so components of soilscape modelling (in Chapters 5–11) are
that some parts of the landform and their soils have similar response discussed independently of landform evolution modelling,
times (Coevolution). but which can then be coupled to a landform evolution
model to simulate those situations where soils coevolve
One example goes to the heart of how the topics in the
with the landform. Because the soils equilibrate quickly
book are organised. In the soil science literature it is often
the landform will evolve in response to the equilibrium
discussed that the soil reflects, among many other inputs,
reached by the soils, and this equilibrium will be the
the topography of its hillslope. The main two historical
geomorphologically effective representation of the soils.
ways of characterising the development of soils are
Slow soils on the other hand can be modelled by a
CLORPT (Jenny, 1941) and SCORPAN (McBratney
landform evolution model that has a time-invariant soil
et al., 2003). The names are both mnemonics, and in both
layer. The soilscape does not evolve, or evolves at a much
cases R stands for topography. Notably in neither case is
slower rate than the landform. This is the rationale for the
there a term for the history of the topography. This is not
standalone chapters on landform evolution in Chapters 4
an oversight since a common assumption is that the soils
and 12–15. From the perspective of the soilscape, the
are in equilibrium with the topography. The question is, in
landforms reach equilibrium much faster than the soils,
the context of a book about landscape and soilscape
so the soils will reflect the response to the long term
evolution, how this can be true. One potential answer is
history of the landform.
the relative time scales of processes. Figure 2.5 shows a
The vertical lines in Figure 2.5 highlight what are
schematic of the relative time scales for soils and land-
‘scale breaks’, which postulate that there is no overlap in
forms and highlights three different types of soils based
the response times of fast soils and landforms, and slow
on their response times relative to the response time of the
soils and landforms. The only time it is essential to model
landform. The three types are the following:
both the soil and landforms simultaneously is when there
• Slow soils: These are soils where the landform responds is an overlap in response times for the soils and the
much faster than the soil. An end member of this case is landforms, the intermediate case of coevolving soils and
where a landform evolves through an unchanging soil landforms. Willgoose et al. (2015) presented results for
profile (e.g. a gully cutting through a soil layer). Many the Tin Camp Creek research site in the Northern
existing LEMs fit into this category because they have Territory, Australia, where they had a standalone and
fixed layers of geology under the landform surface. calibrated landform evolution model, and a standalone
• Fast soils: These are soils where the soils respond much and calibrated soilscape evolution model. They showed
faster than the landform. An end member of this case is that for most of the area the soils are Fast, and there were
where a soil evolves on a landform that doesn’t change only very small parts of the landform where the soil and
(e.g. many constructed landforms are covered with landform response times overlapped.
The relevance of this example is that the results from greater insight and transparency. Accordingly this
this suggest that for fast soils it might be possible to use alternative approach is particularly useful when analysing
equilibrium soilscape model results (Cohen et al., 2009, computational experiments where the governing physics
2010; Welivitiya et al., 2016) as effective parameterisa- are known but the response of the model is so complex as
tions of the soilscape in a landform evolution model, and to be difficult to link to input data and parameters. This is
that it might not be necessary to explicitly couple together particularly true for spatially distributed nonlinear models,
a soilscape and landform evolution model. This would whose behaviour can be nontrivial even for simple math-
allow the LEM to use larger timesteps because the LEM ematics (see, for instance, the extensive Chaos literature).
doesn’t need to model the fast dynamics of the SEM; it Landform evolution models fall into this latter category.
only needs to use the equilibrium soils results from the The alternative approach is called ‘scaling analysis’ and
separate soilscape modelling. In the terminology intro- involves nondimensionalisation of the governing equations
duced at the start of this section, the equilibrium soilscape directly (Daily and Harleman, 1966; Fox and McDonald,
description is a subgrid effective parameterisation of the 1998; Munson et al., 1998).
soils, reflecting the ‘effective’ response of soils at higher
resolution times than modelled. 2.3.2 Scaling Analysis of Governing Equations
Of course, this is only one site, and other sites may
vary, but the point of the example is that it indicates the We will explain scaling analysis using a very simple
potential value of a soilscape model that can be used landform evolution model. Consider a one-dimensional
independently of the landform evolution model. In other hillslope (elevations z changing in the x direction with
chapters in the book we will see other cases where we can time t) where the land-sculpting processes are tectonic
derive subgrid effective parameterisations of processes uplift U and Fickian diffusion (e.g. soil creep with
within the models, and which will allow us to significantly spatially constant diffusivity D; see Chapter 9):
speed up numerical calculations.
The question of time scales and scale breaks will ∂z ∂2 z
¼UþD 2 (2.16)
arise repeatedly in the book in (1) river cross section, ∂t ∂x
meandering and floodplain interactions, (2) vegetation There are three scales in this problem: (1) the horizontal x
and soil interactions and (3) climatic impacts on geomor- scale, (2) the vertical z scale and (3) the time t scale. If we
phologically effective erosion models. think of applying this equation to a specific landform,
then there will be scales for these properties of this
2.3 Nondimensionalisation and Scaling landform (we delay the question of how we define scales
Analysis until later), and we define the following nondimensional
properties:
2.3.1 The Purpose of Nondimensionalisation z x t
z∗ ¼ ; x∗ ¼ ; t∗ ¼ (2.17)
Nondimensionalisation is a technique that allows us to h zi hxi ht i
rescale the physics occurring at one site to that occurring where the superscript star indicates a nondimensional
at another site, solely on the basis on the difference in the property, and the h:i property is the scale of the landform.
scale of the two sites. Almost every book on basic fluid Equation (2.17) can be substituted into Equation (2.16):
mechanics has a discussion of it, typically introducing the
Buckingham-Pi theorem and then deriving, apparently by hzi ∂z∗ h zi ∂ 2 z ∗
∗
¼ U þ D 2 ∗2 (2.18).
black magic, nondimensional numbers that relate aspects hti ∂t hxi ∂x
of the flow and geometry to fluid mechanics regimes (e.g.
In this equation the derivatives are now nondimensional.
the transition from laminar to turbulent flow, using the
We then collect terms to make the equation nondimen-
Reynolds number). From this they then demonstrate
sional to yield
how to rescale model results to the field. While the
Buckingham-Pi theorem is useful for exploratory experi- ∂z∗ ht i h t i ∂ 2 z∗
mental studies, it is severely limited in cases where there ∗
¼U þ D 2 ∗2 (2.19)
∂t h zi hxi ∂x
are multiple length or time scales and Buckingham-Pi is
not very transparent in its application. from which we see we now have two nondimensional
When we know (or can postulate) the mathematics numbers, one associated with each of the additive
of the governing equations, a different approach leads to terms on the right-hand side of the equation. The first
2.3 Nondimensionalisation and Scaling Analysis 21
characterises the uplift term (i.e. Uhzhiti) and the other the soil 2.3.3 Definition of the Appropriate Scales
creep term (i.e. Dhxhit2i). In the spirit of nondimensionalisation The most difficult aspect of nondimensionalisation, the one
nomenclature let’s call these numbers Nu and Nc, respect- that everybody struggles with, is defining the scales in the
ively (u for uplift and c for creep). The most important nondimensionalisation exemplified by Equation (2.17).
thing to note is that provided that Nu and Nc do not The scaling analysis at the end of the section above
change, the nondimensional solution to Equation (2.19) suggests that we wish to define scales that result in the
(i.e. z∗ ðx∗ ; t ∗ Þ where z∗ is a function of x∗ and t ∗ ) will nondimensionalised variable being of O(1).
also not change. So, for instance, if we change the vertical A subtler, and often ignored, issue is that the scale
scale hziof the catchment (e.g. give the catchment higher should be reasonably insensitive to random sampling
relief), then provided U is changed so that Nu remains the error. All scales will have some sampling error in them
same value, then the results from the equation (i.e. the since they are measured from a single realisation of an
change of z∗ with x∗ and t ∗ ) will be the same. This shows inherently random process (e.g. catchments will vary from
that provided the physics of Equation (2.16) is an appro- site to site even if geology, climate and process are iden-
priate representation of the field processes, we can rescale tical at each site). Scales that depend on the mean of what
different catchments having different areas and uplift and is being measuring rather than some extreme of the prob-
compare them. ability distribution are preferred because the sampling
If we rearrange Equation (2.19) we can extract even error is less for mean properties. Even better is something
more information from the governing equation, and dem- that is an integrated measure of the landscape (pedants
onstrate one of the more powerful aspects of scaling will point out that the mean is in fact an integrated
theory, which is to characterise the relative dominance measure by the nature of how it is calculated). Integrated
of different processes in the governing equation: measures smooth out noise. A poor choice is something
! that is a derivative because derivatives accentuate noise
∂z∗ U hti Dhzi ∂2 z∗ (e.g. slope is inherently noisier than elevation, even if you
¼ 1þ (2.20)
∂t ∗ h zi U hxi2 ∂x∗2 take the means).
A good case study is how to define the vertical
The nondimensional number inside the parentheses
elevation scale of a landform hzi. An obvious, but poor,
multiplying the soil creep is just Nu/Nc, but its import-
choice is to use the difference between the minimum
ance is that it determines the relative importance of soil
and maximum elevation of the landform, the ‘relief’.
creep to uplift. If UDhhxzii2 1 then the equation is dominated
Certainly, if the elevations are divided by this scale, the
by the soil creep term, while if UDhhxzii2 1 then the equation nondimensionalised elevations will be O(1). But the relief
is dominated by the uplift term. This interpretation should is an example of an extreme of a probability distribution.
be intuitively obvious from the ratio of D/U in the non- Its value can change dramatically simply if a single isol-
dimensional number, but the nondimensional number ated mountain peak occurs in the catchment, even if the
indicates that the relative dominance is also a function bulk of the landform’s elevations are unaffected. A better
of the vertical and horizontal scales. choice for the vertical scale here would be the mean
Finally, the most subtle aspect, and the one that gives elevations above the minimum elevation, though this is
scaling analysis its name, is when the scales are adopted so still impacted by the minimum elevation, but in practice
that the nondimensional scaled x, z and t are of O(1). The the minimum elevation is much less variable than the
terminology O(1) means that the value is of order 1, that is maximum. An obvious example of a nondimensionalised
approximately equal to 1, as opposed to approximately graph that suffers from this poor choice of vertical scale
equal to either 0.1 or 10. This can be easily achieved by for nondimensionalisation is the hypsometric curve,
selecting the scales hzi,hxi and ht i appropriately. The where elevations are scaled between the minimum and
∗
∂2 z∗
implication of this is that ∂z∂t ∗ ¼ Oð1Þ and ∂x∗2 ¼ Oð1Þ, so
maximum elevation (see Section 3.3.5). Moreover, the
that for Equation (2.20) to be internally consistent if both hypsometric integral, the area under the curve, proposed
processes are important, then Nu ¼ Oð1Þ and Nc ¼ Oð1Þ. as an indicator of landscape development and process
This is consistent with Nu=Nc> < 1 characterising the rela-
(Strahler, 1952; Willgoose and Hancock, 1998) can dra-
tive scale of processes. The reader may need some convin- matically change if even a single isolated peak is added to
cing on this point, but this conclusion is generally true if the the landform.
nondimensionalisation scales are chosen carefully, and is a Even for the horizontal scales of a catchment there are
potentially important consideration in defining scales. better or worse choices. For the catchment scale hxi the
square root of the drainage area appears to be a good we can calculate the long-term average impact of the time-
choice. However, when we define Equation (2.16) in x varying land-sculpting process. The sections below out-
and y rather just x, the opportunity to nondimensionalise line a simple technique that can be used to determine the
differently in the two directions arises so that we have a averages of these time series. In many cases the analyses
horizontal scale in the x direction and a (potentially) that follow give exact answers that allow us to exactly
different horizontal scale in the y direction. The most determine the mean of a process (i.e. this might be thought
obvious case where we might to do this is to characterise of as being the geomorphologically effective average of
catchment and/or channel network shape. One way of the process) in terms of the process parameters and the
determining channel network shape is to calculate the ratio statistics of the forcing inputs. Some more complex forms
of network length to network width, where length is the of expression require approximations, but even in these
length of the longest stream in the network and width is cases we can obtain first-order approximations to the
maximum width of the network width function. Alarm geomorphologically effective average of the process. This
bells ought to be immediately ringing at the mention of discussion provides enough detail to be able to understand
‘maximum’ because again this is an extreme of a probabil- the averaging analyses presented in this book.
ity distribution (in this case one maximum for length and
another for width). It is better to use mean width and mean 2.4.1 Perturbation Analysis Principles
stream length. The shape of the catchment could then be

characterised by the nondimensional number hxi hyi and the It is convenient to split a random process into its mean and

area nondimensionalised as A hxihyi . a random perturbation around that mean so that a random
Finally, there are scales other than time and length. process X can be written as
Taking Equation (2.16) as an example, if the uplift is
spatially varying (i.e. U ¼ U ðxÞ), then it might be appro- X ¼ X þ X0 (2.22)
priate to nondimensionalise the uplift using, for instance, where X is the mean or expectation of X (i.e. E½X ¼ X)
the mean uplift U so that we can define a nondimensional and X 0 is the random perturbation around that mean. We
uplift U ∗ ðxÞ ¼ U ðxÞ=U (effectively in this case the will see below that a useful consequence of this definition
spatial pattern of uplift). Equation (2.20) would then look is that the expectation of the perturbation X 0 is E½X 0 ¼ 0:
like All random variables are split into a mean and perturb-
! ation about the mean using Equation (2.22). The variance
∂z∗ U hti ∗ Dhzi ∂2 z∗
¼ U ðx Þ þ (2.21) of X is then
∂t∗ h zi U hxi2 ∂x∗2 h 2 i h i
¼ E ðX 0 Þ
2
σ 2X ¼ E X X (2.23)
This use of rescaling is quite general, can be applied to
any property in an equation and allows spatially variable
and the covariance between two random processes X and
properties to be included in any nondimensional number
Y is
developed from the governing equations. Of course, if the
spatial pattern of U ∗ ðxÞ changes, then the deterministic σ 2XY ¼ E½ðX 0 Y 0 Þ (2.24)
solution of Equation (2.21) will change. However, in
many cases (particularly if the spatial variability is It is also worth remembering some simple properties of
random) we are interested in characterising statistical the expectation process:
properties, and in these cases the exact spatial pattern
• The expectation
of a nonrandom number is just that
(i.e. U ∗ ðxÞ) may not be critical whereas the magnitude number (e.g. E X ¼ X; E½R ¼ R if R is a nonrandom
(i.e. U) is. number)
• A variable that is not random (i.e. a number) can be
2.4 Perturbation Analysis for Analysis taken outside an expectation (i.e. E½RX ¼ RE½X if R is
of Stochastic Processes a nonrandom number)
• The expectation of the sum of two random variables is
Erosion time series vary randomly with time, yet in most equal to the sum of the expectations of the variables
case we are interested in the cumulative geomorphic taken separately (i.e. E½X þ Y ¼ E½X þ E½Y ) and
effectiveness of these variable time series. In many cases • The expectation of the multiplication of two independent
what we need to know about the geomorphic effective- random variables is the multiplication of the expectations
ness of the time series is simply the mean over time so that (i.e. E½XY ¼ E½X E½Y ).
These properties will allow us to break up more com- of the fractal dimension so the variance approaches infin-
plex problems, that cannot be directly solved using the ity as the length of data increases. Since sediment trans-
techniques below, into a combination of problems that can port can be a function of Q2 and discharge is asserted to
be solved. be fractal over geomorphic time scales, this suggests that
the average sediment transport will approach infinity with
2.4.2 Examples longer periods of record. We discuss the subtleties of this
issue in Chapter 4.
To demonstrate the principles involved in perturbation
analysis, some simple examples are provided below. 2.4.2.2 Example 2: Linear Combination of Three
Random Inputs
2.4.2.1 Example 1: Linear Combination of Two
Example 1 showed a case where the result was exact. We
Random Processes
will now consider a case where an exact answer can be
Consider a random process Z that is the product of two
obtained but where it is more common to use a simplifi-
random inputs X and Y so that Z ¼ XY: Expressing this
cation based on the normal distribution. Consider a
as perturbations yields
random process Z that is the product of three random

Z þ Z 0 ¼ X þ X0 Y þ Y 0 (2.25) inputs U, X and Y as Z ¼ UXY: Expressing this as
perturbations
If we wish to know the mean of Z, we simply take the
expectation of this equation after first expanding the right- Z þ Z0 ¼ U þ U0 X þ X0 Y þ Y 0 (2.28)
hand side
Expanding this and taking the expectation of the equation,

Z ¼ E Z þ Z0 Z ¼ E Z þ Z0

¼ E X þ X0 Y þ Y 0 ¼ E U þ U0 X þ X0 Y þ Y 0

¼ E X Y þ XY 0 þ YX 0 þ X 0 Y 0 ¼ E U X Y þ U XY 0 þ U YX 0 þ UX 0 Y 0 þ X YU 0
þ XU 0 Y 0 þ YU 0 X 0 þ U 0 X 0 Y 0
¼ E X Y þ E XY 0 þ E YX 0 þ E½X 0 Y 0
¼ U X Y þ U X E½Y 0 þ U Y E½X 0 þ U E½X 0 Y 0
¼ X Y þ XE½Y 0 þ YE½X 0 þ E½X 0 Y 0 (2.26)
þ X Y E½U 0 þ X E½U 0 Y 0 þ Y E½U 0 X 0 þ E½U 0 X 0 Y 0
We can further simplify this by recalling that the mean of (2.29)
the perturbation is zero (i.e. E½X 0 ¼ 0) and by using the
In contrast with Example 1 the substitution for the per-
definition of the covariance between X and Y so that
turbations’ mean and covariance
Z ¼ X Y þ σ 2XY (2.27)
Z ¼ U X Y þ Uσ 2XY þ Xσ 2UY þ Yσ 2UX þ E½U 0 X 0 Y 0 (2.30)
The implications of Equation (2.27) are that (1) this equa-
yields a new term that is a triple product of the perturb-
tion is exact, (2) the mean of Z is not, in general, equal to
ations. Its meaning is somewhat clearer if we look at the
the product of the means of the X and Y, (3) the difference
special case when U, X and Y are the same random
from the product of the means is due solely to the covar-
variable, say, W (i.e. Z ¼ W 3 ). In this case the equation
iance between X and Y (note that this equation is equal to
becomes
the result for independent processes when the covariance
is zero) and (4) while Equation (2.27) appears to be 3 3=2
Z ¼ W þ 3Wσ 2W þ σ 2W γ (2.31)
limited to normal distributions, because only mean and
covariance appear, it is in fact true for all distributions for where γ is the skewness of W, suggesting that the third-
which mean and covariance are defined. order perturbation is related to deviations from normality.
An important special case is that covariance is not We can also rearrange the original perturbation equa-
defined for many fractal processes. Moreover, for the tion as
special case where X and Y are the same random process
U0X0Y 0
(so Z ¼ X 2 ), then Equation (2.27) is a function of the Z ¼ U X Y 1 þ rXY þ r UY þ r UX þ E
2 2 2
(2.32)
2 UXY
variance (i.e. Z ¼ X þ σ 2X ) and the variance is also not
defined for a fractal X. Typically σ 2X / T α where T is the where r is the correlation coefficient. If the perturbations
period over which averaging is desired and α is a function around the mean are small, then we can assert that
E½U 0 X 0 Y 0 << U X Y: Alternatively if the probability dis- we make an approximation, which is to expand the power
tributions of U, X and Y are all normal, then E½U 0 X 0 Y 0 ¼ 0: on b using Taylor Series around the mean X :
Either way Equation (2.32) can simplified as b bðb 1Þ 0 2 b2
X þ X 0 ¼ X þ bX 0 X
b b1
þ X X
Z ¼ UXY þ Uσ 2XY þ Xσ 2UY þ Yσ 2UX (2.33) 2!
bðb 1Þðb 2Þ 0 3 b3
þ terms X 0
4
þ X X
In conclusion we note that this result is either 3!
(2.37)
• Exact if the distributions of X, Y and Z are Gaussian or
• An approximation with the error of the approximation where the ‘terms ðX 0 4 Þ’ indicates terms consisting of per-
between a function of the degree of relationship turbations to the fourth power and higher. This equation is
between X, Y and Z in the third- and higher order then substituted into Equation (2.36) and evaluated using
moments (like skewness) and how large the random the standard results for mean, variance and skewness:
perturbations are around the mean. b abðb 1Þ b2 2 bðb 1Þðb 2Þ b3 2 3=2
Z ¼ aX þ X σX þ X σX γ
This approach highlights one of the ways of arriving at 2! 3!
h i
þ E terms X 0
4
an approximate solution by ignoring third-order and (2.38)
higher moments of the probability distribution.
As many higher statistical moments (powers on the
perturbation) as required can be included, but because of
2.4.2.3 Example 3: Mean of a Single the division by the factorial the extra terms in the series
Nonlinear Input rapidly decrease in size. A common assumption is to
Finally we will consider a nonlinear transformation of a ignore all but the mean and variance (called a ‘second
random input. We will consider a power law relationship, moment approximation’) so that we have
because many of the relationships in this book are of that

form, but the generic approach is applicable to more b abðb 1Þ b2 2 b bðb 1Þ σ 2X
Z aX þ X σ X ¼ aX 1 þ 2
complex functions: 2! 2! X
(2.39)
Z ¼ aX b (2.34)
Comparing this to the exact solution for b ¼ 2 (Equa-
where a and b are (nonrandom) parameters for the trans- tion 2.35) there is a striking similarity in the form of the
formation. As a preliminary example we will consider a result. Moreover, the result of Equation (2.35) for b ¼ 2 is
special case where the solution is particularly simple, exact because the skewness and higher-order terms all
b ¼ 2. Substituting for the perturbations and solving for involve ðb 2Þ ¼ 0. The second moment approximation
the mean of Z, of ignoring higher than second-order perturbation terms is
quite powerful when used together with Taylor Series as it
Z ¼ E Z þ Z0
h 2 i allows the development of explicit solutions to general
¼ E a X þ X0 stochastic problems, though as the coefficient of variation
h i of X increases the error in the approximation also
¼ a E X þ 2XX 0 þ X 0
2 2
increases.
2 2 Finally we introduce the concept of an ‘effective’
¼ a X þ σ 2X ¼ aX 1 þ CV 2X (2.35)
parameter. We can rearrange Equation (2.39) as
where CV X is the coefficient of variation of the random
b bðb 1Þ σ 2X b
process X. Note that this is an exact solution that is Z ¼ aX 1 þ 2
¼âX (2.40)
2! X
independent of the form of the probability distribution of
X, even though it uses only the mean and variance of X. where the new equation using parameter ^
a is the same as
σ 2X
We now return to the original general problem in a ¼ a 1 þ bðb1
Equation (2.34) if ^ 2!
Þ
2 : The relationship
Equation (2.34). Substituting in the perturbations yields X
h between a and ^ a is simply a function of the nonlinearity of

b i
Z ¼ a E X þ X0 (2.36) the relationship, b, and the variability of the process. If we
are interested only in the mean effect on Z of the random
where the problem we have solving this is that we cannot input X, then this equation shows that if we use the
evaluate the expectation except in special cases of integer original Equation (2.34) with X and parameter ^ a , then
b, as we did for b ¼ 2 above. To evaluate Equation (2.36) we would obtain the same Z as we would by using the
random input X with parameter a and averaging the above. More complex examples are analysed in the same
resulting random output Z. Accordingly â is called the way. Consider
‘effective parameter’ and a the ‘true parameter’, where ^ a
Z ¼ aX 2 (2.41)
captures all of the effect of randomness in X on the mean
output Z: From the previous example we calculated the mean by
The concept of an effective parameter is an import- substituting in the perturbations and evaluating the
ant and powerful concept. It implies that if we do the expectation
averaging correctly we can model the average effect of h 2 i
Z ¼ E Z þ Z 0 ¼ E a X þ X0
2
a random input by simply adjusting process parameters ¼ a X þ σ 2X
and use the statistics of the input, rather than having to (2.42)
do all the calculations involved in treating input ran- h i
0 2
domness explicitly (e.g. Monte Carlo simulation). The variance of Z is ¼ E ðZ Þ , so the first step in
σ 2X
Moreover, in many cases the underlying functional evaluating the variance is to write an equation for the
form of the relationship is unchanged, though, unlike perturbation component of Z:
the simple example above, this is not always true. In
Z 0 ¼ Z Z ¼ aX 2 a X þ σ 2X
2
(2.43)
many cases such unchanged functional forms have been
implicitly assumed by process geomorphologists when We next substitute in the perturbation expression for X:
using event-scale equations to average up processes
2
Z 0 ¼ a X þ X 0 a X þ σ 2X
2
to geomorphic time scales. Fortunately, they are
often true.
¼ 2aX 0 þ ðX 0 Þ aσ 2X
2
(2.44)
Finally, if a model is calibrated to field data and the
only available data are the averages Z and X, then We then square this equation and take the expectation:
the calibrated parameter in Equation (2.34) will be the h i 2
effective parameter â not the underlying true parameter a. E ðZ 0 Þ ¼ E
2
2aX 0 þ ðX 0 Þ2 aσ 2X
This has implications for scaling up small-scale laboratory 2 h i h i
¼ 4a2 σ 2X 2a σ 2X þ 4aE ðX 0 Þ þ E ðX 0 Þ
3 4
models, typically lacking field scale variability, to field
scale. Equation (2.40) shows that the parameter a for the (2.45)
laboratory and field scales will be different. This differ-
If the skewness and kurtosis of X are zero, then the last
ence results from the difference in variance of the input X
two expectations are zero. If they are not zero, the second
at the laboratory and field scales. Equation (2.40) shows
moment approximation is invoked, and they are set to
that even for modest nonlinearity and field scale variabil-
zero. The variance of Z is then
ity (e.g. b ¼ 2 and CV ¼ 1) the difference between the
true and effective parameters is a factor of 2. 2
σ 2Z ¼ 4a2 σ 2X 2a σ 2X (2.46)
2.4.2.4 Example 4: Variance of a Single

Nonlinear Input 2.4.3 Conclusions
The previous example determined the mean of an input
that is transformed by a simple nonlinear power law The examples above outline the most common techniques
transformation. This example calculates the variance of for analysis of stochastic processes using perturbation
that same series. Determining the variance of a time analysis. Many useful problems can be broken down into
series involves additional steps. These steps are poten- components where the examples can then be used to solve
tially quite messy even for simple relationships, so they the problem, and certainly all applications in this book can
will be demonstrated here with the simple example used be solved using these four case studies.
3 A Brief Hydrology and
Geomorphology Primer
Water is central to many of the processes discussed in this between the energetics of the rainfall (i.e. the kinetic energy
book. Fluvial erosion on hillslopes is driven by overland of the raindrops when they hit the ground) and the amount
flow during rainfall events. Soil weathering, vegetation of rainfall (which drives the amount of runoff and thus the
and organic matter are driven by water in the soil profile. amount of fluvial erosion). We will defer discussion of this
Accordingly a brief summary of the key aspects of issue until Chapter 4, where rainsplash erosion will be
hydrology that are important is warranted. Moreover a discussed.
range of geomorphology concepts have permeated The second factor is the spatial distribution of rainfall,
through the landform evolution literature, either because and the interaction between rainfall location and (dynamic)
they triggered many of the original approaches or because topography, which we will discuss below. Rainfall and
they have been found useful when studying (either evaporation are related to elevation. Rainfall increases
experimentally or mathematically) landform evolution. with elevation, but there is no consensus about the rela-
This chapter will highlight the most important hydrology tionship. Potential evaporation also appears to increase
and geomorphology concepts in soilscape and landform with elevation, primarily because of increases in wind
evolution. For more detail the reader is guided to one of speed, though the number of studies is small.
the many introductory hydrology and geomorphology Anders and Nesbitt (2015) used satellite remote sensing
textbooks. of rainfall to identify trends of rainfall with elevation and
identified four climate regimes. For three of the climate
regimes the rainfall increases up to a height threshold of
3.1 Hydroclimatology
about 1–2 km and then declines for heights above that
3.1.1 Rainfall threshold. For the tradewind regime there seems to be no
trend with height, though this might simply be because the
The primary driver of hydrology is precipitation, and in analysis stopped at an elevation of 1 km (the maximum
particular rainfall. Snowfall is also important, but as a elevation of the land in the two regions studied) and didn’t
primary erosion agent it is the melting of the snow rather reach a threshold height.
than the falling of the snow that is important (i.e. floods Hutchinson (1995, 1998) found a parabolic relation-
ship between elevation and mean rainfall (R =2 ¼ αZ
1
and soil saturation), and we will not discuss this melt
behaviour. Rainfall on the other hand through both its where R is rainfall and Z is elevation), through the coeffi-
long-term mean (yield hydrology, driving soil moisture) cient α varied spatially, and he found aspect to be a
and its peaks (event hydrology, driving flood and erosion second-order factor after elevation. Hutchinson (1995)
events) drives landscape response. In Section 3.2 we analysed an area that is just south of Anders and Nesbitt’s
discuss the impact of (1) short-term, decadal scale, climate EAU region and with a similar exposure to the ocean, and
variability (e.g. El Nino, Indian Ocean Dipole [IOD], about double the elevation range (i.e. maximum elevation
Interdecadal Pacific Oscillation [IPO], North Atlantic about 2 km), so it is likely comparable with their EAU
Oscillation [NAO]) and (2) long-term, millennial to epoch analysis, where Anders and Nesbitt found no trend with
scale, climate variability (e.g. glacial-interglacial). height. Hutchinson (1998) analysed a portion of the
Two further factors are important because they drive the European Alps which has no equivalent in the Anders
relative dominance of landscape and soilscape evolution and Nesbitt analysis. In situations where the atmosphere
26 processes in the field. The first factor is the relationship is highly saturated with water, even small topographic
variations (50–60 m elevation) can lead to a doubling of Taiwan with an upslope lift model and found a good
rainfall (Bergeron, 1961). match for the northern part of the mountain range and a
Kyriakidis et al. (2001) on the other hand found a poor match in the southern (for both the amount of rain
relatively poor relationship with elevation (correlation and spatial distribution). The reason suggested for the
coefficient R ¼ 0:22) when 1 km spatial resolution eleva- poor performance in the south was that the mountain
tion data were used. However, a better correlation was range was too narrow to allow time for the raindrops to
found when the elevation was averaged by a 13 km form and fall in the location at which condensation
13 km window, suggesting that it was mesoscale topog- occurred (i.e. where the lift was occurring). In this case
raphy that drives orography rather than the high- the rainfall fell on the leeward side of the mountain range.
resolution topography. This suggests that a length scale, L, might distinguish
Alpert (1986) and Roe (2005) outline a simple model when the uplift model is appropriate, L ¼ ut, where u is
for precipitation distribution driven by orography and the wind velocity at right angles to the mountain range,
based on an atmospheric lift model: and t is the time for raindrops to form and then fall to the
ground. When L is large relative to the width of the range,
∂z zs=H
R ¼ αρ qu e m (3.1) then the uplift model is a poor approximation. Roe et al.
∂x
(2002, 2003) added a tuneable advection-diffusion term to
where ρ is the depth averaged air density, q is the depth- Equation (3.1) to adjust for the time to form raindrops, the
averaged specific humidity (so ρ q is the depth-averaged advection of these raindrops by the prevailing wind and
mass of water per cubic metre of air), u is the velocity of the spreading by gusting wind patterns. This allowed
∂z
the wind at right angles to the mountain range spine and ∂x rainfall to spill over the mountain ridge into the rain
is the slope of the mountain range at right angles to the shadow side of the range.
mountain range spine. The main assumption of the method One constraint on the uplift model is that the strato-
is that a parcel of air approaching the range is raised at a sphere provides a lid on upward flow of air within the
∂z
velocity w with w ¼ u ∂x : The parameter α is the rainfall troposphere at an elevation of about 10,000–15,000 m.
efficiency, which is the proportion of the condensation (i.e. Some mountain ranges block a significant proportion of
raindrops formed in the cloud) that actually falls to the the lower troposphere (approximate range heights: Hima-
ground (α < 1 and reported values are typically of the laya ~7,000 m, Andes ~5,000 m, Southern Alps NZ and
order of 0.2–0.5; e.g. Alpert, 1986; Yu and Cheng, Rockies ~3,000–4,000 m, Taiwan ~3,000 m) forcing wind
2013). The parameter H m is a maximum precipitation to flow laterally along or around the barriers, reducing the
height length scale that is of the order of 2–4 km. amount of air being uplifted (Garreaud et al., 2009). This is
Equation (3.1) is quite simple and one-dimensional, not accounted for in the uplift model above.
and can predict rainfall only on the windward side of the It is also possible to take an altogether simpler
range. A limitation is that with increasing slope the rain- approach to orographic rainfall and its effect of erosion
fall rate increases without bound, and the equation does rate: this is to simply assume a difference in erosion rate
not allow for the possibility of depletion of water in the between the windward and leeward sides of the range.
parcel of air that is being lifted by the mountain range. Willett (1999) assumed an erosion rate twice as high on
Specifically, ρqu is the mass rate of water entering the the windward size than the leeward side (and independent
ÐX
upwind boundary of the mountain range, while 0 Rdx is of elevation or slope) as a first step to understanding the
the amount of rainfall falling between the upwind bound- feedbacks between tectonics and erosion. However, like
ary of the range and a distance X into the range. At some Equation (3.1), this assumes that there is a known, dom-
distance X the rainfall integral (depending on how fast inant direction from which rainfall comes and thus a
the exponential term in Equation (3.1) declines with known windward and leeward side of the range, and this
increasing elevation) will equal the mass rate of moisture would need to remain true over geologic time since he
entering the upwind boundary of the range, so that no was interested in the effect of erosion on mountain range
more atmospheric moisture can be condensed. This deple- development.
tion of rainfall is commonly observed in the field with the This last point highlights one major constraint on
maximum precipitation occurring on the windward side of rainfall-topography relationships for use in landform
the ridgeline (e.g. Alpert, 1986). evolution studies. This problem is identifying what are
The use of the upslope lifting models like the Alpert the windward and leeward sides of topographic rises
model comes with a caveat. Yu and Cheng (2013) com- (hills, mountains and so on) since this distinction will
pared rainfall (observed using radar and rain gauges) in follow from prevailing wind patterns, which themselves
28 A Brief Hydrology and Geomorphology Primer
may vary with season and local topography (e.g. Espinoza adage in the agricultural land management community
et al., 2015). The recent availability of satellite rainfall that ‘if you can control surface runoff then you can control
products may lead to useful empirical relationships if erosion’, so surface runoff is a central factor for landform
prevailing wind directions are known (e.g. Nesbitt and evolution due to erosion.
Anders, 2009; Anders and Nesbitt, 2015), but the We turn first to the infiltration process. The partition-
observed coupling between wind and topography sug- ing between infiltration and surface runoff is a function of
gests that ultimately a coupled landform-atmospheric the wetness of the soil and the unsaturated conductivity of
model operating over geologic time may be required to the soil. If the initial soil wetness (called the antecedent
fully resolve the relationship for any particular location. wetness) at the start of a rainfall event is low, then the
infiltration rate is high as the soil initially wets up. As the
3.1.2 Evaporation soil wets up the infiltration rate starts to drop, and over
time the infiltration rate asymptotically approaches the
Methods to estimate potential evapotranspiration (i.e. the saturated conductivity of the soil column. If the rainfall
maximum rate of evapotranspiration which occurs if the rate is higher than the infiltration rate, then the soil will
soil is saturated) and its coupling with the elevations of an become saturated. The saturated conductivity is a function
evolving landform are less common than for rainfall. One of the soil properties, and the long-term infiltration rate
of the most common models for potential evapotranspira- increases for coarser soil grading (i.e. more sand and silt,
tion calculation is the Penman-Monteith model, which and less clay), lower bulk density (so pores are more
calculates the energy balance at the land surface. Kalma open), higher soil organic content (i.e. more open soil
et al. (2008) provides a recent summary of how potential structure; see Chapter 10) and a higher density of macro-
evaporation can be determined by using energy balance pores (i.e. old root channels, soil expansion/contraction
methods at a point (including Penman-Monteith), but cracks; Beven and Germann, 1982, 2013). Long-term
concludes that uncertainties in methods for determining multiyear drought has been shown to change runoff rates
the environmental inputs into the energy equation (i.e. (Saft et al., 2015, 2016a), and while the exact mechanism
surface and atmospheric temperatures) limit methods is not well understood, it is suspected to be linked to
for spatially extrapolating evaporation measurements. vegetation changes because of strong relationships of
McVicar et al. (2007) determined the topographic influ- surface runoff rate with spring rainfall and leaf area index
ence on evaporation over the Loess Plateau in China. (Tesemma et al., 2015; Saft et al., 2016b).
They spatially interpolated all of the input variables for In the period after the rainfall event finishes, the soil
Penman-Monteith (maximum and minimum temperature, water drains from the upper soil into the deeper soil
near surface wind speed, atmospheric water content) using layers until the upper soil moisture content reaches the
topography as one of the independent variables. The field capacity. The field capacity is the soil moisture at
correlations of the input variables with topography varied which water can no longer drain downwards under grav-
from summer to winter by factors of two or more, and ity and the water is accessible only to plants. The field
reductions due to the temperature gradient with increasing capacity is primarily a function of the grading of the
elevation were balanced by increases due to higher wind soil and is 15–25% by volume for sandy soils and
speed. Moreover, lapse rates for temperature changed 45–55% for clay soils (NRCCA, 2016). The decrease
from summer to winter. They did not find any consistent of soil moisture immediately after rainfall ceases is a
relationship between evapotranspiration and topography result of drainage and is a function of the saturated
(McVicar, personal communication). A 30 km transect hydraulic conductivity of the soil and may occur quite
showed a factor of two variability in ET along the tran- quickly, in a few days. If there are roots from plants in
sect, a bias compared with field data of less than 5%, and the soil column, then the soil moisture will be further
visually the results showed a stronger correlation with reduced by plants until the soil moisture content reaches
aspect than with elevation. the plant wilting point (a function of plant type, clay
content and soil salinity; typically 5–10% for sandy
3.1.3 Hydrology soils and 15–20% for clay soils; NRCCA, 2016). This
latter plant-driven soil moisture decrease is the water that
When it rains, either the water infiltrates into the soil or it is used by plants for transpiration, and the reduction
runs off as overland flow. Whatever cannot infiltrate either typically takes days to weeks. Any water that penetrates
ponds on the soil surface or generates surface runoff the soil profile below the root zone becomes recharge to
downslope. This surface runoff drives erosion. It is an deeper groundwater.
An important historical distinction is made in the bottom of the soil profile has a permanent groundwater
hydrology literature and divides hydrology into two table. The infiltration process described above occurs in
application areas. The first is ‘event’ hydrology, which the region above the water table, and any excess water
concentrates on the runoff in a single rainfall event to recharges the water table. A key characteristic of this
the exclusion of what happens between events, and the process is that toward the bottom of the hillslope in the
time resolution of modelling is typically of the order of valley bottom there is commonly a region of fully satur-
hours or days depending on the response time of the ated soil. This saturated region is where most surface
catchment to rainfall (sometimes called the ‘time of runoff is generated because rainfall cannot infiltrate into
concentration’). The second is ‘continuous’ or ‘yield’ the soil. The water table moves up and down quite slowly
hydrology, which concentrates on simulation of con- (timescale of weeks) since rises are driven by rainfall
tinuous time series of runoff over (typically) years or infiltration and water table declines are driven by evapor-
decades with time resolution of days or longer. Histor- ation loss and recharge to the river. Thus the percentage of
ically, because of computer limitations, these have been the catchment with the groundwater table at the surface
pursued as separate fields, even though there have been changes quite slowly, so that percentage of catchment
significant, though not complete, overlaps in the tech- generating runoff does not change significantly during a
niques used. Event hydrology has a stronger emphasis single rainfall event. Within the fully saturated zone it is
on the hourly dynamics including infiltration dynamics common for the soil chemistry to be anaerobic.
and river routing. Yield hydrology has focussed more
on the evaporation and transpiration dynamics that 3.1.3.2 Arid Zone Hydrology
dominate in the inter-rainfall event periods at the daily This is generally the case when rainfall is less than
and weekly resolution. evapotranspiration and results in ‘infiltration-excess’
There are also two main types of hydrology process runoff generation (Figure 3.1b). In this case there is a
dominance in the field, and the distinction is a function of net deficit of water within the profile, and there is rarely
how dry the climate is (Figure 3.1). any permanent saturated zone within the profile (though
saturation does occur during rainfall events). In extreme
3.1.3.1 Humid Zone Hydrology cases, other than the water that runs off during the rainfall
This is generally the case when rainfall exceeds evapo- event, all the soil water is returned to the atmosphere
transpiration, and results in ‘saturation-excess’ runoff through evaporation or transpiration (commonly 80–90%
generation (Figure 3.1a). In this case there is a net excess of rainfall). Runoff tends to be generated in those areas
of water in the hillslope so a permanent water table is where soil properties are such that infiltration rates are
formed within the soil profile. Starting at the catchment locally low (e.g. high clay content, low organic content)
divide as you proceed down the hillslope, this water or vegetation is sparse (so that there are few old root
excess increases and an increasing proportion of the channels for water to infiltrate through). It is possible for
FIGURE 3.1: The two main hydrology process types: (a) humid catchment where there is a semipermanent water table within the soil
profile and where any time it rains runoff occurs from those areas where the groundwater table has saturated the entire soil profile, (b) arid
catchment where evaporation is greater than rainfall so a permanent water table cannot be created, and where runoff occurs during
high-intensity rainfall events when the near-surface soil is saturated.
water to run off from localised low-infiltration zones into 3.1.4 Soil Profile Hydrology in the Partially
zones downstream with higher potential infiltration rates, Saturated Zone
and in this case the runoff can infiltrate in these high-
infiltration rate zones. Since runoff is generated by local The soil profile at the start of a rainfall event is normally
saturation at the surface of the soil profile, runoff gener- dry, and the potential infiltration rate at the start of an
ation can be very dynamic with infiltration capacity event is quite high so that most water infiltrates. If the soil
changing within a few minutes. The background soil is wet before the event, then a greater proportion of
moisture condition between rainfall events trends toward rainfall runs off. A simple approximation for what
the wilting point of plants. Rivers and creeks are ephem- happens with the infiltrated water within the profile is
eral, unless fed by deep groundwater sources. It is given by the Green-Ampt equation (Figure 3.2). This
common for the soil profile to be aerobic, but with the approximation assumes that at the start of a rainfall event
soil atmosphere having high carbon dioxide and low the soil profile at the soil surface saturates, and a satur-
oxygen levels in the root zone due to plant respiration ation front propagates down the profile as the infiltration
(Chapter 14) and biodegradation of soil carbon (Chapter 10). at the surface continues. Behind the front the soil is
The microbiology and plant growth dynamics are com- saturated, while in front of it the soil moisture is whatever
monly limited by water availability. it was before the event. A simple implication of this is that
the wetting front propagates down the profile at a velocity,
3.1.3.3 Transitional Hydrology v, determined by the infiltration rate, i, filling that part of
The distinction of hydrology into ‘humid zone’ and ‘arid the soil porosity that didn’t contain water before the event,
zone’ is a simplistic but useful shorthand for describing a ne , so v ¼ i=ne : The wetting front propagation stops when
catchment’s dominant processes. In reality the distinction the infiltration at the surface stops (e.g. at the end of the
is not just about climate. If we consider Figure 3.1a, if the rainfall event). For example, if the available porosity is
(1) soil was much deeper or (2) the hillslope much shorter, 0.2 and the rainfall in the event is 10 mm, then the wetting
then the water table may not have reached the surface so front will propagate down the profile 10/0.2 mm = 50 mm.
that the saturated zone at the bottom of the hillslope may After the rainfall event, the soil water evaporates, is taken
not have formed so that runoff would not be generated by up by plants or drains down into lower parts of the soil
the saturation-excess mechanism. Thus the drivers for profile below (e.g. in the example, below 50 mm). The
saturation-excess runoff generation are not just climate sharp wetting front is maintained during infiltration by the
(a permanent water table will not form without an excess
of rainfall over evaporation) but also shallow soils and
long hillslopes. Thus the drivers are a mixture of climate,
soils and geomorphology.
Moreover, if we consider Figure 3.1b, infiltration-
excess runoff will occur when the rainfall rate is high
enough to saturate the surface of the soil. This process
can also occur in Figure 3.1a in the parts of the hillslope
upstream of the saturated zone shown in the figure gener-
ating runoff. While not shown in Figure 3.1b, it is possible
that a water table exists that never reaches the surface.
Finally in a seasonal climate (where average rainfall
and/or evaporation are seasonally variable) it is possible
that during the seasons of excess rainfall that saturation
excess may dominate, while during periods of excess
evaporation there is a switch to infiltration excess. The
classic example of this was the Tarrawarra catchment in
southeastern Australia (Western et al., 1999) where during
winter (low evaporation) there were persistent wet regions
along the valley bottoms, while during summer (high
evaporation) these patterns disappeared and relatively FIGURE 3.2: A schematic of the Green-Ampt infiltration equation,
wetter regions were spread with no obvious pattern across with the wetting front moving down the profile as the water
all parts of the catchment. infiltrates at the surface.
strong contrast between the low hydraulic conductivity of observed rainfall from a semiarid site in eastern
the dry soil preceding the wetting front and the high Australia. This figure assumes that infiltration obeys the
conductivity in the saturated zone behind the wetting Green-Ampt approximation and ignores any redistribu-
front. A more sophisticated model for infiltration (Philip tion after the rainfall event. The key result is that for
model) will be discussed in Section 14.4. Green-Ampt is a deeper depths in the profile the amount of water passing
good approximation for soil moisture profiles in sandy that depth is reduced. This is also true for the number of
soils during the rainfall event. To do better than this times per year that infiltration events penetrate to wet
typically requires the use of an unsaturated flow model that depth. The functional form of the decrease will no
based on the Richards equation (e.g. the HYDRUS-1D doubt vary from site to site, and with aridity, but in the
model; Simunek et al., 2008), and full characterisation of case exemplified here the decline in flux and number of
the unsaturated conductivity behaviour of the soil through times that depth in the profile is wetted is approximately
the profile (e.g. Chen et al., 2014). exponential. This conceptualisation of the infiltration
The important point is that the upper part of the profile process will be used when we discuss chemical
is flushed with fresh rainwater more often than lower in weathering in soils.
the profile. The deeper part of the soil profile requires A fairly obvious observation is that there is potentially
bigger rainfall (i.e. less frequent) events for the wetting a coupling over pedogenic timescales between soil
front to propagate, and thus the deeper soil is flushed with moisture and the soils developed at a location, though
fresh rainwater less often. soilscape evolution researchers have not yet explored the
This vertical trend in wetting down the profile poten- details of this potential coupling. This coupling may lead
tially drives a depth dependent infiltration term in the to spatial organisation of soils in space, through a linkage
chemical weathering model discussed in the soil chemical between weathering and soil moisture.
weathering chapter (Chapter 8). For arid zones this verti-
cal flushing is the dominant process. For humid areas the 3.1.5 Runoff Generation
base of the soil profile has a groundwater table with lateral
flow down the hillslope. In this case the upper portions of Surface runoff is that rainfall that does not infiltrate. The
the soil profile above the water table are dominated by location where this runoff is generated depends on the
vertical flushing, while the deeper portions of the soil climate-controlled runoff mechanisms and soil properties.
profile within the water table are dominated by lateral Saturation excess tends to generate runoff around the
flushing from upslope. valley bottoms. Infiltration excess tends to generate runoff
An example of this depth-dependent infiltration is more randomly in space as a function of the soil properties
given in Figure 3.3. This figure shows the flux of water (Figure 3.1). It is possible for catchments to exhibit an
per year passing a given depth in the profile using arid soil moisture distribution in summer (when evapor-
ation is high or rainfall is low) and a humid one in winter
(when evaporation is less or rainfall is high) (e.g. Western
et al., 1999).
For saturation excess the rate of change of the eleva-
tion of the groundwater table within the soil can be quite
slow since it is driven by the net water balance of the
infiltration down and the discharge laterally downslope.
The timescales for change of the groundwater elevation
are typically of the order of weeks. Thus during each
rainfall event the elevation change of the groundwater
can be ignored, and the percentage of the catchment that
is saturated (and thus generating 100% runoff from the
rainfall) is the same throughout the event and is a func-
tion of weighted average of the preceding rainfall over
the previous few weeks. The runoff coefficient of the
FIGURE 3.3: Depth-dependent infiltration as predicted by the
catchment (i.e. the proportion of rainfall that is turned
Green-Ampt equation for a semiarid site (Blackville) in eastern
Australia: (solid line and left axis) the flux of water passing that depth
into runoff) is constant through the storm event and is
versus the depth, (dotted line and right axis) the number of wetting just the percentage of saturated area. Using the slope-
events that occur at that depth per year versus the depth. area and cumulative area geomorphology relationships
(see Section 3.3 for these relationships) it is possible to water ponds in the randomly spaced surface indentations
derive this percentage analytically (Willgoose and (called the microtopography). The routing of the water
Perera, 2001). then is a function of the residence time of the water in the
dams and how many of these dams the water needs to
cascade through. This is a basis of the Laurenson non-
3.1.6 Hillslope Runoff Routing
linear cascade (Mein et al., 1976), which approximates a
Once the rainfall is converted into surface runoff it is then series of dams by a single nonlinear storage of the form
routed over the hillslope down to the valley bottom and 1=m
Ay
then into the river network to the catchment outlet. The q¼ (3.3)
k
main factors affecting the routing are (1) the velocities of
the water on the hillslope and in the channel and (2) the where A is the catchment area, y is the average flow depth
length of the pathway that the water takes from the runoff and k and m are calibration coefficients. Experience has
generation location and the catchment outlet. The com- shown m to be in the range 0.6–1.0, giving a power on
bination of these two determines the time it takes for the average flow depth, y, equivalent to em in the range
surface runoff to reach the outlet. 1.0–1.67, which is consistent with derivations using the
One model for overland flow routing that is used in kinematic wave.
distributed hydrology models is the kinematic wave. The It should be obvious from these two approximations
kinematic wave model predicts the discharge of overland that it is difficult to determine desktop values for the rate
flow, q, based on the equations of the routing (i.e. the speed of water downslope) but
easier to determine the nonlinearity on how the rate
q ¼ KS1=2 changes with increases with discharge. Without site-
K ¼ cr Aecsm specific calibration there are few guidelines for what
1 values the parameters should take. For the kinematic wave
q 1= =em ððem 1Þ=em Þ
v¼ ¼ cr S 2 q (3.2) model the speed of the kinematic wave is a function of
Acs
the surface microroughness and microtopography. For the
where Acs is the cross-sectional area of the flow per unit nonlinear storage model, the amount or storage in the
width (which equals the average flow depth), S is the storage cascade determines the rate of routing. In this
slope, K is the conveyance, cr and em are hydraulic par- latter case the rate is a function of the amount of leaf
ameters (Willgoose and Kuczera, 1995) and v is the vel- litter, ability of the vegetation to inhibit movement of the
ocity of the flood. leaf litter and the microtopography of the surface.
One assumption that can be made to determine the An important consideration is that most of the pro-
parameters is to assume the Manning equation for flow cesses above occur at relatively small scale and are com-
depth and a perfectly smooth surface (e.g. a car park), monly not explicitly resolvable in landform evolution
which yields the parameters cr ¼ 1=n and em ¼ 1:67: models. Accordingly effective parameterisations need to
Natural surfaces, however, are rarely perfectly smooth be developed that account for these processes when mod-
and have rills and other microtopography that concentrate elling is performed at spatial and temporal resolutions that
flow. Willgoose and Kuczera (1995) explored the effect cannot explicitly resolve them. These parameterisations
on the kinematic wave parameters using two rill geom- are approximations, and they lie at the heart of how
etries: (1) a series of parallel equally spaced triangular rills hydrology is represented in landform evolution models.
where the rills are only partially full of water and (2) a The other factor impacting on the time that hillslope
natural topography from a plot study of Parsons et al. runoff takes to enter the channel is the distance that the water
(1990). They found for the same frictional surface rough- needs to travel to enter the channel. One estimate of this can
ness (i.e. Manning n) that the kinematic wave from the be determined from assuming that all water flows in a straight
natural and rilled surface was slower than for the perfectly line down the hillslope and then using half the hillslope
smooth surface (cr was 10–20% of the cr for a smooth length as the average travel distance. By definition the hill-
surface) and em ¼ 1:21:33: slope length is the inverse of the drainage density. The higher
An alternative to the kinematic wave for modelling the drainage density the shorter is the hillslope, so the travel
overland flow is to note that on hillslopes with significant time is shorter. Half the hillslope length is a reasonable
leaf litter, the litter is floated down the hillslope until it assumption for infiltration-excess runoff generation since in
gets caught on plants and then forms small dams of water. this case runoff sources are spread over the hillslope. For
These dams can also occur on a rough surface where the saturation-excess, however, runoff is primarily generated
from the saturated regions surrounding the channels, and hillslopes since most hillslopes are downward concave
very little from sites toward the tops of the hillslopes, so that so that velocities will increase down the hillslope. If we
assuming uniformly distributed sources will overestimate the combine the Manning equation for velocity, the
routing delays on the hillslope. discharge-area relationship (Equation (3.19)) and the
slope-area relationship for channel networks we obtain
3.1.7 Channel Runoff Routing an equation for the flood wave velocity with increasing
catchment area
The main difference between hillslope routing and channel
v kA0:1 (3.4)
routing is that flow in the channel is deeper so that in the
kinematic wave equation the flood wave speed is larger. where k is a constant that includes the impact of the
Thus even though the distances the water follows in the Manning roughness. Field observations also indicate that
channel network are longer, the higher velocity partially any area dependence on velocity, if it exists, is quite weak
offsets this. That said in most large catchments it is (e.g. Pilgrim, 1976, 1977). It is illuminating to derive
common to assume that most of the time lag from runoff Equation (3.4) with typical parameters because it shows
source to the catchment outlet is the channel routing term. how geomorphology can theoretically underpin empirical
Most modern flood routing models use either the kin- hydrology observations. If we combine the Manning
ematic wave equation or the diffusion wave routing equation for velocity and discharge
models. Except for very flat slopes (typically in tidal areas
R =3 S =2
2 1
near the coast), and very rapidly rising and falling (‘fla-
v¼
shy’) catchments (e.g. urban or mountainous catchments), n
R =3 S =2 w
the difference between the kinematic wave and diffusion 5 1
wave is small and less than the accuracy with which we Q¼ (3.5)
n
can estimate the parameters of either of the models.
One common approximation used is to assume the then
velocity of the flood wave is constant throughout the " #
1 2= 3=
catchment. In this case the geomorphic width function v¼ Q 5 S 10 (3.6)
n =5 w =5
3 2
(the number of flow paths, typically channels, at a given
distance from the catchment outlet; see Section 3.3.6)
where n is the Manning roughness coefficient, w is the
can be transformed into the unit hydrograph for that
channel width and R is the hydraulic radius. This is the
catchment (the discharge versus time for a unit depth
same as Equation (3.2) when the hydraulics of Equation
of generated runoff over the entire catchment). The
(3.5) is assumed. If we substitute the discharge-area rela-
distance in the width function is transformed into the
tionship in Equation (3.12), then the velocity is
time by use of the velocity, and the width of the flow 2 2 3
paths is turned into discharge by the unit runoff. Sto- =
β 5 2δ= 3=
chastic models for the width function have been v ¼ 4 3 2 5A 5 S 10 (3.7)
n =5 w =5
developed (Mesa, 1986), and the planar shape of the
catchment has been related to the width function and If we now substitute the slope-relationship (using the
the unit hydrograph (Rinaldo et al., 1995). Using Strah- notation of Equation (3.29))
ler network statistics a stochastic model for the instant- 2 2 3 2 2 3
=5 3=10 2δ 3α =5 3=10
aneous unit hydrograph, the geomorphic instantaneous β γ β γ
v ¼ 4 3 2 5A 5 S 10 ¼ 4 3 2 5A =5 þ =10
2δ 3α
unit hydrograph (GIUH), was developed by Rodriguez- =
n w5 =5 =
n w5 =5
Iturbe and Valdes (1979).
(3.8).
It may seem that the kinematic wave model, where
velocity increases with increasing discharge (Equation We can now evaluate this equation for typical param-
(3.2)), and the constant velocity assumption are in con- eters. For α ¼ 0:5 and δ ¼ 1 the exponent on the area
flict, but they may not be so. As the catchment area is 0.25, and for α ¼ 0:5 and δ ¼ 0:5 the exponent is
increases, the slope of the catchment decreases because 0.05. These two parameter sets are the end members of
of the upward concavity of the stream profile. Every- observed behaviours, so the scaling of velocity lies
thing else being equal (i.e. the same discharge), then as somewhere between these values. The exponent in
the slope decreases the velocity also decreases. This is Equation (3.4) was adopted to be midrange between
one significant point of difference between channels and these values.
3.1.8 Dynamics between Rainfall Events leaves per unit area of land surface). The latter two
mechanisms are the drivers for the water demand, while
For surface water (either on hillslopes or in channels) the first is a restriction on how fast water can be extracted.
most of the dynamics between rainfall events is a result Focussing on the soil water restriction for a given soil
of either (1) the draining of the hillslope or channel moisture, plant water uptake varies with depth below the
surface water storage as runoff decreases resulting in an surface of the soil. It is generally considered proportional
exponential or near-exponential decline in runoff after the to the mass of roots per unit volume of soil. Jackson et al.
cessation of rainfall or (2) long-term flow as a result of (1996) found that the best fit for the distribution of root
groundwater springs or groundwater discharge directly mass in the soil (globally across all climate zones and
into rivers or floodplains. vegetation types) was
The dynamics relevant to this book occur within the
soil. The Green-Ampt approximation discussed in the Y ¼ 1 βd (3.9)
previous section is a good approximation of behaviour
where Y is the fraction of the total root mass between the
while there is ongoing infiltration at the soil surface.
soil surface and the depth below the surface d (cm), and β
When infiltration at the surface stops, the surface rapidly
is a parameter that varies from 0.92 to 0.98. This shows
dries as the water in the profile drains down away from the
that most of the root mass is near the surface (so most
surface. The previously sharp wetting front spreads as a
water extraction occurs near the surface) but also that the
result of capillary surface tension forces within the pores
roots also exist quite deep in the profile, though at a low
of the soil. Eventually when the soil moisture equals the
density. Accordingly water may be extracted from deep in
field capacity, water can no longer move because it is held
the soil profile, up to 10’s of metres, albeit at a slow rate
stationary by surface tension forces in the pores between
(Canadell et al., 1996). Transpiration will continue until
the soil particles. The most important factors in the period
the soil water content approaches the wilting point of the
between rainfall events are then evaporation and transpir-
plants (which varies with species and soil water salinity),
ation, which remove water from the profile (until the soil
at which time the plants will either die or senesce.
moisture decreases to the plant wilting capacity, called the
wilting point) and return it to the atmosphere. Legates
et al. (2010) reviews the geomorphic and ecohydrologic 3.1.9 Peak Discharge versus Mean Discharge
significance of these soil moisture dynamics.
Evaporation occurs primarily in the top few milli- Many of the processes we will discuss in the following
metres near the soil surface and is the result of incoming chapters (e.g. erosion) are driven by one of a combination
solar energy being converted into latent heat of vaporisa- of two aspects of runoff: (1) peak exceedance of runoff
tion of the soil water. The water vapour is then transported over some threshold and (2) a function of the discharge
upward away from the surface by turbulence in the atmos- time series (potentially a nonlinear function in the case of
phere. Since a typical midsummer day can evaporate up to erosion time series).
10 mm of water, the top few millimetres of the soil are A practical example of where peak exceedance is
rapidly dried. In addition, surface tension at the water-air important is in the landform evolution of a landscape with
interface within the soil (capillary effects) creates a par- contour banks. If the discharge exceeds the capacity of
tially saturated zone above any zone of saturated soil. This the channel behind the bank, then water will flow over the
partially saturated zone is called the capillary zone. The bank potentially breaching the bank. If the breach is not
thickness of the capillary zone is a function of the size of repaired, this will lead to permanent adjustment of the
the pores in the soil. For sand the thickness might be a few flow pattern on the hillslope, potentially destabilising
10’s of millimetres, while for a loam agricultural soil it otherwise stable contour banks further down the hillslope
can be 2 m or more. This capillary force draws water up in future runoff events.
from deeper in the soil profile to the surface as the surface An example of nonlinear dependence on discharge is
dries. This allows evaporation to dry deeper layers in the transport-limited sediment transport where the sediment
soil, albeit indirectly, as a function of the pore size distri- flux at any time is given by an equation of the form (see
bution of the soil. Chapter 4 for details)
Transpiration extracts water from within the root zone
of the plants. The amount of water transpired is a function Qs ¼ cQ / Qm (3.10)
of the availability of water in the root zone, the potential where m is typically in the range 1.5–2 and m > 1 implies
evaporation rate and the leaf area index (the area of that the concentration of sediment c increases as the
discharge increases. Equation (3.10) is true during a run- and in the river network. This effect is normally modelled
off event. In our evolution modelling, however, we are with either (1) a storage model (e.g. Muskingham-Cunge),
often interested in the cumulative effect of the range of (2) a kinematic wave model (where the flood wave speed
storms, from small to big and from short to long, the is a function of the depth of flow) or (3) a diffusion wave
so-called ‘geomorphically effective’ erosion rate. This model (where flood wave speed is a function of both the
requires averaging Equation (3.10) over time. The larger depth of flow and the longitudinal curvature of the flow
the value of m then the larger the amount of sediment depth profile down the channel). However, the contribu-
transported by the larger discharges relative to smaller tion of this process to the nonlinearity of the discharge-
discharges (Section 2.4.2.3). As an end member we may area relationship is small, with δ > 0:9 for kinematic
ignore the lower discharges because they generate wave routing, using typical surface roughnesses and
insignificant erosion relative to the high discharges. Many channel/rill geometries (Huang and Willgoose, 1993;
traditional agricultural erosion models (e.g. USLE) Willgoose and Kuczera, 1995).
summarise this integration by using the peak 30 minute The main contributor to the nonlinearity in Equation
or 1 hour rainfall in a year (the so-called 1 in 1 year (3.12) is the spatial variability of rainfall. To explain this
extreme rainfall) as a way of characterising rainfall effect, consider a simple example of a thunderstorm
erosivity for a site. Likewise we can do the same for (typically a few square kilometres in area). For a small
discharge (and we will show how this can be used to catchment where the thunderstorm is larger than the entire
simplify erosion modelling in landform evolution models catchment, the peak discharge can increase linearly with
in Chapter 4) so that the erosion potential of a site can be area (i.e. δ ¼ 1) because as we add extra area to the
characterised by the mean annual peak discharge flood catchment this extra area is also covered by the thunder-
(the peak discharge of a 1 in 2.33 year extreme flood) so storm. However, for a large catchment where the thunder-
that the mean geomorphically effective sediment transport storm covers only a small proportion of the catchment
is then area, an increase in the catchment area will not change
the discharge because the extra area will not have any
Qs Qm
2:33 (3.11) extra rainfall falling on it (i.e. δ ¼ 0). Clearly this example
is rather simplistic, but it captures the idea that as the
where Q2:33 is the mean annual peak discharge. The
catchment area increases, the extra area added to the
significance of this equation is that the mean sediment
catchment falls outside the area of maximum rainfall so
transport is no longer a function of the mean discharge of
that, even if rain is falling, it is falling at a lower average
the catchment (i.e. the total volume of water discharged
rate than the average rate over the original, smaller, catch-
divided by the time period) but of the mean peak extreme
ment. In event hydrology this behaviour, called a ‘partial
discharge, the peak discharge of the hydrograph (see
area storm’, is modelled with an ‘areal reduction factor’
examples in Solyum and Tucker, 2004, 2007). This mean
where the peak rainfall generating the storm for a given
peak extreme discharge is almost always nonlinear with
area is reduced as the area of the catchment is increased
area and is commonly modelled by
(e.g. Rodriguez-Iturbe and Mejia, 1974). The example
Q2:33 ¼ βAδ (3.12) above also suggests that the exponent on area may change
with catchment area, being δ ¼ 1 for small catchments and
where δ is empirically observed to be in the range 0.5–0.8. declining with increasing area (Solyum and Tucker, 2004,
It is important to note that δ 6¼ 1 does not violate mass 2007), which suggests that Equation (3.12) with a fixed
continuity. The total volume of water that leaves the exponent on area, while commonly used by hydrology
catchment over the duration of the flood event complies practitioners, might be a poor approximation to the area
with mass continuity (i.e. the total volume is proportional dependence of Q2:33 : This is an area of ongoing research.
to area), it is just that the peak of the hydrograph does not Finally, Equation (3.11) characterises the peak flow
increase linearly with area. There are a number of reasons behaviour of the catchment and does not capture what
for this nonlinearity with area. happens for smaller and larger flows. Leopold et al.
One minor reason is that as the catchment area (1964) talks about the cumulative effect of the range of
increases, the hydrograph tends to become longer in dur- flood events, and introduces the concept of a ‘dominant’
ation as well as higher. Since the volume of runoff from discharge, the discharge that is most important for shaping
the event is the area under the hydrograph, this means that the geomorphology. For river channels this is sometimes
the peak cannot rise linearly with area. This effect is due equated with the ‘bankfull’ discharge, which itself has
to routing and temporary water storage on the hillslope been empirically observed to be about a 1 in 2 year
discharge. While these are all empirical findings, the matching the exact location of a gully with a given geom-
repeated appearance of a 1 in 2 year (or thereabouts) etry is deterministic replication. In contrast, we may not
extreme discharge event suggests that frequent extreme be able replicate the gully location and/or geometry deter-
runoffs are important in shaping the landscape. ministically, but we may be able to replicate the probabil-
ity of a gully of a given size occurs at that location, which
is statistical replication.
3.2 Modelling Climate Temporal Variability More challenging is the case of validating models for
field sites where the evolution is the function of pre-
In many landscape and soilscape evolution problems it
instrumental climate (i.e. instrumental records typically
is sufficient, at least to first order, to assume that the
go back only 100 years, and sometimes less). In this case
climate does not change over time. However, there are
we must rely upon paleo-climate data. Well-known paleo-
applications where explicit modelling of climate vari-
climate data are the ‘Hockey Stick’ curve for the last
ability is warranted. For instance, if we wish to charac-
1,000 years, and the ice core climate data going back
terise the effect of the different response times of the
about 800,000 years.
vegetation, soils and landforms and how their temporal
As an example we will consider the ice core climate
changes will interact, this will depend on the rate of
record. To convert the ice core data into reconstructed
change of climate relative to the landscape response
climate data, concentrations of CO2, isotopes of the gases,
time. This section is a summary of methods for simulat-
salt and dust trapped in the ice are related to climate. This
ing climate variability.
is done with proxy relationships where known climate
In many cases we may know the climate history and
records (typically instrumental but sometimes paleo-
wish to reproduce this exactly; this is deterministic vari-
climate data) are empirically related to atmospheric gas
ability. In other cases we don’t know the exact details of
concentration, and this proxy relationship is used to
the climate history, but we know statistics of the climate
reconstruct the climate from the ice core gas data. The
so that we can reproduce time series that have the correct
accuracy of the reconstruction is limited by (1) the con-
statistics (e.g. mean, variance, autocorrelation) but not the
sistency of the gas record and dating between ice cores
exact year-to-year values. We typically model this latter
(e.g. Bazin et al., 2013), (2) the accuracy of the proxy
case by stochastically generating a number of climate
relationship (e.g. Goodwin et al., 2004 shows correlation
histories (each climate history is called a realisation of
coefficients of 0.3–0.5) and (3) the temporal resolution of
the climate) and then run the model with each of these
the ice core data. For instance, the Vostok record (Antarc-
histories and then assess the statistics of the soilscape and
tica; Petit et al., 1999a, b) goes back to 414,000 years
landform response; this is stochastic variability.
before present and has an average resolution of about
The accepted impacts of climate variability are on
2,000 years, while the NGRIP record (Greenland; NGRIP,
temperature and rainfall/runoff. However, there are sig-
2004a,b) goes back 123,000 years before present and has
nificant, though poorly understood, impacts on vegetation
been gridded to a resolution of 50 years. Because of
and vegetation-fire feedbacks that will be touched upon in
higher snow deposition rates, Greenland cores tend to
Chapter 14 (e.g. Robertson et al., 2016).
have higher temporal resolution but shorter length of
record than Antarctica (Alley, 2014). However, in both
3.2.1 Deterministic Variability cases there is significant subgrid temporal variability (e.g.
resolution finer than 50 years for NGRIP) not extractable
In cases where it is required that an LEM be validated from these records. We will return to this subgrid tem-
against behaviour of a monitored site (e.g. an experimen- poral variability in the next section.
tal site), then the instrumental record (i.e. direct rainfall Other climate histories use tree rings, coral and
and evaporation measurements from instruments) should speleothem proxy data. The hockey stick curve is one
be used to reconstruct the runoff, erosion and soil mois- example of these data (Mann et al., 1999). The principles
ture, and subsequent performance of the site (e.g. Bell and are the same as for the ice core data: that is, some environ-
Willgoose, 1998). However, we know, particularly for mental indicator(s) that can be linked to climate, a proxy
landform evolution, that models are sensitive to initial relationship and the longer term paleo-data record(s) from
conditions and forcings, so we should not necessarily which climate is then reconstructed with some accuracy
expect that the model predictions would deterministically and temporal resolution. They tend to be location specific,
match the evolution observed in the field. This can be so I leave it to the reader to investigate paleo-climate
contrasted with statistical replication. For instance, reconstructions for their site, or for methods to
3.2 Modelling Climate Temporal Variability 37
FIGURE 3.4: Ice core reconstructed temperature data for the last 420,000 years with current day at the extreme right-hand side of the
graph, together with the Cohen saw-tooth approximation to that record (from Cohen, 2010).
geographically translate existing records to their site (e.g. cases where we may have a deterministic record (as in the
Ho et al., 2014). previous section), the low temporal resolution of the record
In landscape evolution studies it has been common to may require infilling or disaggregation between data points
adopt simple sinusoidal or step-change climate change to a higher temporal resolution required by the modelling.
(e.g. Willgoose, 1994b; Moglen et al., 1998; Whipple In these cases a range of stochastic climate simulators have
and Tucker, 1999; Baldwin et al., 2003), either to exam- been used in the hydrology community, and they are
ine the dynamics of numerical models or to develop typically referred to by the generic terms ‘climate simula-
analytical expressions for the dynamics, but it is less tor’ or ‘disaggregation model’, or more specifically by the
common to use reconstructed climates. Cohen (2010) techniques used such as ‘auto-regressive moving average
and Cohen et al. (2013) used the Vostok record to (ARMA)’, ‘point process’ or ‘Markov chain’. This section
examine the dynamics of soilscape evolution on a hill- cannot review all of the techniques, but will focus on the
slope. Cohen (2010) also examined the impact of using two most commonly used for rainfall simulation: ARMA
a saw-toothed approximation to the Vostok record and Markov chain models.
(Figure 3.4), and while short-term (<10,000 years) fluc- A discussion of disaggregation and ARMA models in
tuations were smoothed out the long-term trends were hydrology can be found in Bras and Rodriguez-Iturbe
unchanged. This 10,000 year smoothing was consistent (1985). A key feature of disaggregation models is that
with simulations that used a simple step-change in the they take a low time resolution time series and break it up
imposed climate, which indicated a circa 10,000 year (disaggregate) into a higher resolution series, and the
response time for the soil profile. methods ensure that the high-resolution series replicates
the low-resolution series: that is, if a yearly resolution
3.2.2 Stochastic Variability series is disaggregated to a monthly series, then when
the months are added up the yearly totals will exactly
In many situations we do not have a deterministic climate match the original yearly series. The techniques discussed
record to use, but we may have climate statistics. Even in below do not provide that assurance.
ARMA models are a simple technique for generating Markov chain rainfall models simulate a time series
stochastic time series when the probability distribution where a timestep (for simplicity of explanation let us
function is Gaussian distributed and the time series is assume a day but it can be any period) is randomly
autocorrelated. In many cases the time series will not be simulated to be either wet or dry. If the day is wet, then
Gaussian but can be transformed so that it is Gaussian the amount of rainfall falling on that day is then sampled
after the transformation. A commonly used transformation from a probability distribution (typically exponential or
is the Box-Cox transformation Gamma distributions provide a good fit to daily rainfall).
8 λ If a wet day follows a previous wet day, then the rainfall
< Yt 1
λ 6¼ 0 on the second day may or may not be correlated to the
Xt ¼ λ (3.13)
: rainfall on the first day. If the rainfall is Gamma distrib-
ln λ λ¼0
uted, then this model will have four parameters; the
where the parameter λ is the Box-Cox transformation probabilities that a wet day follows a wet day (pwetwet )
parameter, X t is the transformed value and Y t is the and a dry day follows a dry day (pdrydry ), and the two
original untransformed value at time t. For λ ¼ 0 this parameters of the Gamma distribution. Note that the
equation is a log transformation, while for λ ¼ 1 the time probability that a wet day follows a dry day is pdrywet ¼
series is untransformed other than a change in the mean. 1 pdrydry and likewise pwetdry ¼ 1 pwetwet : The
A simple and commonly used ARMA model is one simulation process is that at every timestep a uniformly
where the value at time t þ 1 is correlated with the value distributed random number between 0 and 1 is generated,
at time t: an autoregressive lag-1 (AR(1)) model. The U tþ1 : If the previous day was wet, then if U t < pwetwet
simulation process is to take the value at time t and then the day will be wet, otherwise it will be dry. Likewise if
simulate the value at time t þ 1 using the lag-1 correlation the previous day was dry, then if U t < pdrydry the day
ρ: If X t is the value at time t, then the value at time t þ 1 will be dry, otherwise it will be wet. It is common to
X tþ1 is observe a seasonal variation through the year for the
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi parameters (e.g. wet seasons typically show a larger
X tþ1 ¼ ð1 ρÞμX þ ρX t þ σ X 2 ð1 ρ2 Þεtþ1 (3.14)
value for pwetwet and a smaller value for pdrydry , so
where εtþ1 is a Gaussian random variable with mean ¼ 0 wet periods last longer and are more frequent, and a
and variance ¼ 1, and μX and σ X 2 are the mean and Gamma distribution with a larger mean, so that wet
variance of X. Higher order autoregressive models (where periods are wetter). In this case the parameters can vary
the value is correlated not just to the previous timestep but from season to season or from month to month (e.g.
other previous times as well) and moving average (i.e. the Chowdhury et al., 2017).
MA part of ARMA) are possible. The reader is referred to
Bras and Rodriguez-Iturbe (1985) for details. 3.3 Geomorphology
It is also possible to model two or more variables
simultaneously (e.g. rainfall and evaporation) where the This section will briefly outline some geomorphology
variables are cross-correlated. In this case X t and εt techniques that are commonly used to describe geo-
become vectors, and the equation involves matrix multi- morphology and LEMs. Some of these topics have
plication on X t and εt where the matrices encapsulate developed in conjunction with LEM research in the last
both the cross-correlations and autocorrelations. This two decades, so to fully understand the background to
general multidimensional extension is the Yule-Walker these concepts requires knowledge from the chapters
equation. that follow. In this case the reader may wish to skim
ARMA models are quite simple to use and calibrate. these sections now for the key concepts and outcomes
MATLAB, Python and R all have packages to calibrate but come back, as they are reading the remainder of the
and simulate ARMA models. The main problem with book, to fully appreciate some of the physics underpin-
ARMA models is the requirement for the variable(s) to ning them. Most of the statistics below have strong
be Gaussian. For instance, while mean annual rainfall is physically-based causal explanations for homogeneous
commonly close to Gaussian, daily rainfall is not. Daily (rainfall, geology) catchments. When there is spatial
rainfall has many days with zero rainfall, and it is not variability, the behaviour departs from the explanations
possible to transform these zeros in such a way that the below, and this departure itself may be a good diagnos-
time series can be made Gaussian. Daily resolution rain- tic of catchment process (e.g. Wu et al., 2006; Solyom
fall and runoff is a case for which Markov chain models and Tucker, 2007; Huang and Niemann, 2008; Roy
are commonly used. et al., 2015).
3.3.1 Channel Network Classification Systems generate networks that were topologically distinct (i.e.
the arrangement of the links was different) random net-
Network classification systems are a means of conceptu- works (TDRNs) with a given magnitude/area. TDRNs
ally organising the arrangement of the channel network. have been the basis of much work on randomness in
Some of the schemes involve physical length and areas, channel networks.
but most of them simply summarise the topology of the The main problem with these schemes is, no matter
network. I will not provide a comprehensive summary their enduring appeal in the geomorphology community,
here but only those that will be useful elsewhere in the that they are coarse measures of network topology and
book. A more comprehensive discussion can be found have large random variability. Willgoose (1989) used a
elsewhere (e.g. Abrahams, 1984). landform evolution model and showed that with random
A definition common to the classification systems initial conditions the bifurcation ratio of the resulting
below is that a ‘link’ is the channel segment between network ranged between 2 and 6, this range overwhelm-
two adjacent tributaries. An ‘exterior’ link is a link ing any signal resulting from changing process param-
between the upstream end of the channel where the eters. Using the Shreve TDRN model Kirchner (1993)
hillslope transitions into a channel and the first tributary reached a similar conclusion that randomness with order
encountered travelling downstream, while an ‘interior’ overwhelmed any deterministic signal.
link is any link between two tributaries.
The Strahler (1952) scheme classifies all exterior 3.3.2 Conceptual Channel Network Models
links as Strahler ‘order’ 1. Moving down a channel if
this channel encounters a tributary of lower order, then Prior to the current research using physically based land-
the order of the next link downstream remains the same form evolution models a range of channel network models
order. If the tributary is the same order, then the next link were used to simulate the blue lines on contour maps.
downstream is incremented by one. If the tributary is of They ranged from simple stochastic models, to models
higher order, then the next link downstream has the based on energy minimisation and entropy maximisation.
order of the higher order tributary. A Strahler ‘stream’ Two models have stood the test of time and are
is then that river segment of connected links that going summarised here.
downstream all have the same order. A catchment or The first model is the headward growth model
drainage network is said to have order k if the link at (Howard, 1971). In this model a grid is defined and a
the catchment outlet has order k. Strahler defined the value of 0 (hillslope) or 1 (channel) allocated to each node
‘bifurcation ratio’ as the average number of (n1) order of the grid. Initially the entire grid is 0 except for one
streams that flow into a single n order stream. Shreve point on the outside edge of the grid that is 1, which is the
(1967) showed that an infinite topologically random starting point for the channel network. Headward growth
network yielded a bifurcation ratio equal to 4. Tokunaga of the channel network proceeds in a series of cycles. At
(1978) showed that the bifurcation ratio was scale each cycle one node that is adjacent to the existing chan-
dependent and that the bifurcation ratio changed with nel network (i.e. that directly abuts to a node that is
the order of the drainage network. He defined a new ratio already 1) is randomly selected that becomes a new node
that defined how many (ni) order streams drained into in the channel network (i.e. goes from 0 to 1). In this way
the single order n stream. Willgoose (1989) showed that the channel network grows outward from the first initial
as the network order increased, the Tokunaga definition starting node. Howard and subsequent authors have exam-
asymptotically converged on the Strahler bifurcation ined changing the rules (e.g. having more than one node
ratio of 4. become channel in each cycle), but the fundamental
Shreve (1967) introduced the concept of stream ‘mag- notion of a channel network growing headward from a
nitude’ and ‘topologically random channel networks starting node remains.
(TDRN)’. Every exterior link has magnitude 1, and the A similar approach, commonly used in the physics and
magnitude of an interior node is the sum of the magnitudes cellular automata community, is based on diffusion-
of the two immediately upstream links. If a network has n limited-aggregation (DLA) (Witten and Sander, 1981),
exterior links, then the magnitude of the network is where particles that are not connected to the network are
ð2n1Þ, and if the area contributing to each link is the moved around on the grid by diffusion until they hit the
same, Al , then the area of the catchment is ð2n1ÞAl : network at which time they become part of the network.
Magnitude has sometimes been used as a proxy for catch- Willgoose (1989) showed that for that part of the region
ment area. Shreve (1967) introduced a methodology to that is not occupied by the network (i.e. where the
particles are moving around by diffusion) this is a discrete hillslope tops are concave down and that channels are
approximation to concave up given the characteristic rolling hills and
2 valleys, and that the rounded hilltops result from a trans-
∂z ∂ z ∂2 z port process that increases with slope downstream (e.g.
¼ D þ (3.15)
∂t ∂x2 ∂y2 Gilbert, 1909). Empirical studies have also long shown
that for rivers an approximately log-log linear relationship
where z is the concentration of particles/unit area, and
exists between distance downstream and slope (Hack,
D is the diffusivity. Looking ahead to the discussion of
1957; Tarboton et al., 1989). The availability of digital
soil creep (Chapter 9) this is the same equation used for
elevation models worldwide and GIS tools has made
a landform dominated by linear creep when z is defined
slope-area analyses easy to perform.
as elevation.
Kirkby (1971) showed mathematically how the rela-
The second model is the Optimal Channel Network
tionship between the distance downslope and the gradient
(OCN) model. In this model a channel network is created
in the transport-limited erosion process determines the
by some method (typically by using a headward growth
concavity of one-dimensional hillslopes (one-dimensional
model, or by simply allocating every node a random flow
hillslopes have no flow divergence or convergence).
direction while ensuring there are no closed loops), and
Willgoose et al. (1991c) extended Kirkby’s work by
then the network topology (i.e. how one node connects to
solving mathematically the catchment problem in two
another node) is modified until the total energy consump-
dimensions (of which Kirkby’s solution is then a special
tion of the network is everywhere equal and minimised.
case for one dimension) for dynamic equilibrium. Howard
The energy consumed by the flow in each node is propor-
(1980) provides an equivalent two-dimensional solution
tional to Q0:5 : The area draining through each node is
for the case of detachment-limited erosion. The conclu-
derived at every point in the grid. The discharge at every
sion of these works is that the slope and drainage area of
node is determined from the area. The total energy con-
P hillslopes and catchment can be plotted against each other
sumed by the network is Ep / Q0:5 where the summa-
for common transport- and detachment-limited erosion
tion is over all the nodes in the grid. The network that
physics (see Chapter 4 for a discussion of erosion phys-
minimises Ep is the OCN for that grid. The exponent of 0.5
ics), and the slope and area are log-log linearly related
is empirical but was derived by using empirical equations
(Figure 3.5). Willgoose (1994a, b) extended his 1991 solu-
for how velocity and channel width change with area/
tion for the cases of (1) zero continuing uplift and (2)
discharge, and the slope-area relationship (Section 3.3.3).
cyclic uplift, and showed that the dynamic equilibrium
If any of these empirical equations change, then the expo-
relationship can be simply extended by including the
nent of 0.5 is changed, and the form of the resulting OCN
average catchment elevation. Except for very low eleva-
will change. Ijjasz-Vasquez et al. (1993) compared the
tion catchments the effect of this average elevation cor-
OCN to field catchments and found the differences
rection is small, and less than the scatter typically
between the field catchments and the OCN were small with
encountered in the slope-area diagrams using field data.
the field catchments having an E p about 5–10% higher than
Finally, Whipple and Tucker (2002) compared the time-
the OCN. Ibbitt et al. (1999) compared networks generated
varying dynamics of elevations for transport- and
by the OCN and the SIBERIA landform evolution model,
detachment-limited erosion. They found that while the
and found that SIBERIA and the OCN systematically had
equilibrium trends were the same, the time-varying
differences in E p of about 10%, and that SIBERIA had an
dynamics were different. Detachment-limited erosion
Ep closer to the field catchment than the OCN. When the
created localised high-slope regions in river channels
OCN was used as input to SIBERIA, the channel network
(with some resemblance to knickpoints) when adjusting
changed and Ep converged to the SIBERIA result. Ijjasz-
between equilibria. For transport-limited erosion the tran-
Vasquez et al. (1993) speculated that their field catchments
sition was smooth and did not create knickpoints.
had not reached optimality, while the results of Ibbitt et al.
We will talk in detail about erosion processes in Chap-
(1999) suggested more fundamental differences.
ter 4, but a simple calculation will highlight the process
dependencies of the slope-area relationship for equilib-
3.3.3 Slope-Area Analysis rium landforms. If we assume a transport-limited sedi-
ment transport model where the sediment transport Qs is
Probably the most significant advance in analysis tech-
Qs ¼ K 1 Qm1 Sn1 (3.16)
niques for landscapes to be developed in recent years is
slope-area analysis. It has long been recognised that and the discharge Q is related to area as
An alternative derivation can be performed for catch-

ments where detachment-limited erosion occurs. For
detachment limitation the erosion rate at every location
of the catchment can be expressed as
E ¼ K 2 Qm2 Sn2 (3.20)
When we substitute Equation (3.17) for discharge and

assume that the erosion rate at every location in the
catchment equals the uplift we obtain (Howard, 1980)
1=n2
U
S¼ Aα2
K 2 β3 m2 m3
m2 m3
FIGURE 3.5: Slope area analysis with typical data derived from a
α2 ¼ (3.21)
n2
ground survey also showing a best fit slope-area curve through
these data for one (the straight lines; see text for explanation) or The similarity between Equations (3.19) and (3.21) is
two processes (the curved line). The dots are the raw data from striking, with the main difference being the change in
analysing a 20 m gridded DEM from the ground survey, while the
the value of the slope-area scaling exponent. Again if
circles are from averages of 20 pixels to better show the mean trend
(after Willgoose, 1994b). The large scatter in the slope for any given we use typical parameters for detachment-limited erosion
area is not a reflection of low spatial resolution elevation data but in rivers, we find that α2 0:5, which is consistent with
seems to be fundamental because recent analyses using high spatial observed exponents.
resolution LIDAR data exhibit similar scatter. The area at the A number of authors have noted the similarity in the log-
intersection point of the fluvial and soil creep lines At is given by
log linear form of Equations (3.19) and (3.21) and the typical
Equation (3.28).
slope-area scaling exponent, which means it is difficult to
differentiate river process (i.e. transport- versus detachment-
limited) solely from slope-area analyses (e.g. Whipple and
Q ¼ β3 Am3 (3.17) Tucker, 2002). However, the physically based derivations
then it is possible to write an equation where the sediment above have provided a strong physical rationale for using
transport out of the catchment is equal to the uplift U over slope-area analyses for analysing field catchments, and the
the catchment (i.e. QsðoutletÞ ¼ UA). The transport-limited application literature is now too big to summarise here.
form of the slope-area relationship is then (Willgoose When there is more than one process, there is an area
et al., 1991c) threshold above and below which the different processes
will be dominant (Willgoose et al., 1991c), leading to the
K 1 β3 m1 Am1 m3 Sn1 ¼ UA (3.18) possibility of there being different log-log linear relation-
ships applying for different ranges of catchment area. We
Rearranging this yields the classic form for the slope-area
will see examples later in the section. Finally, it is possible
relationship
to obtain slope-area relationships (albeit probably not log-
1=n1 log linear) even if the erosion processes don’t have the
U
S¼ Aα1 nice power law relationship in Equations (3.16) and
K 1 β3 m1 m3
(3.20). All that is required is that the erosion processes

1 m1 m3
α1 ¼ (3.19) depend only on area and slope.
n1 If the erosion processes are not a function of slope and
This is a log-log linear relationship between area and are only a function of area, then it is possible to determine
hillslope gradient. The parameter α1 is called the slope- a threshold uplift rate where for uplift rates higher than the
area ‘scaling exponent’. The term inside the parentheses threshold no equilibrium is possible. The elevations will
reflects the balance between uplift and erosion rates and increase without bound. Likewise, below this threshold
material erodibility, and determines the average slope of elevations will decrease without bound. Fortunately for
the catchment. Typical fluvial erosion parameters yield the case of high uplift, when this happens other processes
α1 0:5, which is approximately consistent with ana- such as landsliding (with its strong threshold slope
lyses of fluvial erosion–dominated river networks. dependence) and glacial erosion (with its strong threshold
elevation dependence) (Mitchell and Humphries, 2015) define spatial trends in uplift (Willgoose, 2001) and run-
become dominant, so mountains do not rise without limit. off/erosion (Roe et al., 2003) with varying success.
One of the main problems with slope-area analysis is A recent methodological development of slope-area
that there is always considerable scatter in the slopes of analysis is chi plots (not to be confused with the chi
individual pixels for a specific area (typically one or more statistic from the statistics literature) (Perron and Royden,
orders of magnitude; see Figure 3.5), and we still don’t 2013). This technique addresses the scatter in the slope-
understand what the driver for this variability is. The area by calculating an integrated measure of slope-area
variability can also make it statistically difficult to com- from the outlet of the catchment to some point up a river
pare processes between field catchments, and between profile, to yield average statistics for that reach of the
landform evolution models and the field. A number of river. The description of Perron and Royden unnecessarily
authors have developed spatial averaging methods to limits the method to bedrock rivers at equilibrium. It is
address this (e.g. Cohen et al., 2008), but scatter about sufficient that the log-log linear slope-area relationship
the mean slope-area trend when using field data remains applies:
a problem.
dz
The slope-area relationship has been used in two sig- S¼ ¼ γAα (3.22)
dx
nificant landscape models that don’t model landscape
evolution. The first is the Optimal Channel Network where z is the elevation and x is the distance along the
(OCN) (see the previous section on Conceptual Network river positive in the upstream direction. Integrating this
Models) where it is used to integrate out the elevation in equation up from the catchment outlet yields
the derivation of the spatial distribution of energies in the ðz ðx
energy optimisation when the OCN generates its planar dz ¼ γAα dx (3.23)
zb xb
channel networks (Rodriguez-Iturbe et al., 1992b; Rigon
et al., 1993). The second is the Quick Equilibrium Land- where (x, z) is the location and elevation at some point
forms (QUEL) model (Ibbitt et al., 1999; Willgoose, upstream of the catchment outlet (xb , yb ). If the value of γ
2001), which uses a channel network as an initial condi- (the ratio of uplift to erodibility) is constant along the
tion and then reconstructs the elevations of the catchment river, then γ can be moved outside the integral so that
using the slope-area relationship. ðx
Finally, Tucker and Whipple (2002) reviewed field z ¼ zb þ γ Aα dx (3.24)
xb
slope-area data (mostly for the United States) and used
Tucker’s GOLEM landform evolution model with esti- It is convenient to normalise the area with an area scale
mated process parameters from the equilibrium slope-area ^ so that
(e.g. the area of the catchment at the outlet) A
relationships in Equations (3.19) and (3.21) to generate ðx
^α A α ^ αχ
landforms with different process parameters. They then z ¼ zb þ γA dx ¼ zb þ γA (3.25)
xb A^
examined a range of geomorphic features of these land-
forms to assess whether these landforms were feasible in where χ is the chi integral for the location (x, y) on the
the field. In an earlier and more limited study Willgoose river. The plot of z versus χ is the chi plot and is a straight
and Gyasi-Agyei (1995) showed similar results when they line when Equation (3.22) applies because γA ^ α is a con-
compared the differences in how landforms evolved when stant. This chi plot has the advantage of having much less
they chose process parameters that were consistent with scatter than the traditional slope-area plot (e.g. Figure 3.6)
either the USLE or CREAMS agricultural erosion models. and still has an ability to map spatial variations in γ and α
One problem in interpreting process functionality from that may be indicative of spatial changes in process or
the slope-area relationship is that it is not a one-to-one uplift (Royden and Perron, 2013; Fox et al., 2014) though
relationship. For instance, in Equations (3.19) and (3.21) the quantitative interpretation of variations from linearity
an increase in uplift U can be balanced by an increase in in Equation (3.25) due to spatial changes is not as direct as
erodibility K, leaving slopes changed. Moreover, as will it is with traditional slope-area analysis. As previously
be shown in Chapter 4 (where Equation (3.16) is derived noted, chi plotting is applicable to both transport- and
from first principles) the parameter K includes a range of detachment-limited landforms and requires only that the
channel geometry parameters that are known to change catchment comply with the log-log linear slope-area
with location (e.g. channel width, area, roughness), and relationship.
that may confound any process estimation. Despite these Finally, we return to the physical explanation for the
problems researchers have used maps of the S=Aα to classical slope-area diagram for a short discussion of what
FIGURE 3.6: Chi plot (a) the river profile on traditional elevation and distance axes, (b) the same river profile transformed to chi
coordinates (from Royden and Perron, 2013).
happens when multiple processes operate. Both the trans- shape of this transition depends on the ratio m1 m3 : n1 :
port- and detachment-limited explanations (Equations This explanation is not restricted to the combination of
(3.19) and (3.21), respectively) for fluvial erosion yield a transport-limitation and soil creep, but the exact form of
negative value for α of about 0.5. If this equation is the transition will be different for different combinations
evaluated for decreasing area, the slope will increase, of processes. The key conclusion from this is that at
approaching infinity at the catchment divide. This is some intermediate area there is a transition from concave
clearly unrealistic. Soil-mantled landscapes have rounded down to concave up in the downslope profile of the
hilltops, while mountains have a maximum slope above landform (point B in Figure 3.5), and the area at which
which landslides occur. In both cases for high slopes and this occurs is a function of the relative magnitudes of the
small contributing areas processes other than fluvial ero- competing soil creep (commonly referred to as the ‘dif-
sion become dominant. For soil-mantled landscapes, soil fusion process’) and fluvial erosion (commonly referred
creep (Chapter 9) is important, so that the classical for- to as the ‘incision process’). This competition can be
mulation for a transport-limited catchment (Willgoose seen if Equation (3.27) is solved for the area at point
et al., 1991c) is At in Figure 3.5:

Qs ¼ K 1 Qm1 Sn1 þ DS (3.26) !1 ðn þm m 1Þ
Dn1
1 1 3
and following the derivation above for dynamic At ¼ ðn Þ =n

K 1 β 3 m1 U
1 1
1
equilibrium
1=n1
U
K 1 β3 m1 Am1 m3 Sn1 þ DS ¼ UA (3.27). St ¼ At α1 (3.28).
K 1 β 3 m1 m3
This equation can only be solved analytically for slope for
the values n1 ¼ 1 and 2, but can be solved numerically to For typical parameters the two exponents on the paren-
fit data (Figure 3.5). If fluvial erosion is the dominant theses are positive:
process, then the data follow the line labelled ‘fluvial
n1 2 ½1; 2; m1 2 ½1; 2; m3 2 ½0:5:1;
erosion’, while if soil creep is dominant, then the data .
follow the line labelled ‘soil creep’. When both of these ðn1 1Þ
n1 2 ½0; 0:5;
processes are of similar magnitude, then the data follow
ðn1 þm1 m3 1Þ 2 ½0:3; 2
1
the curved line and (1) for large area the catchment
asymptotically approaches the fluvial erosion line because α1 2 ½0:3; 0:6 (3.29)
the area term in Equation (3.27) dominates, (2) for small
area the catchment asymptotically approaches the soil so that as the ratio of the rate of soil creep D to the rate of
creep line because the area is small so that soil creep erosion K 1 the area increases (the peak of the slope-area
dominates and (3) for intermediate areas there is a curved curve shifts right), and vice versa. When the runoff and
transition between these two end members. The exact erosion rates β3 and K 1 increase, the area decreases. Thus
it can be seen how the dominance of the processes characterise (historically by the mapping of blue lines on
changes the area at which downslope convexity changes contour maps). With the advent of DEMs a finer reso-
sign and thus what percentage of the catchment is ridge/ lution statistic was developed, the cumulative area
hilltop and what percentage is valley (see the cumulative diagram (CAD) (Figure 3.7). The CAD gives the percent-
area diagram in the next section to quantify this percentage of the nodes in the DEM (y axis) that have an area
age). Note also that as the uplift rate increases (so the draining through them greater than a given area (x axis).
catchment becomes steeper) the area decreases. Willgoose Normally (1) the areas on the horizontal axis are normal-
et al. (1991c) provides a more general solution for At for ised by the maximum area in the DEM and (2) the analy-
when both processes are nonlinear functions of both area sis is done on individual catchments rather than DEMs as
and slope. a whole. In the latter case ignoring catchment boundaries
One point that arises repeatedly in the pages that leads to many partial catchments, some of which will have
follow is what happens when (1) erosion rate decreases area draining into them from off-DEM so the drainage
or (2) creep increases, with everything else unchanged. In areas cannot be calculated. Rodriguez-Iturbe et al. (1992a)
these cases the transition area At in Equation (3.28) found that there was a central log-log linear portion that
increases. If either creep rate or the erosion decrease, then had a slope of 0.5, and that this value has subsequently
the slope at the intersection point St will increase been confirmed by a number of authors to be an almost
(Figure 3.5). universal characteristic of catchments worldwide (e.g. de
Vries et al., 1994; Moglen and Bras, 1995; Rodriguez-
3.3.4 Cumulative Area Diagram Iturbe and Rinaldo, 2001).
Perera and Willgoose (1998) showed that the linear
One of the key characteristics of catchments is that for the central region can be derived as an asymptotic solution to
rounded hilltops near the catchment divide the catchment the Tokunaga network classification scheme (Section
flowlines diverge. Moving downstream and entering the 3.3.1), and that the slope of 0.5 follows when the net-
channel network, the flowlines converge. Much of the work complies with the statistics of topologically random
theoretical work for hillslopes in contrast assumes that channel networks (Shreve, 1967). The importance of Per-
flowlines downslope are parallel so that there is no flow era and Willgoose’s work is that it derives the relationship
divergence or convergence. A key challenge has been to between drainage convergence properties and the slope of
characterise statistically the flow divergence and conver- the CAD, and variations from 0.5 are indicative of
gence properties of the flowlines on hillslopes. systematic deviations in the convergence characteristics
The channel network in a catchment is generally con- of the drainage network. For instance, the top left-hand
vergent everywhere (exceptions include braided rivers corner of the CAD corresponds with the divergent hill-
and alluvial fans), creating the network of tributaries that slope, and the CAD shows a S-shaped region of varying
the network classification methods in Section 3.3.1 slope with area that reflects the divergent flow pattern of
FIGURE 3.7: Cumulative area diagram (CAD) as derived from three different data sources for the same landscape as in Figures 3.5 and 3.9.
The dark line is the CAD derived using a ground survey, while the finer lines are the 90% confidence limits from (a) the DEM was derived from
interpolating a contour map, (b) the DEM was derived from digital stereo photogrammetry using stereo aerial photography (from Walker and
Willgoose, 1999).
that is concave down everywhere. For hillslopes where

fluvial erosion and creep are both active, then the upper
part of the hillslope is concave down and lower part is
concave up (as discussed in Section 3.3.3), and the part of
the hypsometric curve above the elevation of the inflection
point in concavity of the long profile is concave down and
the part below that elevation is concave up. One implica-
tion is that the exact shape of the hypsometric curve is a
function of the length of the hillslope, with a shorter
hillslope having a greater proportion of the hypsometric
curve being concave down than a longer hillslope. This
indicates that the shape of the hypsometric curve is scale
dependent even in the absence of a change in process rates.
In two dimensions the shape of the catchment and the
FIGURE 3.8: Hypsometric curve. The classical shape is indicated by
exact branching pattern of the network are important
curves (a)–(c). The different curves are for different ratios of diffusive
erosion to fluvial erosion. Curve (a) is dominated by fluvial erosion, (Willgoose and Hancock, 1998). For long skinny catch-
while curve (e) is dominated by diffusion, the ratio of diffusion to ments the hypsometric curve is similar to that for the one-
fluvial erosion increasing from (a) to (e). Key features are the knee at dimensional hillslope discussed above. For more rounded
the bottom right due to internal hills and catchment divides in the in plan catchments where the channel network branches
lowlands of the catchment, the upward curvature in the top left for
out in all directions, there is the appearance of a knee in
curves (a)–(c) due to fluvial erosion and the downward curvature
everywhere in curve (e) due to the diffusion the bottom right of the hypsometric curve for low eleva-
dominance (from Willgoose and Hancock, 1998). tions. This knee results from the small low internal hills
that are formed in the central part of a catchment. While
Willgoose and Hancock (1998) derived their results for
the rounded tops of the hillslopes. The steep drop-off of transport-limited catchments, since their results depend
the curve at the right-hand side is simply a reflection of only on the slope-area relationship, they are equally
boundary effects when pixel drainage areas become very applicable for detachment-limited catchments.
close to the total area of the domain being analysed. The dependency of the hypsometric curve on catch-
Moglen and Bras (1995) used the CAD to test the ment scale and the branching pattern of the channel net-
validity of a landform evolution model. work may suggest that it is not a good descriptor of
geomorphology, but that is not the case. If the channel
3.3.5 Hypsometric Curve network is perfectly known (i.e. the flow directions at
every node in the catchment are known), then the contrib-
The hypsometric curve (Figure 3.8) characterises the pro- uting area at every point can be determining simply by
portion of a catchment that lies above a given elevation. summing areas downstream from the source. Using the
It has long been used as an empirical measure of the mean trend from the slope-area relationship we can deter-
distribution of landscape elevation (Strahler, 1952) but ministically reconstruct the elevations for the landscape
the linkage with catchment process was weak. It is working up from the elevation at the outlet, so that the
possible to make some statements about the form of the hypsometric curve is a function of the channel network
hypsometric curve for catchments in dynamic equilib- and the slope-area relationship. Moreover, we have found
rium. The hypsometric integral is the area under the that the scatter in the hypsometric curve from catchment
hypsometric curve. to catchment is small because it is an integrated measure
In one dimension (i.e. a hillslope with parallel flow of the catchment elevations. It is useful as an indirect test
lines), the hillslope long profile is equivalent to the hypso- of the slope-area relationship. This reconstruction of the
metric curve, and the one-dimensional solutions of hill- catchment elevations up from the outlet using the drainage
slope profile of Kirkby (1971) provide a direct link pattern and the average slope-area curve is the basis of the
between the hillslope processes and hypsometric curve. QUEL model (Ibbitt et al., 1999; Willgoose, 2001) dis-
A fluvial erosion-dominated hillslope with a concave up cussed in Section 3.3.3.
long profile will have a hypsometric curve that is concave Finally we note some statistical limitations on the
up everywhere, while a creep-dominated hillslope with a rescaling that is necessary to plot the hypsometric curve
concave down long profile will have a hypsometric curve and the subsequent calculation of the hypsometric
integral, which have implications for its reliability as a Since the arrival of digital terrain analysis the width
testing statistic. The use of relief for the vertical rescaling, function has been generalised so that the width is the
while the standard method, is problematic because the number of drainage paths, rather than the number of
highest elevation of the landscape is the highest peak, channels or blue lines. In this latter form the width func-
which may not be strongly related to the average elevation tion provides a high-resolution view of the drainage pat-
of the remainder of the landscape (see Section 2.3.3). tern of the catchment (both the channels and the
hillslopes). However, in this form it is quite sensitive to
random noise effects (e.g. errors in digital elevation
3.3.6 The Width Function models), so typically it requires some smoothing to iden-
tify the overall trend (Figure 3.9) (Walker and Willgoose,
In its classical form the width function is a plot of the 1999). Rinaldo et al. (1995) found that the shape of the
number of channels versus distance along the channel catchment and the shape of the width function are strongly
upstream from the catchment outlet. Historically it has linked and thus the unit hydrograph of the catchment is
been drawn by plotting contours of distance upstream strongly linked to catchment shape.
from the outlet and counting the number of blue lines that Three statistics measured from the width function are
crossed each contour, then plotting distance from the commonly used. The maximum width is called the catch-
outlet versus the number of blue lines (Surkan, 1969). ment width, the maximum length is the catchment length
The main attraction of the width function is that it can and the ratio of catchment width to length is the catchment
be transformed into the unit hydrograph for the catch- roundness. All three are dependent on the extremes of the
ment. If the velocity of the flood wave within the catch- width function and so have significant sampling error and
ment is constant (Section 3.1.7), then the distance from have proven to be relatively poor measures of catchment
the outlet can be transformed into the travel time to the topological shape (see Section 2.3.3)
outlet. If the hillslope lengths are the same everywhere,
then the number of channels is area/unit length of channel 3.4 Conclusions
which when unit runoff is applied to that area is the
volume of water contributed. Thus the width function This section has provided a very brief overview of hydrol-
characterises the runoff routing behaviour of the catch- ogy and geomorphology concepts that will arise later in
ment. Kirkby (1976) used the TDRN random network this book. By necessity the discussion has been mathemat-
model (Section 3.3.1) to generate catchment width func- ically superficial because of the brevity of the treatment,
tions and evaluated the effect of catchment runoff rate and and the presentation has concentrated on providing an
time to peak. intuition of how the processes work.
FIGURE 3.9: Width function as derived from three different sources for the same landscape as in Figures 3.5 and 3.7. The dark line is
the best estimate of the width function derived using a ground survey, while the finer lines are the 90% confidence of width function from
(a) the DEM was derived from interpolating a contour map, (b) the DEM was derived from digital stereo photogrammetry using stereo
aerial photography (from Walker and Willgoose, 1999).
3.4 Conclusions 47
If we focus first on the hydroclimatology, the key The geomorphology topics briefly reviewed here are
common feature of all processes discussed here is that not inputs to LEMs but their outputs. They provide a
they all feed into the long-term behaviour of the soil or framework within which to assess the overall geomorph-
landscape. The discussion of extreme events is important ology of LEMs. There are, of course, many other proper-
because they are involved in the triggering and threshold ties of landscapes (e.g. ones that assess site-specific
behaviour of pedogenic or geomorphic events that have characteristics such as regions of high risk of erosion
long-term consequences (e.g. chemical weathering, land- when examining waste containment structures) that have
slides). Mostly, however, the issues discussed are the been, or are being, used to assess both geomorphology
same ones that hydrologists study as part of long-term and LEMs, but the ones discussed above form a set
catchment yield studies used in water supply assessment, commonly used by researchers.
and these issues likewise determine the rates of geo- One thing to note about these geomorphology prop-
morphic processes once they are triggered. These pro- erties is that they are all statistical properties of the
cesses are the runoff generation during rainfall events landform rather than deterministic descriptors. The
and the inter-event drying that then determines how wet evolving landscapes created by LEMs are in many cases
the landscape is at the start of the next rainfall event. One sensitive to the exact details of unknowable inputs (e.g.
of the most challenging aspects of these hydrology calcu- the initial shape of the landform, climate history). If we
lations, common to many of the calculations in this book, explicitly model these unknowable inputs with random
is that it is rarely feasible to directly model all the pro- inputs (replicating their statistical characteristics), we
cesses at the level of detail at which we would like to find that the LEM-generated landscapes can be determin-
model them because (1) the computational demands are istically different (e.g. hills and valleys are in different
excessive and (2) the inputs for these detailed models are locations) but that the statistical properties of the hills
climate series that in general we don’t know deterministic- and valleys are the same. Geographers have a concept
ally. In the latter case we would have to generate several called ‘uniqueness of place’ to describe this (see most
parallel stochastic time series (e.g. rainfall and evapor- introductory physical geography texts, or page 5 in
ation) that are generally autocorrelated in time and space, Beven, 2000 and Petersen et al., 2012 for a deeper
and cross-correlated between each other. In many cases discussion). If you imagine standing at two different
hydrologists know that these correlations are important, lookouts in the Yorkshire Dales, then it is possible just
but the correct modelling of them is problematic (e.g. by looking to say that both views are of the Yorkshire
correct modelling of spatially distributed rainfall time Dales, even though the exact details of the views are
series with their combination of time-varying spatial and different. Thus the statistics of the views are the same
temporal correlations is still a work in progress), so there (and somehow uniquely identify the view as ‘Yorkshire
may be low confidence in the validity of hydrology Dales’, and not, e.g. ‘Colorado Rockies’) while the
time series modelled by this direct modelling approach, deterministic details are different.
quite apart from the mathematical and computational The follow-on question is ‘How well does the statistic
complexity. distinguish the “Yorkshire Dales” from the “Colorado
Accordingly, many existing landform evolution and Rockies”?’ This is a question of the signal-to-noise in
soil evolution models use simplified conceptualisations the statistic. The noise is how much of the deterministic
of hydrological behaviour, commonly called ‘effective detail exhibits itself in the statistical value, while the
parameterisations’. In the view of the author these effect- signal is how much of the generic Dales-iness or
ive parameterisations are likely to be more reliable for Rockies-iness characteristics are exhibited. The classic
prediction (or at least more transparent about their reli- technique to reduce noise is to take many samples (i.e.
ability) because of their observed empirical basis. Com- lots of sites) and average the results. A good statistic is
bining these effective parameterisations with the one that requires only a few samples to distinguish the
physical intuition provided in the sections above should sites. But it is also a function of how different the land-
allow us to test the sensitivity of these effective para- scapes are. It would be a poor statistic that couldn’t
meterisations to changes in the environmental and cli- distinguish the Dales from the Rockies with a small
mate drivers. The challenge then is finding the right number of sites, but a more challenging task would be
balance of direct versus effective modelling. And, of to statistically distinguish the Yorkshire Dales from the
course, this doesn’t preclude further research aimed at North York Moors, adjacent regions that are distinguish-
elucidating the physical and stochastic drivers of these able by eye but similar in geology and climate. Even if the
effective parameterisations. reader is not familiar with the beauty of North Yorkshire,
I’m sure you can construct similar examples from your modelling rather than the continuous yield modelling we
own locale. are interested in. It also has a stronger emphasis on the
When we discuss model testing these issues of signal- hydrology of humid climates than arid climates. The
to-noise and what makes a good geomorphology statistic review of Roe (2005) provides an excellent treatment of
will become very important. orographic rainfall. Ruddiman (2013) provides a good
overview of paleoclimate and its reconstruction. Unfortu-
3.5 Further Reading nately there is no up-to-date reference that can be recom-
mended for the modern geomorphology techniques
Many introductory hydrology textbooks cover parts of discussed here. Abrahams (1984) provides a good review
this chapter, though they do not emphasize the aspects of the older work, but it lacks the recent insights into the
of yield hydrology we are interested in. A good overall effectiveness of these older statistics that has come from
hydrology textbook which is full of sage practical advice work in digital elevation model analysis and landform
is Beven (2012), though it concentrates on event evolution modelling.
4 Erosion and Other Water-Driven
Processes
The chapter discusses processes that primarily result from explanation are (1) soil particles are launched off the soil
surface water runoff: fluvial erosion in all its variants surface with a given speed and angle to the horizontal and
(both erosion and deposition), and including the direct in all directions (in plan) equally, (2) the number of
action of rainfall on the eroding surface. Key features of particles and their launch speed are a function of the
fluvial erosion are (1) it is driven by rainfall, (2) not all of kinetic energy of the raindrop and (3) the kinetic energy
that rainfall ends up as surface runoff (some evaporates or of the raindrop is a function of its diameter and fall speed.
infiltrates into the groundwater), (3) the conversion from There are two components to calculating rainsplash
rainfall to runoff is nonlinear, patchy in space, and event transport:
driven, (4) the conversion from runoff to erosion rate is
(1) the energy and rate with which the raindrops impact
also nonlinear and likewise patchy in space and event
the surface and
driven (so cumulative erosion does not have a one-to-
(2) the net transport of sediment downslope by the
one transformation to cumulative runoff) and (5) while
splashed up water and sediment.
the runoff provides the energetic potential for erosion,
other factors may constrain erosion so that it is less than The discussion below focuses on the mechanisms driving
its potential. While this chapter is about water-driven the net transport downslope (point (2)).
processes, we will not discuss infiltration or groundwater The classic rainsplash formulation follows. If atmos-
in this chapter (see Chapter 3 for these details). pheric drag on the launched particles is ignored, then
solving the equations of motion yields the upslope and
downslope travel distances, xu and xd , respectively,
4.1 Rainsplash and Rainflow
2½vi cos α2
Raindrops, when they hit the soil surface, transfer part of xd ¼ ½ tan α þ tan θ
g
the falling raindrop’s kinetic energy into splashing up 8
>
< 2½vi cos α
2 (4.1)
water and soil particles off the soil surface after the ½ tan α tan θ α>θ
xu ¼ g
raindrop’s impact on the soil surface. This process, >
:
‘rainsplash’, drives two processes. The first process is 0 αθ
that the splashed-up soil particles are spread around the where vi is the initial launch velocity and α and θ are the
site where the raindrop hits the soil surface. The second angles to the horizontal (in the upslope and downslope
process is that the raindrop fragments the soil aggregates direction) of the particle launch and the hillslope. The net
at the surface, making them more readily transportable transport downslope is then (for α > θ)
by water flowing over the surface. In this section we will
focus on the first process, rainsplash. The second pro- 4½vi cos α2 4½vi cos α2
xd xu ¼ tan θ ¼ S (4.2)
cess, often called ‘rainflow’ or ‘interrill erosion’, will be g g
discussed in more detail later in this section when we talk where S is the gradient of the hillslope. If all particles have
about the mechanics of fluvial sediment transport com- the same angle and velocity of launch, then the downslope
bined with rainsplash. transport rate per unit width is
The classic physical explanation for rainsplash
transport and its preferential movement of particles down- M M 4½vi cos α2
qs ¼ ðxd xu Þ ¼ S (4.3)
slope is shown in Figure 4.1a. The key components of this 2 2 g 49
50 Erosion and Other Water-Driven Processes
FIGURE 4.1: Schematic of splashed up particle trajectories: (a) classical model with launch angle of trajectory being relative to the
horizontal, (b) Furbish model with launch angle of trajectory being relative to the inclined plane.
where M is the mass of particles displaced by rainsplash is marked difference in the dependence of the transport
per unit time and width. Since this equation is always rate on launch angle with the classical solution having a
positive, there is a net transport downhill, and the depend- weak decrease with increase in angle, while the Furbish
ency of this transport on the hillslope gradient is a linear solution has a strong increase in transport with increasing
function of slope with a multiplicative constant on the launch angle. For high launch angles on flat slopes, the
equation. This multiplicative constant is a function of the two approaches yield asymptotically identical relation-
angle of launch and energy of the launched soil particles. ships but diverge as the launch angle decreases. Empiric-
Recent work using high-speed photography (Furbish ally both results in Figure 4.2 are well fit by
et al., 2007, 2009) has suggested that this conceptualisa-
qs ¼ KS1:1 (4.6)
tion of the processes is incorrect and that the launch angle
of the particles is relative to the soil surface, rather than which is approximately consistent with Fickian diffusion.
the horizontal (Figure 4.1b). While the data were not If we assume linearity, as is normally done, substituting
comprehensive (Furbish, personal communication), if we Equation (4.6) into the continuity equation in the x (posi-
assume that soil particles are launched at a fixed angle to tive downslope) direction, we obtain Fick’s 2nd Law:
the surface rather than the horizontal, then the downslope
and upslope travel distances become ∂z ∂ ∂ ∂z ∂2 z
¼ ðKSÞ ¼ K ¼K 2 (4.7)
∂t ∂z ∂z ∂x ∂x
2
2½vi cos ðα θÞ
xd ¼ ½ tan ðα θÞ þ tan θ This result is also consistent with field and experimental
g
2
(4.4) data that almost universally suggest a near linear relation-
2½vi cos ðα þ θÞ
xu ¼ ½ tan ðα þ θÞ tan θ ship with slope (e.g. Dunne et al., 2010).
g
These derivations are for a single particle with a
and the net downslope transport is known launch angle in the downslope direction and
2vi 2 launch velocity. The extension to many particles with a
xd xu ¼ cos 2 ðα θÞ½ tan ðα θÞ þ tan θ range of launch velocities and angles and angles to the
g
steepest downslope direction is conceptually straightfor-
cos 2 ðα þ θÞ½ tan ðα þ θÞ þ tan θ (4.5)
ward (van Dijk et al., 2002b) but in practice difficult to
where we note that tan θ ¼ S. The form of the depend- estimate. Even within a single event there is distribution
ency of Equation (4.5) on hillslope gradient, S, is more of raindrop diameters (typically a gamma distribution;
complex than that in Equation (4.3). Figure 4.2 shows Uijlenhoet, 2001), and this distribution changes with rain-
both the classic and Furbish solutions for a range of fall rate (mean drop diameter ðmmÞ ¼ 0:24R0:21 , R is the
hillslope gradients and particle launch angles. Both rainfall rate in mm/hr; Marshall and Palmer, 1948). If the
Equations (4.2) and (4.5) have a linear dependence on launch angle is the same for all drop sizes, then these
the hillslope slope, S, for low slopes. For higher slopes a variables involve integration over the range of launch
better fit is obtained by qs / S1:1 , and the power of 1.1 is velocities, so the only impact is on the coefficient K in
true for both the classical and Furbish solutions. The Equation (4.7) and not the slope dependence. The distri-
classic solution shows a stronger sensitivity to changes bution of launch velocities is a function of raindrop kin-
in slope below the threshold when α < θ but similar etic energy. Van Dijk et al. (2002a) review the factors that
sensitivity above the hillslope gradient threshold. There feed into the energy of the raindrops, and thus the rate of
4.1 Rainsplash and Rainflow 51
FIGURE 4.2: A comparison of the transport rates for the (a) classical and (b) Furbish models for 2Mv g ¼ 1 in Equation (4.3). The different
i
lines are for different launch angles. In the classical solution the kink in the curve for low launch angles is where the angle of launch is lower than
the gradient of the hillslope, so upslope transport is zero.
transport by rainsplash (point (1)). Gabet and Dunne qs ¼ γMxd ð1 γÞMxu (4.9)
(2003) provide an example of calculating this rainsplash
rate constant K. Methods for directly estimating the where γ is the proportion of the total mass displaced
distribution of rainfall kinetic energy from radar data are downslope, which was greater than 0.5 in the Furbish
emerging (Yu et al., 2012). A very simple empirical experiments.
equation used to estimate the change of rainsplash Before finishing we return briefly to the question of the
transport rate with rainfall rate is (Lane et al., 1988) raindrop energy and processes that dissipate this energy
before it impacts on the soil.
E I ¼ aI 2 (4.8) It is normally assumed that the raindrop is falling at
where I is the rainfall intensity (mm/hr) and a is a rate terminal velocity, which for most raindrops is reached
constant. after about 20 m of fall. However, if the raindrops hit the
More critically Furbish et al. (2007) found that in forest canopy, then this energy is dissipated on the leaves,
addition to the differences from the classic solution of and it is only the velocity attained in the distance between
launch angle, that there are differences in the momentum the canopy and the ground that is the relevant raindrop
of the particles between the upslope and downslope direc- energy. The smaller this distance, the less the raindrop
tions with particles having greater downslope momentum. energy. Accordingly, vegetation provides protection
Momentum is the product of mass and velocity, so against rainsplash transport, and this protection is a func-
increases in momentum downslope can arise from tion of the height of the canopy and what proportion
increases in launch velocity downslope (i.e. distance trav- coverage of the ground the leaves provide. This effect is
elled), mass of particles launched downslope (i.e. number incorporated as a reduction factor in most agricultural
of soil particles) or a combination of both. While their erosion models, and the reduction is a function of the
experimental data were not definitive (Furbish, personal height of the canopy and percentage plan coverage pro-
communication), they were suggestive of a combination vided by the canopy (see Section 14.2.1 for more details).
of velocity and mass. For changes in velocity, Equations This reduction factor for forest canopy is rarely greater
(4.2) and (4.5) show that asymmetry in velocity, if it is than 50% and typically less, but can be 100% for a
independent of slope, will appear as a multiplicative con- complete ground level coverage by leaf litter and mulch
stant on the equations increasing the downslope transport (Figure 14.1; the Universal Soil Loss Equation (USLE)
rate. For changes in mass, Equation (4.3) shows that we vegetation cover factor tables used in all USDA erosion
can adjust for the changes in the mass displaced down- models: USLE, RUSLE, MUSLE, CREAMS, WEPP;
slope relative to upslope: Wischmeier and Smith, 1978).
The other energy dissipation process is water on the the central core of existing agricultural erosion models
soil surface. The general rule of thumb is that once the and landform evolution models so these processes are
depth of water is equal to diameter of the raindrop, then all well understood, particularly over the short term (a few
of the energy of the raindrop is dissipated within the water years). The focus here, however, will be on those com-
layer, and rainsplash of sediment is zero. When the depth ponents of erosion and deposition that have an impact
of water is less than the drop size, the combination of over the long term. While it is clear that the long-term
water flow and rainsplash is called ‘rainflow’ (Coventry impacts are the result of the accumulation of all the short-
et al., 1988). In this case sediment transport is a combin- term impacts, a number of feedbacks are not normally
ation of the reduced rainsplash and the shear stress–driven important over the short term but are over the longer term.
sediment transport processes that we will discuss in the Moreover, studies of short-term impacts are normally well
next sections (Gabet and Dunne, 2003). modelled by modern agricultural erosion models such as
In conclusion, it should be clear that there is strong WEPP, SWAT and PESERA. However, we will show
evidence (both experimental and theoretical) that the that long-term erosion modelling is more than simply
transport flux of rainsplash is proportional to slope (i.e. applying these same models over long simulation times.
Equation (4.6)). A consequence of this is that it acts on the Traditionally fluvial sediment transport processes are
surface elevations as a diffusive process tending to classified into two types:
smooth the surface (Section 3.3.3). The rate at which this
1. Transport-limited: This is where the amount of sedi-
process operates (i.e. the coefficient K in Equations (4.6)
ment carried in the water is always at the maximum
and (4.7)) is more difficult to estimate but is a function of
capacity that the water can carry. If when travelling
raindrop size, impact velocity and rainfall rate. Once the
downstream the transport capacity increases, then soil
raindrop hits the surface, there are also uncertainties about
is eroded so as to entrain the material needed to
how the raindrop energy is converted to the momentum of
maintain the flow at its transport capacity. Likewise,
the ejected soil particles, and these processes are currently
if the transport capacity decreases in the downstream
not well understood or parameterised. Recent work
direction, then sediment from the flow is deposited on
(Dunne et al., 2016) simulated the transport of rainsplash
the soil surface. This definition should not be con-
using a calibrated rainsplash model and observed rainfall
fused with ‘transport-limited chemical weathering’,
time series, and confirmed previous work that indicated
which is entirely unrelated and which is discussed in
that rainsplash alone cannot account for observed convex-
Chapter 8.
down hilltops (see the slope-area analysis in Section
2. Detachment-limited: This is where the flowing water
3.3.3); other ‘diffusive’ processes must operate to make
is carrying less sediment than the transport capacity,
up the transport shortfall. These processes will be dis-
but where the ability to entrain material from the
cussed in Chapter 9.
eroding surface is restricted by the ability of the flow
Rainsplash/rainflow diffusion actively suppresses ril-
to detach the sediment from the surface. Since the flow
ling in those areas where the diffusive transport is greater
is always carrying less sediment than it has the cap-
than the fluvial erosion. Because it suppresses rill erosion,
acity to carry, it is not possible to deposit material.
rainsplash is the main component of ‘interrill erosion’.
Because it can only erode, it is sometimes referred to
Typically those regions are near the slope divide, and as
as an ‘incision model’.
you proceed downstream, and the area draining through
that point increases (so fluvial erosion and water depth This differentiation overly simplifies a number of
increases), eventually fluvial erosion dominates and rills important processes and feedbacks. The simple definition
are created (Smith and Bretherton, 1972; Loewenherz- above for transport-limited ignores that the transport cap-
Lawrence, 1994; McGuire et al., 2013). Thus the balance acity of a flow is a function of the grading of the material
between the rates of rainsplash and fluvial erosion is carried. If the sediment is coarse, the transport capacity of
central to modelling the development of rilling and related the flow is less than if the sediment is fine. Depending on
microtopography on hillslopes. the grading of the material entrained the transport capacity
is no longer simply a function of the flow processes, but is
also a function of the material being entrained and carried.
4.2 Erosion and Deposition The detachment-limited definition is likewise simplified.
The definition above hides the range of different detach-
This section discusses erosion and deposition processes ment processes in different geologic and hydraulic envir-
that result from fluvial processes. These processes form onments (e.g. cohesive clays versus bedrock rivers versus
soils cemented by secondary minerals) and entrainment 4.2.1 Transport-Limited Erosion

mechanisms.
It is convenient to define a third type of sediment 4.2.1.1 Equations
transport that does not fall neatly under either of the The fundamental processes that determine the transport
definitions above: capacity of a flow have been studied for many years. Many
specific forms of the sediment transport capacity have been
3. Source-limited: This is where the rate of erosion over developed, so we will focus here on the common funda-
some period is limited by the ability of other processes mental features. The sediment transported by the flow
(typically weathering) to produce transportable mater- increases when (1) the shear stress applied to the soil by
ial for the flow to entrain. For example, in a highly the water at the soil-water interface increases, (2) the diam-
seasonal monsoonal environment it is possible for the eter of the transported sediment decreases and (3) the
transportable material to be created during the dry specific gravity of the sediment particles decreases. More-
season (e.g. rocks being transformed into sandy debris over, there is a shear stress threshold, normally called the
from weathering), and this new transportable material Shield’s stress threshold, below which sediment cannot be
is a store of sediment ready to be transported at the entrained into the flow. The equation normally used to
start of the next wet season. Once this material is model transport-limited sediment transport in LEMs is a
stripped away in the early part of the wet season form that does not explicitly include shear stress
(typically by transport-limited erosion), the store is
depleted. At this stage the surface will be covered by qsc ¼ KQm1 Sn1 (4.10)
a coarse armour that cannot be transported, so erosion where K is the erodibility, m1 and n1 are parameters and
stops or slows down. Under some circumstances the the subscript c on qsc indicates that this is the sediment
flow may become detachment-limited, but more transport capacity to distinguish it from the actual sedi-
typically either (1) the armour is too coarse to be ment transport qs which in later sections in this chapter
transported even if its entrained in the flow (see the will not necessarily be equal to sediment transport cap-
Shield’s stress threshold in the following sections) or acity. The form of this equation has the advantage of
(2) the entrained material is so coarse that the transport having a direct and simple dependency on the discharge
capacity of the flow drops to a very low value given Q (and thus drainage area) and slope S, but hides many
the coarse particle size grading that is entrained into important aspects of the physics of sediment transport.
the flow. Accordingly we will show how this equation can be
Finally, it should be noted that mass is transported out obtained from a more fundamental representation of
of a catchment by three main mechanisms. Their esti- fluvial sediment transport.
mated global rates of delivery to the ocean (Syvitski The nondimensional sediment transport qc sc is related
et al., 2003) are (1) as suspended load within the flowing to the nondimensional shear stress ^τ by (Vanoni, 1975)
water column (18 Gt/year), (2) as bedload where the bed qc
sc ¼ K^
τp (4.11)
of the river itself is moving downstream (2 Gt/year) and
(3) as dissolved load where it is dissolved in the water where K and p are parameters, and the hat indicates the
(5 Gt/year). A further 3.1 Gt/year is delivered to the nondimensional quantities
oceans from ice, wind and coastal erosion. Given the qsc
relative magnitudes of the processes it is unsurprising that qc
sc ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
most research has concentrated on suspended sediment ρw gsF gðs 1Þds 3
transport. The Syvitski et al. load estimates are based on τ0 (4.12)
sediment gauging stations, and so include anthropogenic ^τ ¼
ρw gðs 1Þd s
effects and may not be typical of the long-term geologic
history. Reliable long-term data on total sediment load where ρw is the density of water, g acceleration due to
(both suspended and bed load) is difficult to quantify, and gravity, s the specific gravity of the sediment, qs the
catchment-scale data are scarce in the literature (Hancock sediment transport in mass/time/unit width of flow, τ 0
et al., 2017a). the bottom shear stress at the soil-water interface and d s
The sections below fill in the details of how the three the representative diameter of the sediment particles
transport-, detachment- and source-limited mechanisms (e.g. median or mean diameter). The parameter F is a
work and the mathematics used to model them. The value weakly dependent on the diameter of the sediment
treatment below assumes that you have read Chapter 3. and is about 2/3. For a given sediment density and size,
Equations (4.11) and (4.12) can be simplified (noting that and regime equations (see Section 4.5.1) have shown a
τ 0 ¼ ρw gRS) to dependency of P on discharge of the form
P ¼ K w Qα (4.18)
qsc ¼ K 0 ðRSÞp
p qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi where K w and α are coefficients. This results in the expo-
0 K (4.13)
K ¼ ρw gsF gðs 1Þd s 3 nent on the discharge term in Equations (4.15) and (4.17)
ðs 1Þd s
being reduced to 3p 5 ð1 αÞ ¼ 1:8ð1 αÞ. Published
where R is the hydraulic radius of the flow, S is the slope regime equations based on field data suggest α is in the
of the flow and K 0 is a multiplicative constant dependent range 0.4–0.5 (Graf, 1984). For instance, for α ¼ 0:5 the
on sediment grading and density. Henderson (1966) indi- exponent on discharge in Equation (4.17) is 0:50p ¼ 0:9.
cates that p 3 so sediment transport capacity increases One of the problems with the empirical equations for
with decreasing diameter and specific gravity of the sedi- wetted perimeter like in Equation (4.18) is that they typ-
ment grains. To replace the dependency of Equation ically have a dependence solely on discharge, yet there
(4.13) on R with a dependency on Q we use the Manning should also be a dependence on slope. If discharge is fixed
equation for flow as the slope decreases, the depth of water in the channel
will increase, so the wetted perimeter will also increase.
R =3 S =2 P
5 1
An equation for P that includes slope can be derived using
Q¼ (4.14)
n the Manning equation (following Julien, 2014). If the
wetted perimeter of the channel is given by the equation
where P is the wetted perimeter of the flow, and n is the
(and hydraulic radius R follows because R ¼ Acs =P),
Manning roughness coefficient for the soil surface.
Rearranging this equation we substitute for R in Equation P ¼ βAcs δ
(4.13): 1 1δ
(4.19)
R¼βδP δ
qsc ¼ K 00 Q =5 S =10
3p 7p
where Acs is the cross-sectional area. Substituting into the
Manning equation (Equation (4.14)) yields
K 0 n =5
3p (4.15)
K 00 ¼ h 5 i 3δ
P =5 P ¼ βð52δÞ nð52δÞ Qð52δÞ Sð2ð52δÞÞ
3p 3δ 3δ
(4.20)
For p ¼ 3 (the value for Einstein-Brown sediment trans- If the geometry of the channel is such that δ ¼ 0:5 (e.g.
port) the equation becomes triangular channel cross section), then this reduces to
qsc ¼ K 00 Q1:8 S2:1 (4.16)
P ¼ β1:25 n0:375 Q0:375 S0:188 (4.21)
00
If we expand the constant K and make the terms for
specific gravity, grain diameter and soil surface roughness Julien (2014) presented an equation very similar to this
explicit, then (and tested it against field measurements for 835 sites) and
1 s which also included an empirical factor allowing for the
qsc / n1:8 Q1:8 S2:1 (4.17) d 50 of the sediment (the grain diameter 50% of the mass is
d s 1:5 ðs 1Þ2:5
less than the median grain diameter by mass) in the
Equation (4.17) shows that the sediment flux increases if channel. Equation (4.21) can be simplified if the log-log
the sediment diameter decreases, the specific gravity linear slope-catchment area relationship of Equations
decreases and the roughness increases. The trend with (3.14) and (3.16) is used with a typical observed scaling
roughness appears at first to be anomalous, but it arises exponent on area (i.e. S A0:5 Q0:5 where A is the
because as the roughness increases, the velocity catchment area). This allows us to eliminate the slope
decreases, and for a given discharge this means the depth dependence and calculate an exponent on Q of 0.47 for
of flow increases, leading to an increase in the bottom Equation (4.18), which is within the range of empirical
shear stress τ 0 . These results are specific for the Einstein- values in equations that ignore slope. However, this latter
Brown sediment transport equation, but the general trends result will be true only when the catchment complies with
with sediment characteristics and bed roughness are the the log-log linear slope-catchment area relationship,
same for many sediment transport equations. Finally, otherwise Equation (4.20) applies.
Equation (4.15) shows a dependency on the wetted per- Most of the theory here was been explicitly developed
imeter of the flow P. Normally P increases with discharge, for channels. However, it is also broadly true for
hillslopes. Hillslopes are rarely perfectly flat, so flow con- slope (see Chapter 7; Cohen et al., 2009, 2010; Welivi-
centrates into rivulets and the hydraulics of each of those tiya, 2017) as part of the erosion model or (3) use an
rivulets is the same as the larger channels. The wetted equation with parameters as in Equation (4.22) while
perimeter increases as the discharge and the depth of flow being mindful that Equation (4.22) was derived for only
increases until for extreme flows all the microtopography two sites. The observed short response time (25 years) of a
on the hillslope is submerged. The main difference few years of changes in m1 and n1 to the equilibrium
between hillslopes and rivers is the importance and per- between armour and erosion suggests a strong dynamical
manence of microtopography (e.g. leaf litter debris dams, response to short-term climate variability, which has also
animal hoof prints) on hillslopes, and the lack of a mobile been observed in our coupled soilscape-landscape evolu-
bed on the hillslope so there is no bedload. The hydraulics tion model SSSPAM (Welivitiya, 2017).
and erosion characteristics of hillslopes have been exten- Before we begin to use these equations, it should be
sively studied (Abrahams and Parsons, 1990, 1991a,b, noted that there are many equations for sediment trans-
1994; Parsons et al., 1990, 1992, 1997; Willgoose and port. They vary in their rates and their dependency on
Kuczera, 1995; Willgoose and Sharmeen, 2006), and discharge, slope and sediment characteristics. However,
much of the fundamental physics of transport-limited they are all fundamentally based on underlying principles
fluvial erosion on hillslopes is the same as rivers. similar to those outlined here. There is considerable
However, there is an import caveat for hillslopes. All scatter in the predictions of these equations. Without
of the equations above were derived from experiments calibration to field or experimental data it is rare for any
with sediments that were all one diameter. When the soil of these sediment transport equations to be able to predict
being eroded has a range of sizes, there is the possibility sediment transport rates to better than a factor of 2 or 3
of selective entrainment of the finer fractions, and the (e.g. Hancock et al., 2015a). Thus the sediment transport
creation of a coarser armour that resists erosion on the models used in professional practice all have a strong
surface. Section 4.2.4 discusses this process in detail, but empirical basis, underpinned by extensive field data col-
it is important to note here that long-term erosion can sort lection aimed at reducing this predictive uncertainty.
the hillslope surface such that steeper slopes are covered
by coarser material than flatter slopes, so that relative to 4.2.1.2 Time and Space Variation
an unsorted hillslope (e.g. Equation (4.16)) for a sorted The equations above give the sediment transport capacity
slope the dependence on slope is reduced. Willgoose and at a given point in space and time. During a rainfall event
Riley (1998a) observed this for degraded mine waste the discharge will be varying on a second/minute/hourly
slopes where the equation they fitted to data was time scale. We typically cannot resolve this temporal
resolution since we run LEMs for years to millennia.
qsc ¼ 3:55Q1:42 S0:66 (4.22)
However, the discussion in the previous section suggests
where it can be seen that the slope dependency is signifi- that accurate modelling requires (1) modelling the evolu-
cantly less than Equation (4.16). Subsequent field work to tion of landforms over thousands of years at minute
assess the surface grading on the slopes and the use of resolution and (2) the need to be able to estimate the water
other erosion models (Evans and Loch, 1996) confirmed and sediment discharges at minute resolution may involve
this behaviour. Further study by Sharmeen and Willgoose rainfall prediction at minute resolution. Even if we can
(2007) with their ARMOUR soil armouring model compute the erosion loss at minute resolution, the gener-
showed that if the physics of Equation (4.16) was allowed ation of a minute resolution rainfall series that has the
to evolve over time after 25 years (the age of the Will- correct correlation structure in space to drive a runoff
goose and Riley experimental sites), the discharge and generation model is beyond the state-of-the-art. We return
slope exponents were about those fitted in Equation to this difficult challenge later in this chapter after we have
(4.22), and that at 100 years the equations stabilised to discussed the other types of erosion mechanisms, because
Equation (4.10) where m1 and n1 were about 1–1.2 and it is a challenge common to all erosion mechanisms.
0.5–0.7, respectively, for two different contrasting spoils. The next section deals with question of what happens
This implies that the use of the unsorted sediment param- as the sediment transport varies in space (i.e. downstream
eters in Equation (4.16) for hillslope erosion on old (more along a drainage path).
than a few years) hillslopes is inappropriate, and the user
either needs to either (1) measure surface armour diameter 4.2.1.3 Entrainment- and Deposition-Limited Systems
versus area and slope in the field and include this in the The discussion above suggests that when the flow has
model, (2) use generalised results for diameter, area and a transport capacity that is more than the amount of
sediment in the flow, then the flow erodes to instantan- and the distance that the flow travels to drop all sediment
eously make up the deficit. Likewise when the transport (e.g. in a purely depositional environment like an alluvial
capacity is less than the amount of sediment in the flow, fan) is Lmax ¼ vy=w (Figure 4.3). In a numerical model, if L
sediment is deposited. However, the flow cannot instant- and Lmax ¼ vy=w are less than the grid resolution, then from
aneously entrain or deposit all of the sediment required, so the viewpoint of the modelling, the deposition is all inside
there is a limitation on the rate at which the transport the model pixel (i.e. it is a subgrid process) and deposition
capacity can be satisfied when the transport capacity is is not deposition-limited. If L is longer than the grid
changing downslope. resolution, then this process may need to be modelled
Deposition limitation is where the rate of deposition is explicitly (e.g. a large alluvial fan with a coarse spatial
limited by the fall rate of particles. For example, let’s discretisation).
assume that the sediment is equally distributed through A model implementation issue is illustrated in
the depth of the flow with concentration c (in units of Figure 4.4. Assume that the sediment concentration pro-
mass/volume of water, e.g. mg/l). We will also initially file at point A is well mixed vertically. If all sediment
assume that all the particles have the same fall velocity in settles with settling velocity w, then at the downstream
water w. Then the maximum possible rate of deposition end of the pixel A-B in Figure 4.4a the top of the flow
per unit bed area is D ¼ wc, where D is the deposition rate down to depth wΔt will contain no sediment. Accordingly
in mass/unit time/unit area. If we assume that during the the concentration profile entering pixel B-D is no longer
time that the sediment is being deposited that no sediment well mixed but is clear at the top, and the deposition rate
is also being entrained, then this means that the distance at the bottom of the flow is unchanged from pixel A-B
the flow travels to deposit a mass of sediment ΔD is since the concentration at the bottom of the flow does not
change. However, if there is vertical mixing by turbulence
vΔD
L¼ (4.23) (Figure 4.4b), then the concentration profile at C will be
wc
the same through the profile, but the concentration will

have decreased by the factor Df wΔt =Df where Df
is the depth of the flow, and as a result the deposition rate
in pixel C-D is reduced from that occurring in node A-C.
Typically models don’t track the concentration profile and
implicitly assume that the profile is always well mixed at
all times (i.e. Figure 4.4b). For Figure 4.4a the deposition
downstream will be a constant deposition rate until we
reach point E, and then deposition stops because all sedi-
ment has been removed from the flow. For Figure 4.4b the
deposition downstream will initially be the same rate as
FIGURE 4.3: Deposition of particles showing how a particle that
Figure 4.4a but will exponentially decline with distance
starts at the water surface settles linearly with time and distance
downslope. Lmax is the longest distance that a particle can travel, and downstream.
defines the maximum settling distance for a particle with settling There are a number of complications of this simple
velocity w in a flow with velocity v. conceptualisation. The concentration profile is not
FIGURE 4.4: Concentration profile of sediment down the profile in a purely depositional environment. (a) At point A the profile is fully
mixed from top to bottom. At point B after time Δt the top of the profile down to wΔt is clear of sediment since the whole profile has settled by
wΔt. (b) However, if point B in (a) is the edge of one node and the entry to another, it is common to consider the profile well mixed at entry to
the next downstream node so that the profile at C is what the model actually assumes.
normally the same through the profile, but rather the after a change in the sediment transport capacity (with
concentration is higher at the bottom of the flow, and is a consequent change in c∗ )
commonly assumed to be exponentially increasing with
entrainment ¼ wc∗
depth (Vanoni, 1975). In that case the concentration in
Equation (4.23) is the concentration of sediment in the deposition ¼ wc (4.24)
flow at the soil-water interface, and the rate of deposition dc
y ¼ wc∗ wc ¼ wðc∗ cÞ
will initially be fast and decrease with time as the concen- dt
tration at the interface decreases with time. Finally the fall where y is the depth of the flow, c∗ is the new equilibrium
velocity of the sediment particle in the river is not the concentration and c is the concentration in the flow at any
same as in a still water settling column in the laboratory time after the change in transport capacity. Solving this
because of the effect of turbulence (Dietrich, 1982; Boilatt differential equation yields the solution for how concen-
and Graf, 1982; Willgoose, 1997). tration (and thus the sediment load) in the flow changes
The application to a mixed sediment load is conceptu- after the change in transport capacity
ally straightforward with larger diameter or denser par-
c ¼ c∗ þ ðc0 c∗ Þe =y
wt
ticles with higher settling velocity settling out faster than (4.25)
smaller less dense particles. Thus what happens with a where c0 is the concentration before the change in
flow that is carrying a mixed sediment load is that the transport capacity. This shows that the concentration
deposition process not only reduces the load in the flow, asymptotically approaches the new equilibrium concen-
but it also preferentially reduces the proportion of coarse tration. Our interest is to determine the net erosion or
dense particles in the mixture because they settle out deposition (in units of mass/unit time/unit plan area).
faster. If entrainment is defined positive and deposition nega-
We now turn to the altogether more difficult issue of tive, then
entrainment limitation. As with deposition, entrainment
of sediment does not happen instantaneously. In con- net entrainment ¼ wðc∗ cÞ ¼ wðc∗ c0 Þe =y
wt
(4.26)
trast to deposition there is no generally applicable
which is the rate of erosion or deposition downstream of
physically based means to calculate the rate of entrain-
the change in sediment transport capacity. The time scale
ment. However, at transport capacity sediment is
of the change is t ¼ y=w, which can be used to calculate
depositing from the flow as discussed above, and in
the distance downstream, L, the adjustment propagates,
Equation (4.23) and Figure 4.5, and this loss of sedi-
L ¼ vt ¼ vy=w, where v is the velocity of the flow. If the
ment from the flow is exactly balanced by new sedi-
velocity of flow is high, the depth of the flow is high, or
ment entrained into the flow. If we assume that the
the settling velocity is small, then the distance of adjust-
entrainment process is solely driven by the shear stress
ment is large.
of the flow, then the rate of entrainment for a given
Equation (4.24) can be rearranged to show how the
discharge q, which has a corresponding sediment dis-
rate of detachment changes with time (noting that
charge qs and sediment concentration c in the flow
qs ¼ cq)
(c ¼ qs =q), is equal to the rate of deposition of sedi-
ment when the sediment load is at equilibrium, and this dqs
rate of detachment ¼ y ¼ wðqsc qs Þ
rate is independent of the concentration of sediment in dt
q
the flow at that current time (which in general is not the ¼ wqsc 1 s (4.27)
qsc
equilibrium concentration) (Figure 4.5). Accordingly
This form of the equation emphasizes that the rate of
detachment is a function of the difference between the
actual sediment transport and the sediment transport
capacity.
Some example time scales are (1) for a 0.1 mm fine
sand particle (settling velocity is about 0.01 m/s) in a 1 m
deep flow (e.g. a channel) the time scale is just over 1 day
and (2) for a 10 mm flow (e.g. a hillslope) the correspond-
FIGURE 4.5: Schematic showing the (im)balance between the rate
ing time scale is about 20 minutes. In both examples the
of deposition from the bottom of the profile, wc, and the rate of time and the distance that will be travelled during that
entrainment of sediment, wc*. time by the flow is large. Finer sediments will have even
longer time scales. It is thus reasonable to assert, given ∂z

¼ K t ðK z Qm2 Sn2 τ c Þ (4.29)
how variable sediment transport capacity is downslope, ∂t
that it will be rare for the sediment concentrations in the 3=
ρw gn 5
where K z ¼ , m2 ¼ 0:6ð1 αÞ, n2 ¼ 0:7 and K w
flow to achieve the sediment transport capacity, though, of 3=
Kw 5
course, the details will depend on the sediment, down- and α are defined in Equation (4.18). Tomkin et al. (2003)
stream long profile and flow hydraulics. suggested that m2 ¼ n2 and are in the range 2/3–5/3. This
Equation (4.27) might appear at first glance to be in equation of incision as a function of discharge (or area)
conflict with Equation (4.26), but Equation (4.26) has and slope has become known as the stream power equa-
been derived for the case where there is no entrainment tion (e.g. Sklar and Dietrich, 1998), and should not be
and the deposition rate is given by the initial rate of confused with the unit stream power equation for sedi-
deposition, and so has been derived for a simpler ment transport capacity of Yang (1973).
scenario. Now we can note that the detachment-limited equation
It is possible to construct both the transport-limited is not incompatible with the transport-limited equation
and the detachment-limited sediment transport equations since, as is clear in Equation (4.29), both mechanisms
from this conceptualisation. In fact, this approach was must comply with the mass balance equation. A simple
how the Einstein (1950) transport-limited sediment form of the combination of transport- and detachment-
transport equation was originally formulated (e.g. Hen- limited transport can then be written as
8
> ∂qsc when Dt > ∂qsc
derson, 1966). Davy and Lague (2009) explored the >
implications of this approach in the field. They found ∂z ∂qs < ∂x ∂x
¼ ¼ (4.30)
that the main advantage of this approach is that the ∂t ∂x >
> ∂q
: Dt when Dt sc
transition from transport- to detachment-limited trans- ∂x
port was more gradual and less abrupt, which may also
where
explain why researchers have failed to find an abrupt
change in the field. K t ðτ 0 τ c Þ for τ 0 > τ c
Dt ¼ (4.31)
0 for τ 0 τ c
4.2.2 Detachment-Limited Erosion and Bedrock
This is not a comprehensive form for the equations because
Incision
it assumes in the transport capacity term in Equation (4.30)
Detachment-limited erosion occurs when the constraint on that the upstream flow is also at transport capacity but
sediment transport is the ability to detach particles from regardless it captures the first-order idea that detachment
the surface being eroded. In contrast with transport- from the surface may limit the sediment that can potentially
limited erosion, once detachment occurs, there is no con- be transported by the flow. This limitation is in addition to
straint on the ability of the flow to transport sediment any entrainment limitation discussed in the transport-
downstream. These models are sometimes called incision limited erosion section, which is specific to the potential
models because they only downcut and cannot deposit effects of downstream changes in transport capacity.
sediment. Howard and Kerby (1983) and Howard (1994) Also note that while Equation (4.30) has been written
defined a detachment-limited erosion model for rivers using the detachment-limitation formulation of Howard,
flowing over rock or regolith the equation is easily generalised to any type of
detachment-limitation Dt by substituting a different equa-
∂z ∂q K t ðτ 0 τ c Þ for τ 0 > τ c tion for Dt (e.g. rainsplash).
¼ s¼ (4.28)
∂t ∂x 0 for τ 0 τ c A slightly different model of detachment limitation
used in the WEPP agricultural erosion model (and has
where it is the incision rate rather than the sediment
thus been optimised for hillslope erosion rather than chan-
transport rate that is defined explicitly, K t is the incision
nel erosion) combines the transport capacity of the flow
rate parameter and τ c is the critical shear stress threshold
with the detachment capacity of the shear stress of the
below which sediment cannot be detached from the bed
flow (Nearing et al., 1989)
and incision does not occur. This equation is defined for a

river, and so has a downstream direction x but no cross- Df ¼ Dc 1 qs=qsc
stream component (Chapter 2). This can be converted into (4.32)
D c ¼ K r ðτ 0 τ c Þ
a relationship dependent on discharge and slope using the
same methodology used for the transport-limited erosion where Df is the detachment rate for the flow (equivalent
equations so that to Dt in Equation (4.30)), K r is the soil erodibility of the
flow, and noting that the sign convention here is that Df is The volume eroded per impact is a function of the
negative where the surface is eroding for consistency kinetic energy of the sediment particle and how much of
with Equation (4.30) (but of opposite sign to that used that kinetic energy is transferred to the bed surface, and
in Nearing et al., 1989) and Dc is the detachment capacity the ability of the bedrock to absorb energy before fractur-
for a clear flow with no sediment. The major difference ing under tensile stress so that
between Equation (4.30) and (4.32) is that in Equation
M p wz 2 E
(4.32) the amount of detachment is a function of the Vs ¼ (4.33)
kv σ T 2
difference between the actual sediment transport rate
and capacity, while Equation (4.30) has a detachment where M p is the mass of the sediment particles impacting
rate that is a constant rate independent of the amount of the surface, wz is the vertical component of the particle
sediment in the flow already. The formulation in Equa- velocity, E is the Young’s modulus for the bedrock
tion (4.32) is conceptually similar to the entrainment matrix, σ T is the tensile strength of the bedrock matrix
limitation for transport-limited erosion in Equation and k v is an empirical parameter that relates the rate of
(4.25). Though the rate constants may be different, in particle generation relative to the energy per unit volume
both cases the rate of detachment is modified by the stored by the bedrock at the moment of tensile failure.
difference between actual sediment transport and the The impact rate is
sediment transport capacity. qs
The main weakness of the detachment-limited erosion Ir ¼ (4.34)
M p Ls
equation dependent on shear stress as formulated by
Howard for bedrock rivers is that it was developed empir- where Ls is the average hop length for the particle, the
ically from fieldwork at a small number of sites, and it has average distance between impacts with the stream bed by
lacked a fundamental process-based cause and effect. a sediment particle.
A large number of studies have tested the model in the These equations yield the rate of incision as a result of
field, and Lague (2014) provides a comprehensive sum- ballistic impact
mary of these works. A major research effort has gone
∂z ws 2 E
into developing process-based models based on impact ¼ V sIr ¼ q ¼ K b qs (4.35)
∂t k v σ T 2 Ls s
dynamics (the effect of impacts between particles in the
flow and particles on the bed at the water-soil interface) to where K b is a rate constant for the ballistics impact pro-
give Howard’s shear stress model, in addition to its con- cess and is a function of sediment characteristics (ws ),
ceptual appeal, a stronger physical basis (e.g. Whipple bedrock characteristics (k v , E, σ T ) and sediment-flow
et al., 2000). interactions (Ls ).
The two mechanisms responsible for incision are Returning to Howard’s shear incision equation one of
(Lamb et al., 2008, 2016) (1) rock particles, which are the first improvements to Howard’s model was the ‘tools’
plucked from the bedrock surface by the flowing water, model (Sklar and Dietrich, 2004, 2012). The idea of this
and (2) sediment in the flow, which impacts the bedrock mechanism is that even for detachment-limited transport,
surface and detaches fragments from the surface (the sediment is being carried in the flow (but the rate is less
‘ballistics model’). It is generally accepted that mechan- than the sediment transport capacity) and is being
ism (1), on its own, cannot generate the rates of incision exchanged between suspension/saltation and the bed.
that are observed in the field. Thus most recent research The cover of the bed by sediment is a function of the
has been on the second mechanism, the ballistics model. sediment being transported: low at low transport rates and
There is a long history of research into erosion of high at high transport rates, peaking at 1.0 when the
surfaces resulting from ballistics outside the geomorph- transport capacity limit is reached. At low sediment loads
ology community (e.g. the erosion of pipelines carrying the number of particles (the ‘tools’) impacting the bed-
particulate load). In the context of bedrock river erosion, rock surface is low even though the amount of bedrock
Sklar and Dietrich (2004) summarise the fundamentals of exposed is high and incision rates are low. At high
the ballistics model. The erosion by ballistics consists of sediment loads, near the transport capacity limit, the
two components: (1) the volume eroded per impact of a bed is mostly covered by sediment, so while there are
sediment particle with the bed, V s , and (2) the number of plentiful particles/tools, they rarely impact exposed
impacts per unit time, the impact rate, I r . Sklar and bedrock on the channel bottom because it is covered by
Dietrich derive the ballistics transport equation assuming sediment most of the time. The maximum incision rate
all sediment is one size and density. occurs at an intermediate load when there are plentiful
sediment particles to impact the bedrock and the bed is Whipple and Tucker (2002) and Sklar and Dietrich
not always covered by sediment. (2008) explored the slope-area properties of rivers
The cover effect is modelled by a linear function of the subjected to their model. They found that the scaling
actual sediment transport divided by the sediment trans- exponent α in the slope-area relationship decreased with
port capacity increasing area so that the log-log slope-area relation-
η ship was concave upward rather than linear. Using a
F e ¼ 1 qs=qsc (4.36)
sensitivity study, they showed that this change in α with
where F e is the fraction of exposed bedrock, and Sklar area could potentially be offset by orographic changes
and Dietrich assumed η ¼ 1. Sklar and Dietrich assumed in rainfall, downstream fining and changes in the bed-
that incision occurs only when the bedrock is exposed, so load grading. However, it would appear that a number
Fe is a multiplicative factor on the incision rule (e.g. of compensating processes need to be ‘just right’ to
Equation (4.35)). Chatanantavet and Parker (2009) com- generate the log-log linear slope-area relationship. For
pared Equation (4.36) against river abrasion data and this to be broadly true across many catchments suggests
found the best fit was for η ¼ 1. that there is a self-organisation principle driving this
Combining Equations (4.35) and (4.36) yields the balance. This begs the question of what the principle is
incision rate in the presence of the cover effect and why the log-log linear slope-area relationship is
central to this perfect balance (e.g. Rodriguez-Iturbe
∂z
¼ K b qs 1 qs=qsc (4.37) and Rinaldo, 2001).
∂t Another challenge in using the ‘tools’ model is that the
when η ¼ 1. This shows that when qs ¼ 0 and qs ¼ qsc impact energy is a function of the mass of the particles, so
the incision rate is zero and the incision rate is maximum there is a need to estimate the grading of the material
when qs ¼ 0:5qsc . Note that this equation is fundamen- being transported. In general this material will be material
tally different from Howard’s Equation (4.28) because of from the hillslopes, since if erosion is the same every-
the dependence on the sediment flux, which involves an where, then almost 100% of the material in the channel
integration of the incision/erosion rates for the entire must have come from the hillslopes. Thus it appears that
upstream catchment, while Howard’s equation depends the only way to parameterise the in-channel grading is to
only on the flow rate from upstream. Building on Sklar couple the channel model with the hillslopes, and for the
and Dietrich’s work there have been other models for the hillslope model to be able to predict the grading of the
cover effect (e.g. Lamb et al., 2008; Chatanantavet and material it delivers to the channel (e.g. Michaelides and
Parker, 2009, 2011), experiments to better characterise the Singer, 2014). This will likely involve coupling with a
empirical parameters for bedrock erosion (Sklar and Die- soils model much like will be discussed in Chapter 7,
trich, 2001), experiments to characterise particle fragmen- where the grading is a function of the in-profile soil
tation characteristics leading to different functional forms weathering processes breaking down the saprolite into
for Equation (4.35) and data on the change in cover and smaller fragments with time. Perhaps some of the
relative magnitude of tools versus cover with sediment grading-slope-area relationships derived by Cohen et al.
supply (Nelson et al., 2009; Turowski and Rickenmann, (2009, 2010, 2013) and Welivitiya et al. (2016) may be
2009). All maintain the underlying relationship between useful effective parameterisations of this process. Sklar
sediment load and transport capacity, with a midrange and Marshall (2016) have proposed a model based on
high incision rate. empirical data for soil grading from landslide deposits.
An unresolved problem with both the shear stress These works suggest that as incision rate increases (and
detachment and the tools-cover models is that they do erosion rate on the hillslope consequently increases), then
not appear to be able to reproduce the log-log linear the grading of the material delivered to the channel will
slope-area relationship of catchments (Section 3.3.3). coarsen. Welivitiya (2017) found that as the hillslope
Tomkin et al. (2003) evaluated several variants of the erosion rate increased, the grading of the hillslope surface,
shear stress model against field data using slope-area and thus the eroded sediment, also increased because the
profile data and found that none of the models fitted the resultant thinner soils were subjected to less weathering.
observations. They assumed that discharge was propor- Brocard et al. (2016) found that the fine nature of the
tional to area. They attribute the differences to incorrect eroded hillslope deposits in Puerto Rico (due to the wet
representation of the discharge relationship with area, and humid climate enhancing soil weathering) resulted in a
their results were consistent with a power law relationship low rate of downcutting in a bedrock channel because of
with power less than 1 (e.g. Equation (3.10)). the lack of abrasive power of the fine sediments. To date
this coupling between hillslope delivery and tools conceptual connection between the Lagrangian approach
abrasion has not been studied numerically. and the entrainment- and deposition-limited erosion in the
Finally, we observe one feature of some of the previous section follows.
detachment-limited stream incision models that become Furbish et al. (2012a, b, c) and Roseberry et al. (2012)
obvious only when incorporated into a LEM. Equation extend this approach to model bed load transport by
(4.39) shows that the rate of incision is a function of integrating in time and space the full probability distribu-
sediment from upstream. In the tools model if this sedi- tion of particle motions to yield a set of two-dimensional
ment transport is low for some reason and the tectonic (downstream and cross-stream directions) Eulerian
uplift is high, then it is possible that the incision rate equations. The downstream and cross-stream velocity dis-
cannot keep up with the rate of uplift and the landform at tributions are approximately exponentially distributed
that location will rise until potentially some other pro- (Furbish et al., 2012c), and the downstream velocities of
cess (that is likely to have a strong slope dependence, particles once entrained in the flow are significantly less
landsliding etc.) becomes dominant (Gasparini et al., than the flow velocity. This results in the probability
2007). distribution of the hop distance being nonlinearly depend-
ent on the time in the flow (Lt t =3 , where t is the time in
5
4.2.3 The Travel Distance Lagrangian the flow for each hop).
Conceptualisation One advantage of the Lagrangian approach is that it is
possible to track different populations of particles. For
The discussion of the previous section about entrainment instance, in Chapter 8 we talk about differentiating
and deposition limitation using an Eulerian formulation between particles and their chemistry. The Lagrangian
is clearer when we look in a little more detail at how approach is the logical extension for erosion to handle
sediment is actually transported in the flow. In this explan- this. One specific advantage is that the rate of entrainment
ation we examine the path followed by particles. Tracking p can be made proportional to the exposure of the differ-
the individual sediment particles uses a Lagrangian ent types of particle on the surface. If the concentration on
formulation (Section 2.2.3). the surface is low, then p is also low so that, everything
In the Lagrangian formulation a particle is entrained else being equal, the transport rate of that particle type is
into the flow, and then travels a set distance when it is low, and vice versa.
deposited from the flow. As a first approximation, if we The distance Lt in Equation (4.29) and L in Equation
assume that all the particles travel a distance Lt once they (4.23) are related because in the Lagrangian approach if
are entrained into the flow, then the number of particles all entrainment stops (i.e. a purely depositional environ-
passing a point per unit time can be converted to a mass of ment), then after distance Lt all particles will have
particles per unit time if the mass of a single sediment deposited. The relationship between these two distances
particle is known so that (Figure 2.3) provides intuitive support for the idea that Lt scales with
velocity and settling velocity because of the expression
qs ¼ mpLt (4.38)
for Lt .
where m is the mass of an individual sediment particle, p Equation (4.38) presents one challenge for modelling.
is the rate/unit plan area/unit time of particles being If the travel distance Lt is a function of flow velocity and
entrained into the flow and Lt is the travel distance. Thus is larger than the distance between computational nodes,
it can be seen that the sediment transport is increased if then the upstream history of the particle is needed at each
bigger particles are entrained (i.e. larger m), the rate of node to determine how much further downstream that
entrainment of particles p increases or the travel distance particle will travel before depositing. In principle, this
Lt increases. The travel distance Lt is likely to scale with means tracking the trajectory of every particle from
the flow velocity v, depth of the flow y and the inverse of entrainment until it is deposited. As discussed in Section
the settling velocity w, vy=w. As the flow velocity 2.2.3 this makes the process nonlocal.
increases, the sediment particle travels proportionately Furthermore if the travel distance Lt is longer than the
farther, while if the depth increases, then it takes longer length of the hillslope, then the Eulerian equations for
to settle out through the depth of the profile. This leaves transport-limited sediment flux may not be valid because
the rate of entrainment to adjust to ensure that Equation the hillslope will never reach the transport capacity of the
(4.38) equals Equation (4.17). In Equation (4.38) mp is flow. In this case at the bottom of the hillslope the sedi-
the mass rate of entrainment of particles per unit area per ment transport rate will simply be the detachment rate
unit time, which is also wc in Equation (4.24). The multiplied by the area of the hillslope.
Wainwright et al. (2015) used the Lagrangian repre- modelling. The Lagrangian approach provides very useful
sentation to question the very concept of transport theoretical insights, but ultimately it is the Eulerian
capacity. If we consider m, p and Lt in Equation (4.38) models, heavily based on experimental and field evidence,
to be random variables, then the flux qs is also a random that have been the workhouse tools for LEM modelling.
variable, and this would suggest that there is no maximum
value for qs (i.e. the concept of ‘transport capacity’, the
maximum amount of sediment that can be carried, is 4.2.4 Sediment Mixtures
incorrect), only statistics (e.g. mean) for the probability
distribution of qs . This Lagrangian conceptualisation and The discussion of the previous sections has largely
its implications (Wainwright et al., 2008a, b, c) have been ignored the impact of the grading of the transported
somewhat controversial (Hairsine and Sander, 2009; sediment on the rate at which the surface is eroding and
Wainwright et al., 2009). I note in passing that the concept the characteristics of the sediment entrained in the flow.
of ‘transport capacity’ is a fundamentally Eulerian The sediment grading has been discussed only in so far as
concept (being defined at a location), and this is one of it has controlled the total sediment load of the flow: the
the areas where the Eulerian and Lagrangian formulations finer the grading, the higher the load that can be carried by
for sediment transport are in fundamental conflict, and the flow. In this section we will talk specifically about the
more work is needed to resolve this conflict. approach to estimate the full particle size distribution
Furbish et al. ( 2012a, b, c) and Roseberry et al. (2012) (PSD) of the sediment entrained in the flow.
develop a Lagrangian theory to describe bedload sediment Two approaches are commonly used to model the
transport and test it experimentally. They show, for transport of sediment mixtures:
instance, that the travel distance of entrained particles 1. The transport capacity of the flow is calculated using
follows an exponential probability distribution (typically statistics of the sediment particle size distribution
an assumption of previous workers, e.g. Wainwright et al., in the flow (typically mean or median grain size). If
2008a, b, c). They also show that the entrainment-depos- the transport capacity of the flow is greater than the
ition-entrainment cycle is more complex than outlined amount of sediment in the flow, then the flow will try
above since (1) the initial entrainment preferentially to entrain sediment from the eroding surface until
selects particles from the least stable places on the bed, the transport capacity is reached. Depending on the
(2) then the particle preferentially deposits itself into the grading of the entrained sediment, this may modify
more stable places of the bed, (3) so that in the second the grading of the sediment in the flow and thus the
entrainment cycle the same particle is more difficult to transport capacity. These are called mixed size sedi-
entrain because it is in a more stable place of the bed ment transport models.
where particles are less likely than average to be 2. The transport rate of the different size fractions in the
entrained. This hysteresis in the suspension-resuspension flow is calculated directly (e.g. Wilcock and Crowe,
process was explicitly recognised in the Eulerian erosion 2003). Sediment is entrained into the different size
model developed by Hairsine and Rose (1992a, b). fractions from the bed to make up the difference
Apart from these detailed insights Furbish and his between the transport capacity in any specific size
coauthors’ work also highlights another practical diffi- fraction and the actual amount of sediment being
culty of using a Lagrangian approach to model transport. transported in the flow. These are fractional transport
The total sediment transport is a function of the aggregate models.
behaviour of the sediment particles, and the intricate
details of everything about each sediment particle’s In both cases, if there is too much sediment in the flow,
trajectory (and how that trajectory may influence the then deposition occurs by a settling velocity dependent
hydraulics, particularly for high concentrations) may be mechanism discussed previously, which will change the
crucial and this typically results in (1) exceptionally particle size distribution of the flow and thus the transport
complex models that are overparameterised with respect capacity because coarser particles (with higher settling
to available data or (2) simple models that might ignore velocities) will preferentially deposit, making the trans-
subtle but crucial aspects of sediment transport-hydraulics ported sediment finer and the transport capacity higher.
feedbacks.
The author’s personal view is that simple models, 4.2.4.1 Armouring
guided by theoretical insights but based heavily on experi- If we consider initially an ideal perfectly flat surface, the
mental and field evidence, are the most reliable tools for rate of entrainment of any given size fraction is a function
of the transport rate of that fraction if the flow contained where β is in the range 0–1. When β ¼ 0 all size fractions
100% of that size (this is called the relative mobility of are equally mobile, and as β increases in value the relative
that size class), and the proportion of the surface exposed mobility of the fine fraction decreases relative to the larger
to the bottom shear stress applied by that flow. End size fractions. There are numerous functions in the litera-
members cases of this arrangement are (1) if the surface ture (Parker and Klingeman, 1982; Proffitt and Suther-
contains 0% of the size class, then the entrainment rate land, 1983; Parker, 1990), and they all capture the
for that size class is zero and (2) if the surface contains empirical observation that finer particles are less entrain-
100% of the size class, then the entrainment rate is the able than would be the case if the shear stress were
rate that would apply if the sediment were 100% of applied to the surface and distributed according to the
the entrained sediment. It is commonly assumed that the proportion of the surface covered by that size particle.
entrainment rate is linearly related to the proportion of The published hiding functions have typically been
the surface covered by that size class, since for a perfectly developed using data from flumes or rivers, and so their
flat surface the bottom shear stress will be applied to each applicability to hillslope erosion might then be question-
size class in proportion to the fraction of the surface able. Hillslope surfaces are different from river channel
covered by that size class. The consequence of this is that bottoms (they are dry most of the time, vegetation is more
as the surface becomes coarser or has a greater percentage important and the ratio of the surface roughness height to
covered by immobile stony fraction, then the entrainment depth of flow is higher), so likely to have different surface
rate (and thus erosion) will decrease (Rieke-Zapp et al., sorting and hiding behaviour (e.g. Michaelides and
2007). Martin, 2012). Willgoose and Sharmeen (2006) examined
A side note here is that the shear force applied to field plot erosion data and found that the Parker and
immobile stone cover by the flow is transferred into the Klingeman (1982) hiding function (Equation (4.39)) gave
soil profile underneath as a result of the flow-resistant good results, but that the hiding parameter β was low, and
force applied by the stone particles into the bed, and in less sensitive to the particle size distribution of the surface
extreme cases this force may destabilise the bed particles armour than for rivers. This highlighted the need for site-
resisting the force applied by the stone particle. specific calibration of the hiding function for hillslope
However, most eroding surfaces are not perfectly flat erosion and armouring studies.
but armoured as a result of the preferential removal of the Wilcock and Crowe (2003) (hereafter referred to as
fine fraction. This preferential removal of fines is a func- W&C) summarise the fractional transport approach. The
tion of the higher relative mobility of fine particles. For transport rate of the flow in size class i is
the ideal perfectly flat surface, finer particles will be
W i F i u∗ 3
entrained at a rate that is higher than their proportion on qsi ¼
gðs 1Þ (4.40)
the surface because of their higher relative mobility,
resulting in the depletion of the finer sediments from the W i ¼ f ðϕ Þ
surface. The larger particles then hide the remaining where qsi is the sediment transport rate of grading fraction
smaller particles from being entrained. The mechanism i (in volume/unit width/unit time), s is the specific gravity
for this is that the larger particles project up out of the of the particles, F i is the proportion of the eroding surface
surface into the flow, and the large particles are exposed to covered by grading range i, u∗ is the shear velocity
a greater proportion of the bottom shear stress than would (u∗ ¼ ðτ 0 =ρw Þ, ρw is the water density), W i is the trans-
be the case if the shear stress distribution was based on the port function, ϕ ¼ τ 0 =τ ri , τ 0 is the bottom shear stress of
proportion of the surface occupied by large particles the flow and τ ri is a reference shear stress (which experi-
alone. This redistribution of the shear stress is character- ments show varies with sediment size) that is used to
ised by a ‘hiding’ function, which relates the diameter nondimensionalise the bottom shear stress. From experi-
of the particle fraction to the proportion reduction or ments W&C derived an equation for the transport function
increase in entrainment rate triggered by this shear stress
8
redistribution. >
< 0:002ϕ7:5 ϕ < 1:35
The main difficulty with using a mixed size transport Wi ¼ 4:5 (4.41).
> 0:894
model is then in defining this hiding function. Its generic : 14 1 0:5 ϕ 1:35
ϕ
form is
β The key result from this model is that given the proportion
di
Hi ¼ (4.39) of the sediment fractions on the surface, the hydraulics of
d 50 the overland flow (so that τ 0 and u∗ can be determined),
and a relationship for the reference shear stress τ ri , then in-stream grading at the catchment scale (Gasparini
the sediment transport rate and the size distribution of the et al., 1999, 2000, 2004; Coulthard et al., 2002).
sediment in the flow can be determined without any Before we finish specifically discussing erosion and
apparent need for calibration. W&C provide a method- sediment mixtures, there is one observation of specific
ology for determining the reference shear stress, and this relevance to soils and entrainment processes. In rivers it
was based on analysis of flume experiments using a sedi- is common to note that for a constant discharge over time
ment mixture of sand and gravel. They showed a signifi- the evolutionary end point of selective transport of fines is
cant change in behaviour based on whether the proportion to develop a river bed armour that is completely depleted
of sand was insufficient or sufficient to fill the pores of transportable sediment so that erosion stops (e.g. Mao
between the gravel particles (called a ‘framework-sup- et al., 2011; and many other authors). There is another
ported’ or ‘matrix-supported’ river bed, respectively). evolutionary end point.
While Equation (4.41) gives the appearance of a Consider a two-layer soil or river bed where the (thin)
model that does not require calibration, W&C note that armour layer overlies other material. If the armour layer is
the transport function can be conceptually related to the not weathering, then, for the case when the armour is not
hiding function so has the same empirical basis and changing with time (i.e. constant armour thickness and
limitations as the hiding function. For instance, Nelson grading), the sediment entrained into the flow by erosion
et al. (2009) found that the W&C model was a poor must have the same grading as the material underlying the
match to their flume experiments and postulated that their armour. This statement assumes only that no weathering
flume may have been operating outside the range of breakdown of the grading occurs within the armour layer.
conditions W&C was calibrated to. Recking (2010) com- It should be clear from the discussion about hiding that for
pared predicted (using W&C as well as other models) this to be possible requires a very specific (perhaps fortuit-
versus measured (flume and field data) sediment load and ous) balance between five processes: (1, 2) the grading
found a scatter of about a factor of 10 for W&C. Note that and rate of transport of the sediment in the flow from
in Equation (4.40) the transport function W i acts to dis- upstream, (3) the rate of erosion at that location, (4) the
tribute the shear stress of the flow between the size hiding arising from the armour and (5) the grading of the
classes i. For instance, if W i is constant for all grading underlying layer. This balance may not be fortuitous,
sizes, then the sediment is transported in the same pro- however, and can occur as a result of self-organisation.
portions as are exposed on the eroding surface, F i . This In rivers, where the couplings between those five pro-
intuitive argument provides a direct link back to the cesses are weak, particularly between the grading of the
hiding function. underlying material and the other processes, this balance
Finally, because of the significance of the transition may not occur.
between framework- and matrix-supported beds (and thus For soils, however, weathering within the soil profile
the significance of the contrast in particle diameters and the couplings between soil production, erosion and
between the sand and the gravel) it is not clear that the soil depth provide a mechanism to achieve this balance.
W&C model can be generalised to a continuous grading For example, if erosion is low, then the soil deepens, the
and whether other factors may then become important. grading of the material just below the armour becomes
The implication of the preceding discussion is that any finer (because with deeper soil there is greater time for
LEM that uses W&C without calibration to site-specific weathering action on the soil particles during exhumation;
data will likely have a predictive uncertainty of plus or Section 7.3.3 and Figure 7.5), the sediment transport
minus about a factor of 3. This is consistent with the capacity increases (because the grading in the sediment
reported predictive uncertainty of other erosion models in the flow is finer as a result of entraining finer sediment)
such as WEPP when book values are used. and the erosion increases until there is a balance between
Models of selective entrainment of sediment have the soil depth-weathering system and erosion. The
been used by a number of authors to predict spatial grading in the soil just below the surface acts to modify
organisation of surface grading. Chapter 7, on physical the transport capacity (Welivitiya, 2017). The coupling
weathering, will discuss the formation of an armour layer between the depth of soil and the cumulative weathering
on the soil surface and how this interacts with soil effect on the soil grading during soil exhumation is key,
development and organisation of soil grading in space and if this coupling is weak, then this balance is difficult
(Willgoose and Sharmeen, 2006; Sharmeen and Will- to achieve. It is also worth noting that this explanation
goose, 2006, 2007; Cohen et al., 2009, 2010, 2013). does not require slopes to change (e.g. in response to
Others have looked at the spatial organisation of erosion), and so does not require landform evolution to
achieve this balance, and the balance is a function solely particles. Using Equation (4.13) the transport rate for soil
of the soilscape evolution models to be described later in aggregates relative to dispersed soil is
this book. 1:5p
qsA d sA sA sA 1 0:5p
Finally, conceptual models for armouring are also ¼ (4.42)
possible. A simple model employed within the SIBERIA qsD d sD s D sD 1
LEM uses a reduction in erodibility that is a function of where the notation is the same as in Equation (4.13) except
cumulative erosion since the start of the simulation that the subscripts A and D indicate the aggregates and
(cumulative erosion being a proxy for the amount of fines dispersed particles, respectively. Figure 4.6 plots Equation
stripped from the surface) to estimate the erosion rate (4.42) for p ¼ 3 to show the sensitivity of the sediment
reduction. Hancock et al. (2017c) demonstrated that this transport rate to the diameter and density of the aggregates,
approach works well for a mine trial plot site. and it shows how sensitive the transport capacity of the
aggregates is to the density of the aggregates. The size
4.2.4.2 Particle Density and Enrichment range used for the diameter axis uses aggregates of 2.5 mm
One aspect of sediment transport that is unique to diameter, and the dispersed soil ranges from silt to sand.
hillslope erosion is that many of the particles that are Figure 4.6 shows that for most of the range of values of
transported are soil aggregates. As discussed in Section diameter and density, the water stable aggregates are more
10.6.1 soil organic matter (mostly exudates from plant transportable than the dispersed particles so that the con-
roots and fungal hyphae), clay, fine roots and fungal tents of the aggregates (and thus soil organic matter) will
hyphae combine to form larger particles called water- be more concentrated in the flow than the soil surface. The
stable aggregates. These aggregates are also less dense ratio of the concentration (mass of the constituent per unit
(~1,200–1,300 kg/m3) than disaggregated (also referred mass of sediment) in the flow relative to the concentration
to as ‘dispersed’) soil particles (~2,500–2,700 kg/m3). in the surface is called the ‘enrichment factor’.
Aggregates are enriched in soil organic carbon so that
erosion of these is an important component of carbon loss 4.2.5 The Transition from Detachment-Limited to
from soils. Examining Equation (4.13) it can be seen that Transport-Limited
depending on the relative impact of increased particle size
(reducing particle transport rate) and decreased density Howard (1998) discusses what happens in the field when
(increasing particle transport rate) these aggregates may rivers transition from transport-limited to detachment-
be more or less transportable than the dispersed soil limited and vice versa, and how transport and detachment
FIGURE 4.6: Contours of

relative mobility showing the
relative entrainment of sediment
for varying diameter (vertical axis)
and varying specific gravity
(horizontal axis). The top right
corner is heavy large particles, and
the bottom left corner is light
small particles.
limitation can coexist in the same stream. Patches of if the amount eroded on the hillslopes upstream increases
sediment cover the streambed, and between the patches (and assuming none of this hillslope material is deposited
the bare bedrock is exposed. The bare rock will be incised upstream). The amount of sediment exiting the catchment
by the flow while the sediment patches protect the (i.e. Equation (4.44)) must also increase. Since the trans-
bedrock (and presumably result from either localised port terms are fixed, the only way this can occur is if p
deposition or waves of sediment bedforms moving down increases (since if p increases the sediment load increases,
the stream). The similarity with the discussion in Section as a result of Equation (4.45)) and an increasing propor-
4.2.2 regarding the tools and cover model for bedrock tion of the time is being spent in the transport-limited
rivers should be clear, though in this section the sediment condition.
patches simply cover the underlying bedrock and do not The pure detachment-limited equation (Equation
contribute to bedrock incision. (4.29)) applies no matter what the upstream transport is,
The key issue then is determining what percentage of provided only that the sediment load is less than the
the bed, p, is covered by sediment. If the patches of transport capacity. Accordingly it must be true when
sediment occur randomly in space, then a time-space there is no hillslope erosion and the only incision is
substitution is possible, and we can then say that (1 p) occurring in the stream, and this is the case for p ¼ 0.
is the proportion of the time that the bed is detachment- As hillslope erosion increases then p increases until
limited, while the remainder of the time the sediment p ¼ 1. If the hillslope erosion continues to increase, then
transport is transport capacity-limited so that the lowering deposition must occur upstream because the stream sedi-
rate of the stream bed then becomes (from Equations ment transport capacity is insufficient to transport all the
(4.10) and (4.29)) sediment out of the catchment. Furthermore by the time-
space substitution, this argument also applies for the
∂z
¼ ð1 pÞK t ðK z Qm2 Sn2 τ c Þ (4.43) proportion of the bed covered by sediment with it
∂t
increasing from p ¼ 0 (for no hillslope erosion) up to
and the actual sediment transport averaged over time is p ¼ 1. The relationship between the hillslope erosion rate
ðð and p is linear. We thus conclude that the value of p
qsa ¼ ð1 pÞK t ðK z Qm2 Sn2 τ c ÞdA follows directly from the ratio of hillslope erosion rate
Ac
ðð relative to the channel transport capacity. This linear
þ ð1 pÞEdA þ pKQm1 Sn1 (4.44) variation is consistent with experiments that have specif-
ically examined the variation of channel coverage with
AAc
bedforms versus changing sediment inputs into the flume
where the first integral is the lowering/erosion over the (e.g. Nelson et al., 2009).
area of channel during the period of bedrock erosion, the One intriguing consequence of this relationship is that if
second integral is the erosion over the hillslopes in the hillslopes steepen and erosion subsequently increases,
the period of bedrock erosion (E is the hillslope erosion then p will increase and the average rate of incision in the
rate) and the third term is the sediment transport capacity channel will decrease (Equation (4.43)), tending to flatten
during the period between bedrock erosion events. Note the hillslopes. In many cases it is active channel incision
that (1 p) is inside the integral as, in general, p will vary that destabilises or steepens the hillslope leading to high
spatially. At first sight it would appear that p is free and hillslope erosion, so a reduction in the incision rate will
cannot be determined, but this is not the case. First, note result in a reduced hillslope erosion rate. This negative
that by definition feedback means that over time the channel incision rate
ðð and the hillslope erosion rate will reach an equilibrium,
ð1 pÞK t ðK z Qm2 Sn2 τ c ÞdA where the rate of downcutting of the river equals the rate of
Ac erosion from the hillslopes (Kirkby and Willgoose, 2005).
ðð
However, Gasparini et al. (2007) used a LEM to show that
þ ð1 pÞEdA < KQm1 Sn1 (4.45)
the transient behaviour before equilibrium is reached and
AAc
where equilibrium is disturbed by changes in external
otherwise the catchment will be transport-limited. Both forcings (e.g. climate, tectonics) of the landforms where
the detachment-limited and the transport-limited (the left- both detachment- and transport-limited are operating can
hand and right-hand side of Equation (4.45), respectively) be quite complex. The complexity arises because the inci-
terms do not change in response to changes in the sion by detachment-limitation upstream drives what parts
upstream geomorphology. Consider now what happens of the downstream reaches are transport-limited.
The derivations above do not include the tools-cover limitation modes for changes in the weathering/erosion
model extension to the detachment-limited model, but an rates is abrupt (type I weathering limitation in Figure 4.7),
extension to this case can be done using the same but where the weathering products are similar to the size
approach. of the original particles (e.g. when particles break in
half), then there is a more gradual transition (type II
4.2.6 Weathering-Limited Erosion weathering in Figure 4.7). Type II weathering limitation
occurs because the weathering results only in a gradual
The third type of sediment transport limitation is where change in transportability of the material so that the
the source of transportable sediment provides a limit on change in the transport rate is more driven by the gradual
how much sediment can be transported. The classic case change in the grading of the surface as a result of
of this is where weathering breaks down material at a rate weathering. In contrast, for type I weathering limitation
that is less than the sediment transport capacity. In this there is an immediate source of very highly transportable
case the surface of the soil is covered by a coarse armour material which is either available or not, leading to a
that is too coarse to be transported, but weathering frag- threshold behaviour.
mentation creates smaller particles that are fine enough to Type II weathering limitation can also lead to source
be transported. In Chapter 7 a range of potential fragmen- limitation that is a function of chemical weathering (Riebe
tation mechanisms will be discussed. et al., 2003, 2004). In this case the chemical weathering
Sharmeen and Willgoose (2006) discussed a case through the profile (that generates fine material when the
where this occurs. They studied a mine site consisting particles are exhumed at the surface) is in balance with the
of a rapidly weathering waste rock from a mine. In the erosion. If erosion for some reason increases, then the soil
monsoonal environment they found that during the dry will thin, soil particles are exposed to less time weathering
season the rock fragmented from physical weathering within the soil profile and the grading of the material on
due to magnesium sulphate precipitation in micro-cracks the surface will become coarser stabilising the erosion
in rock fragments, and then during the wet season the fine rate. Likewise if erosion drops, then the soils will thicken,
fragmentation products were transported during the early time spent weathering in the profile will increase and the
part of the wet season. The erosion didn’t stop com- soil surface will become finer so that the soil erodibility
pletely when the fine particles were depleted, but the will become higher, stabilising the erosion rate from fur-
decline in transport was quite rapid. They coupled a ther reductions. Thus there is a balance between the
fragmentation model, as will be discussed in Chapter 7, source of the potentially transportable material (i.e.
with an erosion and armouring model based on the hiding weathering) and the erosion and armouring process. Riebe
model of Parker and Klingeman (1982). By varying the concluded that field data supported this hypothesis
rate of weathering they showed a transition between (Figure 4.8). In Chapter 7 we will discuss how this pro-
transport limitation and weathering limitation cess can be modelled using a soil depth and a physical
(Figure 4.7). When the fines generated by weathering weathering fragmentation model and how simulation
were dramatically smaller in size than the armour (e.g. results with these models also support this mechanism
spalling-like behaviour), then the transition between (Cohen et al., 2010; Welivitiya et al., 2016).
FIGURE 4.7: Physical weathering-limited erosion showing (a) type I (spalling) weathering limitation, and (b) type II (body fracture)
weathering limitation (from Sharmeen and Willgoose, 2006).
experiments (Bagnold, 1936). The main difference is that

the viscosity and density of the atmosphere are different
so that the coefficients on transport rates scale differently
(Sørensen, 2004).
As far as deposition is concerned, the same issues with
process rates arise as for fluvial sediment transport. The
impact of aeolian deposition on soils has been shown to
vary. For small deposition rates relative to fluvial erosion
rates there may be little impact, but when rates of depos-
ition of fine sediment are comparable with rates of fluvial
erosion, it is possible for the armours that develop on the
surface due to fluvial erosion to be drowned and for the
erosion rates to proceed at rates much in excess of the
rates that would apply in the absence of the deposition of
the fine material (Cohen et al., 2015, 2016).
4.3 Temporal Averaging of Erosion
The transport-limited and detachment-limited erosion

mechanisms discussed previously apply for the instant-
aneous rate of sediment transport during a runoff event,
and the dependence on instantaneous discharge and/or
bottom shear stress in the equations characterises this
temporal dependence. Accordingly the sediment load
varies during the runoff event in response to the changing
discharge. The discharge varies in space and time both
within and between rainfall events as outlined in
Chapter 3. In landform evolution we are interested in
the net effect of these processes over long periods (years
to centuries and longer), and in many cases the details of
the intra-event processes is only a necessary step toward
FIGURE 4.8: Chemical weathering-limited erosion showing how determining the ‘geomorphologically effective’ equation,
the erosion rate (denudation rate in the figure) is limited by the which is a temporal integral across the distribution of all
chemical weathering (from Riebe et al., 2004).
the runoff events.
If we are simply running a computer model, and not
4.2.7 Aeolian Erosion and Deposition needing any deeper analytical insight, then a simple,
though rather brutal, solution would be to use a hydrology
In this book we will not discuss aeolian processes in model and a rainfall time series to generate a runoff time
detail. However, many predominantly fluvially domin- series at each node in the landscape discretisation and then
ated sites have aeolian inputs that cannot be completely use this time-varying hydrology as input to the erosion
ignored. These sites do not need to have fields of sand models in this chapter. This time series of erosion at every
dunes, but it may simply be that a portion of the miner- pixel can then be used to drive landform evolution
alogy of the field site may have been generated or (Coulthard and Skinner, 2016). This approach avoids the
sourced in dry past climates. Even under current-day need to explicitly develop a geomorphologically effective
climate it is believed that dust from Saharan Africa equation because the cumulative erosion is simply the
provides essential nutrients for the soils of the Amazon result of the time-varying simulation, provided the time
(Gläser et al., 2015). resolution of the simulation is sufficient to capture the
It has been long recognised that the principles of dynamics of the runoff and erosion processes. This is the
erosion for atmospheric dust are much the same as for approach that has been adopted by many modern agricul-
sediment in rivers. Much of the original research on tural erosion models (e.g. WEPP, SWAT), but is funda-
sediment transport was done with atmospheric mentally dependent on having either (1) high-quality, high
temporal resolution observed rainfall to drive the hydrol- where the functional relationship of the geomorphologi-
ogy model or (2) a high-quality stochastic rainfall simula- cally effective sediment transport equation is the same
tor that can generate rainfall and runoff series that are as the instantaneous sediment transport with the same
statistically valid. Many ‘climate simulators’ exist that exponents on discharge and slope, but with a different
can satisfactorily model rainfall for a small catchment, erodibility (K σ ) and the discharge is now the mean peak
where the rainfall is the same everywhere and the model discharge qp (approximately the 1 in 2.33 year extreme
needs only to replicate the observed rainfall’s probability discharge, Equation (3.10)). This shows that the geomor-
distribution and temporal correlation. However, for large phologically effective erodibility K σ is a function of a
catchments the problem is more difficult because of the number of variables including (1) the instantaneous
spatial variability of rainfall. Most climate simulators erodibility K, (2) the rate of recurrence of events λ, (3)
struggle to generate high temporal resolution rainfall with the normalised shape of the hydrograph (the integral), (4)
the correct spatial correlation statistics, the latter of which the mean time to peak of the event T p and (5) a complex
is important to get the relationship between discharge and relationship involving the coefficient of variation of the
catchment area correct (Section 3.1.9). peak discharge (the second term in second square brackets
An alternative approach is to use the simulated runoff in the equation for K σ ), the skewness of the peak discharge
and erosion time series to directly develop a ‘geomorpho- and the correlation between the peak discharge and time to
logically effective’ equation from erosion simulations for peak of the hydrograph. A simple and commonly used
a range of synthetic catchments (Willgoose and Riley, approximation of the probability distribution for the peak
1998a). In this case it is helpful to have some quantitative discharge is the log-normal distribution, in which case the
insight into how the physics of the intra-event erosion is skewness γqp can be directly expressed as a function of
related to the geomorphologically effective erosion, so variance σ qp . The importance of this equation is that it
that we know what the geomorphologically effective ero- shows that the geomorphologically effective sediment
sion will depend on. It is clear that it will depend on the transport can be directly related to the instantaneous sedi-
erodibility and the discharge, slope and shear stress rela- ment transport and that the functional form of the relation-
tionships, but it will also be dependent on the hydrologic ship (i.e. the power law relationship with discharge and
characteristics of the runoff (and thus rainfall), the intra- slope, and their exponents) remains largely unchanged.
event dynamics (peaky storms versus slowly rising The main point of difference is that the discharge is now
storms) and the probability distribution of runoff/erosion no longer the discharge at that instantaneous point in time
events. but is the mean peak discharge. As will be discussed for
the channels in Section 4.5 the mean peak discharge is
related to two common concepts in river analysis, the
4.3.1 Similitude Averaging dominant discharge and the bankfull discharge.
In practice Equation (4.47) is rather cumbersome to
Willgoose et al. (1989, 1991a) provides a simple simili- use, and involves a number of approximations, but it does
tude derivation that highlights these relationships, and provide a justification for the normal way that long-term
uses the principles of scaling analysis in Section 2.3. If erosion modelling is done. That is, the functional depend-
the instantaneous sediment transport relationship is ency of qs on area (mean peak discharge is normally
qs ¼ Kq S m n
(4.46) related to area by a power law relationship, Equation
(3.12)) and slope is used, and the rate parameter K σ is
then the geomorphologically effective sediment transport is calibrated by fitting the equation to a high-resolution
qs ¼K σ qp m Sn erosion series or monitoring data. This erosion series can
ð∞ be created by using Equation (4.46) with a high-resolution
ð^q ð^t Þ d^t Þ
m
K σ ¼K T p λ hydrology series (either measured or simulated) or by
∞
2 3=
2
3 using one of the many physically based agricultural ero-
6 σ q
2 γ q p
σ q p
2
σ q T
2
p 7
sion models that use runoff series (e.g. WEPP).
41þmðm1Þ 2 þmðm1Þðm2Þ þm p 5
p
qp qp 3 qp T p It should be noted that while Willgoose et al. (1989,

1991a) derived Equation (4.47) for transport-limited ero-
qð^t Þ sion, it is also equally applicable for the Howard
q ð^t Þ¼
^
qp detachment-limitation model since the derivation leading
t to Equation (4.47) requires only that Equation (4.46) be
^t ¼ (4.47)
Tp valid and the power on discharge be positive (so that
erosion increases with discharge, and small erosion events Equation (4.50) is parameterised by the mean discharge Q
can be ignored). Howard’s model is not expressed in the Lague’s results indicate that it is extreme discharges that
form of Equation (4.46), but in the case of dynamic control the incision rate, just as is the case in Equation
equilibrium where erosion is the same everywhere, then (4.47). Lague does not provide analyses (e.g. with differ-
the flux out of the catchment is ent values of m) to indicate whether this dependency on
the extreme discharge has an exponent of m as in Equation
∂z
qs ¼ A ¼ K t K z Qm2 Sn2 A (4.48) (4.47). That caveat aside, the full probability distribution
∂t
approach of Lague does provide indirect confirmation that
which can be reformulated into Equation (4.46) using the the temporal averaging of discharge leads to a dependency
peak discharge-area relationship in Equation (3.12). For on the extreme discharge statistics of the catchment and
the situation where the catchment is not in dynamic equi- not the average discharge of the catchment, and that the
librium, then this geomorphologically effective transport functional form of the discharge dependency remains
cannot be predicted by Equation (4.48). unchanged after averaging.
Both Equations (4.47) and (4.50) are dependent on the
4.3.2 Full Probability Distribution Averaging variability of the flow, and this variability will change
with the area of the catchment. As catchments become
Lague et al. (2005) present a more comprehensive aver- larger, the variability of the flow relative to the mean (i.e.
aging solution for the geomorphologically effective inci- the coefficient of variation) is reduced. The highest vari-
sion rate for the shear stress detachment-limited erosion ability occurs for the smaller catchments upstream. LEMs
using the full probability distribution of the flow, which transform the discharge dependency into an area depend-
they assume to have a gamma distribution. In their case ency so that they can calculate the sediment transport/
they average the incision equation incision at each node in the model. In Chapter 3, Equation
(3.12) shows how extreme discharges vary with catch-
I ¼ KQm Sn φc (4.49)
ment area and shows that they do not scale linearly with
where I is the instantaneous incision rate, Q is the instant- area. This reduction in variability is a result of two factors:
aneous river discharge and φc is the shear stress detach- (1) increased water storage due to flow routing down the
ment threshold below which incision does not occur. Note catchment, which smooths the hydrograph peaks, and (2)
the similarity of Equation (4.49) with (4.46) and (4.48), the spatial variability of rainfall for larger catchments is
particularly if we set φc ¼ 0. The dependence on dis- averaged out over the catchment. This means that for both
charge is the same so that the averaging techniques here geomorphologically effective equations the area depend-
are equally applicable to the transport- and detachment- ency of the equation will have an exponent that is less
limited erosion. Even if φc 6¼ 0 in the field, setting φc ¼ 0 than m. For Equation (4.47) the area term is
means only including the smallest runoff events, which
Amδ
also have the smallest incision rates. After averaging and qp m ¼ β (4.51)
setting φc ¼ 0 Lague’s geomorphologically effective inci- W
sion rate I was where W is the width of the channel and the width appears
because Equation (4.47) uses discharge per unit width,
Γðk þ 1 γÞ γ m n
I ¼ K kQ S (4.50) while Equation (3.12) uses area. For Equation (4.50) the
Γðk þ 1Þ
adjustment will likely be of the form
where Q is the mean average discharge (not the mean
k ¼ k0 Aðδ1Þm (4.52)
annual peak discharge as in Equation (4.47)), Γðk Þ is the
gamma function for k and k is a function of the variability where k 0 is a multiplicative constant independent of area,
of the daily discharge with higher variability for smaller k and the area-dependent variability adjustment is done on k
(Lague’s k is a combination of the gamma distribution because k characterises the variability of the flow.
shape and scale parameters, and is not the gamma shape
parameter normally denoted by symbol k). As the vari- 4.3.3 Perturbation Analysis Averaging
ability of the flow increases, Equation (4.50) increases,
potentially by orders of magnitude, because of the term k γ A third approach to averaging the erosion is based on
(see Figure 2 in Lague et al., 2005), and this can occur stochastic perturbation analysis (see Section 2.4 for a
only if extreme high runoff events have an increasingly extended discussion of this method). Some LEMs model
more significant effect as variability increases. Thus while the temporal variability of the runoff/erosion time series
explicitly rather than use averaged geomorphologically where cv is the coefficient of variation of the discharge
effective equations as above. The LEM explicitly models about the daily average value. Now Kq2 is the sediment
the variability of the runoff, and uses that variable runoff transport rate that we would calculate if we simply used
series to generate the sediment transport or erosion series. the mean value of the daily discharge and ignored
The temporally varying sediment transport series will the subdaily variability of discharge. Allowing for this
have an average erosion or transport rate. The runoff will subdaily variability about the daily mean shows that
have been modelled with a specified resolution, and impli- the correct sediment transport is higher by a factor of
citly it ignores any variability at finer time resolution than b is called the effective parameter for K.
ð1 þ cv 2 Þ and K
the modelled resolution (this finer resolution variability is It is common in small catchments for the cv of subdaily
called subgrid variability). The question is how accurate is discharge to be 1 or more so that the corrected sediment
the average of the generated erosion series if this process transport can be 100% more than that calculated using the
is followed, as it will be the average of the transport/ daily average runoffs. Some notable features of this
erosion that will drive the evolution of the landform. We derivation are:
can estimate the effect on the average transport rate by
using a first-order second-moment perturbation analysis. • The solution is exact for the integer exponent m ¼ 2.
• If we replace K with K b we can use the daily discharge
We will start with the simple erosion model
data and get the correct average transport rate since the
qs ¼ Kqm (4.53) functional form of the relationship between sediment
where initially we will assume an integer exponent of 2 on transport and discharge is unchanged.
discharge. This result will then be generalised for all • In this example we assumed that the modelling was
values of m later. For the sake of the example let’s done with daily average discharges, but any resolution
consider that we model the erosion with a daily timestep mean discharge can be used (e.g. hourly, weekly,
(i.e. we explicitly model the day-to-day variation in dis- yearly), and the coefficient of variation is calculated
charge and erosion, but average away everything occur- for the variability around the mean discharge for that
ring within the day). We know that there is significant resolution. As the resolution of the modelling is
variation in discharge and erosion during the day that is increased (e.g. changing from daily to hourly timesteps)
lost in the daily averaging. We can model the transport then the variability about the mean within each timestep
and discharge during the day as being the sum of the mean will be less, the effective parameter correction based on
for that day and the variability around that mean, where the coefficient of variation will be less and the error in
the overbar below indicates the mean value and the using the average discharge at the modelling resolution
primed value is the variability around the daily mean will be reduced.
(since we are doing a perturbation analysis here, technic- • While this derivation is done for transport limitation, the
ally the primed term is the perturbation about the mean) approach is equally valid for Howard’s detachment-
limited incision rate equation, though the value of m is
q ¼ q þ q0 ; qs ¼ qs þ qs 0 (4.54). smaller (typically m ¼ 0:5 1).
If the expectation (E[]) is taken of the discharge and • A parallel exists in the Lague et al. (2005) effective
transport, then by definition incision rate (Equation (4.50)) where K b in Equation
γ
h i (4.57) is equivalent to k in Equation (4.50).
E½q ¼ q; E½q0 ¼ E½qq0 ¼ qE½q0 ¼ 0; E q0 ¼ σ 2q
2
The solution can be generalised to the case where m 6¼ 2

E½qs ¼ qs ; E½qs 0 ¼ 0 (4.55) in Equation (4.53). Taylor series is required to expand the
where σ 2 is the variance. If we substitute Equation (4.54) discharge term when the perturbations in Equation (4.54)
into Equation (4.53), we obtain are substituted into Equation (4.53). If the first three terms
of the Taylor series are kept and higher order terms
qs þ qs 0 ¼ K q2 þ 2qq0 þ q0
2
(4.56) dropped, then the solution for the mean transport rate is
" #
and if we take the expectation of this equation and use mðm 1Þ σ 2q mðm 1Þðm 2Þ γq
Equation (4.55), then qs ¼ Kqm 1 þ þ
2 q2 6 q3
!
σ 2q (4.58)
qs ¼ K q þ σ q ¼ Kq 1 þ 2 ¼ Kq2 1 þ cv 2 ¼ K
2 2 2 b q2
q
where γq is the skewness of the subgrid variability. For
(4.57) m ¼ 2 this equation is exactly Equation (4.57). Again we
can do the effective parameter K b substitution. The only criteria that differentiated between those areas of the
change from before is that the correction for the subgrid catchment where a small surface perturbation would
variability is no longer exact (Taylor series terms of fourth grow (e.g. a rill, once formed, would get bigger) and
order and higher have been dropped), and the subgrid those areas where a small perturbation would disappear
correction now involves skewness. Note also that for over time (e.g. a rill if formed would tend to fill in and
m ¼ 1 that K b ¼ K so that for a linear transport-discharge disappear). Using Equation (4.10) the criteria is
relationship no correction is required, and this result is α ¼ m1n1
1
¼ 0. This is the same α defined for the slope-
exact because the dropped terms in the Taylor series are all area relationship at dynamic equilibrium (Section 3.3.3),
zero; this shows that it is the nonlinearity in the transport- and defines the break between those regions where slope
discharge relationship that introduces the effective param- increases with increasing area downstream (i.e. concave
eter correction. down, diffusive processes) and regions where slope is
Finally Huang and Niemann (2006) demonstrated that decreasing with increasing area downstream (i.e. con-
the definition of the geomorphological effective event cave up, incisive processes). Recent experimental work
(and thus the geomorphologically effective transport rate) (Sweeney et al., 2015) confirmed this result for diffusive
is a function of the dominant process (e.g. the value of m and incisive processes that were constant in time. There
in Equation (4.53)) so that when process dominance is still an unresolved question of channel head identifi-
changes (e.g. diffusive soil creep transport in upslope cation for temporally variable processes where the rela-
areas with m ¼ 0, and fluvial transport in the downslope tive regions of process dominance wax and wane
regions with m ¼ 1 to 2), then the definition of geomor- depending on climate variability, but where the land-
phologically effective processes (e.g. Equation (4.58)) form shape and regions of convexity lag these process
will also change. This complicates the calibration of geo- variations (Bull and Kirkby, 2002). It is important to
morphologically effective transport equations, and specif- note that Smith and Bretherton’s result is a stability
ically the evaluation of the process rate constant K b. analysis on the process dominance not landform shape,
and equating regions of process dominance with regions
4.4 Hillslope to Channel/Gully/Rill of convexity can be guaranteed only for landform equi-
Transitions librium. Sweeney’s experiments were performed for
dynamic equilibrium.
One of the most obvious features in geomorphology is
the distinction between channels and hillslopes. How-
ever, one of the most difficult challenges of the digital 4.4.2 Threshold Criteria
terrain data era has been how to identify where channels
start: the hillslope-channel transition. Likewise when An alternative approach is to define channels as being
constructing LEMs and modelling the different pro- formed when some threshold is exceeded (Bull and
cesses that occur on the channels and the hillslopes, a Kirkby, 1997). Initial work (Willgoose 1989) focused on
significant and only partially resolved issue is to define, velocity and shear stress thresholds that can be expressed
using LEM physics, the hillslope-channel transition in the form
location. Agricultural scientists face a similar problem β5 Qm5 Sn5
in determining when, where and how many rills (i.e. hillslope <1
threshold
small channels) will form on agricultural land, and what (4.59)
β5 Qm5 Sn5
the sediment load is in an actively rilling catchment. In channel >1
threshold
this section we use the terms channel, gully and rill
interchangeably because we are talking about the transi- where using the Manning equation and assuming a wide
tion from sheetflow to concentrated flow, and this con- channel of width w, then for velocity, β5 , m5 and n5 are

cept applies equally in all three cases, and it is simply a 1= w0:4 n0:6 , 0.4 and 0.3, respectively, while for shear
change in scale and potential permanence that differenti- stress, they are ρgn0:6 =w0:6 , 0.6 and 0.7, respectively,
ates the three cases. where n is the Manning roughness coefficient, ρ is the
density of water and g is the acceleration due to gravity.
4.4.1 Stability Criteria The dependence on discharge and slope is quite similar
for velocity and shear stress (Q=S constant with
The earliest work to address this transition was Smith m5 =n5 1), so it is difficult to distinguish the appropriate
and Bretherton (1972) who determined a stability equation from field data. The main difference is in the
4.4 Hillslope to Channel/Gully/Rill Transitions 73
sensitivity to Manning n. If one of these equations defines 4.4.3 Temporal Dynamics

the threshold of transition to channel, then the only
remaining question is to define the discharge used in the It is common to observe that the gully/channel network
equation, that is, mean discharge, dominant discharge, expands rapidly (i.e. drainage density increases) after land
one-in-two-year discharge or bankfull discharge (Board- disturbance and during this extension most sediment
man et al., 2003). This discharge definition is an unre- comes from gully excavation rather than sheet erosion
solved issue, but will impact on both β5 and m5 (see (Prosser et al., 1995; Wasson et al., 1998; Poesen et al.,
Section 3.1.9). 2003). This extension after disturbance is consistent with
Montgomery and Dietrich (1988, 1989) plotted field both gully head criteria. For the threshold criteria the land
channel data against area and slope, and found results that disturbance may result in either a higher discharge or a
are functionally consistent with Equation (4.59) if the lower threshold. For the stability approach, if the dis-
common assumption of discharge linear with area is charge increases, then the region dominated by incisive
made. While functionally consistent, the parameters were processes expands with a larger region that can potentially
very different from those above as they found m5 =n5 0:5 channelize.
for their Tennessee Valley site. Montgomery and The discussion above assumes that the channel heads
Foufoula-Georgiou (1993) present a landform slope-area are always in equilibrium with the landform and its
analysis (see Section 3.3.3) for the same Tennessee Valley hydrology and erosion. If channels heads are in disequi-
site, and this analysis shows, within the accuracy of the librium and are advancing upslope as a result of a change
DEM data they used, that the channel head area and slope in the environment, it is generally observed in the field
data plots parallel to the slope-area relationship for the (Radoane et al., 1995; Vandekerckhove et al., 2001;
topography, so their channel head relationship is likely a Vandekerckhove et al., 2003; Rengers and Tucker,
reflection of the erosion processes driving the slope-area 2014) that the strongest covariate for gully advance rate
relationship for the catchment rather than any fundamen- is the contributing area to the gully head and that the form
tal relationship for channel head dynamics (e.g. Kirkby of the relationship is
et al., 2003; Figure 8 in Poesen et al., 2003). Most subse- Er ¼ KAα (4.60)
quent studies have found it is difficult to identify a chan-
nel head relationship, but if any relationship can be where Er is the extension rate or velocity of the gully head
identified, it is typically in terms of contributing area upslope, α is an area exponent and K is a rate constant
and slope, and of the form of Equation (4.59) (e.g. Torri dependent on runoff rate (Vanmaercke et al., 2016). The
and Poesen, 2014; Dewitte et al., 2015). Moreover, these value of α is observed to be in the range 0.38–0.79
studies have typically failed to show that the relationship (Radoane et al., 1995; Vandekerckhove et al., 2003;
is different from the landform slope-area relationship, so it Rengers and Tucker, 2014; Vanmaercke et al., 2016).
is possible their relationships simply reflect points from
the landform slope-area relationship. Almost all studies 4.4.4 Mass Balance Considerations
indicate a strong relationship with area (e.g. Placzkowska
et al., 2015) and use of a ‘support area’ alone with no Ultimately the balance of erosion dynamics (typically the
slope dependency to define the hillslope-channel transi- transport-limited sediment transport physics described in
tion is common in LEMs. Prosser and Abernethy (1996) Section 4.2.1) at the headcut must drive (1) the velocity of
found a good correlation between channel heads and the the hillslope-channel transition upslope for extending net-
shear stress definition if they predicted runoff using works and (2) rates of infilling for shrinking or retreating
saturation-excess runoff generation (see Section 3.1.3.1) networks. To date there is no consensus about the details
rather than infiltration-excess runoff so that discharge was of gully extension dynamics (Poesen et al., 2003; Valentin
not proportional to area. et al., 2005). For instance, for a gully to extend and
Willgoose et al. (1990) suggested a solution to the maintain a sharp gully head, two things must happen:
apparent contradiction between the stability and threshold (1) there must be a step change in sediment transport at
approaches to defining the channel head. The Smith and the gully head to maintain the distinct head and (2) the
Bretherton stability criterion defines the regions where sediment transport processes in the gully downstream of
channels can potentially form, while the threshold criter- the gully head must be more efficient than upstream so
ion determines where they can actually occur within that that the sediment being generated by the excavation
potential region. The threshold criterion remains poorly of the extending gully head can be removed. Typical
resolved. explanations for the step change involve breaching of
the vegetation cover (e.g. Prosser et al., 1995), but this as being transport-limited, though we will see when we
doesn’t explain other cases, such as gully formation in look at the long-term evolution of floodplains that this is
bare agricultural fields. Finally, when the gully stops unlikely to be true at all times. As we have seen in
advancing, there must be a change in the transport previous sections, whether for transport- or detachment-
capacity in the gully downstream (e.g. gully widening limited conditions, a description of the channel geometry
downstream of the gully head), otherwise the gully will is required. At a minimum, the channel width is needed.
erode the gully downstream of the gully head to compen-
sate for the reduction of sediment delivery by the gully 4.5.1 Channel Geometry
head that is no longer extending.
Finally, there is one aspect of the hillslope-channel Two approaches can be taken to describe channel geom-
transition that is of specific interest to LEM users: the etry. They are (1) use empirical equations for channel
mass balance impacts of the transition from hillslope to geometry (e.g. regime equations) that are typically para-
gully/channel. In the LEM it is not simply a matter of meterised by discharge and channel slope or (2) use
identifying where channels occur and then changing the physically based dynamical models to explicitly model
dominant process modelled in the model depending on the evolution of channel geometry and couple it to the
whether that location is hillslope or channel. The channel evolution of the broader landform and contributing hill-
also has width and depth dimensions. Typically the width slopes. The first approach has the advantage of simplicity,
of the channel is less than the grid resolution of the model but regime-like equations are implicitly assumed to be
so is not explicitly resolved. As a consequence the change equilibrium properties of the channel, so it is not possible
in mass in the transition between hillslope to channel to model dynamic interactions between channel geometry
is not resolved (i.e. mass ¼ width depth density; and landform evolution using them.
Figure 4.9). Generally a model without some explicit An example where a dynamic description of channel
allowance for this effect will overestimate the amount of geometry might be useful is the common observation of
sediment lost in the transition from hillslope to channel. In changes in channel width with reductions in sediment
the agricultural erosion community it is generally believed load, and those reductions in sediment load could result
that the mass of sediment eroded by rill and gully erosion from changes in the erosion rate (and thus landform
is of a comparable magnitude to sheet erosion (e.g. Van- evolution) on the hillslopes. As the previous derivations
daele and Poesen, 1995; Martinez-Casasnovas, 2003; have shown, a change in channel width will result in a
Poesen et al., 2003), so this mass balance error may be change in sediment transport capacity, a subsequent
significant for cases with active gully development. change in channel evolution, which will feedback into
hillslope evolution. An obvious question is whether it is
necessary to model the relatively ‘fast’ evolution of the
4.5 Channel-Specific Issues channel geometry if we are interested only in the rela-
This section addresses issues of channels in alluvial flood- tively ‘slower’ evolution of the landform. The downside
plains. It is sufficient at this stage to think of these channels of using the second, dynamical approach is that (a) the
processes controlling geometry dynamics are not well
understood quantitatively and (b) physically based
dynamical models will need to be run at the time scale
of the evolution of the channel cross section, which will
typically result in very small model timesteps and long
compute times.
We will not discuss regime equations in detail here
(Parker et al., 2007 revisits both the data and the physics
underlying them) other than to note that they characterise
the bankfull dimensions of the channel as a function of the
bankfull discharge Q and are of the form
FIGURE 4.9: Channel and hillslope cross section at right angles
to the flow. The heavy line shows the actual profile while the light B ¼ K b Qb
(4.61)
lines shows how it is interpolated in a digital elevation model with H ¼ K h Qf
resolution coarser than the width of the channel. The hatched
cross-sectional area is material that the LEM will erode that is not where B is the bankfull width, H is the bankfull depth, K b
actually eroded in the field. and K h are proportionality constants and the exponents b
and f are typically about 0.5 and 0.4, respectively (Millar,

2005 in a review quotes ranges 0.45–0.55 and 0.33–0.40,
respectively). It has been noted by many authors over
many years (e.g. Henderson, 1966) that these equations
are not dimensionally consistent (e.g. both K b and K h
have dimensions that change with b and f). Also the
exponents b and f are not independent since Q ¼ BHv
where v is bankfull velocity.
Under some circumstances these equations can be
dimensionally consistent. As shown in Section 3.1.7 for
catchments at dynamic equilibrium (so slope approxi- FIGURE 4.10: Schematic of shear stresses being applied by (a)
mately declines with the square root of area), velocity is the flow to the particles on the bed and (b) gravity down the side
weakly dependent on discharge when travelling down- of the channel.
stream in the channel (v K v Q0:1 ; Equation 3.4). Substi-

tuting this into the equation for discharge and summing where the cross section is divided into vertical slices as
the exponents from the regime equations shown. If we consider each slice independently of the
others, then the shear stress applied by the flow on the
b þ f þ 0:1 ¼ 1 (4.62)
channel boundary is
we see that the typical values for b and f are approxi- ,
ρgyS
mately dimensionally consistent. The need to assume the τ¼ dy (4.63)
slope-area curve (which determines discharge dependence dx
of the velocity equation) at dynamic equilibrium to
where the absolute value on the derivative is to remove
achieve dimensional consistency highlights the depend-
the change in sign that occurs at the channel centreline. At
ence of the regime equations on channel slope, an output
the edges of the channel where the flow is shallower, this
of LEMs. When regime equations are used in river engin-
longitudinal shear stress is less. However, there is also a
eering practice, two further equations, one for sediment
shear stress being applied to the particles on the boundary
grain size and one for channel slope, are normally
by gravity driving particles to move down the channel
specified to address this potential dimensional inconsist-
boundary at right angles to the flow toward the bottom of
ency. Both slope and grain size are an output from LEMs
the channel. The total shear stress on the boundary is then
rather than specified by empirical equations. This may
the vector sum of (1) the downstream shear stress applied
lead to dimensional inconsistencies in channel geometry
by the flow and (2) the gravity force applied at right
if the regime equations in Equation (4.61) are used in
angles to the flow. At equilibrium the channel cross
LEMs (where slopes are an output rather than specified),
section geometry is such that the magnitude of the total
a potential issue that is typically ignored by LEM
shear stress is the same everywhere. This approach is
developers.
the basis of a number of simple models for optimal chan-
We will now discuss the dynamical channel geometry
nel geometry (e.g. Lane, 1955). However, this simple
models in detail. Later in this chapter we will construct a
example ignores a number of important processes. For
model for river meandering and floodplain evolution, and
example, each slice has a different average velocity, so
this meandering model is built on the principles used for
there are shear stresses (as a result of the velocity gradient
the channel geometry models. While the mathematics of
across the slice boundary) applied at the vertical interface
channel geometry evolution can be quite challenging, the
between each slice. However, to first order it captures the
underlying principle is relatively simple. The underlying
accepted drivers of channel geometry. This simple model
principle is that channels evolve until the shear stress
also ignores the stabilising effect of vegetation and
applied to the boundary of the channel (i.e. both the
increased shear resistance of vegetated banks (Pollen-
channel bottom and the sides) is constant. This means that
Bankhead and Simon, 2010; Camporeale et al., 2013).
the mobility of the sediment forming the channel bound-
This approach can be extended to transient conditions:
ary is equal and that sediment transport occurs at the same
rate on all parts (i.e. both the bed and the banks) of the • If there are regions on the channel cross section with
channel boundary. localised high shear stresses (high relative to the rest of
A simple example will demonstrate this equal shear the channel cross section), then these regions will erode
stress principle. Consider the cross section in Figure 4.10 until the shear stress is reduced back to that being
applied to the rest of the cross section. Likewise for sediment is being delivered from upstream to make a
localised low shear stresses the erosion will be reduced cross section transport-limited).
relative to the rest of the cross section (deposition may
even occur) until the cross section is in equilibrium.
4.5.2 Floodplains and Meandering
• If the entire cross section has a higher shear stress than
the equilibrium shear stress, then the channel geometry We now build on the channel geometry models to build a
will erode until the shear stress is reduced to the equi- physically based model for river meandering. Consider a
librium value. Generally this involves widening of the minor lateral (i.e. at right angles to the direction of flow)
channel. In the derivation for transport-limited erosion perturbation of flow such that the highest velocity is no
(Equation (4.15)) the wetted perimeter (approximately longer exactly in the centre line but is slight offset to one
equal to channel width) appears in the denominator of side. In this case the shear stresses will now be slightly
the sediment transport rate constant so that for constant higher on the side of the channel with the higher velocity.
discharge and longitudinal slope the sediment transport This results in higher erosion on that side of the channel
rate decreases (and as a consequence so does the erosion so that the channel cross section moves laterally in the
rate) as the channel widens. Thus the channel geometry same direction as the perturbation in the velocity, a posi-
response to excess shear stress and so excess sediment tive feedback. These perturbations are the trigger for the
transport capacity is for the channel to widen, flow to beginnings of meandering in what was initially a straight
become shallower and the average cross-sectional vel- channel. This is the basis of all the physically based
ocity to decrease. models of channel meandering within a floodplain. The
• Likewise if the shear stress is lower than the equilibrium full mathematics for this process is quite challenging, and
shear stress, the cross section will narrow and deepen to date no general solution has been developed. The
until the depth of flow is such that the shear stress differences between the approaches adopted in the litera-
applied to the channel bed is equal to the equilibrium ture are in the approximations in the equations used for
value. the modelling, and the field and experimental data to
justify those approximations. Because of this mathemat-
The derivations above are normally discussed in the ical complexity, this section concentrates on the under-
context of alluvial channels and transport-limited sedi- lying concepts, and the reader is referred to the key
ment transport. However, nothing in the derivation literature for the detailed derivations and field data. The
invalidates its use for detachment-limited conditions in two main approaches are (1) the HIPS equation (named
bedrock channels. The shear stress threshold could as for the developers; Hasegawa, 1977; Ikeda et al., 1981;
easily be the threshold for plucking of particles from the Parker et al., 1982) and (2) topographic steering (Dietrich
rock (since the rate of plucking in bed rock rivers is and Whiting, 1989).
normally parameterised by use of shear stress, with or Parker et al. (2011) summarises the HIPS approach.
without a shear stress threshold below which plucking Figure 4.11 shows the plan of a river bend. The key point
does not occur; Whipple et al., 2000; Lamb et al., 2015). is that the average velocity on the outside of the channel
One interesting aside for bedrock rivers is that using centreline is larger than the average velocity on the inside
flume studies, Johnson and Whipple (2007) found that so that
the channel cross section evolved from initially being
detachment-limited to a condition where the roughness 1
Δu ¼ ðuo ui Þ (4.64)
of the channel was sufficiently high that the transport 2
capacity dropped and the channel became transport- where the subscripts o and i refer to the outside and inside
limited. When they increased the sediment load to the of the meander bend, respectively, and the rate of migra-
channel, sediment deposition occurred in the scour holes tion of the meander is then
of the river, smoothing the roughness. We speculate that
ζ ¼ KΔu (4.65)
this feedback between detachment limitation increasing
bedrock roughness and transport limitation smoothing the where ζ is the rate of migration of the meander bend and
roughness may lead to bedrock channel cross sections, K is an ‘appropriately chosen dimensionless coefficient’
equilibrating at the boundary between detachment and of the order of 108–107 (Parker et al., 2011). Parker
transport limitation. This suggests a coupling between notes that while the mathematics leading to Equation
these, local, cross-sectional dynamics with upstream (4.65) yields the physical dependency of K, experience
sediment processes (which determine whether enough has found that the value needs to be determined by
calibration in the field. That still leaves the issue of deter- equation on either side of the centre line. The key concept
mining the velocities in Equation (4.64). A variety of is that the velocity of the outside of the cross section is
means are available of determining velocities, but all higher than on the inside. The HIPS model is based on a
require cross-sectional geometry. This cross-sectional linearisation of the flow equations. Güeralp and Marston
geometry can either be assumed or derived with a coupled (2012) review a number of extensions that use higher
erosion model (Parker, personal communication). Once a order approximations and that claim to better model
geometry is determined, the velocities can be calculated details of the shape of the meander bends.
by solution of the shallow water equations, or, given the The topographic steering model depends on the differ-
empirical basis of Equation (4.65) which requires calibra- ence in the depth of the flow from the inside and outside
tion in the field (Parker et al., 2011), something more of the bend (Figure 4.12). As the cross section changes
approximate such as the application of the Manning from being symmetrical at the approach to a meander
bend to being asymmetrical within the bend (with the
deepest depth on the outside of the bend), the river flow
(and the thalwag) is steered toward the outer bank. This
steering causes the flow to hit the outside of the bend at an
angle so that erosion occurs on the outer bank. It is this
steering of the flow toward the outer bank that triggers
erosion. Parker et al. (2011) notes that the main LEM
implementations of topographic steering (Lancaster and
Bras, 2002; Tucker et al., 2001a), while based on different
physics from HIPS, are functionally equivalent to the
HIPS equations as summarised in Equation (4.64).
The main functional difference between HIPs and
topographic steering is that HIPS can generate meander-
ing from a symmetric channel cross section, while topo-
graphic steering requires the asymmetry to start the
FIGURE 4.11: Schematic in plan of the forces and velocities in a meandering. The debate about which of the HIPs and
meander bend: (a) plan, (b) cross section (after Parker et al., 2011). topographic steering models better simulates meander
z
s
(a) (b) n i+1

b/2
H
i+1
i ds
i
(c)
Acs
qnV
i+1 i dAcs
FIGURE 4.12: Schematic of the flows directions in the topographic steering mechanism (from Lancaster and Bras, 2002).
dynamics is now more or less settled in favour of HIPS so of the average velocity of the cross section). Lancaster and
that river velocity differences rather than depth differ- Bras (2002) and Tucker et al. (2001a), while they used the
ences are believed to be the dominant driver of meander topographic steering mechanism, also noted that the peak
development (Parker et al., 2011). erosion needs to be downstream of the peak curvature of
Howard (1992) suggested an equation for meander the meander bend, and they offset the location of the peak
lateral movement based on a combination of HIPS vel- migration rate downstream by a term that was a function
ocity and topographic steering cross-sectional change of the average velocity of flow.
drivers that was Van De Wiel et al. (2007) and Coulthard et al. (2007)
describe a meandering model in their CASEAR LEM
ζ ¼ K aΔu þ bΔh (4.66)
which is a function of the radius of curvature of the flow
where Δh is the difference in the flow depth between the direction and where the maximum rate of migration is at
outer and inner half of the cross section, and a and b are the location of the maximum curvature. These authors
parameters. The HIPs component is the first term inside have not done any studies of meander migration or shape
the parentheses, while the topographic steering compon- with their model. Given that Howard and Knutson (1984)
ent is the second term. Subsequent analyses of field and concluded that a model based solely on radius of curva-
laboratory data (for a discussion of this see Howard, 1996 ture and ignoring the upstream geometry (and the lags
and Parker et al., 2011) has shown the topographic introduced) did not produce downstream migrating
steering term is dominated by the HIPs velocity term meanders, this would suggest that the Coulthard model
(i.e. b ¼ 0) leading to Equation (4.65). would be unable to simulate down-valley migration of
Howard and Knutson (1984) presented a meandering meanders.
model that captures many aspects of the physics discussed Güeralp and Marston (2012) noted that, in general,
above. Even though it predates much of the more recent model-generated meanders are too regular compared with
theoretical work it still provides insight into how to observed meanders, and that there is some evidence that at
construct a meandering model. There are a number of least part of this field irregularity is a function of variabil-
competing processes. The channel is modelled as a series ity of soil and vegetation on the floodplain, both of which
of links downstream that can move in response to bank are ignored in the models above. One of the implicit
erosion. The rate of migration is a function of the radius of assumptions of both HIPS and topographic steering is that
curvature of the meander at that location and the upstream the dominant mechanism for meander migration is the
geometry of flow. A model based solely on radius of erosion rate of the bank. If the erodibility of the bank is
curvature yielded unrealistic meanders. An implicit spatially variable, then meander migration will be con-
assumption is that the channel geometry and meandering strained. This variability could be due to vegetation (e.g.
evolve much faster than the floodplain so that the Pollen-Bankhead and Somin, 2010), floodplain soil vari-
meandering can be considered to occur on a fixed eleva- ability from previous meander migration or floodplain
tion floodplain. As the sinuosity of the stream increases, topography. For floodplain topography, if the channel
the streamwise length of the river increases while bank is high, then it is possible that the bank erosion will
the valley slope remains unchanged. Thus the slope, and undercut the bank and a large amount of sediment from
the erosive potential of the river decreases, and the flow the collapsed bank will slump into the river and protect
depth increases in proportion to the decrease in the the bank from further migration until the slump is eroded
streamwise slope. away. Thus the migration rate is a function not just of the
Howard (1996) and Lancaster and Bras (2002) note bank erosion rate but also of the ability of the flow to
that for meanders to migrate down-valley as observed in remove the full height of the eroding bank (Howard and
the field, then the location of maximum ζ must be offset Knutson, 1984; Parker et al., 2011; Bufe et al., 2016).
downstream of the location of maximum river curvature. Finally, in almost all cases the modelling has assumed
If there is no offset, then the meanders migrate only at that the width of the channel does not change with
right angles to the direction down-valley. In HIPS this meandering. Changing width complicates the theory, but
offset downstream occurs because the velocity does not a number of authors have attempted to extend the HIPS
respond instantaneously to the change in channel cross theory for variable width (Luchi et al., 2010; Zolezzi
section, and the distance from the minimum radius down- et al., 2012). The main impact of varying width is to make
stream to the peak velocity is a function of the rate of meander bends wider and shorter than is the case for
acceleration of the flow and distance travelled down- constant width channels. Parker et al. (2011) present a
stream during that acceleration (and is itself a function framework where the outer bank is eroding and the inner
bank is depositing independently and where both are requires event by event modelling of runoff, and prob-
explicitly modelled so that width variations can arise ably good intra-event hydrology to capture the portion of
naturally out of the formulation. With the exception of the hydrograph that generates overbank flow and the
Parker et al. (2011) the models discussed above assume hysteretic sediment transport of the rising and falling
that the deposition on the inner bank keeps pace with the limbs of the hydrograph, and (2) even for a single flood
erosion on the outer bend because they implicitly assume event, the computations for these types of models are
that the channel width doesn’t change. An advantage of high, so modelling multiple events over 100’s to 1,000’s
the Parker formulation is that it is then quite natural to of years is likely to be infeasible. Lauer and Parker
allow the outer bank to be more or less resistant due to (2004) present a box model that explicitly captures the
vegetation or slumping without implicitly assuming an event dynamics but without the complex floodplain
exactly compensating change in the deposition for the hydraulics. Typically, however, the few LEMs that
point bar on the inner bank. explicitly model floodplain dynamics use empirical
One of the problems with the modelling of meandering models for floodplain deposition and are based on gen-
is the difficulty of testing the models against data, and eral observations of floodplain deposition patterns.
thus determining what is the best modelling approach. For Existing models are based on models proposed by
instance, Parker et al. (2011) noted that none of the Howard (1992, 1996). Howard’s model captures two
models work well without significant site-specific calibra- empirical observations:
tion, and in many cases the criterion for success is simply
replication of the rates and direction of the meander • The rate of overland deposition diminishes with flood-
plain age and floodplain elevation, and this is because
movement (e.g. from aerial photos over 10’s of years),
the higher elevation locations on floodplains are flooded
which may not be sufficient to model long-term behav-
less frequently.
iour. Teles et al. (1998) propose a test using the strati-
graphic properties of the floodplain deposits. While their • Floodplain deposition rates diminish with distance from
the channel in the floodplain.
work was very preliminary, there seems some potential in
this approach if large enough areas of floodplain can be To capture these observations Howard (1996) pro-
analysed. To collect data for the large areas required posed an equation for deposition
ground penetrating radar or recent developments in the h i
ϕ ¼ ν þ μe =λ e½γðzfp zbed Þ
d w
use of seismic surveying for shallow depths might be (4.67)
suitable.
where ϕ is the overbank sedimentation rate, ν and μ are rate
4.5.3 Alluvial Floodplains parameters for sediment deposition for fine and coarse
sediment respectively, λ is a length scale controlling the
The modelling of meandering rivers in LEMs discussed in rate of decline of coarse sediment deposition away from the
the previous section deals almost entirely with the river river, d is the distance to the closest part of the channel,
itself and does not explicitly model the evolution of the zfp zbed is the difference in height between the channel
floodplain within which the meandering river is situated bed and the floodplain and γ and w are parameters control-
(e.g. Howard, 1992). The meandering river simply moves ling the rate of deposition with floodplain depth. The first
from side to side within the preexisting floodplain, and the term (in the square brackets) controls how the sedimenta-
elevation of the floodplain outside the channel does not tion changes with distance from the river. The second term
change. That leaves two unresolved modelling tasks: (1) (the exponentiation) controls how the sedimentation
aggradation or degradation of the floodplain and (2) changes with increasing height of the floodplain above
widening of the floodplain. the river bed, and implicitly reflects the probability distri-
Floodplain widening can be modelled by adding a bution of the river discharges, the bankfull conveyance
multiplicative term in Equation (4.65) to account for capacity of the river and thus how often the river floods
differences in the bank erosion resistance of the rock the floodplain. The main difference of Equation (4.67) from
outcrops surrounding the floodplain relative to the flood- that proposed earlier by Howard (1992) is that the earlier
plain material (Howard, 1996). work had a linear function for the distance decay, rather
The overbank deposition and scouring on the flood- than an exponential term.
plain could be calculated by doing a two-dimensional Tucker et al. (2001a) modified the Howard (1996)
coupled floodplain hydraulics and sediment transport model by removing the distinction between fine and
model, but this has two disadvantages for a LEM: (1) it coarse sediment and simplifying the function for
floodplain elevation relative to river elevation. Their models that model transport-limited sediment transport
relationship was (i.e. processes that allow deposition) generate structures
d
whose elevations look like alluvial fans (e.g. Willgoose
ϕ ¼ μe =λ zfp zwl (4.68) and Riley, 1998; Coulthard et al., 2002). Recent work
(Welivitiya, 2017) has developed a coupled soilscape
where the notation is the same as in Equation (4.67) and
and landform evolution model that implements (1) mixed
zwl is the elevation of the water level in the river. The
sediment size erosion and deposition and (2) tracks the
water level in the river is determined by
full particle size distribution in the flow. This model
δs
zwl ¼ zbed þ k h Qb δb Q=Qb (4.69) generated realistic landforms (cross- and long-section
concavities) and realistic patterns of particle size distribu-
where the parameters k h , δb and δs can be determined tions that qualitatively match observations including
from regime equations as discussed in Section 4.5.1. In downstream fining, and lenses of coarse and fine materials
contrast with Howard’s equation, Equation (4.68) requires within the fan itself (Figure 16.5). The internal structure of
knowledge of the probability distribution of discharges in their fan appears qualitatively consistent with previous
the river to determine the distribution of Q and thus zwl . modelling work of submarine fans that used an empirical
Finally, we observe that the elevation of the floodplain model for sediment size fractions (Koltermann and Gor-
bottom (i.e. the floodplain deposit-saprolite interface) is not elick, 1992). Gasparini et al. (1999, 2004) also observed
normally modelled. We might reasonably assume that the downstream fining within an erosional drainage network.
processes that convert saprolite to soil in Chapter 6 are Welivitiya (2017) also qualitatively observed other fea-
inactive because the floodplain deposits are too deep for tures of the simulated alluvial fan such as longitudinal
them to be active. The floodplain bottom can then only radial filaments of coarse material. These match features
decline only as a result of incision of the saprolite interface observed in the field (Gómez-Villar and García-Ruiz,
by the flowing water in the river. The implication of this 2000; Clarke, 2015).
incision control is that floodplain lowering is a function of
the rate at which the bottom of the meandering river is able 4.6 Some Synthesis of Fluvial Erosion
to incise the floodplain bottom as it meanders back and
forth across the floodplain (called ‘bevelling’; Bufe et al., Before discussing the details of numerical implementa-
2016). While for much of the time alluvial rivers have a tions and the issues they address, it is worth summarising
sand or gravel bottom (so the bedrock is protected from how the many topics discussed in this chapter fit together.
incision), during floods the channel cross section is excav- It will then be clear how many of the concerns expressed
ated exposing the underlying bedrock and the exposed by Chen et al. (2014), while true in some cases for the
bedrock base is incised. Thus bedrock lowering of the early versions of LEMS, have been, or are in the process
floodplain bottom is driven by the small proportion of time of being, addressed in recent years. The reader might also
that the bedrock is exposed by the cross-sectional changes like to look at the discussion about erosion in LEMs in
during flood events, and the small proportion of the area of Tucker and Hancock (2010), which complements the
the floodplain covered by the incising channel. These discussion below.
regions of bedrock incision will control the rate of lowering When modelling transport-, detachment- and
of the floodplain. If these locations are distributed down the weathering-limited transport mechanisms the early imple-
river valley, then they will also control the evolution of the mentations of the LEMS typically only allowed one of
valley long profile. These cross-sectional dynamics are the these mechanisms to be active at any specific time and at
domain of mobile bed channel models, where the channel any specific node, and if more than one process was
cross section dynamically responds to the time-varying modelled, there was a switch between them based on a
flow (Julien and Wargadalam, 1995; Julien, 2014). To date threshold of some kind. For instance, early versions of
no LEM models this process, so no current LEM can model SIBERIA (Willgoose et al., 1991a) had a switch that
floodplain decline. changed the erosion model applied on hillslope and chan-
nels based on whether a node was a channel of hillslope,
4.5.4 Alluvial Fans, Deltas and Downstream Fining or whether a node was transport-limited or detachment-
limited. Most modern models have abandoned that
Alluvial fans are primarily depositional structures, with approach because it can lead to mass balance problems
entrenched streams that avulse across the surface of the when the dominant process switches. Nowadays they
fan as the fan builds up. Most existing landform evolution track the amount of sediment being carried in the flow
and use a formal detachment process to model entrain- distributed rainfall-runoff model) can take 1–2 weeks with
ment, and compare the potential load at every node CAESAR (fully coupled with the LISFLOOD rainfall-
against the transport capacity (which may itself change runoff model) with little difference in the net erosion
based on the grading of the material being carried) to estimates (Hancock et al., 2010). But as we saw in the
determine the amount of erosion or deposition. This expli- discussion of meandering processes, the behaviour of the
cit entrainment mechanism is consistent with modern floodplain deposition is dependent upon correctly model-
agricultural erosion models. ling the probability distribution of the runoff events that
Some models track the grading of the sediment being created flow depths in the channel that spilled over onto
carried, but the tracking can introduce severe nonlineari- the floodplain, so to model floodplain dynamics, temporal
ties that significantly slow numerical solvers. For variability needs to be explicitly modelled.
instance, SIBERIA V8 (i.e. after 2001; Willgoose, It is not just floodplains that require a probability
2005b) optionally tracks the mean diameter (version distribution for discharge. Other possible applications
7 and earlier did not track sediment grading), CHILD include post-failure behaviour after breaching of flow or
(Tucker et al., 2001a) tracks the relative proportion of erosion control structures (e.g. contour banks, farm dams,
two particle size fractions (sand and gravel), while armoured constructed channels, drop structures, moon-
ARMOUR (Willgoose and Sharmeen, 2006), mARM scaping on mine sites). In many of these cases it is
(Cohen et al., 2009, 2010), SSSPAM (Welivitiya et al., extreme floods, ones that occur only every few years or
2016) and CAESAR (Coulthard et al., 2002) track the full even less frequently, that are critical. Typically once fail-
particle size distribution. All the models mentioned here ure occurs, the erosion protection they provide is removed
allow the fluvial sediment transport capacity to respond to (and in some cases, such as contour bank and moonscap-
changes in the grading of the sediment in the water since, ing failure, erosion may be worse than without the contour
as we saw in Section 4.2, there are relationships that bank being built in the first place), and the long-term
express how the sediment transport rate changes with erosion response of the landform may change markedly
sediment d 50 . Even if grading is tracked, there are com- post-failure (Willgoose and Gyasi-Agyei, 1995).
plexities because the grading of the material available for For the conceptual and calibrated models (e.g.
erosion (the source for the material tracked downstream) SIBERIA, CHILD) this random discharge is achieved by
then needs to be known, necessitating that either (1) the randomly varying in time the runoff rate in the discharge-
soil surface grading is specified a priori or (2) a soil area relationship. It is important to understand that this is
submodel is incorporated that estimates the surface different from fully coupled discharge modelling (using a
grading over time and that responds to erosion and depos- time-varying rainfall input) because the former do not
ition (e.g. the SEM models that will be discussed in explicitly model the runoff-routing process (even though
Chapters 5–11) for the calibrated models it will be ‘cooked into’ the
To average over time, two main approaches have been calibration parameters, specifically those relating dis-
adopted. Most models determine a relationship between charge to area). Some authors (e.g. Chen et al., 2014)
the geomorphologically effective discharge/transport and have misunderstood the use of geomorphologically effect-
catchment area, and use this relationship to determine the ive discharge and confuse it with the average discharge
geomorphologically effective discharge or geomorpholo- (i.e. the annual discharge divided by 365 days), but
gically effective sediment transport rate using the analysis Section 4.3 shows how the concept of geomorphologi-
for area draining through each node. Some explicitly cally effective discharge is related to extreme flow events
model the intra-event dynamics using a coupled rainfall- rather than the mean flow.
runoff model. The level of sophistication of this relation-
ship varies between models (1) purely conceptual (e.g. 4.7 Numerical Issues
DELIM, Howard, 1994), (2) calibrated to output from a
distributed rainfall-runoff model (e.g. SIBERIA, Many of the concepts discussed here, such as spatial and
Willgoose and Riley, 1998) and (3) fully coupled that temporal discretisation and numerical solvers, have
are coupled to a distributed rainfall-runoff model (e.g. significant implications for the validity of results from
CAESAR, Coulthard et al., 2007). Only the last kind LEMs. In some cases incorrect or poor choices may
can fully model the intra-event flood dynamics (e.g. flood invalidate the results of the modelling entirely, even if
wave routing down the channel), but it comes at a signifi- the underlying science is completely adequate. The
cant computational cost: runs that take 15 minutes in issues discussed briefly below are, in the author’s view,
SIBERIA (discharge-area relationship calibrated to a so important, and so widely misunderstood, that a
complete chapter (Chapter 15) is devoted to them. Issues with its response to changing elevations. Inherent in the
specific to erosion modelling, such as numerical stiff- method is a threshold where the drainage direction does
ness, are discussed below, and in Chapter 15 they will be not change immediately with incremental changes in
placed into the context of a larger coupled landscape adjacent elevations but will change suddenly when the
evolution model. adjacent node with the steepest downslope direction
changes. So for small changes in elevations the drainage
4.7.1 Spatial Discretisation, Drainage Analysis direction (and thus areas draining through nodes down-
Algorithms stream) is less responsive, and then when the threshold is
exceeded there is a sudden change in area. Since
The first of the numerical challenges is to determine the excavation of a valley has as a key component the con-
catchment, or discharge at every point of the domain, so vergence of flow into the valley, this means that because
the erosion equations can be applied. This is normally D8 is initially unresponsive to flow convergence, valley
done with a drainage direction search algorithm on a formation is slower than it should be. Sediment transport
spatial discretisation. capacity is proportional to Qm Am where m is 1.5–2, so
Spatial discretisation has been done in two ways: that when, for example, two nodes of equal area converge
into a single node downstream, the sediment transport
• A rectangular grid with uniform spacing in both direc- capacity at the downstream node is twice (for m ¼ 2)
tions. There may be an implicit assumption that the grid
the sediment draining into that node, and the downstream
is oriented in the E-W and N-S directions. Many codes
node has to erode the deficit, excavating a valley in the
have used square grids because of their simplicity.
process.
However, many field digital elevation data sets are
An alternative drainage direction algorithm is D∞
delivered on a degree, minute, second (DMS) grid
(Tarboton, 1997), which weights the drainage between
(e.g. the Shuttle Radar Topography Mission [SRTM]
the adjacent node with the steepest downslope direction
data). which means that a square grid in DMS units will
and the node next to it that has the next steepest direction
be rectangular in units of metres depending on the
(Figure 4.13). D∞ yields a more accurate definition of the
latitude of the site, so there is benefit in using rectangu-
drainage direction and areas, but for LEMs it has a more
lar grids. For instance, at 40 S or N a 1 second square
important property. The weighting of the proportion of
grid (the highest resolution SRTM data) is 30 m in the
area/discharge/sediment flux between the two nodes
N-S direction and 23 m in the E-W direction.
downstream immediately responds to changes in eleva-
• Irregularly spaced data that are discretised with triangles tions, so even small changes in elevations result in
into triangulated irregular networks (TIN). While more
changes in the areas downstream. Willgoose (2005a)
flexible than grids, significantly more information needs
compared D8 and D∞ on a constructed landform and
to be stored to do with how the triangles are connected
found that over 1,000 years the D∞ simulations evolved
(Tucker et al., 2001b).
Once the grid discretisation is determined, the next

decision is how to determine the area draining through
each node. The default, though not necessarily the best,
approach is to determine at every node the steepest down-
slope direction to an adjacent node and rout all flow from
the current node to this adjacent node. Once the directions
are known at every node, it is a simple matter of simply
summing up the areas contributing to the current node.
For gridded DEMs this is referred to as the D8 algorithm
because it deterministically routs all flow to one of the
eight adjacent nodes. There is no equivalent name for D8
on TINs, but this approach of routing all flow in the
steepest downslope direction is used in the CHILD
LEM, where all the flow is routed along the boundary
between the triangles that the current node is a vertex of.
The main problem with D8 (O’Callaghan and Mark,
1984) (and presumably the equivalent on TINs) is to do FIGURE 4.13: The weighting scheme for D∞.
about 15% faster than the D8 simulation. There were a 4.7.2 Timestepping, Landform Time Scales and
large number of valleys generated on the landform, and Numerical Stiffness
their excavation dominated the sediment mass balance
over the 1,000 years of the simulation. The only apparent The erosion models we have discussed here have some
explanation was the threshold in the D8 that suppresses characteristics that make them expensive to solve in time,
flow convergence and valley formation. A side benefit of and difficult to achieve an accurate mass conservation.
D∞ was that the valleys curved more smoothly in plan The dependence of sediment transport on area (through
rather than looking blocky and gridded. the discharge dependence in Equations (4.10) and (4.29))
It is likely that the TIN equivalent of D8 and D∞ will means that the rate of change of elevation (which is
also show this latter blocky behaviour, though the random related to the response time) for large areas near the
location of the grid points likely obscures the ‘gridded’ catchment outlet is fast and for small areas on the tops
flow patterns. No such comparison (i.e. between TIN of hillslopes is slow. This range of response times means
versions of D8 and D∞) has been done with TINs that small timesteps are required to solve the equation for
(Tucker, personal communication), but it seems reason- elevation change near the catchment outlet while large
able to suspect that a similar threshold convergence mech- timesteps can be used for hilltops, while the amount of
anism may occur for TINs as well. time to run a landform to equilibrium is determined by the
The area at every node needs to be recalculated at slowest part of the landform, the hilltops. Numerical
every timestep, and a LEM simulation may involve a problems with large differences in response times are
million or more timesteps so efficient calculation of area grouped under the heading of ‘stiff problems’. The prob-
is required. During the original construction of SIBERIA lem can be quantified by examining a simple problem
(i.e. Willgoose, 1989) a large number of algorithms for with transport-limited erosion and tectonic uplift in one
calculating area were tested and the fastest one for D8 was dimension (Equations (2.1), (3.12) and (4.10)):
1. Determine the D8 directions ∂z ∂q ∂ ∂ m mδ n

¼ U s ¼ U ðKQm Sn Þ ¼ U Kβ A S
2. Search for all sources of drainage lines in the grid. ∂t ∂x ∂x ∂x
∂ mδ n
These are the nodes that don’t have any nodes draining ¼ U Kβm A S (4.70)
∂x
into them.
3. Starting from each source in turn, step down the drain- If we nondimensionalise this equation using
age lines summing the area of all nodes draining into
z ¼ zz0
each node (i.e. including any tributaries) as you follow
the drainage directions downstream until you come a A ¼ AA0 (4.71)
node where one or more of the tributaries have not yet t ¼ tt 0
had areas calculated for them. At this point stop and
where z is a vertical length scale characterising the eleva-
repeat step 3 for the next source in the list.
tion of the problem (e.g. mean elevation, relief), A is a
4. When you do step 3 for the last source, you are
horizontal length scale characterising the area of the prob-
guaranteed to have calculated the area for every node.
lem (e.g. catchment area) and t is some time scale of the
Our experience is that this algorithm is not just fast, problem (e.g. rate of change of elevation), then we can
but that it scales well to large domains (we have used it for rewrite Equation (4.70) as
up to 3,000,000 node domains). Braun and Sambridge h i∂
∂z0 t m ðmδÞ ðn1Þ
A0 Sn
mδ
(1997) independently arrived at a very similar algorithm ¼ U Kβ A z t (4.72)
∂t 0 z ∂x
for their TIN-based DEM and called it CASCADE.
For the calculation of areas with D∞ there seems to be where the term inside the square brackets parameterises
no alternative to using the recursive algorithm of Tarboton the scaling of the rate of change of elevation for different
(1997). Informal benchmarking using SIBERIA indicates area and elevation from erosion. For typical values of
that the area calculation with D∞ is about four times m 2 and mδ 2 then as the area of the catchment
slower than with the D8 algorithm above. More critical increases (and with slope constant) the rate of change of
is that it is not possible to parallelise the recursive solution elevation on the left-hand side increases with the square of
on modern multicore processors, whereas the D8 algo- area. This is shown graphically in Figure 4.14a where a
rithm can be. Calculating areas using D∞ is the only node with large area changes elevation relatively faster
component of SIBERIA that has not been able to be than one with a smaller area. The implication is that if an
parallelised in some form. Euler timestepping algorithm is used for large timesteps,
FIGURE 4.14: The schematic of

the approximate analytic solution
for the time evolution of
elevations for fluvial erosion (from
Willgoose, 1989).
oscillations will be created in the solution because the introduces mass conservation problems (since pit-filling
projection forward in time will overshoot the correct fills up the pits with sediment that is not accounted for in
answer for nodes with large areas. the mass conservation equations).
The significance of this overshoot is for the analysis of The common, brute force, solution to these over-
the drainage pattern at the next timestep. All current shoots and oscillations is to use smaller timesteps, but
LEMS use drainage direction algorithms (D8, D∞, as since the power on area is about 2, this means that even a
discussed in the previous section), and the overshoot modest-sized problem (e.g. 100 100 nodes, with a
creates a localised pit in the elevations, which breaks the single catchment outlet so the maximum catchment area
drainage line so that flow and sediment transport is inter- is 104 nodes) will have a range of time scales (and thus
rupted. This means that areas draining through the nodes rates of change in elevations) of 108, and this range
downstream of the pit drop dramatically, so that the sedi- increases for bigger problems. This scaling problem is
ment transport balance in these downstream nodes is not as severe in the case of stream power detachment-
changed dramatically. An inevitable consequence is to limited Equation (4.29) because the power on the area in
trigger elevation instabilities downstream, and cata- this case is about 1–1.5, so in the example the time scale
strophic failure of the elevation solver follows very range is only 104–106.
quickly. Some users have suggested pit-filling these pits, Willgoose (1989) proposed a solution using an
but this only wallpapers over the underlying problem and approximation to the analytical solution of elevation with
time (Figure 4.14b). He noted that for a one-dimensional but like the SIBERIA method it assumes that the drain-
problem, if we assumed that the elevations immediately age network does not change during the timestep, ultim-
upstream and downstream were fixed, then with time the ately limiting the maximum timestep that can be used
elevations will converge to an equilibrium solution where just as it does for the SIBERIA method. Comparing the
the elevation is not changing. For a one-dimensional implicit method and a simple Euler solution, they found
problem this equilibrium solution can be derived analytic- a factor of 100 times speed-up with less than a 5% mass
ally by equating the sediment inflow with the sediment balance error.
outflow so that In Chapter 9 we will talk about diffusive transport
processes (processes that don’t have an area dependence,
Qm ðzi1 z∗ i Þ ¼ ðQ þ ΔQÞm ðz∗ i ziþ1 Þ
n n
(4.73) only slope), but it is convenient to jump ahead a bit at this
stage. Both Willgoose’s and Fagherazzi’s methods are
where the asterisk superscript indicates the equilibrium
applicable to diffusive transport as well as the fluvial
value and ΔQ is the extra increment of discharge contrib-
erosion, but without the area dependency the stiffness
uted by node i, and
! problems disappear, and many of the complexities and
Q =n zi1 þ ðQ þ ΔQÞ =n ziþ1
m m
∗
approximations of these methods become unnecessary.
z i¼ (4.74).
Q =n þ ðQ þ ΔQÞ =n Perron (2011) proposed an implicit solver for nonlinear
m m
diffusion (Equation 9.20). Perron used a Taylor series

The approximate solution that is then used has a slope at expansion of the nonlinear diffusion equation and then
the start of the timestep that is given by Equation (4.70) formulated an implicit solver based on this linearization.
and that converges to Equation (4.74) over time. In two The nonlinear diffusion problem does not have an area
dimensions the approximate analytic solution requires the dependence, so there is no need for drainage directions to
use of Taylor series to solve the equivalent equation to be defined to determine area, and accordingly Perron’s
Equation (4.73) where there are multiple inflows of sedi- solution does not assume a fixed drainage network during
ment from upstream nodes, but the principle remains. the timestep. Perron does solve an example involving
Experience in using this solution technique in SIBERIA, fluvial erosion, but at each node he assumes that either
and incremental improvements since its original deriv- diffusion or fluvial erosion is active but not both (the
ation in Willgoose (1989), indicates that the criterion that switching criticised by Chen et al. (2014) discussed in
limits the size of timesteps is no longer stability and Section 4.6) and then solves the fluvial erosion equations
potential oscillations, but acceptable mass balance with a simple Euler explicit solver using operator split-
(Section 15.3.2). The mass balance errors arise because ting (see Section 2.2.2). Thus Perron’s solution is not
the equilibrium solution in Equation (4.74) assumes that designed to address the stiffness problems of the fluvial
(1) the elevations of the upstream and downstream nodes erosion process.
are not changing with time and (2) the drainage directions
do not change during the timestep. This is typical of Euler 4.7.3 Catchment Divide Migration
numerical solvers, but with increasing timestep size the
approximate analytical solution is an increasingly poor LEMs approximate the location of the catchment divide.
approximation. However, the mass balance error does In D8 for each node the area of the node is assumed to
not scale so severely as the Euler solver with the domain flow is a single downslope direction, and for D∞ two
size making very large grids and higher spatial resolutions downslope directions. The drainage divide is then along
more feasible. the boundary of the cells halfway between nodes. In
An alternative solution to the stiffness problem is to reality the drainage divide will not be exactly halfway
use implicit solvers. Implicit solvers are more stable between the nodes. Stream capture occurs when a node
allowing larger timesteps. Fagherazzi et al. (2002) and switches its drainage directions from one subcatchment to
Perron (2011) outline implicit solvers for the transport- another. For D8 all the area of a node flows into one or
limited sediment transport equations. the other subcatchment. More realistically as the divide
Fagherazzi et al. (2002) used Taylor series to linear- moves there will be gradual change over time in the
ise Equation (4.70) in slope S and then solved for the apportioning of the nodal area from one subcatchment
future elevation in the classical way that implicit solvers to another.
are formulated including the need to invert an N N Goren et al. (2014) proposed a method for modelling
matrix where N is the number of nodes. The method is the gradual change in the location of the drainage divide.
unconditionally stable (i.e. there will be no oscillations), It modelled the evolution of the rivers and then applied an
analytical solution for the long profile of the hillslope where d is the average diameter of the sediment, q is the
between the rivers, giving a precise location for the divide discharge and c is the sediment concentration. The trans-
based on the hillslope profiles. As the river elevations and port capacity of the flow in SIBERIA changes with the
locations changed, so did the hillslope profile and the mean diameter of the sediment, as given in Equation
location of the catchment divide. Goren and his coworkers (4.12). For the eroding surface the average grading of
tested it against a traditional LEM. They found that it the surface sediment is a model input, and it is assumed
modelled the catchment divide better at coarse resolution that the surface grading is not changed by selective
(8 km), and that the results of the two models were similar entrainment. Likewise for deposition, selective deposition
at higher resolutions (1 km, with a hillslope length in both of coarse particles does not occur, so both the grading in
cases of 500 m). The method was not tested at the reso- the flow and the deposited sediment is unchanged by the
lutions normally used in LEMS that resolve the hillslopes deposition process.
(10–100 m, or 0.01–0.2 times the hillslope length). Without the ability to model the selective entrainment
of fines, this approach cannot armour the eroding surface.
This results in erosion depths (both sheet and gully ero-
4.7.4 Tracking Sediment Grading
sion) that are too high. To address this, SIBERIA
It should be clear that to simulate the sediment selectivity implements a conceptual armouring model. The erodibil-
of armouring and deposition it is necessary to track the ity is modelled to decrease with increasing cumulative
entire grading distribution of the sediment. For erosion, depth of erosion. The rationale for this is that erosion
without the full grading distribution it is not possible to initially removes the finest particle size fraction from the
model the entrainment of the finest fractions and the surface layer. When 100% of this fraction is removed,
subsequent reduction in mobility of remaining sediment then the next coarser fraction is removed in turn. As this
on the armouring surface to erosion. For deposition to process proceeds, the eroding surface is gradually
model the faster deposition of coarse fractions (and the coarsened (the mean diameter is increasing), and this
spatial patterns of deposited sediment on alluvial fans), coarsening is determined by the cumulative amount of
and the subsequent relative increase in the mobility of the erosion. This relationship between erodibility and cumu-
sediments remaining in the flow, the deposition of the lative erosion is easily determined from the full grading
coarsest fractions need to be tracked. distribution of the surface material, and the rate of
However, tracking the full grading introduces nonli- ‘armouring’ is dependent only on one parameter, the
nearities in the sediment transport equation that, for a full depth of the surface armour layer assumed. A deeper
physically based treatment, can be computationally over- armour has a greater mass of fine material, and so requires
whelming. For instance, the author’s ARMOUR model more cumulative erosion to reach a given mean diameter
(Willgoose and Sharmeen, 2006) tracks the entire grading armour than a thinner layer. A thinner armour means the
distribution and simulates selective entrainment and surface armours faster.
deposition. ARMOUR runs many orders of magnitude
slower than SIBERIA, which doesn’t track the full
grading distribution. In Chapter 7 we will discuss one 4.8 Further Reading
approximate solution to this problem that allows us to
The best up-to-date discussion of rainsplash mechanics
track the full grading distribution. SIBERIA uses a differ-
and field experimental data is Dunne et al. (2010). South-
ent, more limited, approach.
ard (2006) provides a beautifully intuitive discussion of
The author’s SIBERIA model implements a simple,
sediment transport fundamentals. For those who are inter-
physically based, approximate algorithm to track the
ested in aeolian transport processes, a good place to start
changes in the transport capacity as a result of changes
is the special issue on ‘Aeolian Research: processes,
in the average grading of the sediment. SIBERIA tracks
instrumentation, landforms and palaeoenvironments’,
the mean of the grading so that when two flow paths (let’s
Geomorphology, 59 (1–4), 2004. For agricultural erosion,
say (1) and (2)) merge, then the grading of the sediment
one consistent, coherent reference does not exist, but the
after the merger (3) is
WEPP model documentation is probably the most com-
q1 c1 d 1 þ q1 c1 d1 prehensive up-to-date reference (online, but see Laflen
d3 ¼ (4.75)
q1 c1 þ q1 c1 et al., 1991 for an overview).
5 Soils: Constructing a Soilscape Evolution
Model – Basic Concepts
5.1 Overview dioxide concentrations in the atmosphere ranged from a

very slight increase of about 50 ppm above current day
From the previous chapter on erosion it can be seen (and levels to a more than tripling to more than 1,000 ppm.
will also be seen in later chapters) that information about They concluded that the main reason for this variability
soils is required to fully parameterise many processes. was differences in each model’s projection of soil mois-
Without some soil dynamics model that interacts with ture and knock-on impacts (particularly in semiarid cli-
the landforms, the user of a landform evolution model mates) on vegetation growth and soil microbiology.
needs to specify the soil properties. Likewise, Peng et al. (2014) found that changes in soil
But there are many other reasons above and beyond moisture, after changes in mean temperature, were the
landform evolution to be interested in soils and their most significant factor in determining global soil carbon
dynamics. Soils cover 80% of the earth’s terrestrial sur- storage. Results from soil moisture research generally
face (the remaining 20% is predominantly covered by ice, conclude that the parameterisation of the soil functional
permafrost and alps) (FAO, 2009), and by understanding properties (e.g. hydraulic conductivity) is the single big-
the temporal dynamics and spatial coupling of soils we gest uncertainty after climate in estimating soil moisture
can develop a quantitative understanding of (1) the spatial (e.g. Chen et al., 2015). The conclusion is that improving
distribution of soils, (2) the links between soil properties, our predictions of atmospheric carbon at the end of the
the underlying geology and climate and (3) their resilience century is crucially dependent on improvements in the
to environmental change and human impacts. representation of soils in coupled global biogeochemical
A compelling example of this need for deeper under- and climate models. This recognition is one of the motiv-
standing of soil dynamics is recent work examining the ations for the digital soil mapping (DSM) community to
uncertainties in the models of the global carbon cycle. Our develop global soil maps focussing on functional soil
current best estimate of sequestration of anthropogenic properties (e.g. the GlobalSoilMap initiative). To date
carbon emissions is that 55% of the extra carbon dioxide DSM has been based on interpolation of field data, and
emitted into the atmosphere that is sequestered, is seques- a process-based explanation for the spatial distribution of
tered on the land (Stocker et al., 2013) as enhanced soil soils would be a useful tool to complement existing DSM
carbon, microorganisms and plant biomass. This terres- statistical approaches. Moreover, DSM assumes that the
trial rate of sequestration per square kilometre is more soil properties do not change, and a dynamic soilscape
than three times that of the ocean. Yet our understanding model may provide deeper insight into the impacts of
of the rates of sequestration by these terrestrial processes climate change of the soil properties themselves.
is poor, and we do not have a clear understanding of their One of the challenges of modelling soil evolution is
sensitivities to climate change. It is conceivable that this that soil has many facets that are important (e.g. soil
55% terrestrial sequestration rate may change because of physical properties such as particle size distribution, soil
the impacts of climate variability and climate change, with functional properties such hydraulic conductivity and soil
consequent knock-on effects for atmosphere carbon diox- chemical properties such as pH). For a comprehensive
ide levels and climate change. For example, Friedlingstein review of soil processes and the challenges in modelling
et al. (2006) examined 11 biogeochemical models used to them see Vereecken et al. (2016). Vereecken is a consoli-
model the global carbon cycle and models for global dated overview as perceived by the soil science commu-
climate projections out to 2100. The projected carbon nity, and other fields could add even more aspects that 87
88 Soils: Constructing a Soilscape Evolution Model – Basic Concepts
need to be modelled (e.g. mine waste rehabilitation (e.g. Anderson et al., 2002), this bulk density decrease is
requires pedogenesis models, which has only passing accompanied by a volume increase when saprolite is
mention in Vereecken). In many cases we need to know transformed into soil.
both their distribution in space (e.g. downslope and across The typical mathematical approach for modelling
hillslopes) and with depth down the profile, as well as environmental systems is based on mass balances. How-
how they may change with time. The following soil ever, there has also been a history of research into energy
process chapters will discuss these properties in order of and entropy methods, but they are not as enlightening
increasing complexity. The order in which they will be about soil processes as mass balance methods. Minasny
discussed is et al. (2008) performed an approximate energy balance for
soil formation and estimated that the energy required for
• Soil depth mineral chemical weathering was 0.01% of the total
• Soil physics properties such as particle size distribution, energy for soil formation, which makes it very hard to
clay content, porosity and bulk density
measure in the field, and very hard to formulate defensible
• Soil physics functional properties such as hydraulic mathematical equations using energy methods.
conductivity, porosity, shear strength, rheology and
water-holding capacity
• Soil chemistry and its impacts on weathering processes
• Soil carbon and soil organic matter and
5.2 The Interaction between the Soil
• Soil transport downslope.
Profile and Weathering
The initial chapters will discuss modelling of residual Weathering of rock into its residual soil profile is a com-
soil development, where there is no soil movement across plicated process involving physical (e.g. fracturing by
or downslope. Initially we focus on processes in the soil imposed stresses), chemical (e.g. transformation of min-
profile that operate vertically. We will then move on to erals into clays) and biological processes (e.g. modifying
processes that move soil downslope – erosion at the soil the soil atmosphere by respiration). Any modelling of
surface and movement of the soil mass downslope by weathering is bound to be a simplification of these com-
creep or flow. plex and interacting processes. The chapters that follow
Finally, there is the question of what is soil. Every- will look at these three components in turn.
thing above solid bedrock is ‘regolith’, and if there is We also simplify the discussion by treating physical
biological material as part of the regolith, it is ‘soil’ (Ollier and chemical weathering separately though there are no
and Pain 1996; Scott and Pain, 2009). Regolith is also doubt strong interactions. The justification for this separ-
sometimes referred to as saprolite, and underneath sapro- ation is that physical weathering is the first process to
lite is lightly weathered fractured rock called saprock. The operate on particles because large particles have a low
models described in the following chapters for soil depth specific area (i.e. surface area per unit mass of rock), so
and mineral matter do not explicitly model the biological chemical processes are relatively slow. As the rock breaks
component and so, strictly speaking, should be called down from the action of physical processes, the specific
regolith models rather than soils models, but this distinc- area rises and at some time (depending on the relative
tion will not be stressed. intensities of physical and chemical processes) the spe-
Anderson and Anderson (2010) provide another defin- cific area will be high enough that chemical processes will
ition. Saprolite consists of the rock in its original unweath- dominate the physical processes. Thus initially physical
ered location, shape and organisation. Soil is that same weathering dominates (though chemical weathering is still
material but no longer spatially organised as it was in the occurring at relatively low rates), and then at some stage
underlying saprolite. Using an analogy of children’s chemical weathering dominates (though physical
building blocks, the saprolite has all the blocks nicely weathering is still occurring at a relatively low rate).
packed away with the blocks tightly packed together and To further simplify the discussion that follows, the rate
organised with no space or gaps in between (i.e. porosity of weathering has been disaggregated from the products
is low), while soil is as if those blocks were randomly of weathering. This allows us to simplify the task of
tossed into the box so that there are random gaps and modelling weathering, and this separation follows the
spaces between the blocks (i.e. porosity is high). The approaches and emphases of the literature. We focus on
implication of this is that there is a step increase in the soil grading because many pedotransfer functions of soil
porosity (and without chemical alteration) from saprolite functional properties that are used in environmental
to soil. If the chemical dissolution loss of the rock is small models (e.g. hydraulic conductivity, soil water-holding
5.2 The Interaction between the Soil Profile and Weathering 89
capacity) are based, at least in part, on the particle size The chapter on ‘chemical weathering’ is about the
distribution of the soil. This is not to downgrade the chemical transformation of the rock fragments in the soil
importance of other soil properties (e.g. organic matter, into clays and other secondary minerals. While this is
cation exchange capacity), only that it provides a starting fundamentally a chemical process, the processes
point for modelling. involved may have both chemical (e.g. carbonic acid
With respect to the evolution of the particle size distri- infiltration as a result of rainfall equilibrating with carbon
bution for physical weathering, there is a small amount of dioxide in the atmosphere) and biological components
quantitative data on the rate of weathering but very little on (e.g. carbon dioxide and anoxic soil atmosphere
the products of weathering. However, the recent work of generation within the soil as a result of biodegradation).
Wells et al. (2005, 2006, 2007, 2008) has clarified, for one There may also be physical components if the chemical
rock type, the grading characteristics of the weathering process triggers fragmentation (e.g. many hydration
products. Again, for chemical weathering there is consid- products increase in volume), thus generating more
erable data on the rates of weathering (though there is (chemically) untransformed surfaces available for further
debate about the consistency of the data from laboratory chemical transformation.
and field experiments) but only qualitative data on the In the soils literature the use of ‘physical’ or ‘chem-
grading of the products of the weathering processes. Both ical’ descriptors is normally used to distinguish the pro-
bodies of work provide pathways for weathering but little cesses driving the weathering. Here we are focussed on
information on the grading characteristics of the resulting the results of the processes (and the modelling techniques
particulate matter and its chemical reactivity. to describe these transformations), and it is more con-
Finally I will deviate slightly from traditional termin- venient to distinguish them by how we model the
ology for classifying weathering processes. dynamics of the physical and chemical products of these
The chapter on ‘physical weathering’ is about how the processes.
geologic material in saprolite is physically transformed Because this book is about techniques for process
into finer material, primarily by fragmentation processes. modelling, many soil processes which we know operate
In this physical transformation process we model only the (e.g. soil microbiology) but have not been described
transformation of the physical properties of the soil, for quantitatively (or in some cases even qualitatively
example, particle size distribution, porosity and bulk described) have been dealt with here in a cursory fashion.
density. The mineralogy is unchanged. It is important to Accordingly, the reader may find it useful to read the
note that the processes causing this fragmentation may be soilscape chapters that follow in conjunction with a more
physical (e.g. freeze/thaw), chemical (e.g. volume change holistic description of soil processes such as Paton et al.
as a result of hydration of minerals) or biological (e.g. (1995), or Schaetzl and Thompson (2015) to gain a com-
fungal hyphae physically wedging cracks open). plete view of soil evolution.
6 Soils: Soil Depth
6.1 Introduction this chapter typically assumes that a distinct interface

between saprolite and soil exists.
Soil depth is a fundamental property of soils. Sensitivity
tests have shown that uncertainty in our knowledge of
soil depth is a significant contributor to uncertainty in 6.2 Soil Depth and Bedrock Conversion
estimating carbon and water fluxes between the terres- to Soil
trial environment and the atmosphere (Knorr and
Lakshmi, 2001; Peterman et al., 2014), and thus the soil Ahnert (1976) first described a model for the evolution of
moisture and the biosequestration potential of the terres- soil depth:
trial environment.
dD ρs
Yet even something as apparently simple as soil depth ¼ ðP E Þ (6.1)
dt ρb
can be difficult to define objectively, as the definitions
discussed in the previous chapter for regolith, soils and where D is the depth of the soil, P is the rate of
saprolite indicate. The models described below define soil conversion of bedrock to soil (depth/time), E is the
as all materials above the unweathered bedrock or sapro- erosion rate (depth/time) and ρs and ρb are the bulk
lite, so soil depth is that depth to the bedrock or saprolite densities of the soil averaged over depth and bedrock
interface. In the field such a distinct boundary may or may density at the soil-bedrock interface, respectively. Some
not exist. For instance, the regolith may contain signifi- early equations ignore changes in porosity and bulk
cant corestones surrounded by soil, and the corestones density that occur in the conversion from bedrock to
may become bigger and more frequent with depth until soil. Equation (6.1) includes this bulk density change.
there is a transition to solid bedrock (e.g. Fletcher et al., Ahnert did not consider bulk density changes resulting
2006; Graham et al., 2010). Measurements of changes in from soil production but they are included above for
porosity and bulk density down through the soil profile in consistency with Chapter 2. The general form of Equa-
the field show mixed results even when allowing for the tion (6.1) (either with or without the density change) is
impacts of agricultural landuse. Some profiles show no the basis of all the soil depth modelling that has
significant change down the soil profile suggesting that all followed on from Ahnert.
change occurs at the base of the profile (e.g. Angers et al., This equation ignores the effect of dissolution because,
1997; Corti et al., 2001), while others show a trend of as we see when we discuss bulk density evolution
increasing bulk density with depth suggesting that (Section 7.6), there is the possibility of complex inter-
changes may occur gradually throughout the profile (e.g. actions between dissolution and bulk density changes.
Unger and Jones, 1998; Graham et al., 2010). Ouimet However, in a model of chemical dissolution in soil Mudd
(2008) noted a sharp boundary in bulk density within his and Furbish (2004) and Yoo and Mudd (2008) assumed
profiles and attributed this to tree throw physically mixing that bioturbation would mix the soil from top to bottom
(i.e. bioturbation) the soil. Finally, Richter and Markewitz and ensure that bulk density remained the same through-
(1995) report a gradual transition in the soil biogeochem- out the profile (Brimhall et al., 1992). Brimhall et al.
istry with depth and argue that physical properties under- (1992) found a complex pattern of dilation (e.g. increase
estimate the depth of the biochemically active zone. in porosity) and collapse (decrease in porosity) based on
90 Despite these conflicting data the literature discussed in depth, age and mineralogy, and attributed many of the
changes to bioturbation and related processes, which will

be discussed in Chapter 7.
The function P is (interchangeably) called the bedrock
conversion function, bedrock weathering function, bed-
rock erosion function or soil production function. In
recent years the soil production function (SPF) has been
the most commonly used description, and this is the
terminology used in this book. If, as is commonly
accepted, the SPF declines with the depth of the soil-
bedrock interface below the surface, then an equilibrium
depth can be attained. A particularly simple solution arises
when the ‘exponential’ SPF is used:
P ¼ P0 eλz (6.2)
where P0 is the soil production rate when there is zero soil
coverage, λ is the rate at which the conversion rate
decreases with increasing soil depth and z is the depth
below the soil surface. For equilibrium when the left-hand
side of Equation (6.1) equals zero, FIGURE 6.1: Example exponential and humped soil production
functions (SPFs). Two humped functions are shown: (a) p ¼ 0 is zero
P0 eλD ¼ E (6.3) at the soil surface, (b) p ¼ 0:2 is nonzero at the soil surface; all other
parameters are the same. The depth decay parameter is the same for
where D is the depth below the surface of the bedrock
all three functions (i.e. λ ¼ δ2 ), so, in this example, the exponential
interface (i.e. the soil depth), or rearranging and humped functions converge to the same value for deeper soils.

P0
D ¼ λ ln (6.4)
E
site (Anderson et al., 2002). Recent field research has
This equation shows that if the SPF and erosion rates clarified some causal factors for variations from site to
are the same everywhere, then the soil depth will also be site (Navarre-Sitchler et al., 2013) with
the same everywhere. Also if the SPF rate increases or the
erosion rate decreases, then the soil depth will increase. • the underlying geology and soil moisture being high-
lighted as dominant and
Since it is rare for soils to be constant depth everywhere,
one of the conversion or erosion rates must vary in space • less clear-cut conclusions on the role of temperature,
with some authors showing a strong impact (e.g. Eppes
if Equation (6.4) is to be true. Equation (6.4) also says that
and Griffing, 2010) while others show a relatively
everything else being equal, then higher erosion rates
minor impact for sites below the snowline (e.g. Ras-
should lead to shallower soils, which has been observed
mussen et al., 2010).
in the field (Cox, 1980; Dietrich et al., 1995; Heimsath
et al., 1999). There are other postulated SPFs. One criticism of the
The conceptual appeal of the exponential form of the exponential function is that it has the highest weathering
SPF is that the conversion rate decreases with the depth of rate when there is no soil. If weathering is also driven by
the bedrock interface below the soil surface. If soil pro- the time that the rock is wet (e.g. by chemical reactions),
duction is driven by temperature (e.g. driving cycles of then the rate should be highest just below the surface
internal thermal stresses in rock particles) and soil wetness when there is sufficient soil to store some of the infiltrated
variations (e.g. driving cycles of salt crystallisation), both water (Carson and Kirkby, 1972). To model this, the
of which become less extreme with depth in the soil “humped” function was developed (Figure 6.1).
profile, this function is qualitatively consistent with A variety of mathematical forms have been presented
observed behaviour. Recent research using cosmogenic (e.g. Ahnert, 1977; Anderson, 2002), all of which are
nuclides for the dating of soil profiles has provided strong empirical. While Heimsath et al. (2009) and Stockmann
empirical evidence that, to first order, the exponential et al. (2014) provided experimental evidence which sup-
decline with depth in Equation (6.2) is well founded ports the exponential function, their data are also consist-
(Heimsath et al., 1997, 1999). There do, however, appear ent with humped behaviour because there is significant
to be significant variations in both P0 and λ from site to scatter in their data for shallow soil depths.
92 Soils: Soil Depth
The key differences between the mathematical formu- and showed that the spatial pattern of the bedrock topog-
lations of the humped SPF are whether (1) the function raphy and soil depth appeared to be unconnected to the
goes to zero or not, for zero soil depth (i.e. whether a bare surface topography. The bedrock surface had patterns of
rock surface can weather at all), (2) it asymptotes to zero drainage with valleys and holes in the bedrock surface
for large depths (i.e. weathering declines to zero at depth suggestive of concentrated water flow over the bedrock
like the exponential function) and (3) the function is surface (James et al., 2010). This disconnect between the
continuously differentiable over its range (differentiability surface topography and the bedrock topography, but
can be useful for numerical solvers). Many of the pub- where the patterns were clearly not random nor linked to
lished equations do not meet the third point because they geology, suggested feedbacks in the soil thickness dynam-
provide different equations for above and below the depth ics that cannot be modelled by the simple models above.
of maximum weathering rate. One equation that satisfies Saco et al. (2006) investigated whether the bedrock
all three requirements is a modification of that used by patterns observed at Panola are a result of soil water where
Minasny and McBratney (2006) and Cohen et al. (2010): the lateral soil water drainage patterns are determined by
the bedrock topography, not the soil surface topography.
P ¼ P0 1 eδ1 zPa eδ2 z (6.5)
They extended the exponential SPF by making it a func-
where δ1 , δ2 and Pa are all parameters with positive tion of soil moisture so that the SPF became
values, where δ1 > δ2 , and Pa controls the weathering
P ¼ f ðM ÞP0 eλz (6.7)
rate for zero soil depth. The depth D∗ at which maximum
weathering rate occurs is where M is the soil moisture as determined by using the
wetness index
1 δ1 þ δ2
D∗ ¼ ln Pa (6.6)
δ1 δ2 A
M¼K (6.8)
S
Furbish and Fagherazzi (2001) present another equa-
tion, but it has the disadvantage that it cannot control the but where the contributing area A and slope S used in the
rate of decline with depth (i.e. the δ2 in Equation (6.5)) wetness index formulation were derived from analysing
independently of the weathering rate at the surface (i.e. Pa the bedrock topography rather than, as is traditionally
in Equation (6.5)), so we will not discuss it. Finally some done, the soil surface topography. Depending on the func-
studies suggest that soil production rates are independent tional relationship between soil moisture and weathering
of depth (Wilkinson et al., 2003), though they are in the used (i.e. the functional form of f ðM Þ, e.g. using absolute
minority and the reasons for their observed behaviour is depth of soil water versus percentage saturation of the soil
not known. profile) spatial patterns in the bedrock topography and soil
All these equations are empirical and based on field depth qualitatively similar to that observed by Freer could
experiment data including cosmogenic nuclide dating be generated (Figure 16.1).
(CN), optically stimulated luminescence (OSL) dating There are ecological feedbacks that I will touch on
and geochemical modelling of rivers and soil profiles briefly here but will return to later in the book. In arid
(Minasny et al., 2015). One consideration with the dating zones where the vegetation spatial distribution is banded,
methods is they rely upon assumptions about rates of it has been observed (Ludwig et al., 2005) that infiltration
exposure to the atmosphere. Bioturbation potentially was higher (and therefore so was soil moisture) within the
mixes the soil, bringing deeper material to the surface bands of vegetation (enhanced organic matter changed the
and burying surface material. If the soil is fully mixed soil structure and increased the infiltration capacity) than
from top to bottom, then the cosmogenic dating methods between the vegetation bands. Underneath the bands of
will yield the average age of the soil, whereas OSL will vegetation the soil depth was greater than in the interband
yield the time since the particle(s) were last at the surface areas where there was lower soil moisture and organic
(Dunai, 2010). matter. The model formulations of Saco et al. (2006,
One of the limitations of the SPF above is that they 2007) are consistent with this behaviour.
lack any explicit feedbacks with soil moisture and tem- Gabet and Mudd (2010) have proposed a tree throw
perature. The work of Freer et al. (2002) provides empir- mechanism that results in a distinct soil-saprolite
ical evidence of the importance of spatial feedbacks that boundary, a distinct change in bulk density at this bound-
are not included in the above formulations. They mapped ary and which naturally leads to a humped SPF. In
the bedrock topography and soil depth at the Panola site essence tree roots rip out rock from the saprolite during
FIGURE 6.2: As soil thickness

increases, the amount of bedrock
disrupted by tree roots decreases
because of their limited vertical
Thickness of bedrock extent. Conversely, thicker soils
disrupted by tree throw support higher tree densities.
Tree density The combination of these two
Bedrock erosion rate competing trends is posited to
produce a humped soil production
function (Figure 2 from Gabet and
Mudd, 2010).
0
0 Soil thickness
0.4 FIGURE 6.3: Model results

Model results compared to soil production rates
measured in a Douglas fir forest in
Oregon Coast Range
Soil production (mm/y)
0.3 the Oregon Coast Range. The

(Heimsath et al., 2001)
results of Gabet and Mudd, (2010)
(their Figure 6) bracket the shaded
area. Both the data and the model
0.2
results reveal a humped
relationship between bedrock
erosion rates and soil thickness
0.1 (Figure 7 from Gabet and Mudd,
2010).
0.0
0 20 40 60 80 100 120 140
Soil thickness (cm)
tree throw. This causes the distinct boundary and bulk soil, if at the equilibrium in Equation (6.4), will return to
density change at the soil-saprolite interface. The humped this equilibrium if the soil depth is perturbed slightly (this
SPF results from an interaction between an exponential return to the previous solution is the mathematical defin-
decline in root density in depth, and an increasing density ition of stability, with respect to soil depth). The humped
of trees as soil depth increases (Figure 6.2). The model has function, however, is not unconditionally stable every-
been tested in a heavily forested wet climate (Figure 6.3). where (Figure 6.4). For soil depth greater than the hump
Finke et al. (2013) modelled tree throw as a hypothesis for peak D∗ (Equation (6.6)) the soil depth solution is stable.
explaining the apparent random variability in soil depth and However, for soil depths between zero and the hump
lack of any spatial organisation in soil cores at an aeolian site peak, the equilibrium solution for soil depth is unstable.
in Belgium (Vanwalleghem et al., 2010). A model with tree In this region if the soil depth is slightly increased, then
throw was an improvement over a model without tree throw, the SPF increases and the soil depth increases further.
though the improvement was relatively small. This continues until the soil depth reaches the hump peak,
One final observation on the SPFs is on the stability of and the soil depth will evolve past the peak (because at the
the equilibrium solutions to the soil depth equation. hump peak soil production is greater than the erosion rate)
Kirkby (personal communication) has pointed out that until it reaches the corresponding equilibrium soil depth
the solution in Equation (6.4) for the exponential function greater than D∗ , and vice versa if the soil depth is slightly
is unconditionally stable for all soil depths, so that any decreased, then the soil production function decreases and
94 Soils: Soil Depth
FIGURE 6.4: Trajectories over time of soils with a range of initial soil depths and subjected to the humped soil production function. (a) All
simulations have the same erosion rate for each initial soil depth and trajectory with time; (b) the soil production function applied in this
example and (c) each initial soil depth has an erosion rate just balancing the corresponding soil production rate for that initial soil depth. In panel
(c) a small random perturbation on this erosion rate (0.5% the erosion rate) is then applied, and this panel shows the subsequent trajectory of
the soil depth.
the soil thins further. Again this will continue until the soil from this equilibrium toward either zero depth or the
depth is zero (the rock exposure likely turning that part of stable depth on the other side of D∗ :
the hillslope into a weathering- or detachment-limited Figure 6.4c shows a slightly different but equally
regime). enlightening set of depth trajectories for the same SPF in
Figure 6.4 shows an example of the soil depth evolu- Figure 6.4b. In this figure for each initial depth, the
tion for a range of initial soil depths (in increments of erosion rate that would exactly balance the soil production
0.1 m from 0.1 m to 2.0 m) and the humped SPF in rate has been calculated. If the soil production and erosion
Equation (6.5). The same erosion is applied to all the exactly balance, then the soil depth will not change with
trajectories so soils grow deeper or shallower based on time, but if the equilibrium is unstable, tiny perturbations
the imbalance between the erosion and the soil produc- will grow and eventually cause the depth to diverge from
tion for that depth. In this example most of the initially the equilibrium and converge on the stable equilibrium.
deeper soils evolve to an equilibrium depth of 1.17 m To trigger this divergence, a small (0.05%) random
(the exact value of the equilibrium depth is a function of perturbation on the erosion rate was applied so some
the parameters used in the SPF, 1.17 m is the solution for erosion rates will be slightly higher (i.e. so soils will
the parameters used in Figure 6.4). Figure 6.4a shows if initially thin slightly) while others are slightly smaller
you project across to the SPF in Figure 6.4b, that the (so soils will initially thicken slightly) than the exact
equivalent shallow depth equilibrium with the same soil equilibrium. Figure 6.4c shows that for soils depths
production rate is 0.21 m. All soils with initial depth less greater than the hump, the trajectory is a straight line
than 0.21 m (i.e. in the figure the 0.1 m and 0.2 m initial because they are a stable equilibrium and slight perturb-
soil depths) diverge away from the equilibrium to have ations do not make significant difference in the equilib-
zero depth, while soils with an initial depth of greater rium soil depth. However, for all the depths less than the
than 0.21 m converged past the hump to the deeper hump there is divergence. Some of the trajectories con-
equilibrium. This means that in practice while a solution verge on zero. These are the cases where the random
for soil depth between zero depth and D∗ is mathematic- perturbations produced an erosion rate slightly higher than
ally possible it is unstable, and the solution will diverge equilibrium so soils thin. Some of the other trajectories
increase and converge on the equilibrium soil depth that is general than the exact form of Equation (6.5) and the
at greater depth than the hump. For instance, projecting example in Figure 6.4. To further investigate this hypoth-
across to the SPF in Figure 6.4b, the trajectory starting at esis Furbish and Fagherazzi (2001) coupled the humped
0.4 m converges on the equilibrium at about 0.9 m. The SPF with diffusive and fluvial hillslope transport and
zero depth simulation in also interesting because the par- showed that the spatial coupling does not substantially
ameters chosen for the SPF have a nonzero value at depth change the hillslope profile at the hilltops but may lead to
zero and the random perturbation results in a reduced significant changes downslope where fluvial transport is
erosion rate. Because the erosion is very small, the balan- dominant.
cing soil production rate is also very small, and they only The key characteristics of the models in this chapter
balance when the depth is off-scale at the bottom (about are that they only simulate the processes that are occurring
8 m in this case). This is the reason the zero depth simula- at the top and bottom of the soil profile and use these
tion disappears off-scale at time 1000. processes to determine the mass balance of the soil profile,
Carson and Kirkby (1972, p. 105) and Kirkby (2000) lumped as a whole from top to bottom, to describe
speculated that these stable/unstable equilibria will lead to the dynamics. However, much weathering occurs within
a hillslope that is a patchwork of bare rock and soil with the profile (e.g. Anderson et al., 2002), so focussing on the
depths greater than D∗ and is potentially a factor in bad- top and the bottom of the profile is a simplification. More
land development. Figure 6.4c supports that idea because importantly these models do not describe what the char-
even with identical physics minor random perturbations acteristics are of the soil that is created, and they do not
caused some soils to disappear and some to deepen. This provide any information about what is occurring within
argument is true for any shape SPF where there is a region the soil profile. We will now discuss models that simulate
where the SPF increases with increasing depth, so is more the evolution of soil characteristics within the profile.
7 Soils: Physical Weathering and Soil
Particle Fragmentation
7.1 Physical Weathering and Soil Mineral • Biological: Tree roots wedge apart rock fragments,
Matter while the finer root hairs may wedge apart the smaller
particle size fraction. Fungi secrete acids along hypha
The previous chapter discussed models for soil depth channels to release nutrients that may lead to disinte-
alone and didn’t provide any information about how the gration of minerals (Li et al., 2016).
soil varied within the profile. In this chapter we talk about
models for predicting the particle size grading down the It is obvious from this list that other processes (e.g.
soil profile. chemical transformation, microbiology) may occur in
In this chapter we discuss processes that change the association with the fragmentation that they cause. With
soil particles physically with an emphasis on the particle the use of operator splitting (Section 2.2.2) we can math-
size distribution. Chemical transformations will be dis- ematically model the fragmentation separately from these
cussed in Chapter 8. We will not discuss the evolution other transformations. We will discuss how to model this
of the strength of the rock/soil fragments, though this may coupling in Chapter 11.
be important in some circumstances when overburden
load is high (e.g. deep inside mine spoil waste dumps).
7.2 The Evolution of the Soil Surface
The focus here is on fragmentation processes where
larger rock and soil particles physically break down into We start with models that simulate the evolution of only
smaller particles and where the fragmentation process the surface of the soil profile. It has been long recognised
itself does not change the chemistry of the particles. that one of the long-term impacts of fluvial and aeolian
However, we make no presumption about the processes erosion is that the fine material on the surface is
that cause fragmentation. Fragmentation may be caused winnowed out, leaving behind a layer of coarser material
by physical, chemical or biological processes, including than protects the surface against further erosion. If this
the following: armour is undisturbed, it is common to see erosion drop to
• Physical: cycles of salt crystallisation in cracks may near zero when all the entrainable material from the
wedge open existing cracks or pits, as may cycles of armour is removed and what particles remain covering
freeze-thaw. Rapid heating and cooling (e.g. wildfire) the surface cannot be moved by erosion. This mechanism
may shatter the rocks, or fragment an outer layer of has been widely studied both experimentally and theoret-
rocks. Heating and cooling of hydrated minerals or salt ically for rivers and soil surfaces subjected to aeolian
crystals may change their crystal structure with conse- erosion, but there has been little comparable quantitative
quent changes in volume. work for hillslopes and their soils when subjected to
• Chemical: Chemical weathering may transform min- fluvial erosion.
erals into a form that has a larger volume, causing Willgoose and Sharmeen (2006) tested a number of
wedging fragmentation. For instance, many hydration river armouring models, using their ARMOUR1D model,
products have a higher volume than the original against rainfall simulator field trials at Ranger mine (Riley
untransformed minerals. Chemical weathering along et al., 1991). ARMOUR1D had a discretised particle size
highly reactive mineral grain boundaries may lead to a distribution in a surface layer and a semi-infinite layer
breakdown of the ‘glue’ holding the minerals together underneath, and simulated the runoff on the hillslope
96 (sometimes called ‘rotting’). using a saturation-excess hydrology model and observed
rainfall. From this runoff time series they used a variety of laboratory weathering experiments (Wells et al., 2005,
published detachment and selective entrainment mechan- 2006, 2007, 2008) and found that the equilibrium time
isms to entrain the size fractions into the flow while for the surface was longer than for the no weathering case,
recharging the surface layer from the layer below to on the order of 500 years.
maintain a mass balance in the surface layer. In this way
they allowed the surface grading to evolve as a result of
runoff events. 7.3 The Evolution of the Full Soil Profile
They identified that the Parker and Klingeman (1982)
model best fit the data, and Sharmeen and Willgoose We now turn to models of the grading properties of the
(2006, 2007) explored the long-term implications of this full soil profile, rather than only the surface.
model for erosion rates for this site. Fitting the empirical There are a range of analytical particle size distribu-
fluvial transport-limited equation (Equation (4.10)) tion functions that have been used to fit experimental
particle size distributions, but it is difficult to distinguish
Qs ¼ KQm Sn (7.1) them on causal grounds so typically authors have used
they found that not only did the erodibility K decline with the functions that fit their data best. For instance, San-
time, as expected, but the parameters m, n and m1 chidrian et al. (2014) compared 17 particle size distribu-
n also
changed significantly with time. The parameter m tion functions against 1,234 data sets and found different
declined from 1.8 to a minimum of 1.0, then stabilised functions fit different parts of the distribution best, but
at around 1.2, while n declined from 2.1 to a minimum of none did best overall. Accordingly our focus has been on
0.5, then stabilised at around 0.5, both over 100 years. modelling the particle size distribution explicitly using
The parameter trends matched field data from 20-year-old physical principles, and that is the basis of the treatment
sites (Evans and Loch, 1996; Willgoose and Riley, that follows.
1998a). The changes in m and n resulted from longer Legros and Pedro (1985) modelled the evolution of the
and steeper slopes developing a coarser armour (so grading of a soil column by physical and chemical
becoming relatively less erodible), and meant that fluvial weathering. They modelled the soil column as one lumped
erosion model parameters derived for unsorted whole and compared the pedogenesis trajectories with
sediments in flume studies (as is normally done; e.g. the field data. They modelled the soil as being broken up into
initial m ¼ 1:8 and n ¼ 2:1 above resulted from using the 1,000 particle size fractions from 2 μm to 2,000 μm diam-
Einstein-Brown sediment transport model on unsorted eter. They ignored size fractions greater than 2 mm, and
sediments) may not be appropriate for equilibrium soils simulated processes that transformed particles in one size
on hillslopes (Section 4.2.4). class into particles in another smaller size class. These
Their results also indicated that it would take about processes included the following:
200 years to stabilise to these equilibrium values, at which 1. Fragmentation: A particle of a particular size is frag-
stage about 30 mm of cumulative erosion had occurred. mented into a number of smaller particles of the same
We will return to the question of the rate at which these total mass as the original particle and
soils equilibrate later in the book, but we will note here
2. Dissolution: Which dissolved material from the surface
that 30 mm erosion is quite small relative to elevations of of a particle, and when the particle was small enough it
landforms, providing evidence that the surface erosion transitioned to the next smaller size class. The dis-
properties of the soil would equilibrate long before the solved material was lost to the system.
erosion created any significant landform evolution. The
downside of the Sharmeen approach was the computa- Legros and Pedro did not detail their fragmentation and
tions at each time step were very intensive, and the time dissolution processes (e.g. size and number of particles,
resolution required was high (seconds during runoff rate of processes), but their model exhibited a change of
events) so that it was not feasible to simulate more than soil texture over time that they postulated was an analogue
a few short hillslopes for a few hundred years. for field sites they presented.
Cohen et al. (2009) using a new, more efficient, Subsequent work in the field has either explicitly
approach, mARM1D (to be described in detail in the next followed the methodology of Cohen et al. (2009, 2010)
section), was able to replicate the results of Sharmeen (i.e. mARM and mARM3D) or can be recast into Cohen’s
when only armouring occurred on the surface. In an framework. The remainder of this section draws from
extension he included a weathering model that broke Cohen’s mARM3D model unless otherwise stated. We
down the armour particles that was calibrated to will start with a qualitative description of how the model
98 Soils: Physical Weathering and Soil Particle Fragmentation
is constructed, and only then will we fill in the mathemat- semi-infinite layer of fractured rock underlying the pro-
ical details of how the model is solved. file. Each layer has a particle size distribution describing
A soil profile is broken into a series of layers the mass proportion of each size within the soil grading.
(Figure 7.1), with a thin layer on the surface that directly Pedogenic processes are modelled within each layer, and
interacts with water flowing over the surface, and a vertical interactions between the layers are modelled so
that material can be moved between layers. To model a
catchment the catchment is discretised into a spatial grid
of nodes so that the catchment consists of a number of
Surface
profiles, one profile for each node. Each layer is fully
Layer Surface mixed vertically. Using the surface elevations of each
node, the surface water drainage pattern is modelled and
surface water flows from node to node. This surface
Layer 1 Soil water erodes and/or deposits sediment at each node
based on the geometry of the surface flow network, the
sediment being transported within the flowing water and
the local transport capacity at that node. This used the
erosion models discussed in Chapter 4. mARM3D
Layer 2 ignores groundwater flows between the nodes (though,
Soil
in principle, there is nothing to stop this being mod-
elled), and there was no interaction between the soil
layers at one node with the soil layers at another node.
The main limits on the number of soil layers and the
number of nodes spatially are computer storage and
Layer 3 Soil
compute times.
Weathered 7.3.1 Dynamics of a Single Soil Layer

Bedrock
Layer 4
We will first describe how mARM3D simulates each
layer, and then we will discuss how the layers interact
vertically. We start with the thin surface layer. This layer
is the layer that interacts with the water flowing over the
surface. In the most general case if the sediment transport
Layer 5 Bedrock
capacity of the flow is more than the amount of sediment
being carried by the flow, then the flow will erode material
from the surface, while if the transport capacity is less
than the sediment in the flow, then it will deposit sedi-
ment. If erosion occurs, then there is preferential entrain-
ment of the finest fractions from the surface layer into the
flow. To maintain the mass of the surface layer when there
Layer n Bedrock is erosion, material is transferred from the layer directly
below to exactly balance the material being removed by
erosion. If deposition occurs, then there is preferential
deposition of the coarsest fractions (the coarsest particles
settle out fastest) from the flow into the surface layer. To
maintain the mass balance in the surface layer during
Layer ¥ deposition, material from the surface layer is transferred
Bedrock into the layer directly below to balance that material being
deposited.
For the layers below the surface layer the mass balance
FIGURE 7.1: Schematic of the discretisation of the mARM and is maintained at every time step. Thus if material is
SSSPAM soil profile pedogenesis models (after Cohen et al., 2010). transferred from the layer to be put into another higher
layer, then material is entrained from the layer below to and the entries in the state vector, gi , are the mass of
balance the mass removed. Likewise if material is trans- sediment in each grading size range i where there are m
ferred into the layer from the layer above, then material size fractions for the grading. Cohen et al. (2010) used the
from that layer is pushed down into the layer below. In the proportion by mass of the layer in each size range as the
simplest case, if material is removed from the top layer by state, but subsequent experience has found that using
erosion, then material from the next layer below is trans- the actual mass in the layer in each size range makes it
ferred up to balance the mass lost to erosion. The next easier to apply mass conservation principles directly to
layer below that then has material transferred upwards to the construction of the transition matrices (see below).
balance the material lost from the layer above, and this The transition from the grading at any given time to the
processes cascades down the profile to the bottom of the grading at the next time step is described by a matrix
soil profile so that the soil-bedrock interface moves closer equation. It describes both how the grading changes with
to the surface and the soil thins. time and how the grading at one time, t1, is related to the
Within each layer weathering can occur and that grading at some time in the future t2:
weathering changes the grading distribution of that layer.
Each layer weathers independently of the other layers. gðt 2 Þ ¼ Rgðt 1 Þ (7.3)
Mass conservative (i.e. physical weathering) and non-
where in the state-space literature the matrix R is called
mass conservative (i.e. dissolution) weathering can be
the transition matrix and is typically a function of the size
modelled, but to date full chemical weathering where
of the timestep ðt 2 t 1 Þ: Note that any set of coupled
dissolution is followed by precipitation of secondary min-
differential equations can also be expressed with Equation
erals has not been modelled, because mARM does not
(7.3). The advantage of the matrix formulation is that we
have a coupled biogeochemical model to simulate in-
can explicitly formulate all the processes that change the
profile geochemistry (Chapter 8). The weathering rate is
grading with the same methodology.
typically depth dependent and can also be a function of
To represent the soil profile, we write one of these
particle grading and time. The model does not have a
equations for each layer so that we can write Equation
coupled groundwater model and so cannot model the
(7.3) for each layer down the profile, where Rj is different
effects of the interaction between soil moisture, soil
in each layer reflecting how weathering processes change
grading and weathering rates, though a known, specified
down the profile. For each layer j we then have
soil moisture spatial pattern (and thus a specified spatial
weathering pattern) can be input. g ðt 2 Þ ¼ R j g ðt 1 Þ (7.4)
j j
The bottom layer of the soil profile has the grading of
the underlying rock, rock being defined as material that is The discussion in the remainder of this section will
100% the coarsest size fraction in the particle size distri- consider only what happens in a single layer, so we will
bution. That layer becomes soil as the weathering function drop the j subscript for the moment.
breaks down the material into smaller size fractions. Any We simulate each physical process as a multiplicative
layers below cannot be soil until a specified proportion of change to the state. If we have a single process, let’s call it
the rock fraction in the layer above is broken down. Thus A, with a corresponding transition matrix R, then the
the bedrock-soil interface arises naturally as a result of the evolution of the grading vector over one timestep is (using
weathering process and is not explicitly modelled by the subscripts for the timestep)
soil production function from Chapter 6.
Having outlined the conceptual approach used in g ¼ Rg (7.5)
tþ1 t
mARM3D the mathematical details follow. The approach
which is simply Equation (7.3) where one timestep is
draws heavily from the state-space literature, and some of
ðt 2 t 1 Þ: To demonstrate how the method works for more
the terminology of this literature will be used where
than one process, consider the case where there are two
appropriate.
independent physical processes, called A and B with cor-
The soil grading at any given time and in any given
responding transition matrices in Equation (7.5) of R and
layer is represented by a vector called the state vector g :
S; then the combination of these processes on the soil
2 3 grading is
g1
6 7
g2
6 7 g ¼ S Rg ¼ SRg
6 7.. (7.6)
g¼6 7 . (7.2) tþ1 t t
6 7
4g 5 where the order of the matrix multiplication implies that
m1
gm process A acts on the grading first and B operates second
on the result of process A. This can be generalized to any weathered into medium sized particles and 1% of the
number of processes, and is the key to the simplicity and medium into fine, then the marginal transitional matrix is
generality of the approach. This splitting of processes into 2 3
0 0:01 0
independent operations is ‘operator splitting’ (Section
A ¼ 4 0 0:01 0:02 5 (7.10)
2.2.2), and this splitting considerably simplifies the task 0 0 0:02
of constructing the matrices for the combined effect of
simultaneous physical processes. where the diagonal element says how much the mass
Equations (7.3) to (7.6) follow the traditional presen- changes in that grading range and the off-diagonal terms
tation of transition matrices. In the discussion that follows say how much of a different size range is added to it. For
it is more convenient to work with the marginal transition instance, for the medium (i.e. i ¼ 2) size range
matrix than the actual transition matrix. In this case
g2,tþ1 ¼ ð1 0:01Þg2,t þ 0:02g3,t (7.11)
R¼IþA (7.7)
where A2,2 ¼ 0:01 indicates that 1% of the mass in the
where I is the identity matrix (where g ¼ Ig) and A is the medium size range is removed from it and A3,2 ¼ 0:02
marginal transition matrix for process A. The change in indicates that 2% of the mass in the large size range is
grading over a timestep is proportional to matrix A. The added to the medium size range.
advantage of formulating the matrices in this form is that Note that for mass conservation each column of the
the rate of the process represented by A (e.g. the marginal transition matrix must add to zero. If mass is lost
weathering rate) can be changed simply by multiplying (e.g. by dissolution of particles), then the column(s) will
the marginal matrix by a scaling factor. For example, sum to less than zero. Also note that none of the diagonal
doubling the process rate is achieved by applying Equa- terms can be less than 1 because otherwise the equation
tion (7.5) twice (the same as doing two timesteps), and if is transforming more mass in the layer than actually exists
the timestep is small, then this is the same as multiplying in that layer. In this latter case a smaller timestep is
the marginal transition matrix by 2 so that required so that all the elements of the marginal transition
matrix are smaller.
process Aðnominal rateÞ ¼ R ¼ I þ A
Finally it should be noted that, within the physical
process Aðtwice nominal rateÞ ¼ R2 ¼ ðI þ AÞ2 ffi I þ 2A
constraints above, there is considerable flexibility for the
(7.8)
contents of the matrix, and therefore what processes can
where the A2 term is dropped since it is small for a be simulated. The entries below the diagonal will nor-
small timestep. In the discussion that follows, the mar- mally all be zero because otherwise the process being
ginal transition matrix is used unless otherwise noted. represented by the matrix will be making larger particles
In marginal transitional matrix form, Equation (7.6) is from smaller particles (albeit if you are modelling soil
g ¼ SRg ¼ ðI þ BÞðI þ AÞg (7.9) aggregation or particle cementation this may be entirely
tþ1 t t reasonable). In the example in Equation (7.10) particles
where A and B are the marginal transition matrices changed only to particles in the next size class down,
for processes A and B, respectively, and correspond which may not be the case, for instance, for particles that
to the transition matrices R and S in Equation (7.6). fragment into many smaller particles, or for spalling
Representing weathering and other processes within a where there is a single large particle and many smaller
layer is then a matter of constructing the marginal particles resulting from the weathering. We will discuss
transition matrices and repeatedly applying the these cases later.
equations above. While not presented this way the work of Salvador-
Before we discuss how soil layers interact, a simple Blanes et al. (2007) can be cast into this matrix formula-
example will be useful to explain how the details of this tion. Salvador-Blanes modified the approach of Legros
process work. Consider an example where there are only and Pedro (1985) to include the modelling of the break-
three grading size ranges (for convenience let’s call them down of particles less than 2 mm in diameter and used
i ¼ 1, 2, 3, ‘small’, ‘medium’ and ‘large’ grain sizes), and Legros’s method whereby a particle was transformed into
we will construct a marginal transition matrix for a a particle in the next size class down (he defined that as
weathering process that breaks large particles into 2 μm smaller). Both Legros and Salvador-Blanes used
medium particles, medium particles into small particles, 1,000 size fractions for particles less than 2 mm, and this
and leaves the small particles unchanged. If at any one is easily transformed into Equation (7.10) (the A matrix
timestep 2% of the mass of the large particles are will be 1,000 1,000) with only the diagonal terms and
that entry directly above the diagonal term being nonzero, where the state vector is the gradings for all the layers
and the values being the rates at which one size fraction is from the surface armouring layer (subscript 1 indicates the
transformed to the next smaller size fraction per timestep. surface armouring layer), through the profile layers and
The recasting of Salvador-Blanes’s approach high- including the semi-infinite underlying layer (subscript ∞).
lights an important but implied aspect of the matrix meth- The state vector is thus a vector of length mðn þ 1Þ and
odology. The matrix A is not describing how an the matrix is of dimension ðmðn þ 1ÞÞ ðmðn þ 1ÞÞ: The
individual particle is breaking down, but the aggregate notation ½0 indicates a matrix that is m m and where all
result of the breakdown of the many particles in a size matrix entries are zeros. Equation (7.13) shows that all
fraction. It is not possible to move a single particle to the layers can interact with all other layers both above and
next size class smaller and maintain mass conservation below, including the semi-infinite underlying layer.
without generating some other smaller particles. How- Looking at the bottom row of the matrix, the semi-infinite
ever, the matrix method describes the aggregate of a given layer can change through time (as a result of the matrix
mass of particles, and it is possible to have many particles A∞,∞ ), but it cannot be influenced by any of the overlying
transform to the next size class lower without generating layers (since all the entries are ½0). Equation (7.13) can be
an array of fine particles. written in a more compact form
g ¼ ðI þ BÞg (7.14)
═tþ1 ═t
7.3.2 Interactions between Soil Layers
where the double underbar notation distinguishes the
We now describe how one layer within the soil profile grading vector for an individual layer (one underbar) from
interacts with another layer within the soil profile. Let us the vector for the gradings for all the layers (two under-
consider two layers j and k. We can construct a matrix bars). The latter is sometimes called a supervector in the
equation that describes how layer j changes layer k in one modelling literature (because it is a vector of vectors).
timestep: The construction of matrix B appears at face value to be
rather daunting simply because of its size. However, Equa-
g ¼ Lk,j g (7.12) tion (7.13) is the most general statement of the problem, and
k ,tþ1 j, t
in many cases simplifications are possible. For example,
This equation says how much mass in layer j in each of
the grading size fractions is added to or subtracted from • If interactions between layers occur only between adja-
layer k in one timestep. Equation (7.12) models only how cent layers: In this case for layer j, the only matrices
material is moved between layers and not weathering. We that are nonzero are the matrices Aj,j for the weathering
have used the matrix notation L here to distinguish this within that layer, Lj,j1 and Lj,jþ1 which describe how
movement between layers from the weathering process the layer above and below, respectively change layer j,
matrices (that transform the particle size distribution and Lj1,j and Ljþ1,j which describe how layer j changes
within a single layer). For mass conservation in layer k, the layer above and below, respectively:
if mass is added to layer k from layer j, then an equal 2 3
A1,1 L1,2 ½0 ½0 ½0 ½0
amount of mass must be removed from layer k and moved 6L
6 2,1 A2,2 L2,3 ½0 ½0 ½0 7 7
to another layer. The combination of the weathering and 6 7
6 ½0 L3,2 A3,3 ½ 0 ½0 ½ 0 7
movement can be expressed in matrix form if we con- 6 7
6 . . . . . . . 7
struct a grading vector that merges all the grading vectors 6 . . . . . . . 7
6 . . . . . . . 7
for each of the individual layers, and the marginal transi- 6 7
6 ½0
6 ½0 ½0 An1,n1 Ln1,n ½0 7 7
tion matrix is constructed from the layer transition matri- 6 7
ces and interlayer movement transition matrices so that 4 ½0 ½0 ½0 Ln,n1 An,n Ln,∞ 5
½0 ½0 ½0 ½0 ½0 A∞,∞
2 3 2 2 3 32 g 3 (7.15)
g g A1,1 L1,2 L1,n L1,∞
6 17 6 17 6 6 17
6 g2 7
6 7
6 g2 7 6 L2,1
6 7 6 A2,2 L2,n L2,∞ 7
76 g 7
27
6 . 7 6 . 7 6 . 76 • If in addition to interactions only between adjacent
6 . 7 ¼6 7 .. .. .. .. 76 .. 7
6
7
6 . 7 6 .. 7 þ 6 .. . . . . 776 . 7 layers, there is no change in the grading over time
6 7 6 7 6 76 7
6g 7 6g 7 6 An,n Ln,∞ 56 7 within a layer: This might happen when there is only
4 n5 4 n 5 4 Ln,1 Ln,2 4 gn 5 mixing of the soil (e.g. bioturbation) and no breakdown
g g ½0 ½0 ½0 A∞,∞ g
∞ tþ1 ∞ t ∞ t of the mineral matter. The matrix in Equation (7.15)
(7.13) simplifies even further to
2 3
½0 L1,2 ½0 ½0 ½0 ½0
6L
6 2,1 ½0 L2,3 ½0 ½0 ½0 7
7
6 7
6 ½0
6 L3,2 ½0 ½0 ½0 ½0 7
7
6 .
6 . .. .. .. .. .. .. 7
7 (7.16)
6 . . . . . . . 7
6 7
6 ½0
6 ½0 ½0 ½0 Ln1,n ½0 7
7
6 7
4 ½0 ½0 ½0 Ln,n1 ½0 Ln,∞ 5
½0 ½0 ½0 ½0 ½0 ½0
• Adjusting the layers in response to erosion at the sur-

face: When material is removed from the surface layer,
then material to balance that lost to erosion must be
removed from the layer below to make sure the mass in
the layer does not change. This cascades down through
all layers in the profile:
2 E E 3
A1,1 I ½0 ½0 ½0 ½0 FIGURE 7.2: How soil grading is conceptualised in mARM and
6 d1 d1 7 SSSPAM as uniformly distributed within size classes. Data are the
6 7
6 E E 7 Ranger Mine grading used in Willgoose and Sharmeen (2006) and
6 ½0 I I ½ 0 ½ 0 ½ 0 7
6 d d 7 Cohen et al. (2009, 2010).
6 2 2 7
6 E 7
6 ½0
6 ½0 I ½0 ½0 ½0 7 7
6 d 3 7 The basis of the conceptualisation of weathering is that
6 7
6 .. .. .. .. .. .. .. 7 particles are spread uniformly within each size range
6 . . . . . . . 7
6 7 within which they fall (Figure 7.2). Thus some particles
6 E E 7
6 ½0 ½0 ½0 I I ½0 7 will be at the lower boundary of the size, some in the
6 d d 7
6 n1 n1 7 middle and some at the top of the size range. We concep-
6 E E 7
6 ½0 ½ 0 ½ 0 ½ 0 I I 7
4 dn dn 5
tualise the fragmentation process as a parent particle
½0 ½0 ½0 ½0 ½0 ½0 breaking into a number of daughter particles (Figure 7.3).
Figure 7.3 shows that depending on the fragmentation
(7.17)
mechanism there may be a range of different types of
daughter particles created. If we assume that the density
where E is the erosion in one timestep in depth units, di is
distribution within a single particle size fraction is con-
the depth of layer i (which converts sediment mass due to
stant, then mass conservation implies volume conserva-
erosion E into a proportion of the layer mass) and I is a k k
tion. In this case the total volume of the daughter particles
identity matrix. The matrix A is the armouring transition
is the same as the volume of the parent particle.
matrix for the surface layer and determines the size selectiv-
We now extend our discussion from what happens to a
ity of the sediment entrainment due to erosion. As an aside,
single particle to what happens to all the particles within a
this is the first time that bulk density of the soil appears in the
size class range. Figure 7.4 shows what will happen to the
matrix methodology. Bulk density is the conversion factor
larger size range when the particles break into two par-
between depth of soil eroded and mass of soil eroded. If soil
ticles of different diameter, if they all follow the rules for
erosion is expressed in mass rather than depth units, then the
breaking of a single particle (i.e. all particles break with
bulk density conversion is not required.
exactly the same geometry). In the discussions that follow
we assume that all particles are spherical. Initially we will
7.3.3 Constructing the A Matrix for Weathering consider the case where a parent particle breaks into two
daughter particles. Figure 7.4 shows how the geometry of
We will now discuss how the matrix A is formulated to a single particle breaking allows us to map the parent
model weathering. Unless otherwise stated, the discussion particle size grade to the smaller dimensions of the two
in this section is for one layer, and we drop the layer daughter particles’ size grades. Likewise we can map the
subscript j for clarity of discussion. We will first discuss largest size from the parent size gradings to the largest
mass conservative weathering and then generalise the dis- size of the daughter size grading. From the geometry of
cussion to non-mass conservative weathering afterwards. breaking of the single particle we know what the
proportions of the mass of daughter particles should be in typically neatly fit into a single size fraction, but will need
each of the two daughter particle gradings ranges. to be interpolated across a number of size fractions. These
Figure 7.4 shows that if we take one parent size fraction, proportions will be a function of the diameters of the
then the daughter particle’s size distribution will not daughter particles and of the size grading fractions
adopted by the user in the modelling. Finally, if the
(a) distribution between the lower and upper size of
d = do – 2dw the parent particles is uniform, then the distribution
dw dw
do
of the daughter particles is also uniform between the
d lower and upper size grading.
time
Using these assumptions it is relatively straightfor-
ward, though tedious, to construct a matrix to simulate
(b) weathering. The calculations in Figure 7.4 are done for
Model 1. Symmetrical cracking each grading range and the results summed.
N particles
However, a conceptual simplification is possible.
v1 There is a combination of grading size fractions and
V v2 ... vN
weathering processes that leads to particularly simple
Vi = V results, and which allow us to derive some analytical
N
results that provide useful insight into the time variation
Model 2. Asymmetric cracking
N = 2 particles of grading under the action of weathering. We will use
this to demonstrate how the weathering matrix works and
(1–a)×V a×V derive some simple weathering results. We will then show
V
that this simple model can be used to derive more com-
plex models.
Model 3. Whitworth cracking N particles For this example we will define the fractions in the
grading size distributionsuch that
the lower diameter limit
v3 1=
v1 ... v of each size fraction is 0:5 3 times the upper diameter
V N
v2 limit for that same size fraction. If we take the maximum
N 1
size limit as 2 mm, then this yields the limits for the size
Vi = V × Σ
N i=1 (N–i+1) fractions as in Table 7.1. There is no special significance
to the upper and lower limits of 2 mm and 0.125 mm in the
FIGURE 7.3: Schematic of fracturing models for rock particles table; they are simply used to show what this grading
(a) breaking of small particles from a rind and (b) body fragmentation looks like. It is useful to examine this size grading because
(from Sharmeen and Willgoose, 2006). this grading fractionation is different from that normally
FIGURE 7.4: Conceptualisation of how size fractions in the grading are transformed in each time step of mARM (from Cohen et al., 2009).
TABLE 7.1: Size grading classes for different weathering The diagonal and off-diagonal α terms mean that α of
mechanisms for two equal particles and three equal particles the mass in the larger size grading is all added to the next
fracturing smaller size grading. The diagonal element for the
smallest fraction is zero because we assume that particles
Upper-lower size limit (mm) this size cannot weather any smaller. Equation (7.18) is
Size grading class
2 daughters 3 daughters mass conservative, so all the columns of the matrix add
to 0.
12 2.0–1.587 2.0–1.387 It is possible to calculate the time variation of the mean
11 1.587–1.260 1.387–0.962 diameter of the particle size distribution purely from the
10 1.260–1.000 0.962–0.666 I þ A matrix using Equation (7.18). With the exception of
9 1.000–0.794 0.666–0.462 the smallest size grading, the mass gi that is in any given
8 0.794–0.630 0.462–0.321 size fraction i will change after one timestep so that
7 0.630–0.500 0.321–0.222 ð1 αÞ remains in size fraction i while α will now be in
6 0.500–0.397 0.222–0.154 the next size fraction
1=
i 1, which in the case of
smaller
5 0.397–0.315 0.154–0.107 Equation (7.18) is 0:5 3 smaller than the size fraction
4 0.315–0.250 0.107–0.074 above, and from the way the size fractions are defined in
3 0.250–0.198 0.074–0.051 Table 7.1 that is true for all size fractions i. For the
2 0.198–0.158 0.051–0.037 smallest size fraction there is no change. If we consider
1 0.158–0.125 0.037–0.025 the case where all the mass is initially concentrated in the
largest size fractions and there is an insignificant mass in
the smallest fraction (this can be achieved by defining the
used for soils analysis (e.g. phi grading) and thus will be smallest size fraction to be very small), then in one time-
unfamiliar to readers. The significance of the grading in step (time changing from t to t þ 1) the mean diameter of
this example is that if a spherical particle splits into two the soil changes from
equal volume spherical daughter particles, then for conser- 1=

vation of volume all the particles in one size grading will d mean,tþ1 ¼ ð1 αÞ þ α0:5 3 d mean,t
fall exactly into the size grading of the next size fraction ¼ ð1 0:2063αÞd mean,t (7.19)
smaller. For example, for the grading class 12 in Table 7.1
a particle with the largest diameter of 2 mm will split into which allows us to derive an equation for the evolution of
two particles of diameter 1.587 mm, and a particle with the the mean diameter of the soil T timesteps into the future:
smallest diameter of 1.587 mm will split into two particles dmean,t¼T ¼ ð1 0:2063αÞT d mean,t¼0 (7.20)
of diameter 1.260 mm. These upper and lower size limits
define the size range for the next smaller Grading Class 11. which is a semi-log linear relationship between diameter
This is true for all Grading Class ranges in Table 7.1. and time, where the slope on a semi-log plot is 0:2063α.
Accordingly, to assemble the weathering matrix with This can also be expressed as an exponential so that
this grading when a parent particle fractures into two
d mean,t¼T ¼ d mean,t¼0 e ln ð10:2063αÞT (7.21)
equally sized daughter particles is straightforward because
the daughter particles only ever fall into the size fraction Figure 7.5 shows this curve for a number of values of α
below, and this size fraction below can receive only and with a starting mean diameter of 3.5 mm.
daughter particles that fracture from the size fraction The exact details of Equations (7.19) to (7.21) are a
directly above. Thus if we say that the proportion of the function of the specific assumptions in Equations (7.18),
particles that fracture for one timestep is α, then the but it is straightforward to extend this analysis to any
weathering matrix is distribution of daughter products, and the only part of
Equation (7.19) that will change is the number inside the
2 3
0 α 0 0 0 0 parentheses; the relationship itself will still be log-log
6 0 α α 0 0 7 0 linear. This is true provided only that (1) the definition
6 7
60 0 α 0 0 7 0 of the size fractions is defined as in Table 7.1 for two
6. .. .. .. .. .. 7 ..
A¼6 . . . . . 7 .
6. . 7 (7.18) daughter particles, (2) the fracture model is independent
60 0
6 0 α α 0 7
7 of the diameter of the particle so that the daughter par-
40 0 0 0 α α 5 ticles are always the same size relative to the parent
0 0 0 0 0 α particle and that this fracture model is independent of
FIGURE 7.5: Numerical solution of the time evolution of the mean of the particle size distribution for a range of weathering rates. The
parameter ‘rate’ is rate ¼ 0:2063α in Equation (7.19). The levelling off on the right-hand side of each graph is because at this stage all soil is in
the finest size fraction in the modelled soil grading.
the diameter of the parent particle and (3) the smallest size given size fraction that breaks down each timestep is α as
fraction is small enough that only a minor amount of the before, then Equation (7.18) is still applicable except now
mass occurs within the smallest fraction (otherwise the the definition of the size grading ranges is different and
curves will start to level out as the particles age as seen in the matrix will be of a different dimension. The equivalent
Figure 7.5 on the right-hand side). three daughters result to Equation (7.19) is
The assumption that particles split into two equally 1=

sized particles leading to Equation (7.21) is rather restrict- dmean,tþ1 ¼ ð1 αÞ þ 0:333 3 α d mean,t
ive, but this result can be generalised as follows. Consider ¼ ð1 0:3066αÞd mean,t (7.22)
first the case where N equally sized particles are generated
at each time step instead of the two in the example above. and Equation (7.20) becomes
This means that the diameter of the daughter particles is
1
now N =3 the diameter of the parent particle. In the same d mean,t¼T ¼ ð1 0:3066αÞT dmean,t¼0 (7.23)
way that we developed a grading range definition for two In general for N daughter particles
daughter particles fracturing in Table 7.1, we can do this T
1
in general for all values of N. In Table 7.1 fracturing for d mean,t¼T ¼ 1 1 N =3 α d mean,t¼0 (7.24)
three equal daughters is shown. Note that the total size
range covered by the 12 size fractions in Table 7.1 is so it is clear that if particles break into a larger number of
different, with the three daughters column going down smaller particles at each timestep, then the soil will
to 0.025 mm, while the two daughters range goes down to become finer at a faster rate, and Equation (7.24) defines
only 0.125 mm. Thus if you wish to cover a specific total what that speedup will be.
size range (e.g. for a given soil), the two daughters size In the discussion that follows we further generalise the
range requires more size fractions. If the proportion of a fracture geometries considered. To simplify the discussion
TABLE 7.2: Fragmentation notation examples half the weathered volume goes into the next class down
from the parent, while half the weathered volume goes into
Fragmentation Physical example for daughter particle the class 2 classes below the parent so the A matrix is
notation geometry 2 3
α=
0 α 2 0 0 0
p2-0–100 2 particles of 50% volume of parent. 6 0 α α= 7
6 2 0 0 0 7
The split in two geometry of Wells 6 7
6 0 0 α 0 0 0 7
et al. (2008). 6 7
6. .. .. .. .. .. .. 7
A ¼ 6 .. . . . . . . 7 (7.25)
p2-0–50-50 1 particle of 50% volume, and 6 7
60 0 α= 7
2 particles of 25% volume of parent. 6 0 α α=2 2 7
6 α= 7
p2-0–50-0–50 1 particle of 50% volume, 0 particles 40 0 0 0 α 2 5
of 25% volume and 4 particles of 0 0 0 0 0 α
12.5% volume of parent. Spalling-
like behaviour. and Equation (7.19) becomes
p2-0-0-0–100 0 particles of 50% and 25% volume, 1= 1=
d mean,tþ1 ¼ ð1 αÞ þ α=2 0:5 3 þ α=2 0:25 3 Þd mean,t
and 8 particles of 12.5% volume of
the parent. Used by Finke (2012) in ¼ ð1 0:2882αÞd mean,t (7.26)
the SOILGEN model.
As expected this case weathers faster than the two
p2-0–50-25- 1 particle of 50%, 1 particle of 25%,
equal daughters case. In a similar fashion you could
12.5-12.5 1 particle of 12.5% and 2 particles
consider that this three daughters case can be extended
of 6.25% volume. A scaling
where one of the two smaller particles itself breaks into
fracturing similar to Whitworth
two, so the daughter particles are 1 ð1=2Þ volume, 1
cracking (Figure 7.3).
ð1=4Þ volume and 2 ð1=8Þ volume and so on.
Similarly the three equal daughters particle fracturing
could be extended by having one of the daughters break
of geometry we introduce a shorthand for fragmentation into three equal volume smaller particles, so that the
geometry, which indicates how, on average, the volume of daughters are two particles with 1/3 the parent volume
particles will be distributed after a fragmentation event (i.e. 2/3 of the parent volume) and three particles with 1/9
assuming the size grading definitions in Table 7.1. The the parent volume (i.e. 1/3 of the parent volume), yielding
general form is pX-AAA-BBB-CCC-DDD where X is the p3-67.7-33.3. In this case Equation (7.25) becomes
number of daughter particles the grading fractions have 2 α 3
0 α =3 0 0 0
been defined for (i.e. X ¼ 2 for the two particle grading in 6 0 α 2α= 0
6 0 0 77
Table 7.1, X ¼ 3 for the three particle grading), AAA is 6
3
7
6 0 0 α 0 0 0 7
the percentage of volume that remains in the parent size 6 7
6. .. .. .. .. .. .. 7
grading fraction after fragmentation, BBB is the propor- A ¼ 6 .. . . . . . . 7 (7.27)
6 7
60 0 α 7
tion in the next size grading smaller, CCC in the next size 6 0 α 2α
=3 =3 7
6 7
down again and so on. In this notation where all of the 40 0 0 0 α 2α=3 5
particles split into two with two equally sized particles, 0 0 0 0 0 α
then the shorthand is p2-0–100. Some geometry examples
are listed in Table 7.2. Some simple examples of fracture and Equation (7.26) becomes
geometries and their fragmentation notation are also 1=
dmean,tþ1 ¼ ð1 αÞ þ 2α=3 0:333 3 þ α=3 0:333 =3 Þd mean,t
2
shown. While all the examples in Table 7.1 have no mass

left in the parent grading after fragmentation (they all have ¼ ð1 0:3775αÞdmean,t (7.28)
a leading zero), at the end of this section an important, but
If two of the three daughters break into three (i.e.
more complex, fragmentation geometry that leads to a
p3-33.3–66.7), then
nonzero percentage in the parent grading will be discussed.
2=
d mean,tþ1 ¼ ð1 αÞ þ α=3 0:333 =3 þ 2α=3 0:333 3 d mean,t
1
Instead of two equally sized daughter particles, let us now
break the particle into three daughters where one daughter is
¼ ð1 0:4484αÞd mean,t (7.29)
half the volume of the original and the other two daughters
are a quarter of the volume of the original (i.e. p2-0–50-50). and as for the two daughters cases, both of these cases
Mass conservation still applies, but what now happens is that weather faster than the case where all the daughters are of
equal sizes, and when more fine particles are generated and if the values of ai,j are small when j is small (i.e. large
(i.e. Equation (7.29) versus (7.28), or Equation (7.28) parent particles do not generate many very small daughter
versus (7.26)) then the soil weathers faster even when particles), then we can write
the rate of breakdown per timestep is unchanged.
d mean,tþ1 ffi admean,t (7.32)
It should be clear from these examples that the com-
binations of sizes of daughter particles possible are quite where a is a constant reflecting the structure of A so that
extensive. The only constraint is a geometry constraint we can write a general equation for how the mean diam-
driven by the grading size definition adopted from eter of the soil will change over time
Table 7.1. If the two daughter grading range in Table 7.1
d mean,t¼T ¼ ðaÞT d mean,t¼0 (7.33)
is adopted, then particles must break into particles where
the volumes of the daughters are related to the parent which shows that the mean of the soil grading is an
particle volume by integer powers of 1/2. Likewise for exponential function with time of the initial grading irre-
the three daughter grading range in Table 7.1, then all spective of the structure of A. Note that the restrictions on
the daughters must have volumes that are integer powers the form of A are relatively modest (just that weathering
of 1/3. This integer power constraint is so that all the cannot generate a large mass proportion of small par-
daughters fall entirely and only into one of the grading ticles), so this is quite a powerful and general conclusion.
size ranges and do not span two grading size ranges. Note also that the use of the two daughter size fraction
This section has shown how to construct the definition in Table 7.1 is a convenience rather than a
weathering A matrix using a conceptualisation of the necessity, as we will see in the next section. The fragmen-
fracturing of individual parent particles into a range of tation notation for this case is p2-(1-ai,i)-ai,i-1- . . . -ai,1.
daughter particles. Thus everything in this section has The discussion above presumes that all particles frac-
been strongly physically based on fracturing mechanisms ture with exactly the same geometry. This is not essential.
that are, in principle, observable in the laboratory or the One simple extension is to allow particles to break with
field. This presentation was intentional because it high- two possible geometries at different rates. The A matrix
lights the link between fracturing of individual particles for the combined process is then
and the rather less tangible mathematics of the A matrix.
A ¼ w1 A1 þ w2 A2 (7.34)
We will deviate from this philosophy for a moment to
generalise Equation (7.18), which will then allow us to where the subscripts are the weathering rate w and
provide a general analytical solution to the change in grading weathering matrix A for each of the two processes. This
over time. When constructing the A matrix, to ensure mass can be generalised to as many processes as required.
conservation we need to ensure only that (1) the summation Figure 7.6 extends the discussion by examining a
of each column should be zero, (2) the diagonal term is number of fracture geometries, and shows how (1) the
between 1 and 0 and (3) the off-diagonal terms are greater different fracture geometries yield different rates of evo-
than or equal to 0. Thus if we can construct a generic column lution of the mean grading and (2) all the geometries
for matrix A, that is the same for all grading fractions evolve as a semi-log of time as in Equation (7.33). The
2 a 3 meaning of the coding used for describing fracture geom-
i, 1 etry is explained in the figure caption, but is based on the
6 .. 7
6 . 7 fracture geometry where particles are 1/2, 1/4, 1/8 and so
6 7
6 ai,i2 7 on of the volume of the parent particle.
6 7
6 ai,i1 7 One of the fracture geometries in Figure 7.6 is philo-
6 7
column i in A ¼ 6 ai,i 7 (7.30)
6 7 sophically different from the ones described above. In all
6 0 7
6 7 the fracture geometries above the geometry of the daugh-
6 0 7
6 . 7 ter particles is deterministically fixed as a proportion of
4 . 5
. the volume of the parent particle. The aggregated behav-
0 iour of all the particles is then just the sum of all the
particles in that grading range. However, consider the case
we can then write Equation (7.19) as (using the two
where a single particle breaks into two particles but where
daughter size grading)
the fracture location in the particle is randomly distributed
1=
dmean,tþ1 ¼ ð1ai,i Þþai,i1 0:5 3 within the particle (e.g. a schist where the particle cleaves
along the layering but where the cleavage plane is ran-
þai,i2 0:5 3 þ...þai,1 0:5 =3 d mean,t
2= ði1Þ
(7.31) domly located within the particle). The probability
(a)
(b)
FIGURE 7.6: (a) The average particle size distribution of the daughter particles resulting from random fragmentation (P2-stoch), dparent is the
diameter of the parent particle, (b) Numerical solution of the time evolution of the mean of the particle size distribution for the same
weathering rate (as used in Figure 7.5) and a range of fragmentation models. The weathering model p2-0–50-50 is the same as ‘rate = 0.03’ in
Figure 7.5.
FIGURE 7.7: These are the same

simulations as in Figure 7.6 (i.e.
same weathering rate, different
fragmentation models) but
showing the particle size
distribution when the d50 of the all
fragmentation models is equal to
0.1 mm. Note that Figure 7.6
shows that the age of each of the
distributions will be different since
the different fragmentation models
generate a different rate of change
in the d50 for the same
weathering rate.
distribution of the volume of daughter particles derived particles more than three grading fractions smaller into the
from particles in a specified particle range is shown in third grading fraction (i.e. the 17.4). Other than the pro-
Figure 7.6a. It is convenient here to define the x axis as the portion remaining in the parent grading range, the relativ-
volume of particles rather than their diameter because the ities between the daughter size fractions are the same as
probability distribution of the daughter particles is then the scaling model p2-0–50-25-12.5-12.6.
particularly simple. If a large number of particles fracture, Figure 7.6 shows that those fractions that generate a
then the probability distribution function gives the mean greater proportion of volume in smaller size fractions fine
of the volume distribution of the daughter particles, from faster, which is consistent with Equation (7.24). Figure 7.7
which A can be easily derived. In fragmentation notation shows the particle size distribution resulting from using a
it is p2-30.7-34.6-17.3-17.4, if we lump the volume of all number of different fragmentation models. Initially all
particles were 4.0 mm in diameter, and Figure 7.7 shows This shows that we can populate the A matrix using
the particle size distribution when the fragmentation any size class definition we like and then later transform it
models have weathered the particles to a median diameter into any other size class definition that is convenient. One
0.1 mm. Two conclusions of this figure are that (1) differ- major consequence of this is that the semi-log evolution of
ent fragmentation models yield slightly different particle the mean diameter we have derived in the previous section
size distributions even though the median diameter of the is not unique to the size class distribution used to derive it,
particle size distribution is the same and (2) while the and that even if the class definition is different (provided
different fragmentation models may yield finer particle the physics is the same), then the semi-log relationship
size distributions at different rates, they yield particle size with time will still be observed.
distributions that are only slightly different (e.g. the p2-
0–100 and the p2-stoch random fracture models are indis-
tinguishable in Figure 7.7, but p2-0–100 fines a factor of 7.3.5 Constructing the A Matrix for
two times faster in Figure 7.6). Pedoturbation
Pedoturbation is any process that disturbs the horizonation

7.3.4 Transforming the Size Grading of soils. Some pedoturbation processes assist in the differ-
Definition entiation of soils into layers, while others mix the soils,
tending to destroy the layers (Hupy and Schaetzl, 2006).
The definition of the size fractions necessary to derive
Bioturbation is a pedoturbation process resulting from bio-
Equations (7.18) to (7.29) is rather restrictive, and is not
logical activity and is a significant process in soil profile
necessarily consistent with size gradings used in soil
development (Hole, 1981). One of the main impacts of
science practice (e.g. sieve sizes). Accordingly, we will
bioturbation is to physically mix soil from top to bottom
now show how to generalise the definition of diameters
(Heimsath et al., 2002; Ahr et al., 2013), and transport
defining the size fractions while still maintaining the
material from the surface into the profile (Astete et al.,
simplicity of the equation and the semi-log relationship
2015). Soil fauna such as ants and termites bring fine mater-
for mean size with time. It is possible to transform the
ial from deep in the soil to the surface (Lobry de Bruyn and
mass fractions from one size fraction definition to another
Conacher, 1990), earthworms likewise mix the profile (Van
size fraction definition. Consider g to be the grading by
Hooff, 1983; McKenzie et al., 1993), while tunnelling
one grading class size definition and h to be the desired
animals such as gophers (Huntley and Inouye, 1988; Gabet,
mass fraction equivalent to g in another grading class size
2000) and wombats (Paton et al., 1995; Field and Anderson,
definition. For generality let us consider that there are n
2003) may also transfer material laterally (via their horizon-
size classes in g and m size classes in h. We can transform
tal tunnels) as well as from within the profile to the surface.
g into h with the matrix equation
Flora such as trees mix the soil through tree throw (Gabet
h ¼ Tg (7.35) et al., 2003; Field and Anderson, 2003). In this section we
will consider only vertical transport of material. Horizontal
where T is a n m matrix that distributes the mass in the movement will be discussed in Chapter 9.
n classes in g into the m size classes in h. Many of these processes leave macropores within the
We can now transform the weathering (or any other profile when material is removed from that layer, which
process for that matter) matrix equation that is originally means that the porosity is increased and the bulk density
g ¼ ðI þ AÞg (7.36) of the soil layers is reduced. If the mass in each soil layer
tþ1 t
in the model is the same and bulk density decreases, then
By substituting in the inverse of Equation (7.35) we get the layers will dilate (or expand), so that the soil surface
the transition matrix in the new grading rises. For instance, Wilkinson et al. (2009) provides data
for bioturbation rates that decline with depth roughly
T1 htþ1 ¼ ðI þ AÞT1 ht exponentially from the surface, which they attribute pri-

^ h
htþ1 ¼ TðI þ AÞT1 ht ¼ I þ A (7.37) marily to ants and worms. They also measured bulk
t
density and found increases with depth (1.1 g/cm3 near
so that in the new grading definition to the surface to 1.8 g/cm3 at a depth of 0.8 m) mirroring
^ ¼ TðI þ AÞT1 ¼ I þ TAT1 the decline in the rate of bioturbation and number of
IþA (7.38)
macropores generated. Shull (2001) presents a very simi-
noting that TT1 ¼ I: lar matrix model for bioturbation in sea bottom sediments.
One unusual form of pedoturbation was the soil mixing

resulting from artillery shells in France during World War
I (‘bombturbation’, Hupy and Schaetzl, 2006, 2008), and
these and other warfare sites provide unique field experi-
ments in pedogenesis (Certini et al., 2013).
7.3.5.1 Simple Biological Mixing

A simple model for bioturbation mixes the soil vertically
randomly, and this can be modelled as a vertical diffusive
process. The idea is that microfauna (e.g. earthworms)
mix the soil vertically. This process could be modelled
by a Fickian diffusion-type equation. A simple method,
which asymptotes to Fickian diffusion (i.e. Qs ¼ D ∂C ∂x
where Qs is the rate of vertical transport per unit area FIGURE 7.8: Definitions of layer thickness and node discretisation
and time, ∂C∂x is the gradient of the property C being mixed for Equation (7.41).
and D is the diffusivity), is to exchange material with the
layers above and below (Williams, 2006):
nodes i þ 1 and i, and α ¼ Δz1 Δz2 . The relationship in
2 3
R1 I R2 I ½0 ½0 ½0 Equation (7.41) is derived from a central finite difference
6 7 approximation to Fickian diffusion. When the thicknesses
6 R1 I 2R2 I R3 I ½0 ½0 7
6 7 of the three layers are the same (i.e. Δz1 ¼ T i and α ¼ 1)
6 ½0
6 R2 I 2R3 I R4 I ½0 7
7
6 .. 7 Equation (7.41) reduces to Equation (7.40).
B¼6 7
6 ½0 ½0 R3 I 2R4 I . 7
6 7
6 .. .. 7
6 ½0
6 ½0 ½0 R4 I . .77 7.3.5.2 Tree Throw
4 5
.. .. .. .. .. .. Tree throw brings to the surface all the soil in the root ball of
. . . . . . the tree leaving behind a large hole in the ground. The soil
(7.39) falls off the roots over time, and the mixed soil in the root
where Ri is the rate of bioturbation mixing at the depth of ball refills the hole in the ground. In addition to the mixing,
layer i. Equation (7.39) assumes that all layers have the there is potentially an impact on the porosity of the soil with
same thickness. Following Wilkinson et al. (2009) the the soil becoming more disordered. Tree throw is an event:
bioturbation rate Ri declines with depth approximately a tree falls, and the consequences happen immediately. It is
exponentially. The identity matrix I means that the mixing not a slow ongoing process, though the cumulative impact
is not size selective. For constant thickness T layers, this of many tree throw events is an ongoing process. Tree throw
converges to Fickian diffusion asymptotically with time can be modelled as an event if it is assumed that the refilling
of the hole happens quickly relative to the timestep of the
Δt modelling. At a given time a tree falls at a particular location
Ri ¼ D (7.40)
T2 or it doesn’t. Thus we do not use Equation (7.14) as the
If the layers are of different thickness, then the ith column model of what happens from t to t þ 1, but we can use it to
of tridiagonal elements changes to model what happens before and after the tree fall event
2 3 g ¼ ðI þ BÞg (7.42)
2DΔt
6 7
═after ═before
6 ðΔz1 Þ2 1 þ 1 7
6 α 7 We simply need to know, when a tree fall event
6 7
6 7
6 2Dð1 þ αÞΔt 7 occurs, what the B matrix looks like for a single tree fall
6 7
6 7 event, and apply Equation (7.42) for that time (and apply
6 ðΔz1 Þ2 1 þ 1 7 (7.41)
6
6 α 77
the equation repeatedly if there are many tree fall events at
6 7 the same time). The mixing of the soil over the depth of
6 2DαΔt 7
6 7
4 2 1 5 the root zone is modelled using Equation (7.42) if we
ðΔz1 Þ 1 þ assume that the root ball rips up the top n layers so the
α
top n layers are mixed together and the n þ 1 and lower
where (see definitions in Figure 7.8) Δz1 is the distance layers are undisturbed. This is an averaging of the grading
between the nodes i and i 1, Δz2 is the distance between of the top n layers so that the B matrix can be expressed as
20 1 3
a uniformly distributed random number between 0 and 1
6B C 7 for each node. If the random number is greater than P, then
6B m1 C
1CI
m1
I ½0 ½0 ½0 7
6BX X 7
6@ n
A n
7 a tree throw event occurs in that node, while if it is less than
6 mk mk 7
6 7 P, a tree throw event does not occur. If the area of a node is
6 k¼1 k¼1 7
6 7 large, and/or the timestep is long, then it is possible for P to
6
6 ... ..
.
..
. ½0 ½0
7
½0 7 be greater than 1 (which indicates more than one tree throw
6 0 1 7
6 7 event). In this case either one of the timestep or node area
B¼6 7
6 B m C 7
6 mn B n C 7 needs to be decreased.
6
6 I BX 1CI ½0 ½0 ½0 7
7
6 Xn
@ n
A 7
The second approach calculates the average effect of
6 mk mk 7 tree throw over many events and models the mean behav-
6 7
6 k¼1 k¼1 7 iour rather than the effect of individual events. In this case
6
6 ½0 ½0 ½0 ½0 ½0 ½0 7
7
6 7 the mean of the tree throw events in any given node is the
4 ½0 ½0 ½0 ½0 ½0 ½0 5 rate of tree throw occurrence per unit time, which is P, so
½0 ½0 ½0 ½0 ½0 ½0 that the equation for the effect of tree throw per unit time,
(7.43) by modifying Equation (7.44), becomes
where mj is the total mass of layer j. The zero matrices for i g ¼ ðð1 αPÞI þ αPBÞg (7.45)
═tþ1 ═t
greater and equal to n þ 1 indicate that the tree throw does
not change the grading for layers n þ 1 and deeper. The term The advantage of the first method is that you can cap-
Pmn j is included to ensure that the mass in layer j is the ture the stochastic effect (in both space and time) of tree
k¼1
mk
throw, whereas the second method gives only the average
same before and after the tree fall event. This is not critical
effects. This stochastic effect may be important in the
but is convenient, because it preserves the thinner armour
understanding of the spatial variability of the soil proper-
layer, which would otherwise become the same thickness as
ties (e.g. Samonil et al., 2010; Gabet and Mudd, 2010). The
the underlying layers if the mass were distributed equally.
average of the first method gives the second method.
The 1 in the diagonal term ensures that the grading in that
Because both these methods independently examine each
layer is entirely replaced by the new grading to negate the
node, the user can impose spatial variability on tree throw
effect of the identity matrix in Equation (7.42).
reflecting that tree throw is spatially clustered around the
When modelling soil spatially it will often be the case
edges of forests and in openings within the forest, and that
that a single tree throw event disturbs only a proportion of
throw events are clustered in time during high-wind events.
the area associated with each computational node. If we
In this case the user can construct a wind event model, where
call this proportion α, then the equation for that node is a
wind is randomly generated across the domain, and based on
modification of Equation (7.42) so that
the wind speeds impose a spatial pattern of tree throw occur-
g ¼ ðð1 αÞI þ αBÞg (7.44) rence (e.g. Espirito-Santo et al., 2014) and study its effect on
═after ═before
the spatial variability of soil development (Finke et al., 2013).
where the ð1 αÞ term is the proportion of the nodal area
that is unchanged by the tree throw event while the α term is 7.3.5.3 Termites and Ants
that proportion of the nodal area impacted by tree throw. The Termites and ants move fine soil material from the sub-
parameter α might change from tree throw event to tree surface to the surface as part of nest construction (Paton
throw event reflecting the size of the tree (and thus the root et al., 1995; Zaitlin and Hayashi, 2012). The surface
ball) impacted. Likewise the depth of mixing might change expression of these nests is then, over time, eroded (typ-
between events to reflect tree throw for different size trees. ically by rainsplash) so that the material moved to the
When modelling the temporal effect of tree throw (i.e. surface of the nest is then spread as a surface veneer over
over a number of years rather than over a single tree throw the soil surface, over an area that might be larger than that
event), there are two main alternative approaches. contained by the nest itself. The underground galleries of
The first approach is to model tree throw as occurring the nests are the material that is moved to the surface and
randomly in time and space (i.e. a Poisson random pro- may be quite extensive laterally (e.g. termite nests may
cess) and modelling every tree throw event using the extend 10’s of metres laterally) and vertically. Neither the
equations above. If tree throw has a probability of occur- lateral nor vertical density of the galleries is well defined
ring of p per unit area per timestep, then the probability P (and likely to be species, climate and soil specific), nor is
of a tree throw event in a given timestep in a node of area A the spatial distribution of nests (all of which determine the
is P ¼ pA: The simplest way to simulate this is to cycle mass of material moved). This also applies to the length of
through every node at each timestep and randomly generate time before the galleries collapse and the soil above them
subsides (which determines the rate of change of the por- layers, and the layers below are untouched. Then material
osity of the soil in those layers that contain the galleries). will be excavated from layers 2 through 4 and deposited
The general principle here is that soil with a given size into layer 1. To ensure that the mass in each of the layers
grading range (though the evidence for size selectivity for is unchanged, we then need to transfer the excess material
ants is not definitive, and nest-building behaviour appears to in top layer down into layer 2, from layer 2 to layer 3 and
be strongly species dependent; Lobry de Bruyn and Cona- so on. (Figure 7.9). The mass and grading of the material
cher, 1990, 1994) is removed from each subsurface layer and brought to the surface is
deposited on the surface. To ensure that the mass in each
layer does not change, we transfer material down from each X
4
Bφi Ag (7.50)
overlying layer into the layer below until we reach the bottom i
i¼2
of the bioturbed region. First, we will look at how to calculate
the amount and grading of material transported to the surface The fluxes of soil between the layers are given in
by the ants. So for layer i the material removed is Figure 7.9 and are based on a fully explicit numerical
representation of the process where each of the soil layers
Δg ¼ Bφi Ag (7.46) is fully mixed. Importantly in Figure 7.9 the grading of the
i i
where Δg is the change in the soil grading vector in this material from one layer to the layer below is based on
i the grading of the layer prior to the redistribution of the
time step, B is the rate constant determining the total
amount of material transported to the surface (not to be soil into the layer above. This is the explicit numerical
confused with B the matrix), φi is a normalised function of approximation of the time discretisation. There is no
depth describing the relative rate at which material is conceptual reason why the derivation can’t use the
transported from each layer i (i.e. φi is the depth depend- grading in the layer above after the soil has been redistrib-
ency of ant-turbation) and matrix A is a normalised matrix uted to it; the arithmetic in the equation below is just more
describing the size selectivity of the ant-turbation mech- complex.
anism. An example of matrix A is We can write the equations for the new grading in each
of the four layers. Starting at the surface and working
2 3
a1,1 0 0 0 0 down the profile,
6 0 a2,2 0 0 0 7
6 7 0 1
6 0 0 a3,3 0 0 7 X
4
A¼6 7 (7.47) Δmi
6 0 0 0 0 0 7 B C X
6 7 B i¼2 C 4
4 0 0 0 0 0 5 g ¼g þ0B
B
Cg þ
C 1 Bφi Ag
.. .. .. .. .. . . 1,new 1
@ X
4
A i
. . . . . . m1 þ Δmi
i¼2
i¼2
where the diagonal elements aj,j are the proportion of the 0 1 0 1
X
4 X
4
total mass in grading size range j that is transported B ΔmiC B C Δmi
upwards by the bioturbation (i.e. aj,j ¼ 0 if nothing is B i¼2 C B C
g ¼g þB
B
Cg B i¼3
C 1 B
C
C
removed, aj,j ¼ 1 if it is all removed). In this example 2,new 2
@ X
4
A @ X
4
A
m1 þ Δmi m2 þ Δmi
a4,4 ¼ 0 because for grading size range 4 it is assumed
i¼2 i¼3
that none of that size range is transported upward (e.g.
g Bφ2 Ag
because it is too coarse). I have shown only the first five 2 2
0 1
rows and columns of the matrix for brevity. X
4
B Δmi C
It also turns out to be convenient to define the total mass B i¼3 C Δm4
(i.e. across all grading size classes) in layer i, mi, and the total g ¼g þBB
Cg
C 2
3,new 3
@ X
4
A m3 þ Δm4
mass of the change in the soil grading vector for layer i, Δmi , m2 þ Δmi
X i¼3
mi ¼ gi,k (7.48) g Bφ3 Ag
3 3
k
X
Δm4
Δmi ¼ Δgi,k (7.49) g ¼g þ g 0 Bφ4 Ag (7.51)
k
4,new 4 m3 þ Δm4 3 4
where the subscript i indicates layer i and the subscript k is where the first term on the right-hand side of each equa-
the kth grading class in layer i. tion is the grading before ant-turbation, the second term is
For simplicity of explanation let us consider a case the soil being transferred down from the layer above, the
where the ant galleries are excavated in the top four third term is the soil being transferred down to the layer
FIGURE 7.9: Schematic of how layers interact in ant and termite bioturbation. While the calculations are done using mass, the figure shows the
layer thicknesses: (a) the soil profile before ant bioturbation, including the thin surface armour layer, (b) the movement of material by ants, from the
top three layers into the surface armour layer (those layers above the thick line are subject to ant transport, those below not; the grey region is the
surface armour layer; note that the only layer that changes thickness is the surface armour layer, so the bulk density in the lower layers is reduced), (c)
the redistribution of excess sediment from the surface armour layer (the thickness of the lines and arrows indicate quantities of soil being moved
between layers) and (d) the reinstatement of the armour layer, which is now a mix of the original armour layer and the ant transport material. The soil
surface has risen because of the reduction of the bulk density of the layers from which ants removed materials.
below and the fourth term is the material moved by the processes in all four equations. Constructing the transition
ants/termites. The zeros in two of the equations are pla- matrix from these equations involves filling the matrix
ceholders so that the third and fourth terms are the same according to Equation (7.51):
2 0 1 3
X
4
6 B Δmi C 7
6 B i¼2 C 7
6 B
6 B
CI
C Bφ2 A Bφ3 A Bφ4 A ½0 7
7
6 @ X 4
A 7
6 m1 þ Δmi 7
6 7
6 i¼2 7
6 0 1 0 1 7
6 X 4 X4 7
6 Δmi C Δmi 7
6 B B C 7
6 B i¼2 C B C 7
6 B CI B i¼3 I þ Bφ2 AC ½0 ½0 ½0 7
6 B X 4 C B X 4 C 7
6 @ A @ A 7
6 m1 þ Δmi m2 þ Δmi 7
6 7
B¼6 i¼2
0
i¼3
1 7 (7.52)
6 7
6 X4 7
6 B Δm C 7
6 B i¼3
i
C 7
6 B CI Δm4 7
6 ½0 B C I þ Bφ A ½0 ½0 7
6 @ X 4
A m3 þ Δm4 3 7
6 þ Δm 7
6 m 2 i 7
6 7
6 i¼3
7
6 Δm4 7
6
6 ½0 ½0 I Bφ4 A ½0 7
7
6 m3 þ Δm4 7
6 7
6 ½0 ½0 ½0 ½0 ½0 7
4 5
.. .. .. .. .. . .
. . . . . .
where, for brevity, only the first five rows and columns of approach used in tree throw (Equation (7.43)) seems
the transition matrix are shown. appropriate.
7.3.5.4 Gophers, Wombats and Prairie Dogs

Burrowing animals also move material from within the 7.4 Sediment Deposition
profile to the surface. There are strong similarities
between the ant/termite representation in the previous Sediment deposition was discussed in Chapter 4 in the
section and that for larger burrowing animals. The main context of the sediment in the flow and how it impacts on
differences are (1) the size and extent of the tunnels, (2) transport capacity. In this section we discuss what
the size selectivity of the material excavated and (3) that happens when that sediment is deposited. As discussed
there tends to be more lateral (either across or up/down previously there are two aspects to sediment to be
slope) movement of material because tunnels are predom- deposited: (1) the amount to be deposited at that point
inantly horizontal. and (2) the grading that will be deposited at that point.
In North America pocket gophers are an important The grading changes due to selective deposition are dis-
mechanism for laterally moving soil (Gabet, 2000) cussed in Chapter 4. Here we simply take the grading
impacting on vegetation (Huntly and Inouye, 1988). vector that gives us the mass of the deposited material, g ,
d
Their burrows extend up to 100 m laterally though they in each of the size fractions (as derived in Chapter 4), and
P g
do not dig very deep (<1 m). It thus seems reasonable the total mass of deposited sediment is md ¼ i d,i
to assume that they would also mix the soil vertically where the subscript d indicates that the grading vector is
(Hole, 1981), though there is less quantitative literature the deposited material.
on this possible impact (Gabet et al., 2003). This problem bears strong mathematical parallels
In a review of bioturbation in North American soils with the bioturbation by ants, where material was
Zaitlin and Hayashi (2012) concluded that prairie dogs deposited on the surface by the ants. The difference is
had a significant vertical mixing effect down to about 2 m, that there is no balancing loss of mass from the layers
but little impact on hillslope movement because they underneath. The deposition stage of the process is for
prefer to live in flat areas. the surface layer
For other environments we have less data. Field and
g ¼g þg (7.53)
Anderson (2003) provide comparative figures for volumes after before d
of soil bioturbed per hectare for an Australian catchment.

This sediment then needs to be distributed down the
Tree throw was the most significant mechanism, but
profile. Because this sediment is an addition to the profile,
bioturbation by wombats was the second most important
the soil profile will become deeper:
and of a comparable magnitude to termites.
In addition to tunnel digging there was also the 2 md 3
I ½0 ½0 ½0 ½0
impact of surface foraging where the surface soil is 6 m1 7 2 3
6 m 7 I
6 d md 7
mixed by the digging for roots and leaves. Mitchell 6 I I ½0 ½0 ½0 7 6 7
6 m1 m2 7 6 ½0 7
(1985) (in Paton et al., 1995) found that the soil volume 6
6
7
7
6 7
md md 6 7
moved during foraging (down to depths of 15 cm) by 6 ½0
6 I I ½0 ½0 77 6 ½0 7
6 m2 m3 7 g þ6 7
g ¼6 7═ 6 7g
t 6 ½0 7 d
wombats was 20–50 times the volume moved per year md md
7
═tþ1 6
6 ½0 ½0 I I ½0 7 6 7
by tunnel digging. The depth of disturbance by 6 m3 m4 7 6 ½0 7
6 7 6 7
wombats by tunnelling (down to 2.4 m) was signifi- 6
6 ½0 ½0 ½0
md md
I I 7
7 4 5
6 m4 m5 7 ..
cantly higher than tree throw (down to 0.42 m), sug- 4 5 .
gesting a greater impact from wombats than tree throw .. .. .. .. .. ..
. . . . . .
on vertical mixing of soils. (7.54)
In the absence of more definitive data it is suggested
that the vertical mixing of the profile from burrowing where the definitions of mi are the same as that used for
animals should be handled with the same approach Equation (7.48). The diagonal and off-diagonal terms
as for ants and termites (Equation (7.52)). The one repeat all the way to the bottom corner (i.e. bottom
difference is that there is likely to be less grain size layer) of the matrix because the deposition pushes down
selectivity for the burrowing animals because of the the material in all layers by an amount equal to the mass
animals’ large size. To model the surface foraging of deposition. The second term in the equation in g
d
where the surface materials are mixed together, the simply ensures that all the deposited material goes into
7.5 Differentiating between Mineral Components 115
2 3
the top layer and nowhere else during that timestep. The g1,Q
similarity with the erosion equation in Equation (7.17) 6 7g2,Q
6 7
6 7 ..
should be clear. 6 7 .
6 7
The preceding discussion makes no assumptions about 6 g 7
6 m1,Q 7
the process driver of deposition. Classically it is sediment 6 g 7
6 m,Q 7
being deposited out of fluvial transport. It could also be 6 7
67
6 g 7 2 3
from aeolian transport (Cohen et al., 2015). In this case 6 1,K 7 gQ
6 g 7
(aeolian) deposition and fluvial (erosion) can be occurring 6 2,K 7 67
6 .. 7 6 7
simultaneously. Cohen showed that if the aeolian material g¼6 7 ¼ 6 gK 7 (7.55)
6 . 7 6 7
6 7 4
is very fine it can enhance erosion and destroy the armour 6 gm1,K 7 5
6 7 gB
that might have formed in the absence of aeolian depos- 6 gm,K 7
6 7
ition. No matter what the grading of the aeolian material 67
6 7
is, it changes the characteristics of the armour layer and 6 g1,B 7
6 7
potentially changes the fluvial erosion characteristics of 6 g2,B 7
6 7
6 .. 7
the surface. 6 . 7
6 7
4 gm1,B 5
gm,B
7.5 Differentiating between Mineral
so that using m size fractions the vector will be 3m in
Components
dimension. The horizontal dotted lines in the vector of
The discussion above has been about differentiating the Equation (7.55) are added simply to highlight the organ-
behaviour of the different fractions within the soil isation of the vector with the grading of the first mineral
solely on the basis of size. However, the state-space first, the second mineral second and the third mineral last.
matrix approach can be easily extended to include dif- Thus the full g vector is split into three parts with the first
ferentiation based on some other property of the soil. third being the gQ grading vector for the Q mineral, gK the
The example we will use here is mineral fractions (e.g. grading vector for the K mineral and gB the grading vector
quartz, k-feldspar, biotite etc.), but it will become clear for the B mineral. This organisation first by mineral and
that the techniques outlined below are suitable for label- then by grading will be convenient shortly. The A matrix
ling any characteristic of the soil (e.g. particle density), will also be correspondingly bigger and will be of dimen-
or for explicitly modelling nonmineral fractions (e.g. sion 3m 3m. While the vector and matrix are larger, the
particulate organic matter). Finally the terminology of formulation of the A matrix is no more difficult than it
referring to the g state vector as the grading vector, was for the nonmineral case. For physical weathering
while convenient so far, will become a bit misleading there is no transfer of mass between the mineral fractions,
because the state vector will now contain other, add- so each mineral fraction is mass conservative in its own
itional, information about the soil, so we will adopt right, and then for each mineral fraction we can write the
terminology used in the state-space literature and here- weathering transition matrices
after refer to the grading vector g as the soil state gQ ¼ ðI þ AQ ÞgQ
vector. tþ1 t
In the previous sections the soil state vector g con- gK ¼ ðI þ AK ÞgK
tþ1 t
tains the mass of material in each size range without gB ¼ ðI þ AB ÞgB (7.56)
tþ1 t
regard to what the characteristics of that material are
other than its size. Consider the case where we have where the A matrices are the matrices that would apply for
three different minerals in the soil that we wish to that mineral if that mineral was 100% of the mass content
distinguish (e.g. perhaps they weather at different rates of the soil layer. We can then write the matrix equation for
and we would like to track the effect of these different the full soil state vector for a single soil layer
weathering rates on the soil grading as it evolves). Let’s 2 3
ðI þ AQ Þ ½0 ½0
call these minerals Q, K and B to distinguish them. We 6 7
g ¼ 4 ½0 ðI þ AK Þ ½0 5g (7.57)
can then extend our definition of g so that it now gives tþ1 t
the mass in each size range for each of the three min- ½0 ½0 ðI þ A B Þ
erals for each layer in the soil profile. The vector will The zero off-diagonal matrices indicate that none of the
then be mineral fractions transform into a different mineral
fraction during weathering. For physical weathering that specific gravity particles (e.g. high in iron) from midrange
is the expected behaviour, but for chemical weathering specific gravity (e.g. silicates), and low specific gravity
this may not be the case. If, for instance, one of the particles (e.g. organic matter and soil aggregates).
mineral fractions is a chemical weathering product of
another fraction, then there will be off-diagonal matrices
7.6 The Evolution of Porosity
that parameterise the rate and grading characteristics of
that weathering process. We can show how this would The layers in the matrix model described above are defined
work in principle by extending our example in Equation based on the mass per unit area in the layer. To convert these
(7.57). Imagine that mineral component Q is a weathering masses per unit area to a depth of soil we need to use the
product of mineral K, then it is possible to write a matrix bulk density of the soil in that layer. If the density of
equation relating the effect of weather from K to Q on the particles doesn’t change, then this is equal to the change
grading of both Q (the destination of the chemical in the porosity. The main drivers of changes in the density
weathering) and K (the source for the chemical of particles are chemical weathering and the amount of soil
weathering) so that carbon. In general the porosity of the soil will change over
2 3 time simply from the change in the grading and the particle-
ðI þ AQ Þ ðI þ AQK Þ ½0
6 7 packing arrangement that is possible with that grading. But
g ¼ 4 ðI þ AKQ Þ ðI þ AK Þ ½0 5g (7.58)
tþ1 t porosity will also change with macropore development and
½0 ½0 ðI þ AB Þ
changing mineral constituency. If either of these latter
where the transition matrix AQK calculates how much of properties is being tracked, then these effects can be mod-
mineral K is transformed into mineral Q, what the grading of elled. For instance, the bioturbation by ants discussed above
the source material K was and how it is transformed into the involves the removal of fine materials from a layer leaving
grading of the destination mineral Q, while AKQ calculates behind voids between the large particles, and will accord-
the inverse of how K is transformed into chemical ingly increase the porosity of the layer from which material
weathering product Q. In general this chemical weathering is being removed. Note that by defining the soil state vector
transformation will not be mass conservative with respect to by mass, this porosity change does not appear in the matrix
the soil components because the chemical reactions will equations, and appears only at the end when a conversion
typically involve constituents and reactions that have not from mass of soil to depth of soil is required.
been tracked in the matrix equations such as hydration (i.e. For a granular media with two size particles (e.g. a
with water), oxidation (i.e. with oxygen) and carbonation binary media) a relationship due to Fraser (1935) and
(i.e. with carbon dioxide). All these reactions will change the Clarke (1979) based on the percentage of the two sizes
mass of the soil mineral constituents and can be accounted is commonly used. This equation has been used (e.g.
for with AQ , AK and AB . We will focus on chemical Morin, 2005) and studied by a number of subsequent
weathering in detail in the next chapter (Chapter 8). researchers (Zhang et al., 2009). For a granular media
In a model for the evolution of organic matter within made up of a range of particle sizes, a number of empirical
soils Kirkby (1977) proposed a similar method using extensions of this model (see a comparison by Tranter
matrices to differentiate the organic fractions (carbohy- et al., 2007) can be used to estimate the porosity, ϕ,
drates, amino acids and lignins) and mathematically including (Koltermann and Gorelick, 1995)
described how they transformed from one fraction to
(
another. He also proposed a specific structure for the ϕc cy1 1 ϕf þ ð1 y1 Þϕf c < ϕc
transition matrices for the different types of soil humus, ϕ¼ (7.59)
ϕc ð1 y2 Þ þ cϕf c ϕc
organic sourcing from leaf fall, and examined what the
long-term equilibrium organic content of the soil was. where
Finally, we note that this section has focussed on the
y 1
weathering process and differences in weathering rates, y1 ¼ c min þ1
ϕc
but this approach can also be taken with erosion and
1 ymin
armouring processes. For instance, it is common for agri- y2 ¼ ðc 1Þ þ1
1 ϕc (7.60)
cultural erosion models to distinguish sediment classes ϕ
based on their relative transportability (e.g. CREAMS, ymin ¼ 1 þ ϕf min
ϕc
WEPP; Knisel, 1980; Laflen et al., 1991). This transport-
ϕmin ¼ ϕc ð1 ymin Þ þ ϕc ϕf
ability is a function of the diameter of the particles and
their specific gravity, so these erosion models are poten- which is applicable for gradings where there are two dis-
tially capable of distinguishing erosion rates for high tinctly different size fractions (e.g. sand/clay, gravel/clay)
7.6 The Evolution of Porosity 117
where ϕf and ϕc are the porosities of the fine and coarse rock fragments that result from physical weathering will
fraction, respectively, and c is the volume fraction of lead to a decrease in bulk density. This bulk density
fines. Koltermann and Gorelick showed that this equa- increase will be reflected in an increase in the porosity.
tion fit data better than the ‘perfect packing’ model Thus if the fragmented rock particles from physical
(Clarke, 1979). weathering have a porosity of 0.3 and the saprolite has
The Koltermann and Gorelick model is only for a porosity of 0.0, then the bulk density will decrease
unstructured granular material and does not account for by 30%, and the volume of the space consumed by
soil structure effects and organic matter. Tranter et al. these rock fragments will increase by 43% after
(2007) compared a number of empirical equations for fragmentation.
the fraction of the soil smaller than 2 mm (i.e. sand and • The rock fragments themselves may change in their
finer) and found the best fit for the bulk density of the density. Chemical weathering may eat away at the
mineral fraction of the soil was inside of the rock fragments without changing the
exterior geometry of the particles. This will reduce
ρbðmineralÞ ¼ 1:35 þ 0:0045 ps 6 106 ð44:7 ps Þ2 the bulk density of the soil layer because there is no
þ 0:06 depth change in the volume occupied by the particle con-
(7.61) sumed. Thus there will be no change in the volume of
the layer even if there is a change in the mass within
where ps is the percentage of sand by mass, depth is the that layer. If, however, chemical weathering removes
distance below the soil surface in metres and ρb is in units only the outermost rinds of the particles so that the
of g/cm3 or T/m3. To allow for organic matter in the particles get smaller and the soil particles can
profile it was found sufficient to adjust for its mass frac- rearrange themselves to adjust to this change in
tion (Stewart et al., 1970) sizing, then the only change in bulk density will be
100 that given by Equation (7.59) (i.e. only reflecting any
ρbð<2mmÞ ¼ (7.62) change in the grading of the soil). These cases will be
OM 100 OM
þ discussed in detail in Chapter 8 when chemical
ρbðOMÞ ρbðmineralÞ
weathering is discussed.
where OM is the percentage of organic matter by mass,
and where Tranter used a bulk density of organic matter of At one field site Anderson et al. (2002) found the
0.224 g/cm3. Finally, Vanwalleghem et al. (2013) bulk density of the soil at the base of the soil profile
extended this to include the rock fraction (i.e. particles decreased by more than 50% relative to the parent rock,
greater than 2 mm) by (Vincent and Chadwick, 1994) and ascribed the cause to be a variety of physical
weathering and bioturbation (e.g. tree throw)
100 mechanisms. They also performed a chemical mass
ρb ¼ (7.63)
p<2mm 100 p>2mm balance and concluded that an upper bound on the
þ
ρbð<2mmÞ ρbð>2mmÞ dissolution of minerals could account for only a max-
imum of a 5% decrease in bulk density. At this site they
Tranter noted that the root mean square errors of
concluded that the effect of chemical weathering on
both Equation (7.61), and Equations (7.61) and (7.62)
bulk density is small.
combined, was about 0.15–0.18 g/cm3, or about
Finally, in an interesting observation Brimhall et al.
10–15%.
(1992) noted the possibility that macropores generated
However, we still find issues specifically related to
by biological activity may fill with deposited material in
the soil history and weathering process that are
the long term. They mention tree roots and speculate
neglected:
that after the decay of the roots the empty macropore
• While the Tranter comparison examined soils with sig- may fill with mineral or organic matter. Likewise ant
nificant soil structure, none of the equations in this and termite galleries might not collapse (and thus might
section explicitly allow for the effect of soil structure not return the soil bulk density to near its pre-
(as a result of either clay or organic matter content) on bioturbation value) but may fill up with sediment
bulk density. Tranter et al. recognised that as a major instead. It is certainly clear that macropores are signifi-
limitation, but noted that soil grading alone was prob- cant preferential pathways for water flow in the soil
ably not enough to do better. (Beven and Germann, 1982, 2013), and it seems rea-
• The transformation from the native rock in the saprolite, sonable to suggest they would also carry sediment that
which is by and large unfragmented, to the fragmented could fill the macropore in the long term.
7.7 Numerical Issues within the layer. The mARM model defined the layers
on thickness, but we have subsequently found that this
Most models break the soil into a series of layers as definition makes formulation of the transition matrices
discussed above. Many of them fix the thickness of, or complex, particularly if the layers are different thickness,
mass in, the layers. With the exception of mARM, most so we advise using mass as the definition of the layer. This
do not do anything special with the uppermost layer. They means that you need to have a model for porosity so that
simply consider that erosion is a loss of material from the this mass can be converted to soil depth, but this is
top layer and don’t model the change in the grading that implicit in the soil thickness formulations.
results in armouring. MILESD (Vanwalleghem et al., A related issue is how to define the size fractions
2013) uses a different vertical discretisation for the layers, within the layer. In mARM we defined them on the basis
modelling just three layers, corresponding to soil hori- of the proportion by mass within each size fraction. This is
zons, and models the evolution of the thickness and consistent with how particle size distributions are nor-
internal characteristics of these layers. This has the advan- mally expressed in the soil science community. However,
tage of allowing them to specialise the physics and chem- subsequent experience has been that formulation using the
istry to what happens in each of the horizons, but has the mass (rather than proportion) in each size within each
disadvantage of not allowing them to investigate what in layer is computationally and conceptually easier, and
the physics and chemistry quantitatively drives the devel- consistent with the mass per layer formulation.
opment of the horizons in the first place. Subsequently, Defining each layer by mass avoids another issue,
Temme et al. (2016) implements a series of processes that which is whether to define the layer boundaries by (1)
have no regard for the layers themselves but only the elevations relative to the soil surface (which itself is
depth of the layers below the surface. changing as a result of the erosion and landform evolu-
The main numerical issue is the coupling of the surface tion) or (2) absolute elevations (e.g. above sea level).
layer in the physical model with the erosion mechanics. Defining the layers with mass avoids the need to define
A key feature of the surface is that if there is a coarse the elevation of the boundaries in the soil model except
weathering-resistant component of the soil, then a surface when weathering rates need to be determined relative to
armour will be formed. This armour consists of the material the surface, in which cases all layers are relative to the
that cannot be moved by the overland flow, and its exist- evolving soil surface.
ence reduces the erosion rate. Typical armour layer depths If porosity is changing, then the thickness of each layer
in the field are of the order of 1–2 times the diameter of the will naturally change relative to how much mass is in each
armour layer, so it is quite thin relative to the depth of the layer. It is then conceivable that the soil surface elevation
soil. Thus we have found that you need a thin surface layer will rise if the right combination of changes in porosity
to correctly capture the armouring process. Thicker layers and mass within all the layers occurs. For instance, con-
then underlie this, so that there aren’t an unnecessarily sider the case of the dilation of a soil due to tree throw, but
large number of layers within the profile. where erosion is low. The porosity increase as a result of
In the existing models of physical weathering some the tearing out of the saprolite of the rock will inevitably
have defined the layers based on thickness of the layer, lead to a soil thickening and a rise in elevation of the soil
while others have defined them on the basis of mass surface.
8 Soils: Chemical Weathering
8.1 Introduction 8.2 Chemical Weathering Principles
Chemical weathering is the chemical transformation of The current approach to measuring and modelling
minerals in the saprolite and soil (collectively these the cumulative impact of soil chemical weathering is
minerals form the soil matrix) from one form to another dominated by the idea of quantifying the proportion of
under the action of (typically but not always) solutions the saprolite chemistry that remains unaltered in the soil.
that contain water and dissolved acids. The reactions A lesser percentage of unaltered material indicates a more
between the solution and the minerals occur at the matrix weathered profile. This would be a relatively straightfor-
surface and can result in (1) dissolved products, (2) ward measurement to make were it not that a significant
precipitates and (3) in situ transformation of the (pri- amount of material is removed as dissolved load so that
mary) minerals into secondary minerals. The dissolved the mass of soil matrix is reduced over time, so that
products typically end up in groundwater and surface minerals resistant to weathering can actually increase their
water and exit the catchment as part of the dissolved percentage content over time. The percentage mass of
load of the rivers and streams (of the order of 20% of elements or minerals resistant to weathering (mostly Ti
the mass transport into the oceans globally; see Section and Zr but others have been suggested) can be used as
4.2), though they can be precipitated elsewhere in the fixed datums against which to compare concentrations of
soil profile and/or the broader catchment if the chemical other minerals and their degree of weathering. In this way
environment changes (e.g. pH changes). The precipitates corrections for dissolution and soil collapse or dilation can
may end up coating the existing matrix particles (and be made. Thus, dissolution complexities aside, the main
thereby inhibiting further reactions), or they may form method for assessing progress of weathering is to examine
cement between particles (increasing the strength of the the percentages of minerals or chemical elements
soil). The rate of the reactions can be a function of many (Brimhall and Dietrich, 1987; Chadwick et al., 1990;
factors, but the most important are (1) the surface area of Brimhall et al., 1992; Riebe et al., 2003) in the source
the matrix exposed to the solution and (2) the difference rock, saprolite and soil and calculate the proportion trans-
in the concentration in the solution near the matrix formed as an indicator of weathering. This is the approach
surface between the equilibrium concentration of the used by most modellers and experimentalists and which
chemical components (both the inputs and outputs of will be presented here. Unfortunately, as compared with
the chemical reactions) and the actual concentration of the approaches in Chapter 7, this does not directly provide
the chemical components. The matrix is heterogeneous us with information about the soil particle size distribution.
with a range of minerals and different exposures of these Before we discuss the modelling of chemical
minerals to the solution, and the different minerals will weathering, a brief summary of the current understanding
react at different rates. Furthermore, mineral impurities of the reaction processes and issues to do with their
are concentrated at grain boundaries making the grain parameterisation is warranted. This will also provide some
boundaries relatively more reactive than the grain proper context for the sensitivity studies performed later in this
so there are differences in the rates of reaction at grain chapter.
boundaries relative to the grain interior leading to The reaction rate is a function of the area available for
marked increases in surface area for relatively little interaction with the fluid and the concentration of acid in
change in grain size. the fluid so that 119
M ¼ R SA t (8.1) particle diameters were between 0.25 mm and 0.85 mm.
Using these data, Equation (8.2) yields SSAsphere of
where M is the cumulative number of moles of mineral
0.027–0.090 m2/g with a value of 0.041 m2/g for the
reacted per unit time, R is the weathering rate, SA is the
mid-range value of 0.55 mm. For the freshly ground-up
surface area of the reaction and t is time (White and
rock their measured SSA was twice that of an equivalent
Brantley, 2003). The acid is normally carbonic acid in
sphere, while for weathered rock it was as much as 100
rainwater, where atmospheric carbon dioxide has equili-
times larger (i.e. the particle roughness was about 100).
brated with the rainwater, or from carbon dioxide gener-
Furthermore as weathering progressed the SSA and
ated within the profile by decomposition of soil organic
roughness increased quite markedly over a relatively short
matter and by plant respiration (Section 14.3). Acid rain
time. The reason for this increase in SSA over time is
caused by atmospheric pollution and dominated by sul-
increased surface roughness of the particle surface
phate and nitrate species, however, has been a potential
resulting from pitting and preferential weathering along
factor in the last few decades (e.g. de Vries et al., 1995).
grain boundaries (which tend to be more reactive than the
West (2012) analysed field silicate weathering data that
central part of the mineral grain).
suggested that the reaction rate is nonlinearly temperature
White and Brantley (2003) also compared particle
dependent with the rate of reaction not only increasing as
roughness for numerous field studies of different minerals
temperature increases, but the rate of increase also
and weathering environments (Table 8.2). They found a
increasing with increasing temperature.
similar trend where weathered materials had a higher
In everything that follows we will be using mass
surface roughness than fresh materials. The roughness
balances, so it should be noted that moles can be con-
for biotite, which has platy grains rather than spherical,
verted to mass using the molecular weight of the mineral:
was much higher than the other minerals. White et al.
mass ¼ Mðmolecular weightÞ.
(1996) found a trend in roughness with soil age for soils at
The surface area is normally parameterised as the
a Merced site. Their roughness started at about 20–30 for
specific surface area, SSA, which is the area per unit
soils about 3,000 years old, and increased to 50–600 for
volume or unit mass of rock. Unless otherwise noted,
soils about 3 million years old. White and Brantley (2003)
SSA will be defined here per unit mass to allow direct
derived an equation for surface roughness with time at
comparison with experimental data. Let us first consider a
their site for silicate minerals
crushed rock consisting of only one mineral. The simplest
approximation for surface area is to assume all particles λ ¼ 13:6t 0:20 (8.3)
are spheres, which yields a specific surface area of
where t is time in years. Using this equation and infor-

SSAsphere ¼ 6 dρ (8.2) mation about how the grain size grading is changing with
time allows the calculation of SSA.
where d is the diameter of the particles and ρ is the density Pardini (2003) fitted a fractal model to soil particles
of the rock particle. This, however, is a poor approxima- along a chronosequence and found that the fractal dimen-
tion for platy minerals such as micas. sion also increased with the age of the soil, which is
Commonly SSA is determined experimentally using consistent with an increasing surface roughness.
the BET technique. The surface roughness of a particle, λ, The second variable in the reaction rate is the rate of
is the ratio of the measured SSA divided by SSAsphere . reaction per unit surface area for a given concentration of
White and Brantley (2003) presented SSA data from acid in the water. As for the SSA, there are significant
their weathering experiments (Table 8.1). They crushed
their granite source material and then sieved it so the
TABLE 8.2 Typical values for weathered particle surface
TABLE 8.1 SSA changes due to weathering (from White and roughness λ for different minerals and different environmental
Brantley, 2003) conditions (from White and Brantley, 2003)
Time from start of Fresh granite Weathered granite Mineral Laboratory Field
experiment (years) SSA (m2/g), (λ) SSA (m2/g) (λ)
Plagioclase 2–20 40–1,000
0 0.084 (2.0) 3.85 (94) K-feldspar 30–150 50–700
1.66 0.126 (3.1) 3.46 (84) Hornblende 3–50 50–250
7.21 0.298 (7.3) 5.33 (130) Biotite 50–150 200–3,000
8.2 Chemical Weathering Principles 121
TABLE 8.3 Typical values for log(reaction rate), log(R), for TABLE 8.4 Regression coefficients and standard error for
different minerals and different environmental conditions (from reaction rates with time (from White and Brantley, 2003)
White and Brantley, 2003)
Mineral a b
Laboratory Field
Mineral (moles/m2/s) (moles/m2/s) Plagioclase 12.5 0.2 0.56 0.05
K-feldspar 12.5 0.3 0.65 0.08
Plagioclase 12.6 to 10.6 17.4 to 12.1 Hornblende 12.7 0.2 0.62 0.07
K-feldspar 12.8 to 10.5 17.8 to 12.9 Biotite 12.3 0.3 0.60 0.07
Hornblende 13.0 to 10.7 17.0 to 13.6
Biotite 13.2 to 10.7 17.5 to 12.9
of about one-third of the reduction could be due to errors
in SSA estimation (i.e. the exponent b in Table 8.4 might
differences in reaction rates between laboratory and field be as low as 0.41).
that are yet to be satisfactorily explained. White and The second possible explanation for the reduction with
Brantley (2003) found that the field rates were about 3–4 time is that fresh rock fragments used in laboratory experi-
orders of magnitude less than the laboratory experiments ments are covered in defects due to the fragmentation
using equivalent minerals (Table 8.3). Table 8.3 shows process (e.g. hammer mill crushing for laboratory experi-
that there is almost no overlap between the ranges of the ments), and that these defect sites are typically more
laboratory values and field values. Furthermore, for all reactive. With time as these reactive sites are removed by
minerals, the differences between the laboratory rates and weathering, the remaining mineral surface has fewer of
field rates are much larger than the differences between these highly reactive sites, and the reaction rate drops. This
the minerals. This latter observation will be relevant later explanation is also consistent with experiments in physical
when we construct our model, and consider possible ways weathering where a similar time reduction of fragmenta-
to simplify the modelling. tion rate has been found (e.g. Wells et al., 2008).
White and Brantley (2003) fitted equations for the The third possible explanation is due to the different
variation with time of the reaction rate for different min- environmental conditions in the laboratory experiments
erals of the form (typically in stirred batch reactors) and field experiments.
In the field there is wetting and drying as well as decaying
R ¼ atb (8.4)
organic matter. Experiments typically have a higher ratio
where the values of the regression coefficients a and b are of water to rock than in the field, so solute concentrations
listed in Table 8.4. The differences between the minerals are lower, and ‘transport-limited’ weathering (see later for
in this regression are well within the error bounds of the a chemical definition of transport-limited) occurs more
fit. Table 8.4 thus suggests that there is no statistical readily in the field, reducing field reaction rates.
difference in the rate of reduction with time of the reaction It should be clear from this very brief discussion of the
rate between the different minerals. Other authors have processes in chemical weathering that it is a very compli-
also observed a reduction in reaction rate with time (Van cated subject, and this is reflected in the literature at the
Grinsven and van Riemsdijk, 1992). current time. There is a considerable literature on the
The reason for the reduction in reaction rate with time pathways of chemical weathering, there are many unre-
is the subject of considerable debate. The first possible solved issues in the quantification of the processes.
explanation is that the reaction rate is calculated from Accordingly in this section we will provide a simple
chemical depletion assessments (e.g. Anderson et al., framework as a way forward in modelling chemical
2002) using Equation (8.1) to estimate the reaction rate. weathering and allow us to elucidate first order behav-
This technique uses the (1) age of the soil, (2) amount of a iours. This simple framework is based on a more compre-
mineral that has been transformed and (3) estimates of hensive geochemical transport model, CrunchFlow
SSA. Any error is SSA will propagate into the reaction (Steefel, 2009). For simplicity in the explanation the
rate estimate. Moreover, if not all the surface area is model below simulates only one mineral component
chemically reactive (e.g. inaccessible cracks that cannot rather than the many components that CrunchFlow can
easily exchange water with infiltration, surfaces covered simulate. The relatively small differences between mineral
with precipitates; Heister, 2016) then the SSA is a poor, chemical properties (Tables 8.28.4) suggest that the
upper bound, estimate of the chemically reactive surface results from the single-component model will provide
area. White and Brantley (2003) showed that a maximum valuable insight into the multi-component model.
The following sections will present simplifications and this section a full profile model is proposed, and some
generalisations to explicate the core underlying issues. simple sensitivity studies carried out. This will explicate
Having presented a framework we will then explore some some of the behaviours of these physically based models,
of the key findings of the geochemical weathering litera- and, as will be seen, allow some assessment of the
ture and how they can be used in the framework, while approximations entailed in some of the simpler chemical
referring the reader to the detailed geochemistry literature weathering models used in existing LEMs.
for the gruesome details of geochemical process quantifi- We begin the discussion of chemical weathering with a
cation of any specific rock and mineral. simple one-dimensional model for mineral weathering.
Finally there is a matter of some jargon, which will be While this model is a simplification of the field processes,
avoided here, but which is used in the geochemistry the simplicity of the model, and the surprisingly large
literature. Congruent weathering is weathering where spectrum of behaviour the model displays, will provide
the only products of weathering are dissolved, whereas understanding of the first order process physics.
incongruent weathering is where some of the weathering A schematic of the model is provided in Figure 8.1.
products are solid. The essence of the model is that acidic water (typically it
will be water with dissolved carbon dioxide but that may
8.3 Full Profile Models not always be the case as we will discuss later) introduced
from the soil surface reacts with the rock, to generate two
The model in this section breaks up the soil profile into products, one dissolved and the other the chemically
layers and explicitly models the interactions between transformed rock. Acid can also be generated in-profile
these layers to predict the time evolution of the layers. as a result of decomposition of soil organic matter and by
There are two approaches to the layer discretisation. The plant root respiration. The model is general enough to
first is the traditional approach used in numerical model- model congruent weathering; however, for simplicity in
ling, which is to simple break it up into a discretisation of this section we consider only incongruent weathering. The
(typically) equally spaced nodes that is (typically) fixed in rock is assumed to be chemically homogeneous so that
time and then model the dynamics between these nodes. there is only one chemical reaction occurring. We also
The second approach explicitly models the soil horizons
(i.e. O, A, B etc.) as the layers and allows the thickness of
these horizons, and their properties, to vary over time. The
advantage of the second approach is that it exploits a
known property of most soil profiles, the horizons. On
the other hand, the first approach allows us to observe
whether horizons arise naturally out of the model dynam-
ics, and if they do, then this is a strong validation of the
model dynamics (McBratney, personal communication).
8.3.1 A Simple One-Dimensional One-Mineral

Weathering Model
In constructing a soilscape model there is a trade-off

between a fully physically based model that explicitly
models processes through the profile, versus models
that (typically) model some conceptualisation of the
weathering for a soil as a single layer. The physically
based models are more computationally complex and
more data intensive, but are better able to (if not better,
then at least more explicitly) model the processes that
occur within the profile. On the other hand, within the
context of the large number of other complex interacting
processes in the LEM, the conceptual models are compu-
tationally simpler, and it is potentially easier to understand FIGURE 8.1: Schematic of the chemical weathering model.
their interactions with other components of the LEMs. In The number of layers used in model is user defined.
assume that there is only one dissolved product and one concentration, the second term is the advection term that
solid product. gives the rate at which the acid is moving down the
The physics of the model below follows that in the profile, and the third term is a diffusion/dispersivity term
CrunchFlow reactive transport model (Steefel, 2009). that defines the rate at which the acid spreads within the
CrunchFlow is a multicomponent (i.e. more than one profile. The right-hand side, R, is the reaction term dis-
mineral) model. This multicomponent capability compli- cussed below. Equation (8.6a) is a classic advection-
cates the model mathematics considerably, so to highlight diffusion/dispersion-reaction transport equation with
the key concepts the presentation below is simplified to advection occurring down the profile.
model only one component. CrunchFlow is missing one In the substrate, S, equation (Equation (8.6b)), the first
important process, which is the exhumation of fresh term on the left-hand side is the rate of change of the
saprolite from below, and the model here has been substrate, and the second term is the rate of exhumation of
extended to include this process. If the soil depth is the saprolite due to erosion. If the soil depth is constant
constant with time and erosion is occurring, then a conse- with time. then the rate of erosion, E, is equal to the rate of
quence is that an equal amount of saprolite exhumation exhumation of the saprolite into the soil profile. This
must occur to balance erosion. second term thus adds fresh substrate mass to the base
We can write the chemical equation for the weathering of the soil profile, exactly equalling the mass of material
reaction as lost by erosion at the surface. The term on the right-hand
side of this equation is the rate of transformation of the
aA þ S ¼ bP þ cL (8.5)
substrate. This is a classic advection-reaction equation
where A is the acid loaded water, S is the unweathered where advection is occurring up the profile. The only
rock (the substrate), P is the solid product and L is the difference between the right sides of Equations (8.6a)
dissolved product, which we will refer to as leachate. and (8.6b) is the constant of proportionality k 2 , which
Equation (8.5) is a mass balance equation, and the par- reflects the differences in the mass balance in Equation
ameters a, b, c are needed to balance the equation and (8.5b) so that
quantify how much mass of component is created per
1
unit mass of substrate. The parameters a, b, c will be k2 ¼ (8.7)
different for different mineral weathering reactions, so a
we will not be more specific at the moment, and this will The dissolved product, L, equation is not necessary
allow the exploration of the equation behaviour for the at this stage, but it is convenient to present it now; it
full range of parameter values. For congruent weathering will become important later. If the dissolved product-
b ¼ 0. loaded water leaves the profile unaltered, then this will
The governing equation we will solve for the chemical be the leachate from the profile. The velocity and
weathering reaction will be one-dimensional in the direc- diffusivity terms are assumed to be the same as for
tion of gravity, with the positive z direction vertically the acid solution A since they are both transported by
upwards: the same solution (and we assume they have the same
sorptivity potential so their advection velocities are the
∂A ∂A ∂2 A
þ v D 2 ¼ k 2 R same). The rate of reaction on the right-hand side is
∂t ∂z ∂z
the same except for the mass balance in Equation (8.5)
∂S ∂S
E ¼ R so that
∂t ∂z
(8.6) 1
∂L ∂L ∂2 L k3 ¼ (8.8)
þ v D 2 ¼ k3 R c
∂t ∂z ∂z
P ¼ k 4 ðS0 SÞ Note that in Equation (8.6) the value of L does not
α influence the values of A and S. Later we will extend
R ¼ k 1 A ðk 5 ðS Smin ÞÞ
these equations to cases where this is not the case. As
where t is time and z is distance from the bottom of the for the acid equation, this is a classic advection-diffusion/
soil profile (i.e. positive upwards). The velocity v is the dispersion-reaction equation with advection occurring
planar average velocity per unit area, called the specific down the profile.
discharge in the groundwater literature. The solid weathering product, P, is simply defined by
In the acidic water, A, equation (Equation (8.6a)), the the amount of the original rock that is weathered, where
first term of the left-hand side is the rate of change of the the amount of original rock before weathering is S0
resulting from the rock properties at the saprolite-soil substrate that cannot be weathered so that as the substrate
boundary, and the mass balance in Equation (8.5) so that approaches the value Smin the reaction shuts down. In
1 terms of physical processes this may represent substrate
k4 ¼ (8.9) that is not accessible to weathering. For instance, in arid
b
areas it is common to find minerals coated in an iron
This equation assumes that the solid weathering products
oxide, so they will not be accessible to environmental
move upwards through the soil profile (by exhumation) at
processes (this process is called occlusion), and the sur-
the same rate as the unweathered rock, S. Thus this equa-
face area associated with those minerals is not available
tion does not model cases where dissolved products may
for reaction. In the discussion below we will assume
be precipitated away from the site of the chemical reac-
α ¼ 1, though some authors have suggested other values.
tion. In the figures in this chapter the values of P are not
Solving this set of equations with time leads to an
plotted because of the simplicity of the relation between S
equilibrium balance between the reactive loss of S, the
and P, where trends in P can be derived by inspection
erosion of S from the surface and the addition of S from
from the plots of S.
exhumation. An example of this evolution is shown in
Finally the reaction term, R, is a combination of the
Figure 8.2, which uses the nominal parameter set in
acid concentration, A, and the reactive surface area of
Table 8.5. The general trends are the following:
substrate, k 5 ðS Smin Þ. Solid-fluid reaction rates are
determined by the reactivity between the two components, • The concentration of the acidic water, A, decreases
k 1 , the acid concentration in the fluid, A, and the SSA. from 1 to near 0 at the base of the soil profile. This
We will assume that the SSA is the same everywhere, so shows that as the acid reacts with the rock it diminishes
that the surface area dependency can be given by in strength.
SSA ¼ k 3 ðS Smin Þ. We will generalise this surface area • The substrate rock, S, becomes more weathered as you
assumption later. The value Smin is a threshold that serves approach the surface. In this example about 60% of the
several purposes. This represents a proportion of the rock is untransformed, while about 40% has been
FIGURE 8.2: Example output of the profile evolution from the chemical weathering model with the nominal parameters. The arrows indicate
the direction of evolution with time. For a detailed explanation of the panels see Figure 8.4.
TABLE 8.5 Parameter sets for the sensitivity study the sensitivity studies in terms of terminology used in the
chemical weathering literature.
Parameter Nominal Full depletion The first concept is ‘transport-controlled’ or ‘transport-
(Equation (8.6)) parameter set parameter set limited’ profiles. Equation (8.6) can reach an equilibrium
state where the concentrations on the right-hand side of
k1 0.1 0.5
the equation have reached a maximum value (e.g. for
k2 2.0 2.0
dissolution imagine a saturated solution where no more
k3 1.0 1.0
solid can be dissolved in the solution, L ¼ Lequilib ). As a
k4 1.0 1.0
thought experiment, imagine water is stationary within the
k5 1.0 1.0
profile. Once this equilibrium condition had occurred.
V 0.1 0.1
then no more weathering can occur because the
D 1.0e-6 1.0e-6
weathering reaction is limited by the concentration of
E 0.1 0.1
the weathering products in the solution. The only way
α 1 1
that the weathering reaction can continue is if some of the
Smin 0 0
solute (i.e. the leachate) is flushed out of the bottom of the
Lequilib in Equation 1,000 (0.4) 1,000 (0.4)
soil profile by the infiltration, in which case the reaction
(8.13) (Leachate
proceeds enough to replace the leachate lost from the
Control)
profile. A mass balance can be done on the entire profile
by noting that the mass lost from the profile at the
bottom is
transformed. The curve of unweathered substrate with
M ¼ vLequilib (8.11)
soil depth is sometimes called a ‘depletion curve’.
• The temporal trend at any given depth is that the rock where M is the mass of leachate leaving the profile, v is
starts off as being unweathered, and the weathering the velocity in Equation (8.6) and Lequilib is the equilib-
approaches an equilibrium profile. Likewise at early rium concentration of the leachate. For steady state condi-
times the acid is consumed more rapidly (because the tions this mass leaving the profile must be replaced by the
substrate and thus the reaction rates are high) so there is amount generated by the weathering reaction within the
less acid in the water at the base of the profile. The profile, so that the rate of weathering of the profile is
details of the transients are a function of the initial limited by the rate at which leachate is transported out
conditions and so are only indicative of likely behaviour of the profile, thus the name ‘transport-controlled’.
in the field. Whether a profile is transport-controlled is dependent on
• The nominal parameter set in this simulation has a very the relativity between the residence time of the water
low diffusion/dispersivity, D, so mass transport within (related to the velocity through the profile) and the time
the profile by water is primarily by the advection of the it takes for the weathering reaction to reach equilibrium. If
solution down the profile. the residence time is long or the rate of equilibration of the
reaction fast, then transport control will occur (e.g. White
The weathering profile in Figure 8.2 is calculated as
et al., 2008). In a later section we will extend Equation
the rate of change of the substrate:
(8.6) to examine transport control. Finally note that this
W ð zÞ ¼ R (8.10) definition of transport-controlled or transport-limited is
not related in any form to transport-limited sediment
Figure 8.2 is one example of the model behaviour, but transport.
it is clear that the log weathering function is not linear Another condition is the more obvious ‘rate-
with depth, so in this case weathering is neither exponen- controlled’, ‘kinetic-controlled’ or ‘weathering-limited’
tial nor humped with depth. profile (Brantley and Lebedeva, 2011), where the amount
of weathering is controlled by the reaction rate of the
8.3.2 Limits on Profile Behaviour weathering and not by the rate of flow through the profile.
Typically experiments are rate-controlled because they
Before the sensitivity studies are presented in later have high ratios of water to rock in the reactor vessels
sections, it is worth discussing some forms of limitation so that dissolved product concentrations are low and
on profile behaviour that are assumed or have been dem- unlikely to reach the equilibrium concentration. Equation
onstrated in the field. This will allow the interpretation of (8.6) is implicitly rate-controlled because of the
formulation chosen for the reaction term R. We will parameter set will be hereafter called the ‘full-depletion’
generalise Equation (8.6) later. parameter set; see Table 8.5). The S-shaped substrate
A final condition is where the rate of weathering of the curves are typical of field data collected by White et al.
profile is limited by the rate of exhumation/erosion of the (2008) (Figure 8.18), and we find a humped weathering
saprolite. Riebe et al. (2004) called this ‘supply-limited’. function, where the location of the peak of the hump
This condition can occur when the weathering reaction moves down the profile as the substrate is consumed. In
equilibrates quickly and the only way for the weathering this case the hump is a transient feature, but we will show
reaction to proceed is for new material to be made avail- that it is also possible to generate this hump, albeit with
able for weathering. some difficulty, at steady state. The hump will occur when
As discussed in the previous section, weathering rates the parameters are such that both the substrate and the
decline with time. It is thus possible for a profile to be acid are completely consumed within the profile.
transport-limited at early time (when reaction rates are fast Figures 8.4 to 8.7 show the sensitivity of the results
and leachate concentrations high) and transition to a for a change of 50% in the parameters k 1 , k 2 and k 3 . In
weathering-limited profile as the reaction rate declines Figures 8.4 and 8.6 the qualitative trends in A, S and L do
(Brantley and Lebedeva, 2011). not markedly change: (1) A still decreases nonlinearly
Finally there is ‘isovolumetric weathering’. This is the from the surface to the base of the profile, (2) S still
case when the weathering reaction occurs without any increases nonlinearly from the surface to the base though
change in volume. For example, in cases where there are there is a marked reduction in the amount of S at the
large quantities of leachate product and very little solid, it surface as k 2 increases (i.e. how much substrate is used
is possible that the soil may collapse because of the voids per unit reaction) and (3) dissolved product P simply
generated in the particles and the volume will decrease. reflects the changes in A because the relationship
Any change is volume is parameterised by strain εi for between A and P is unchanged. As expected Figure 8.7
mineral i in the soil shows a change only in the dissolved product P and not
in A or S, and this is a consequence of Equation (8.6),
V s,i V p,i
εi ¼ (8.12) where the value of the dissolved product does not influ-
V p,i
ence A and S.
where V s,i is the volume of mineral i in the soil and V p,i is The effect of increasing the weathering rate k 1 is to
the volume of mineral i in the saprolite (Brimhall and sharpen the weathering front. This is particularly obvious
Dietrich, 1987; Brantley and Lebedeva, 2011). Note that in Figure 8.5 using the full depletion parameter set
strain is defined with respect to a mineral fraction, not the (Table 8.5), but Figure 8.4, which is for the nominal
soil as a whole. This definition of strain should not be parameter set, is also consistent with this sharpening.
confused with that used for stress-strain calculations in The appearance of a humped like weathering function
mechanics, despite the superficial similarity. in Figure 8.6 is an important result and highlights one
important aspect of the behaviour of Equation (8.6). In
8.3.3 Parameter Sensitivity Studies Figure 8.6 even a modest increase in k 2 of 50% leads to a
humplike weathering function with the maximum
In the following examples we present only the final equi- weathering rate at a depth of about 0.5 m. This is a result
librium results and not the transient behaviour. The intent of subtle interactions within the reaction term on the
of the following examples is to explore the range of right-hand side of Equation (8.6). The key components
behaviours that the systems of equations in Equation of this interaction are (1) fresh acidic fluid enters at the
(8.6) can exhibit. Changing the values for most of the surface and the acid potential is consumed as it infiltrates
model parameters can dramatically change the qualitative down the profile and (2) fresh unweathered rock is
behaviour in Figure 8.2. exhumed at the profile base and is depleted by the acid
For instance, if we increase the ability of the acid to reaction within the profile. The reaction rate (i.e.
completely consume the substrate we find dramatically weathering rate) is a function of the multiplication of
different transient behaviour. Figure 8.3 has the same the acid concentration, A, and unweathered rock, S. The
parameters as Figure 8.2 (i.e. the nominal parameter set weathering is low if either of them is low and is only
in Table 8.5) with the exception that the rate of reaction, relatively high if both of them are high simultaneously. In
k 1 , is increased by a factor of 5, and the amount of initial Figure 8.6 A is low at the profile base and S is low at the
substrate has been reduced from 1.0 to 0.25, so it is easier surface, leading to low weathering rates at the surface and
for the acid to completely deplete all the substrate (this profile base, and the combination of A and S mid-profile
FIGURE 8.3: Example output of the profile evolution from the chemical weathering model with the full depletion parameters.
The arrows indicate the direction of evolution over time. Unlike the full depletion simulations later in the chapter the initial substrate
concentration was S0 ¼ 0:25, rather than the value S0 ¼ 0:5 used in the remainder of the chapter for depletion parameter simulations.
This smaller value S0 ¼ 0:25 was used because it allows us to show in panel (d) how the weathering function changes over time, from one
where weathering is highest at the soil surface, to one where there is a humped function with a peak in mid-profile, and finally to the equilibrium
where the weathering rate is greatest at the bottom of the profile. For a detailed explanation of the panels see Figure 8.4.
is sufficient to lead to the weathering rate being highest may seem rather unexpected that if the velocity is higher,
there. Moreover, the main effect of modifying k 2 is to then the weathering rate is higher. However, with a fixed
substantially reduce the substrate near the surface (since concentration of reactant at the upper boundary condition
it controls the rate of substrate consumption), so the (i.e. a fixed pH), a higher velocity means that more acid is
reason for the appearance of a humped weathering func- being flushed through the profile, so that the cumulative
tion is the low proportion of substrate at the soil surface acid load through the profile increases with the velocity.
because of the relatively higher rate of substrate A 50% increase in velocity results in a 50% increase in
weathering within the profile. We will repeatedly see this acid load. If the acid concentration is a result of rainfall
behaviour arising out of the multiplicative combination equilibration with atmospheric CO2 then this is correct
of A and S. behaviour. This relationship between velocity and rate of
Figure 8.8 shows the effect of changing the water weathering is consistent with column tests by van Grins-
infiltration velocity. As the velocity increases the acid ven and van Riemsdijk (1992). However, as a thought
has less time to react with the near surface rock and is experiment, it is enlightening to separate the velocity from
able to penetrate deeper into the profile. This means that load effects. Figure 8.9 isolates the effect of increased
the acid is more concentrated when it reaches the deeper velocity from increased load by reducing the concentra-
fresher rock. The effect of this can be seen in the substrate tion of A as velocity increases. For instance, the 50%
where the amount of weathering at depth is higher at increased velocity has a soil surface boundary condition
higher velocities, and that at the higher velocities the of A ¼ 0:6667 the nominal A concentration, so that the
weathering function in the deeper part of the profile is total load is the same. This figure shows that at the surface
higher, while near the surface it is relatively unchanged. It there are differences for the different velocities, but that
FIGURE 8.4: Sensitivity analysis for the weathering model with nominal parameters for changes in the weathering rate parameter k1.
The rapid approach to leachate concentration of 0.5 for the higher reaction rates in panel (c) is because with the parameters k2 ¼ 2:0 and
k3 ¼ 1:0, the maximum amount of leachate that can be generated is L ¼ 0:5 when all the acid is used (i.e. in panel (a) at depths greater
than about 1 m). This is also the reason that the substrate concentration is approximately unchanged for depths greater 1 m.
Explanation of the panels: These are the final equilibrium profiles. For the first five panels the legend gives the multiplier that is used to
modify the parameter being changed in the sensitivity study, so 0.5, for instance, means that the nominal parameter is multiplied by 0.5 for that
run. (top left) The concentration of the acid in the infiltration water; (top middle) the concentration of the substrate (typically saprolite, but it
could be a mineral or cation), and this curve is commonly referred to as a depletion profile; (top right) the concentration of the leachate
(dissolved weathering product); (bottom left) the rate of weathering reaction; (bottom middle) the rate of the weathering reaction plotted
on a semi-log horizontal scale (exponential weathering functions will plot as a straight line on this plot); (bottom right) a time series of substrate
at various locations in the profile. The thin lines are for five layers equidistant from the top of the profile (i ¼ 1 in the legend) to the base
of the profile (i ¼ 50 in the legend). The thick lines are the substrate concentration average over the whole profile for three parameter
multipliers corresponding to the legend in the other five panels.
by the time the water has reached the base of the profile exhumation rate. This is consistent with the age of the
there is almost no difference in acid concentration. There rock being less, and being exposed to weathering for less
is a small difference in the substrate at the soil surface. time, for higher exhumation rates. This is also shown in
However, the weathering functions are very different, the weathering function, which has lower weathering rates
with the slowest velocity having the highest weathering for higher exhumation. Again we see a humped
rate. This is because everything else being equal, the acid weathering function appear for the higher rates, and this
has more time to react with the rock and so is able to is because we have a larger substrate higher in the profile,
transform a greater proportion of the rock, and that is and this coincides with the higher acid concentrations. In
reflected in the slightly lower substrate at the soil surface contrast to the higher velocity case above though, here the
for the lower velocity. higher weathering rates in the hump are largely driven by
Figure 8.10 shows the effect of varying the exhum- the higher substrate, because there is only a small change
ation rate (and, as the soil depth is assumed constant with in the acid concentrations.
time, varying the erosion rate). The amount of substrate Figure 8.11 shows the effect of increasing the disper-
that makes its way to the surface is greater for a greater sivity by six orders of magnitude. Other than a slight
FIGURE 8.5: Sensitivity analysis for the weathering model with the full depletion parameters for changes in the weathering rate
parameter k1. This can be compared with Figure 8.4. The main effect of an increase in the weathering rate is that the weathering front
is sharper and more distinct. The weathering peak also occurs slightly higher in the profile. For a detailed explanation of the panels
see Figure 8.4.
M n
smoothing that makes the profiles slightly more linear,
R ¼ k 1 Aðk 5 ðS Smin ÞÞ 1 L=Lequilib (8.13)
there is no significant difference for different dispersivities.
We conclude this part of the sensitivity study of Equa- where Lequilib is the equilibrium concentration of the
tion (8.6) by noting that in none of the cases shown here leachate in the fluid and L=Lequilib is called the saturation
do we see evidence of a weathering function with an index Ω (White et al., 2008). The parameters M and n are
exponential decline with depth. However, we do see, for fitted to experimental data (Steefel, 2009) but we will
high leaching velocities and uplift rates, that a humped assume, for simplicity, they are both 1, as Steefel indi-
weathering function can occur. cates this is the case for kaolinite. This leachate
concentration-dependent term now couples the leachate
8.3.3.1 Leachate Concentration Rate Control and production with the rate of transformation of S and
Transport-Limitation depletion of A. As L approaches the equilibrium concen-
The main shortcoming of Equation (8.6) is that the reac- tration, the reaction term converges to zero. Note that for
tion rates are unlimited. In reality if rock fragments are low concentrations of leachate the leachate-limitation
placed into an acid solution and the acid solution concen- term is approximately 1.0 and the reaction rates are not
tration is maintained constant throughout the experiment, limited, which is what was assumed in the sensitivity
then the weathering reaction will eventually equilibrate at studies in the previous section. A more sophisticated
an equilibrium leachate concentration. Initially the reac- formulation, based on surface chemistry fundamentals
tion rate will proceed at its maximum rate (as defined in (Lasaga, 1984), highlights that there is also a tempera-
Equation (8.6)) but the reaction rate will slow asymptotic- ture dependence in this equation (both rate and equilib-
ally as the equilibrium is approached. Following Lasaga rium concentration, e.g. silica equilibrium concentration
(1984), Steefel (2009), and Maher (2010) the reaction increases by a factor of two from 10 to 20C), which we
term can be reformulated as will ignore.
FIGURE 8.6: Sensitivity analysis for the weathering model with nominal parameters for changes in the acid consumption rate parameter k2.
For a detailed explanation of the panels see Figure 8.4.
In the sensitivity studies that follow, Lequilib ¼ 0:4. near equilibrium, so that the natural decrease in acid with
This value was chosen solely to demonstrate the effect depth can accelerate because it is not being offset by the
of the leachate-limitation term. The value of 0.4 is feedback that increases leachate. At the bottom of the
below the peak values of leachate concentration in the profile the geochemistry is transport limited because the
previous sensitivity studies, so comparisons to the limit on weathering is the ability to transport the leachate
simulations above will lead to visible effects that can out of the bottom of the soil profile. Finally the reduced
then be interpreted. Figures 8.12 and 8.13 show the scatter between the curves for different values of k 1 indi-
sensitivity analysis for k1 and v, and are the leachate- cates that leachate limitation reduces the sensitivity of the
limited equivalents to Figure 8.4 and 8.8. Leachate model to changes in k 1 , which is surprising because k 1
limitation has a marked difference in the weathering determines the rate of the reaction. This suggests that the
behaviour. model quickly transitions down the profile from reaction
In Figure 8.12 the most obvious difference is that the rate limited to leachate concentration limited (and trans-
rate of consumption of acid and transformation of sub- port limitation), which is consistent with Maher et al.
strate has been substantially reduced, and as expected the (2009).
concentration of leachate levels off near the equilibrium Figure 8.13 shows the sensitivity analysis for vel-
value of 0.4. While the maximum weathering (at the ocity with leachate limitation. Many of the conclusions
surface) is not very different from the unlimited case, from the k 1 sensitivity test above (i.e. Figure 8.12) carry
the weathering rate drops more rapidly with depth than over to this case. However, this figure shows that all
for the unlimited case. The log(weathering) function is signs of the humped weathering function (Figure 8.7)
almost linear, and so is close to exponential with depth. have disappeared. The reduction in the sensitivity of
Finally the change in shape of the acid curve is quite acid concentration with changing velocity is similar to
marked. This is related to the shape of the leachate curve that for k 1 but the reduction in sensitivity in substrate is
where leachate doesn’t increase with depth because it is not as marked. As for k 1 sensitivity plots in the bottom
FIGURE 8.7: Sensitivity analysis for the weathering model with nominal parameters for changes in the leachate generation rate parameter k3.
For a detailed explanation of the panels see Figure 8.4. For A, S and W all values of k3 plot on the same line.
half of the soil profile the leachate concentration is higher up in the profile this depletion occurs. Moreover
limited by the equilibrium value and geochemistry is because of the lower substrate, all of the substrate is
transport limited. transformed by the time it reaches the surface. Thus at
the top of the profile the substrate is zero, and at the
8.3.3.2 Relative Concentrations of Acid bottom of the profile the acid is depleted. Because of the
and Substrate multiplicative term in the reaction equation this means
The simulations above were performed with initial con- that the reaction rate (and thus the weathering rate) is
centrations of acid and substrate both equal to one, and near zero at both the top and bottom of the profile, with a
where the relative reaction rate between them was para- maximum within the profile. Thus the weathering func-
meterised by k 2 ¼ 0:5. These parameters were chosen tion in this case is a humped function with the maximum
because they consumed most, but not all, of the acid and rate within the profile. For the mid-range velocity this
substrate. The equations used are the original Equation humped peak occurs almost in the centre of the profile.
(8.6) without leachate limitation. For a higher velocity the acid doesn’t deplete as quickly
Figures 8.14 and 8.15 show the same velocity sensi- down the profile so the peak is toward the bottom of the
tivity results in Figure 8.8 but with half the substrate and profile, while for the lower velocity the depletion occurs
a higher reaction rate (k1 ¼ 0:5 rather than k1 ¼ 0:1). higher in the profile and the hump occurs higher in the
The difference between Figures 8.14 and 8.15 is that profile. Similar behaviour, not shown here, also occurs
Figure 8.15 zooms in on the parameter values for vel- for different exhumation rates. Note that it is the com-
ocity that yield a hump in the weathering function. Like- bination of (1) depleted substrate at the surface and (2)
wise Figures 8.16 and 8.17 show the same erosion depleted acid at the profile base that creates the hump in
sensitivity results as Figure 8.10 but with the same sub- the weathering profile.
strate and reaction rate changes. Because of the higher The S-shaped substrate curves are consistent with
reaction rate, the acid is depleted by the time it reaches data from a series of chronosequences on terraces in
the bottom of the profile, and the slower the velocity the White et al. (2008). White fitted a multicomponent
FIGURE 8.8: Sensitivity analysis for the weathering model with nominal parameters for changes in infiltration velocity parameter v. For a
detailed explanation of the panels see Figure 8.4.
model very similar to that presented here to the data Figure 8.20 shows the variation with soil depth of the
(Figure 8.18). full depletion parameters as in Figure 8.14. As the soil
deepens, the peak of the hump occurs at a deeper depth.
8.3.3.3 Variation of Soil Depth The weathering rate at the surface is reduced for the
The depth variation of weathering depends on the bal- deeper soils because more of the substrate has been
ance between erosion/exhumation and acid infiltration. weathered. While the depth of the soil ranges from 1.0 m
The exact form of the curves will be dependent on the to 4 m, the depth at which the peak of the hump occurs
depth of the soil and the boundary conditions at the top varies only from 0.5 m to 0.8 m, suggesting that the peak
and bottom of the soil profile. Figure 8.19 shows the is largely tied to the soil surface rather than the base of the
variation in the weathering when the depth of the soil is soil profile. This suggests that the major contributor to the
varied from 1.0 m to 4.0 m. In this case the main differ- peak behaviour is the acid concentration, not the supply of
ence is in the substrate curves. For the shallower soils substrate.
the proportion of substrate weathered at the surface is
less than for the deeper soils, for all depths within the 8.3.3.4 Variability of Specific Surface Area
soil profile, because the shallower soil has been exposed The simulations above assume that the specific area of the
to weathering for less time than the deeper profiles. The substrate is the same throughout the profile. If physical
weathering plots also show that the main differences are weathering is the primary mechanism breaking down the
in the top 0.5 m of the soil, where the acid is freshest, rock fragments, then the rock fragments will become finer
because there is more unweathered substrate to weather as they are exposed for longer to weathering, so that the
for the shallower soils. Below 2 m all the acid is finest particles will be in the layer below the soil surface
depleted, and it makes no difference how much substrate armour layer in the soil (Chapter 7). How the specific area
there is to weather because there is no acid for the changes through the profile will depend on the physical
weathering reaction. weathering function (e.g. exponential or humped), the
FIGURE 8.9: Sensitivity analysis for the weathering model with nominal parameters for changes in infiltration velocity parameter v and
fixed cumulative acid load. Note that the fixed acid load condition means that the concentration of acid at the surface changes with the
inverse of the velocity. Also note that the result for multiplier = 1.0 is the same as the curve for multiplier = 1.0 in Figure 8.8. For a
detailed explanation of the panels see Figure 8.4.
parameters of the physical weathering function and the profile at deeper depth. Also the weathering function
erosion rate (since this determines the rate of exhum- shows the weathering becoming more and more concen-
ation). With these variables in mind we will carry out a trated in the near surface (and at a higher rate within this
simple sensitivity study where specific area is allowed concentrated zone), where specific area is highest. If the
to vary with depth down the profile with a reaction term physical fragmentation of particles is driven by chemical
that is weathering, then there will be a positive feedback
between chemical weathering rate and the fineness of
0:5depth ε M n
R ¼ k 1 A k 5 ðS Smin Þ 1 L=Lequilib the particles.
z
(8.14)
where ε is a parameter we vary to observe the impact on 8.3.3.5 Within Soil Acid Generation
the model results. A large value means that shallow It’s common to observe in agricultural soils that the
depths will have finer soil, higher specific area and high carbon dioxide content of the pore space within the soil
reaction rate. Figure 8.21 shows the results for different is near 100%, compared with the content in the atmos-
values of ε. The results for ε ¼ 0 have a specific area phere of 0.4%. This carbon dioxide is generated by the
independent of depth, and are the same as the nominal breakdown of organic matter in the soil by microorgan-
parameter simulations in Figure 8.2. For all positive isms in the soil (Lucas, 2001) and plant respiration
values the specific area is increasing as you get closer (Chapter 14). If this is the case, then the acid in the soil
to the surface, and that behaviour is more pronounced for water may be dominated by the in-profile generated
larger values of ε. As expected there is more depletion of carbon dioxide rather than the carbon dioxide dissolved
both acid and substrate near the surface, and this is in the rainwater that infiltrates (Little et al., 2005). If the
balanced by less transformation of substrate within the rate of consumption of the carbon dioxide is much less
FIGURE 8.10: Sensitivity analysis for the weathering model with nominal parameters for changes in erosion rate parameter E. For a detailed
explanation of the panels see Figure 8.4.
than the rate of generation by organic matter breakdown, ∂S

¼ ðk 1 k 5 AÞS ¼ KS (8.16)
then Equation (8.6) can be simplified to ∂t
A ¼ Aequilib where all the terms inside the parentheses are constant.
The solution to this equation is S ¼ S0 eKt where S0 is the
∂S ∂S initial concentration in the substrate. If we note that the
e ¼ R
∂t ∂z exhumation rate is equivalent to distance up the profile
∂L ∂L ∂2 L (8.15) from the base, we can substitute for time (z ¼ Et so z is
þ v D 2 ¼ k3 R the distance above the soil-saprolite interface) and the
∂t ∂z ∂z
weathering function is
R ¼ k 1 Aα ðk 5 ðS Smin ÞÞ
∂S ∂ Kz=E S0 K Kz=E
W¼ ¼ S0 e ¼ e (8.17)
P ¼ k4 ðS0 SÞ ∂z ∂z E
where Aequilib is the acid concentration when equilibrated This shows that the reverse exponential weathering
with the soil atmosphere. If, as an example, we set profile is a direct consequence of the fixed acid concen-
Aequilib ¼ 1. then Figures 8.22 and 8.23 shows the results, tration and constant exhumation rate. This reversed expo-
with three different exhumation rates. Figure 8.23 shows nential was used as a weathering function by Welivitiya
the result with leachate control and Figure 8.22 without it. (2017) in his soilscape model.
What is marked about this figure is a reversed exponential More sophisticated approaches to the prediction of the
weathering function with the weathering rate being soil atmosphere are based on mass balances of organic
greatest at the soil-saprolite interface and declining expo- matter decay, root respiration and exchange fluxes
nentially toward the surface. The reason for this is with a between the soil atmosphere and the above-ground atmos-
constant A the reaction with substrate with time can be phere (e.g. Pumpanen et al., 2003) and will not be
simplified to discussed here.
FIGURE 8.11: Sensitivity analysis for the weathering model with nominal parameters for changes in dispersivity D. For a detailed
The discussion above focussed on in-profile gener- generation (which itself is a result of oxygen availabil-
ation of acid as a result of carbon dioxide. Another ity) and dominates in the weathering zone.
mechanism is oxidation of sulphide minerals to generate
sulphuric acid, which reacts with minerals to generate
8.3.4 Bioturbation
sulphate minerals and leachates. This is the mechanism
involved in acid mine drainage (AMD), but it also Bioturbation was discussed in Section 7.3.5 in the context
occurs in natural environments (e.g. Chigira and of physical weathering and evolution of the particle size
Oyama, 1999; Oyama and Chigara, 1999). The distribution. The model discussed in that section was a
weathering products and leachate and rates of simple Fickian diffusion equation for the mixing of the
weathering are different from carbon dioxide, but the layers. It is likewise possible to implement a Fickian
generic governing equations in Equation (8.15) are still diffusion model for the bioturbation of the substrate on
applicable. Chigara and Oyama note that there are two the chemical weathering model. We can modify the sub-
zones in the soil profile. The first zone is the region of strate equation in Equation (8.6) as
active weathering where the strength of rock is reduced
by weathering. Above this zone is an oxidised zone ∂S ∂S ∂2 S
where the iron in the pyrite is precipitated, and precipi- e Ds 2 ¼ R (8.18)
∂t ∂z ∂z
tated iron cements the soil particles so that the soil
strength is increased. The presence of two zones is where Ds is the diffusivity for the mixing of the soil due to
because the AMD process results from the interaction bioturbation. The weathering product equation is similarly
of two chemistries. The first process, dominant in the changed. The equations for acid and leachate remain
upper oxidised zone, is a result of the availability of unchanged. As for the physical weathering case, the sub-
oxygen but depletion of minerals that can react with the strate diffusivity can vary with depth and time depending
produced acid. The second process is a result of the acid on the bioturbation mechanism (Section 7.3.5), and may
FIGURE 8.12: Sensitivity analysis for the leachate-limited weathering model with nominal parameters, for changes of the reaction
rate parameter k1 . This figure is identical to Figure 8.4 other than the leachate limitation. For a detailed explanation of the panels
see Figure 8.4.
interact with physical weathering processes and/or the two parameter sets suggests that these trends with increas-
grading of the soil. ing bioturbation may be robust against changes in the
Figure 8.24 shows a sensitivity run with the full rates of the other in-profile processes.
depletion parameters. The nominal value is Ds ¼ 0:005.
It can be seen that as the bioturbation increases the
weathering function becomes less humped and more 8.3.5 The Hydrology Conceptualisation
exponential (the convergence to linearity in the log
(weathering) panel), because the bioturbation mixes The simple conceptualisation of hydrology in the model
unweathered soil through the profile, so the humping, in the previous sections assumes that water is flowing
which results from depletion of saprolite near the surface through the soil at a constant rate. This is a poor approxi-
(note the larger values for saprolite near the surface for mation to the way infiltration actually occurs, with high
the high diffusivities), does not occur. This reduction flows through the soil during rainfall events and long
in saprolite depletion near the surface means that acid periods of no flow between rainfall events. The conceptu-
consumption rates increase in the surface layers so the alisation also ignores transpiration, which will extract
acid concentration also declines more rapidly near the water from the soil in the root zone. And as noted in
soil surface. Chapter 3, in arid zones it is common that infiltration from
The figure shows that everything else being equal, the small events does not penetrate all the way down the
proportion of unweathered saprolite in the soil profile profile, and it is only large rainfall events that wet the
increases with increasing bioturbation, and the region of entire profile from top to bottom.
most active weathering is closer to the surface. For the To address both of these time- and profile-varying
nominal parameter set similar behaviour is observed processes would require sophisticated unsaturated flow
(Figure 8.25). The commonality of the results for the modelling within the soil layer with a full Richards
FIGURE 8.13: Sensitivity analysis for the leachate-limited weathering model with nominal parameters for changes in the infiltration
velocity parameter, v. This figure is identical to Figure 8.8 other than the leachate limitation. For a detailed explanation of the panels see
Figure 8.4.
equation solver and plant transpiration dynamics (e.g. thin film to the soil particles’ surfaces, called the field
using the HYDRUS model). Such a model is beyond the capacity (typically about 10–30%), which cannot drain
scope of this book, but some simple approximations can out under gravity because it is held onto the soil particles
be made that give insight into what a full modelling by surface tension effects. This water will continue to
exercise might display. react with the substrate until it (1) depletes the acid or the
We will consider first the case of event-based infil- substrate, (2) reaches leachate limitation, (3) is evapor-
tration. The velocity term in Equation (8.6) defines how ated or transpired or (4) is replaced by new water from a
much time the water is in contact with the substrate, and subsequent infiltration event (the water from the event
therefore how much weathering can occur. For instance, if must of course penetrate to that depth). In this case the
we consider a small volume of soil of height Δz, then with residence time of water, T r , is the mean time between
velocity v the time the water is in that section of the profile flushing events at that depth in the soil profile. This mean

is Δt ¼ Δz v . In the existing time-invariant model this is the residence time will be short at the top of the profile
residence time in the Δz volume of soil. (because most rainfall events will saturate the surface)
Consider now event-based infiltration. During the and longer at the base of the profile (because only large
rainfall event water will fill the soil voids with water rainfall events will infiltrate this deeply). To first order
down to a depth determined by the depth of rainfall and the average effect of this behaviour would be the same as
porosity of the soil (Section 3.1.4). Using the Green- if the water were slowly flowing through the profile but
Ampt approximation to infiltration, if the infiltration where the velocity reduces with depth (it becomes slower
from the rainfall event is R, and the porosity of the soil with depth as the flushing events get less frequent)
is n then this depth will be z ¼ R=n . Immediately after the so that
rainfall stops, most of that water drains deeper into the
profile, but a small amount of water remains attached as a v ¼ Δz Tr (8.19)
FIGURE 8.14: Velocity sensitivity analysis for the weathering model with the full depletion parameters showing the generation of a
humped weathering profile for changes in the infiltration velocity parameter, v. This figure can be compared with that for the nominal
parameters in Figure 8.8. For a detailed explanation of the panels see Figure 8.4.
This, of course, averages out the variability between infil- (which is equal to the water extracted from the profile)
tration events, which may be important if the fluid to the volume of infiltration per unit time so that below the
becomes leachate concentration limited between infiltra- root zone
tion events. Note also that the time of residence will be

related to climate, with wetter climates having lower IT
vbelow ¼ vabove (8.20)
residence times and higher velocities. I
We turn now to how transpiration extracts water from
within the soil profile. Plants cannot extract all the water where I is the infiltration (mm/day) and T is the
in the profile, being unable to extract water below the transpiration (mm/day), and the subscripts indicate the
wilting point of the plant. This wilting point soil moisture velocities above and below the root zone.
varies between plants, soil salinity and soil properties. Observed residence times for water can be quite long
Typically it is lower than the field capacity, and of the by hydrologic time scales. For instance, White et al.
order of 1–20%. The effect of plants is to remove water (2009) in a Mediterranean climate calculated 10–25 years
(and some of the dissolved weathering products) and based on infiltration velocities 0.06–0.22 m/year in a 15 m
reduce the amount of water infiltrating below the root thick soil profile.
zone. Thus the effect of the root zone is to reduce the Finally there is the question of whether a uniform flow
number of infiltration events that penetrate beyond the through the profile is an appropriate approximation when
root zone. Thus the mean residence time of the water there is ample evidence of heterogeneities, preferred
below the root zone will be longer than it would be in flowpaths and macropores, so that infiltration is not uni-
the absence of transpiration. A crude first approximation form (Beven and Germann, 2013). Numerical studies
would be to reduce the velocity of the water below the coupling the chemical reaction principles in this chapter
root zone by the ratio of the volume of transpiration with preferential flowpaths are beginning to appear
FIGURE 8.15: Velocity sensitivity analysis for the weathering model with the full depletion parameters showing the generation of a humped
weathering profile. This figure is identical to Figure 8.14 except that it shows greater detail for the parameter values around the value that
generates the mid-profile hump in the weathering profile. For a detailed explanation of the panels see Figure 8.4.
(Pandey and Rajaram, 2016; Rajaram and Arshadi, 2016) initial perturbation in soil or infiltration may result in
but experience from the hydrology community on macro- higher flow at one location, and this higher flow leads
pore flow would suggest it will be a hard issue to resolve. to greater chemical weathering, which in turn increases
However, the complexity of explicitly modelling macro- porosity and permeability. This higher permeability then
pores may not be necessary because Pandey and Rajaram allows more water to flow through this localised highly
(2016) present horizontal averaged results for the weathered region, with a positive feedback on chemical
weathered substrate and weathering front, and these aver- weathering and water flow so that a preferred flow path-
aged results are qualitatively similar to the one- way is created. This is mathematically similar to the
dimensional model in this chapter. For instance, their positive feedback between water flow and erosion that
Figures 6 and 7 are remarkably similar to Figure 8.3 here, leads to channel networks formation in catchments (Will-
despite their results showing strong spatial heterogeneity goose et al., 1991a, b).
in weathering due to the preferential flowpaths. This
suggests that the simple one-dimensional model in this
chapter might be a useful horizontally averaged effective 8.4 Multicomponent Chemical
representation of weathering even in the presence of Weathering
preferential flowpaths.
A further complication regarding macropores and Clearly the single-component model above is very simple,
preferential flow is the possibility that the chemical but even this simple model shows a spectrum of soil
weathering process may coevolve with the soil hetero- profile dynamics. We will not discuss in detail the exten-
geneity so macropore density and pathways evolve over sion of this model to multicomponent problems. This
time (and thus the parameters of an effective parameter- is the realm of sophisticated geochemical modelling
isation will also evolve over time). For instance, a minor tools like CrunchFlow (Steefel, 2009), and PHREEQC
FIGURE 8.16: Erosion sensitivity analysis for the weathering model with the full depletion parameters for changes in the erosion rate
parameter E showing the generation of a humped weathering profile. Compare this figure with that for the nominal parameters in Figure 8.10.
For a detailed explanation of the panels see Figure 8.4.
(USGS, 2015). However, it is worthwhile to qualitatively straightforward, but is dependent on the geochemical
consider some of the complexities introduced by multi- model used and their databases of reaction rates for all
component modelling, many of which are still debated in the constituents in the reactions. This is the key capabil-
the literature. ity of the multicomponent geochemical models men-
The most important complexity is that the weathering tioned above. Figure 8.26 shows calculations for the
reactions from the different minerals may interfere with full depletion parameter set where all the weathering is
each other. If two mineral weathering reactions generate done by a single infiltrating acid acting on two minerals
the same product, it is possible for a leachate limitation to with different reaction rates. Figure 8.26 can be com-
inhibit one or both of the reactions. For instance, for two pared with Figure 8.5, which has the same parameter
minerals sets (i.e. both have the multipliers 0.5 and 2.0), but
where each weathering rate occurs independently of
a1 A þ S1 ¼ b1 P1 þ c1 L1
each other’s rate in Figure 8.5. While the observation
a2 A þ S2 ¼ b2 P2 þ c2 L2 (8.21) for Figure 8.5 that the front becomes sharper with
increasing k 1 is still true, the details of the depth
L1 þ L2 ¼ L
dependence of the weathering are different. When
the leachate product L is produced by both reactions, weathering occurs across several minerals simultan-
and if the L ¼ Lequilib then both reactions are inhibited eously (and where the minerals are differentiated by
compared with the situation if only one reaction was their reaction rate), the peak depth of weathering rate
occurring. There are also issues of the rates of reactions is lower in the profile.
with the faster reaction potentially interfering with the If on the other hand there is no interaction between the
slower reaction. The mathematics of this is relatively various minerals being weathered in the profile, then all
FIGURE 8.17: Erosion sensitivity analysis for the weathering model with the full depletion parameters showing the generation of a humped
weathering profile. This figure is identical to Figure 8.16 except that it shows greater detail for the parameter values around the value that
generates the mid-profile hump in the weathering profile. For a detailed explanation of the panels see Figure 8.4.
that is necessary is to run the single-component equations volume in the rock. Predicting the time evolution of
(i.e. Equation (8.6)) in parallel with each other and just surface area due to chemical weathering is difficult,
sum up all the components to determine the cumulative with no consensus on approach and with conflicting
impact of weathering (e.g. Yoo and Mudd, 2008). The results from different experimental methods, rocks,
only interaction is that the acid is consumed by all of the and acids. Even with the use of the sophisticated geo-
processes so there is the possibility of slower reactions chemical models above this is an area of considerable
being inhibited if the acid is preferentially consumed by predictive uncertainty.
the faster reactions. This would be consistent with apply- Finally, Maher et al. (2009) noted that in her field
ing the calculations leading to Figures 8.4 and 8.5 to each study of a leachate-limited problem that her field data on
of the mineral constituents, respectively. reaction rates required either unrealistically high infiltra-
Another complexity is the surface available for tion velocities, or a secondary reaction that removed
weathering. The rate of the reaction depends on the leachate products from solution (i.e. a precipitation
surface area of the mineral exposed to the reaction. This reaction). This requires the modelling of (at least) two
surface area is more than the relative percentages by chemical reactions (the original weathering reaction, and
volume in the rock because the weathering will occur the secondary precipitation reaction involving the
on the most reactive minerals (and mineral boundaries, dissolved weathering products) and is beyond the scope
defects, etc.) first. For instance, Hodson (2006) found of the model presented here. White et al. (2005) found that
biotite weathered about 100 times faster on grain the solubilities of weathering products gibbsite and kaolin-
boundaries than the grain interior. Accordingly at any ite changed with changing pH (which depends on the soil
time the specific surface area of each mineral compon- atmosphere CO2 concentration) and the change in solubi-
ent may deviate considerably from its percentage by lities depended on the aluminium hydroxides present in the
0 0
(a) (b)
qh/4 kr/2 kr/4
4kr
qh/2 2kr
5 5
2qh
kplag = 4.8 × 10–16

10 qh = 0.058 m yr–1 moles m–2 s–1 10
4qh
Depth (m)
Depth (m)
0 0
(c) (d)
a = 1.0
a = 0.66 kr/4
kr/2 1
5 2kr
4kr 2
a = 0.33
a = 0.16 kKspar = 3.8 × 10–16 3

10 moles m–2 s–1
4
0 25 50 75 100 0 25 50 75 100
Residual Feldspar (wt%)
FIGURE 8.18: Data and model fits to a series of chronosequences in Santa Cruz. The data are the residual feldspar left in the rock after
weathering, so are equivalent to substrate. The model is a multicomponent version of the model in this section. The lines are different parameters
in the model for (a–c) plagioclase and (d) K-feldspar: (a) infiltration velocity, (b) reactivity, (c) surface area-diameter relationship and (d) K-feldspar
reactivity (Figure 14 from White et al., 2008).
soil. Conversely, other mineral solubilities were independ- concentration of each component will vary with depth.
ent of pH and aluminium. Again this suggests the need for Lebedeva found layers of kaolinite that they postulated
multicomponent modelling in some situations. created porosity changes down the profile. It is tempting
Lebedeva et al. (2007) modelled multicomponent to speculate that these results hint at the possibility of
weathering with a model very similar to that proposed horizon development within the soil profile.
here but where transport in the vertical was only by
diffusion and there was no advection. They showed the
erosion is key to developing a steady state chemical 8.5 Hillslope Modelling
profile because of the new material that is introduced at
the base of the profile. This is consistent with the single- The model framework and applications in the previous
component model results in this chapter. A notable aspect sections are for one-dimensional flow vertically down the
of the multicomponent aspect of their modelling was that soil profile. If we refer back to our discussion of dominant
the different components had different stable weathering type of hydrology, this is consistent with arid zone hydrol-
profiles through the soil, so that different weathering ogy, where there is no permanent water table within the
products dominated different depths within the profile. soil profile (infiltration excess runoff generation; see
In Figure 8.6 the substrate profile varies differently with Section 3.1.3.2).
depth depending on the reaction rate, so for a number of In more humid climates there is typically a permanent
components with different reaction rates the percentage water table within the soil and water flows laterally in the
8.5 Hillslope Modelling 143
FIGURE 8.19: The impact of varying the depth of the soil profile from 1 m deep to 4 m deep. The multipliers in the legends indicate the
depth relative to the nominal depth of 2 m. All other parameters are the nominal parameters. For a detailed explanation of the panels
see Figure 8.4.
steepest downslope direction down the hillslope (i.e. sat- observation to develop a model for the dynamics of hill-
uration excess runoff generation; see Section 3.1.3.1). slope soil weathering. The rate of change of elevation of
Above the water table infiltration will flow vertically the soil-saprolite interface is
down the profile until it hits the water table. Thus above
∂S ∂qd,x ∂qd,y
the water the one-dimensional vertical flow model in the ¼ F þ (8.23)
∂t ∂x ∂y
previous sections governs, but below the water table the
dominant flow direction is near horizontal rather than where S is the elevation of the soil-saprolite interface (see
vertical. This horizontal flow case has not been as well Equation (2.7)), and F is factor that relates the amount of
studied as the vertical flow case. loss of mineral mass in the soil (which equals the amount
Some simple models are possible for transport-limited of dissolved load in the groundwater) to the rate of change
weathering. For transport-limited weathering the concen- of the elevation of the soil-saprolite interface. The factor F
tration of the leachate is determined by the equilibrium depends on the mineralogy of the rock and how a kilo-
chemistry (i.e. Equation (8.5)) so that the rate of dissolved gram of loss from weathering equates into the transform-
load transport is ation from rock into soil. One way of characterising this is
by the change in porosity of the saprolite to the porosity of
qd ¼ qLequilib (8.22)
the soil above the zone of active weathering.
where qd is the dissolved load carried by the groundwater A particularly simple result can be derived for a one-
(mass/time), q is the discharge (volume water/time) and dimensional hillslope with parallel flow lines downslope,
Lequilib is the equilibrium concentration (mass/volume so that q ¼ ðI ETÞx where I is the infiltration rate
water) of the leachate. The cumulative amount of leachate (depth/time), ET is the evapotranspiration rate (depth/
carried by the groundwater can be used as a measure of time) and x is the distance downstream from the hillslope
the progress of weathering. Braun et al. (2016) used this divide. Substituting Equation (8.22) into Equation (8.23)
FIGURE 8.20: The impact of varying the depth of the soil profile from 1 m deep to 4 m deep. The multipliers in the legends indicate the
depth relative to the nominal depth of 2 m. All other parameters are the full depletion parameters. Compare this figure with the nominal
parameters (Figure 8.19) and note that the full depletion parameters generate a humped weathering profile. For a detailed explanation
of the panels see Figure 8.4.
leads to an equation that estimates the change in elevation saprolite-soil interface the rate of decline of the interface
of the saprolite-soil interface for a hillslope in the direc- must exactly balance the rate of uplift U so that
tion of steepest descent
∂S ∂q
U¼ ¼ FLequilib ¼ FLequilib ðI ETÞ (8.25)
∂S ∂q ∂t ∂x
¼ FLequilib ¼ FLequilib ðI ETÞ (8.24)
∂t ∂x
It should be noted that everything of the right-hand side of
The implication of this is that if the discharge is increasing this equation is a function of the geology or the climate, so
linearly downslope, then the rate of transformation of there is nothing constraining the right-hand side to be
saprolite to soil is constant across the landscape, or as equal to the uplift. Accordingly if the right-hand side is
some may prefer to express it the rate of elevation change larger, than the uplift then the soil will deepen without
of the saprolite-soil interface. This rate of soil formation is limit, while if it is smaller it will decrease until (presum-
independent of depth, in contrast to, for instance, the ably) there is no soil. Thus it seems unlikely that this
exponential soil production function in Equation (6.2) model will generate equilibrium soil depths that are con-
which has the rate of saprolite to soil transformation as a stant with time for a hillslope at dynamic equilibrium.
declining function of the depth of the soil. In conclusion it is should be clear that there are a
Braun et al. (2016) then concluded that a relationship number of conceptual difficulties with this model includ-
for dynamic equilibrium of soil was possible. For land- ing the inconsistency with the exponential soil production
form topography the balance between a spatially constant function, and the inability to generate an equilibrium soil
tectonic uplift and the exactly balancing erosion is simply depth. Those difficulties aside the approach of Braun et al.
the argument leading to the classical slope-area (2016) is novel and is consistent with the hydrology of
relationship (Section 3.3.3). For a constant elevation hillslopes in humid regions. Accordingly this approach
FIGURE 8.21: Sensitivity study of the effect of specific area. The parameters are the nominal parameters. A value of ε ¼ 0 is
equivalent to being independent of specific area (i.e. equivalent to the nominal parameters with multiplier = 1 in the sensitivity studies of
Figure 8.4 to 8.11). For a detailed explanation of the panels see Figure 8.4.
ð
seems a fruitful starting point for a physically realistic 1 ∂S
vLequilib Δt ¼ ∂z Δt (8.26)
model for soil formation on hillslopes in humid regions. c ∂t
where Δt is the period over which the mass balance is

8.6 Some Final Observations on Chemical performed, c is the mass balance parameter from Equation
Weathering (8.6) and the integral inside the parentheses is the total
rate of weathering of the profile. This leads to an expres-
The discussion below returns to discussion of the vertical sion for the total weathering rate averaged over the profile,
one-dimensional model for chemical weathering that has W, per unit volume (i.e. unit plan area and unit depth of
been widely discussed in the literature. the profile)
cvLequilib
8.6.1 Integrating over the Profile for a W¼ (8.27)
D
Lumped Model
where D is the total depth of the soil. If Lequilib cannot be
Transport limitation leads to a simple lumped model for attributed to the weathering of just one mineral, or if there
chemical weathering in the soil profile. If the leachate can is a secondary weathering reaction (e.g. the leachate may
be used to uniquely identify the mineral that is be involved in a precipitation reaction within the profile),
weathering, then we can do a mass balance on the profile then the value of c may not be known, so this equation,
so the mass of leachate exiting the profile can be used to while correct in principle, may not have a value of c that
calculate the total amount of weathering of that mineral in can be calibrated.
the profile using the chemical mass balance in Equation The dependence of this result on infiltration is well
(8.6). Taking a unit plan area soil profile, then supported by field evidence (e.g. Dessert et al., 2003).
FIGURE 8.22: The weathering profile for varying erosion when the acid concentration is the same value throughout the profile (e.g. driven by
equilibration to a soil atmosphere with constant carbon dioxide concentration throughout the profile depth). All parameters are the nominal
parameters. Note the semi-log linear weathering profile in panel (e) (‘reverse exponential’). For a detailed explanation of the panels see Figure 8.4.
Dessert also found a temperature dependence that is unit surface area (k 1 in Equation (8.6)) and Aj is the
related to the leachate equilibration concentration, Lequilib. surface area of mineral j. The parameter k j is the rate of
the reaction (k 1 Aα k 5 and Smin ¼ 0, and k j should not be
confused with k 1 in Equation (8.6)). Solving this with
8.6.2 Conceptual Lumped Models
time, assuming k j is constant with time and mineral mass,
Most existing soilscape models do not explicitly model gives
the chemistry and the physics discussed in the depth mj ¼ mj ð0Þek1 t (8.29)
explicit model above (with the exception of Finke’s
SOILGEN model; Finke, 2012). Typically they rely on This equation is similar to the time component used by
conceptualisations of the rate of the dissolution over time Minasny and McBratney (2001) but the latter authors
from the entire profile, implicitly considering that the soil added a humped function to simulate a soil depth depend-
profile and the processes operating in it are the same from ency (their equation (4)).
top to bottom. Some of the assumptions built into Equation (8.29)
Yoo and Mudd (2008) is a good example of the are that (1) there is no addition of new saprolite to the
approach where they use the idea of mineral dissolution profile, so eventually the mineral concentration will
in Equation (8.6) to write the equation for dissolution loss decline to 0 (i.e. a nonzero equilibrium is not possible),
of a mineral j as (2) the reaction rate does not change with time (i.e. either
reactivity and surface area do not change or they compen-
∂mj
¼ R j Aj mj ¼ k j mj (8.28) sate each other to keep k j constant) and (3) it is possible
∂t to sum up the profile mineral mass over the profile and
where mj is the mass of mineral j (equivalent to S in treat it as a lumped value. Yoo and Mudd (2008) do have
Equation (8.6)), Rj is the reaction rate of that mineral per a separate soil production function, so by using the SPR
FIGURE 8.23: The weathering profile for varying erosion when the acid concentration is the same value throughout the profile (e.g. driven by
equilibration to a soil atmosphere with constant carbon dioxide concentration throughout the profile depth). This figure is identical to
Figure 8.22 but with leachate control with a maximum leachate concentration = 0.4. Note that the reverse exponential weathering profile no
longer occurs because the leachate limitation at the bottom of the profile limits (panel (c)) the weathering rate rather than the acid
concentration. The initial dips in the time series of the substrate in panel (f) are the initial transients when weathering rate is unlimited and
before leachate limitation begins to limit (compare the early time evolution with that in Figure 8.22 without leachate limitation). The subsequent
rise at later times is a result of new substrate being entrained (and not being highly weathered because of leachate limitation at the soil profile
base) from the underlying saprolite by exhumation. This is also the reason why the dip occurs at earlier time for the soil profile base than at the
soil surface (it takes more time for the surface material to be refreshed than at the base of the soil-saprolite interface). For a detailed
they are able to inject new mineral mass into the profile. have a nonzero saprolite exhumation in them because of
By combining the SPR and their lumped model, Yoo and the parameters chosen.
Mudd can potentially attain a nonzero equilibrium min- In Figure 8.27 the time series of substrate with zero
eral content. If Equation (8.28) is used in the model, the exhumation is shown for the nominal and full depletion
time variation of k 1 is less problematic because it can be parameter sets. The deviation from the exponential with
changed with time during the simulation, just as in the time in Equation (8.29) is related to the exhumation rate.
fully depth-explicit model. The third assumption is more While the exact trend with time is slightly different for the
problematic. In the sensitivity studies above for the two plots, the initial concave section followed by a con-
depth-explicit chemical weathering model, the (f) panel cave up period remains. If Equation (8.29) was a good
of all the figures in this section are the depth-integrated approximation, then the lines for the three different
values for the S, which is equivalent to mj here. It is quite velocity parameters should be a straight line on the
clear that in all the simulations an S-shaped curve is semi-log plot. It is interesting to see that the substrate
almost always generated that starts concave down, tran- trends with time for the five layers within the profile do
sitions to concave up, and then levels out at the equilib- show exponential behaviour, after some time lag. This lag
rium nonzero value. This is not the exponential in is greater the deeper the layer is within the profile.
Equation (8.29). All the sensitivity simulations above This lag is because the weathering front takes some time
FIGURE 8.24: The weathering profile for varying bioturbation where the bioturbation is modelled with Fickian diffusion of the substrate
through the profile for the full depletion parameters. The no bioturbation case is for the sensitivity multiplier = 0.0. For a detailed explanation of
the panels see Figure 8.4.
to propagate down the profile. It is this depth dependency ∂A

v ¼ k 2 R
of the lag that creates the concave down section in the ∂z
(8.31)
lumped weathering plots. ∂S
E ¼ R
∂z
8.6.3 A Simple Steady State Humped Chemical and eliminating R from these equations yields
Weathering Model
∂A ∂S E
One of the conclusions of the sensitivity studies in the ¼ k 2 (8.32)
∂z ∂z v
previous sections is that it is possible to have a steady
state humped weathering profile if (1) there is active We can now see that the slopes of the weathering fronts
erosion so that new substrate is always being exhumed are related to the exhumation rate and the infiltration
from the saprolite, (2) all the acid is depleted by the velocity. It is worth noting that this result is independent
bottom of the soil profile and (3) all the substrate is of the depth of the soil profile, or the depth of the soil
weathered by the time it is exhumed to the surface. The relative to the depth of the weathering front, is true for
governing Equations (8.6) can be rewritten as every point in the soil profile, and is independent of
whether the weathering function is humped or not.
∂A ∂A
þv ¼ k 2 R An intriguing result is possible for the humped
∂t ∂z
weathering function. Note that the acid and substrate
∂S ∂S (8.30)
E ¼ R fronts are approximately linear (Figure 8.28). Over most
∂t ∂z
of their range we see that the linear part of the curve fits
R ¼ k 1 k 5 AS
Equation (8.32), so that the acid and substrate profiles can
assuming that diffusive/dispersive transport is zero. For be approximated by piecewise linear function as shown in
steady state the time derivatives are zero so that Figure 8.28. The approximate equations are
FIGURE 8.25: The weathering profile for bioturbation where the bioturbation is modelled with Fickian diffusion of the substrate through
the profile for the nominal parameters. The no bioturbation case is for the sensitivity multiplier = 0.0. For a detailed explanation of the panels
see Figure 8.4.
8
> A0 z < z1 Equation (8.34) is satisfied, then the weathering front can
>
>
< z1 z be in any location within the profile, and that it is inde-
A¼ A0 z1 z z2 pendent of the slope of the front. This suggests that the
>
> z2 z1
>
: equilibrium in Equation (8.34) is unstable so that the hump
0 z > z2
(8.33) will migrate to be attached either to the top or to the bottom
8
> 0 z < z1 of the profile. This is consistent with the observation in
>
>
< z z1 Figures 8.15 and 8.17, which shows that it is a very small
S¼ S0 z1 z z2
>
> z 2 z1 range of the erosion and velocity parameters that leads to a
>
:
S0 z > z2 humped profile where substrate is near zero at the surface
and acid is near zero at the base of the profile. Outside this
which yields a humped function approximation to that
narrow range the humped weathering profile has a signifi-
observed in the sensitivity studies with zero value for
cant nonzero value at either the top or bottom of the profile.
z < z1 and z > z2 . If we substitute the weathering front
Equation (8.33) is an approximation to the weathering
part of the piecewise approximation into Equation
function and has nonzero values in the transitions at the
(8.32), we get
boundaries of the linear part of the weathering front at z1
.
A0 ¼ k E (8.34) and z2 . To examine the impact of this approximation one
S0 2
v can imagine moving the actual weathering function and its
There are several notable aspects to this solution. This associated weathering fronts up in the profile so part of the
gives a condition within which a steady state humped nonzero part of the function lies above the soil surface. In
weathering function is possible, and the combination of Figure 8.17 this is like changing the parameter multiplier
environmental conditions under which it is possible. The from 0.9 to 1.2. The cumulative weathering on the sub-
second is that this solution is independent of z1 , z2 and strate is the area under the weathering function, so any
ðz1 z2 Þ. This says that provided the condition in part of the weathering function that is cut off by the soil
FIGURE 8.26: The weathering profile where there are two minerals being weathering by acid and where the weathering rate of the second
mineral is 0.5 and 2.0 times that of the first mineral. For a detailed explanation of the panels see Figure 8.4.
FIGURE 8.27: Our distributed depth model results showing how poorly the Yoo and Mudd (2008) lumped model approximates
the depth-averaged weathering from the full chemical weathering model. For the Yoo and Mudd model to be a good approximation to
the depth-averaged results (labelled in the legend in both panels 0.5, 1.0, 2.0; the thick lines) the decline in substrate should be linear
on this semi-log plot (Equation 8.24): (a) nominal parameters, (b) Fully depleted parameters. The legend is the same for both panels.
Note that the thick lines do not level off to an asymptotic value, rather they continue to decline to zero at a much slower rate off
scale to the right. The lines labelled i ¼ 1, 12, 25, 37, 50 are the weathering of substrate for five layers down the profile (i ¼ 1 is
the soil surface, i ¼ 25 is mid-profile, i ¼ 50 is the soil-saprolite interface) for the ‘ave.1.0’ run. The runs in these plots are
equivalent to (a) Figure 8.8 and (b) Figure 8.14 but with zero erosion and exhumation. For a detailed explanation see Figure 8.4
for panel (f).
FIGURE 8.28: The conceptual model approximation to the weathering profile. The acid, substrate and weathering panels are the same as
in Figure 8.18. The piecewise linear curve approximation (Equation (8.20)) is the heavy lines in the acid and substrate panels. The horizontal
lines that cross all three panels (1) emphasise that the depth for the breaks in the linear approximation (i.e. z1 and z2 in Equation (8.20)) occur
at the same depth for both the acid and substrate and (2) show how much of the area for the weathering curve is captured by the linear
approximation. For a detailed explanation of the panels see Figure 8.4.
surface reduces the amount of weathering that can occur else being equal, it would all have been consumed), the
within the profile. If the weathering function is cut off at acid concentration at the surface was too low, the infil-
the soil surface, then this indicates that some of the sub- tration velocity v was too small, or the exhumation rate E
strate has not been completely weathered (this follows was too large. Thus it is clear to see why the humped
from the AS multiplicative form of the reaction equation weathering functions generated are all ‘attached’ to either
in Equation (8.30)). If some of the weathering function is the soil surface or the soil base. Only when the solution to
cut off at the bottom, that indicates that not all acid has Equation (8.34) is exactly satisfied can the steady state
been consumed in the weathering. From Equation (8.34) weathering function lie entirely within the profile, and the
if some substrate is not fully consumed this was because exact location of the peak in the hump profile cannot be
either S0 was too large (if S0 were smaller, then everything determined.
9 Soils: Slow Soil Flow and Creep
9.1 Creep and Soil Velocity Despite Culling’s derivation, creep as a process is a
relatively loosely defined concept. It has been used as a
In the previous soil chapters all processes have been catchall term for all slow flow movement of soil down-
assumed to operate vertically. The only lateral movement slope, where this flow rate is proportional to gradient (or
of soil was through erosion and deposition at the surface perhaps a nonlinear function of gradient), and independ-
as a result of fluvial or aeolian processes. The soil profile ent of distance down the slope (i.e. independent of catch-
itself did not move down or across the slope. In this ment area and/or discharge). It is well established that this
chapter we will begin the discussion of how the soil dependency on slope alone leads to concave down-
profile may move down the slope. Specifically this chap- rounded hilltops (Gilbert, 1909) and that creep’s domin-
ter is about soil creep. Other processes such as debris ance over other processes at the tops of hills results in
flows, slumps and landslides, all of which move soil these rounded hilltops even in the presence of other,
(and potentially the underlying saprolite and bedrock) specifically area-dependent, processes (Willgoose et al.,
downslope at high speed, will be discussed in Chapter 13. 1991c) (see Chapter 3). Accordingly, because of the
Soil creep has historically been modelled in landform catchall nature of creep, a number of process explanations
evolution models as Fickian diffusion acting on the soil have been proposed over the years.
surface as (Culling, 1960, 1963) Field evidence has indicated that creep flux (Equation
(9.2)) varies with soil depth (e.g. Braun et al., 2001) with
∂z ∂2 z
¼K 2 (9.1) deeper soils exhibiting higher creep rates. Higher creep
∂t ∂x rates then occur when either the depth averaged velocity,
where z is the elevation of the soil/landform surface, K is the or the depth of the soil increases. In recent work Anderson
diffusivity of the creep process and x is distance in the (2015) discusses one case where the depth distribution of
downslope direction. It will be convenient for the discussion downslope velocity is also important. In the cases that are
that follows to express this in the flux form of the equation discussed below the differences between the formulations
all have implications for the soil downslope flux, the
∂z
qs ¼ K ¼ KS (9.2) average soil downslope velocity and/or the variation of
∂x soil velocity with depth.
where qs is the flux of soil in the downslope direction, and We start with creep as a flow process, which is widely
S is the slope in m/m. To derive these equations Culling used in the geomechanics community, but which appears
assumed that the transport rate of soil was linearly pro- to be inconsistent with field soil creep data. We then
portional to slope (i.e. Equation (9.2)) in an analogy to the examine downslope movement processes that involve
heat flow equation. Equation (9.1) then follows as a direct climate variability and bioturbation.
consequence of continuity of mass flux downslope if K is To allow us to judge the realism of each of the
the same everywhere. While Culling (1960) compared the approximations for soil flow presented below, field data
results of this equation with results for viscous flow, he on soil flow will be examined. Roering (2004) provides
did not derive these equations from first principles using displacement versus depth profiles for six cases across
fluid mechanics. This formulation does not explicitly clay flow and solifluction (Figure 9.1). The clear trend in
model soil movement, only the role of soil movement in these profiles is that the displacement declines approxi-
152 modifying the topography. mately exponentially with depth. Johnstone and Hilley
9.1 Creep and Soil Velocity 153
FIGURE 9.1: Measured soil

velocity profiles in the downslope
direction. Data are from a number
of sites around by other
researchers presented in Roering
(2004). Note that while these
panels are the displacements
downslope over different periods
of time (so the absolute
displacements are not comparable
between the panels), the
displacements down through the
profile for each panel do correctly
yield the velocity profile when
scaled to the mean velocity.
(2015) provide further data for a number of sites (e.g. velocity distribution within the profile is important, but
Fleming and Johnson, 1975), which are consistent with where, as we have seen from the experimental and field
an approximate exponential decline with depth. The rate data above, we face significant uncertainty about the
of decline with depth in both papers is quite variable with velocity profile resulting from creep. Anderson (2015)
length scales with depth ranging from 20 cm up to 1.2 m. calculated the flow path of individual soil particles from
Before discussing the various processes that generate the moment they are released at the soil-saprolite
creep we will consider some implications of creep. We boundary by the soil production function (Chapter 6) to
have already mentioned above how, in landform evolu- the point at which they exit the bottom of the hillslope at
tion, at catchment divides creep dominance over fluvial the hillslope-channel interface. He examined two different
erosion is important in developing the concave down- velocity profiles (1) constant velocity with depth and (2)
rounded hilltops of soil-mantled landscapes. However, a an exponentially declining velocity with depth much like
number of other applications are found where the exact in Figure 9.1. He also assumed that there was no fluvial
(a) FIGURE 9.2: Particle trajectories and

ages on hillslopes for different distributions
of velocity with depth (from Anderson,
2015). (a) Particle trajectories from the
point of release from the underlying
bedrock (bottom axis) to export from the
bottom of the hillslope (left- and right-hand
vertical axes) assuming a velocity that is
constant with depth (note that this implies
a nonzero velocity at the soil-bedrock
interface); (b) the distribution of ages of
particles from time of release from
bedrock to export from the bottom of the
hillslope for the constant velocity profile in
(b) (a); (c) as for (b) but with a velocity that is
exponentially declining with depth.
(c)
erosion and that the only way that soil exited the hillslope completely mixed removes the age difference from top
was by this soil flow process. For a simple geometry he to bottom, so the two velocity profiles will yield the same
was able to calculate the streamlines (i.e. particle paths) age distribution at the bottom of the hillslope.
for the soil for these two cases (Figure 9.2). Though the Anderson (2015) examined the impact of the profile
streamlines look similar the distributions of the particles’ velocity distribution on cosmogenic nuclide concentra-
age (i.e. time since release from the saprolite) delivered to tions. Cosmogenic nuclide concentrations are a function
the stream are markedly different. The reasons for this of the age of the particle (as well as other effects such as
difference are that (1) the older particles in both cases are the depth below the surface). He compared concentrations
concentrated near the top of the soil and (2) the velocity of sampled from the soil profile and compared them with
particles from the top of the soil into the channel are concentrations sampled from the channel sediments, and
markedly different with the exponential profile having a postulated hillslope profile velocity effects might explain
much higher velocity near the surface. Thus the exponen- observed difference between soil and channel concentra-
tial profile delivers a greater proportion of ‘old’ particles tions. The importance of the velocity though goes beyond
than the constant velocity profile. This conclusion ignores cosmogenic dating. We have shown in previous chapters
the impact of bioturbation where the vertical Fickian how physical and chemical weathering of particles is a
diffusion mixing that is typically assumed for bioturbation function of the time elapsed since the soil particles were
(see Chapters 7 and 8) will tend to make the age down the released from the saprolite. Thus if we are interested in the
profile more uniform by mixing soil particles from top to particle size distribution and/or chemical constituents of
bottom. The end member case when the profile is the soil being delivered to the channel, then just as for the
cosmogenics the channel sediments will differ from the a viscous fluid and is where the rate of shear strain is
soil profile average and the velocity profile will be one of proportional to the applied shear stress. A viscoplastic
the drivers of the difference. rheology is one where the soil does not flow until
Finally, Anderson ignored surface fluvial erosion and a threshold shear stress is reached (the plastic shear stress
its impact on the age characteristics of the channel sedi- threshold) and when the applied shear stress exceeds that
ments. Fluvial sheet erosion will preferentially remove threshold the soil flows in response to the excess shear
soil from the surface (i.e. the oldest particles), and in the stress above the threshold. A specific form of a viscoplastic
extreme case where soil flow is zero all sediment in the rheology is a Bingham fluid, where the relationship
channel will have the age distribution of the soil surface, between stress and strain exceeding the threshold is linear.
weighted by the spatial distribution of the erosion. In the These rheologies are shown in Figure 9.4. Other types of
case of gully or rill erosion, the soil eroded will be fluids are commonly studied in the rheology literature (e.g.
averaged over the depth of the rills/gullies, so the sedi- thixotropic/shear thinning, shear thickening, Herschel-
ment supplied to the channel will reflect the age distribu- Bulkley), but Bingham and Newtonian fluids are the ones
tion of the soil surface down to the depth of the rills/ commonly discussed for modelling slow soil flow (e.g. van
gullies. Ash and van Genuchten, 1990; Ancey, 2007; Balmforth
Having now exemplified the importance of the soil et al., 2007). We will start our discussion using Newtonian
velocity distribution we will now discuss various pro- rheology and then extend this discussion to Bingham
posed physical mechanisms for soil creep. rheology.
9.2.1 Constant Depth, Newtonian Rheology

9.2 Creep as a Slow Flow Process
The first case to consider is the case of a Newtonian fluid
In this section we will discuss the basics of soil flow.
with a constant soil depth down the hillslope (Figure 9.3).
A good analogy for soil flow is honey flowing down a
Summing the forces parallel to the soil-saprolite interface
flat surface; the honey is the soil and the flat surface is the
yields
underlying saprolite. Figure 9.3 shows a schematic of
the process showing the force balance for a small section τΔx
of the soil down the hillslope. If the forces are in balance, T¼ ¼ W sin θ þ ðF u F d Þ cos θ
cos θ
then the soil will move down the slope with a
¼ ρb gðD zÞΔx sin θ þ ðF u F d Þ cos θ (9.3)
space-invariant velocity and will be in steady state. If
the forces are not in balance, then the soil will either where ρb is the bulk density of the soil, g is the acceler-
accelerate or decelerate. We will focus on the steady state ation due to gravity, D is the depth of the soil, θ is the
velocity case. angle of hillslope and T and τ are the shear force and shear
The relationship between the applied shear stress and stress, respectively, acting on the saprolite at the soil-
the resulting shear strain per unit time of a fluid is called saprolite interface. If θ is small, then sin θ ¼ tan θ ¼ S
its rheology. Soil is commonly modelled with a viscous or where S is the slope (in m/m). Since the soil flow direction
viscoplastic rheology. A Newtonian fluid is an example of is parallel to the soil surface, then the pressure distribution
FIGURE 9.3: Schematic of

downslope soil flow for viscous or
viscoplastic flow.
FIGURE 9.4: Shear stress versus strain rate showing some of the FIGURE 9.5: Velocity profiles for different rheologies for
viscosity relationships (the slope of the curve) for rheologies that are viscous flow.
commonly used for fluid flow.
ρb gD2 sin 2θ 2
down the soil profile is hydrostatic, the pressure at v¼ ¼ v0 (9.7)
6μ 3
height above the saprolite z is p ¼ ρb gðD zÞ and
F u ¼ F d ¼ ρb gD
2
2 . Accordingly, since the soil depth is The displacement over time is obtained by multiplying
constant downstream, F u F d ¼ 0. The shear stress par- Equations (9.6) and (9.7) by time. Equation (9.6) is con-
allel to the soil-saprolite interface, at any depth above the trolled by two parameters
soil saprolite interface, z, is then
ρb gD2 sin 2θ
ρ gðD zÞ P1 ¼
τ ¼ ρb gðD zÞ sin θ cos θ ¼ b sin 2θ (9.4) 4μ (9.8)
2 z
P2 ¼
D
The appearance of sin 2θ indicates that the shear stress
increases until the slope is 1:1, and for slopes steeper than where P1 is the parameter that determines the rate of flow
1:1 the shear stress decreases with increasing slope. Since of the profile, while P2 determines how that velocity is
most natural soil mantled hillslopes have slopes that are distributed through the soil profile. Finally it is possible to
less than 1:1, the shear stress increases with an increase in recast the soil discharge calculated using Equation (9.7)
slope. For a linear viscous fluid (a Newtonian fluid) into the form of classical creep (Equation (9.2)) where the
dz where μ is the absolute viscosity, and V is the
τ ¼ μ dV diffusivity K is
downslope velocity at depth z so that
ρb gD3
dV ρb g sin 2θ K¼ (9.9)
¼ ð D zÞ (9.5) 3μ
dz 2μ
if the slope is small, so sin 2θ ’ 2 sin θ ’ 2S and noting
and integrating with depth (noting that V ¼ 0 at z ¼ 0)
that qs ¼ vD: The main conclusion from this formulation
yields the velocity profile (Figure 9.5)
is that the diffusivity in Equation (9.2) is then dependent
ρ g sin 2θ 2 z 2
on the cube of the soil depth.
v¼ b D z2 ¼ v 0 1 (9.6)
4μ D However, since the velocity profile in Equation (9.6)
(Figure 9.5) does not match the observed soil displace-
where v0 is the velocity at the soil surface. The depth- ment profiles in Figure 9.1, we conclude that this model
averaged velocity is does not provide a realistic model of creep.
9.2.2 Constant Depth, Depth-Dependent orders of magnitude change in soil viscosity over the
Newtonian Rheology typical range of soil moisture and for common soils.
A high organic content is associated with decreased
One way to address the unrealistic velocity profile for viscosity, but the effect is small relative to the effect of
Newtonian rheology in the previous section is to allow clay and moisture content (Hallet et al., 2010; Markgraf
the viscosity to vary with depth. Experimental evidence et al., 2012). Since the organic content is concentrated at
shows that the displacement increases with increased soil the soil surface this is consistent with a negative α:
moisture and/or clay content, so a systematic variation of In conclusion, while it is possible to create a soil
viscosity through depth is likely. As a sensitivity study we velocity profile that is consistent with observed soil dis-
will assume that the viscosity changes linearly with depth placements using a depth-dependent viscosity, the depth
so that dependency of soil viscosity required appears not to be
z consistent with observed relationships other than a weak
μ ¼ μ0 1 þ α (9.10)
D trend with organic matter.
where μ0 is the viscosity at the soil surface and α is a
constant controlling the rate at which the viscosity
changes with depth. Substituting this expression for vis- 9.2.3 Constant Depth, Bingham Rheology
cosity into Equation (9.5) and integrating yields the vel-
Bingham rheology has a shear stress threshold below which
ocity profile
the fluid does not flow, and then fluid is Newtonian for shear
stress above that threshold. The viscosity above the thresh-
ρb gD sin 2θ D D D þ αz z
v¼ þ 2 ln (9.11) old is μ. Equation (9.4), which gives the shear stress at any
2μ α α D α
depth within the soil, shows that the shear stress increases
Figure 9.5 shows that for positive values of α the with depth. For the soil above the threshold depth, at which
profile is only marginally different from, and no more τ ¼ τ t , the soil does not shear, but rather flows as a ‘plug’ on
realistic than, the constant with depth viscosity. However, top of the shearing soil (called ‘plug-flow’) underneath.
for negative α the velocity profile approaches a linear Below the threshold depth, zt , the effective shear stress used
relationship with depth, which is visually similar to the to calculate the velocity distribution is
field displacement data of Kirkby (1967) and Fleming and ρb gðD z zt Þ
Johnston (1975). As in the previous section the average τ¼ sin 2θ
2
velocity derived from Equation (9.11) can be used to (9.12)
2τ t
transform it into the form of Equation (9.2) by noting that zt ¼ D
ρb g sin 2θ
the slope term in Equation (9.2) can be derived from the
sin 2θ term as before. After transformation into Equation and the velocity profile is
(9.2) the diffusivity is again approximately dependent on 8 2 !
>
the cube power of soil depth. < ρb g sin 2θ zt 2 z2 ¼ v0 1 z
>
z < zt
A negative value of α implies that the soil viscosity is v¼ 4μ zt
>
>
:
highest at the soil base and lowest at the surface. This v0 z zt
would appear to conflict with most observed relationships
ρb g sin 2θzt 2
for changing viscosity of soil, which include clay content, v0 ¼
4μ
soil moisture and organic matter content. Clay content is
positively correlated with decreased viscosity (and thus (9.13)
increased velocity). Typically the clay content of soil We see that Equation (9.13) is controlled by the same two
increases down the profile, and that would imply decreas- parameters as for the Newtonian fluid, the velocity at the
ing viscosity with depth. Increasing soil moisture is asso- surface determining the rate of flow down the slope, and
ciated with decreased viscosity. The soil moisture the distribution of that velocity through the profile, this
distribution down the soil profile will depend on local time normalised by the threshold depth rather than the
hydrology and climate controls, and it is possible for total depth of the profile. The velocity profile in Equation
soil moisture to either increase or decrease with depth (9.13) (Figure 9.5) clearly does not match the observed
(with decreasing and increasing viscosity, respectively) soil displacement profiles in Figure 9.1. Specifically the
depending on the local conditions. Fleming and Johnson region of plug flow with constant velocity in the top of the
(1975), and Ghezzehei and Or (2001) suggest two to three soil does not appear in any of the field data. We thus
conclude that Bingham rheology does not provide a real- 9.3 Creep as a Cyclic Expansion-
istic model of creep despite the laboratory evidence that Compaction Process
this is the best model for soil flow in the absence of field
processes (i.e. processes not active in the laboratory The previous sections have treated soil as a viscous or
experiments). Like the other viscosity models for low viscoplastic flow. In this section we will examine the
slopes, this can be transformed into Equation (9.2), where consequences of soil expansion and contraction combined
the diffusivity is now dependent on the total soil depth with the action of gravity. This expansion and contraction
and the depth of the threshold but it is still linear in slope. may occur as a result of various processes, but the ones
The dependence on soil depth is now more complicated most commonly discussed in the literature are freezing-
than the cube of the depth because of the form of zt in thawing (e.g. solifluction) (Matsuoka and Moriwaki,
Equation (9.12). For τ t ¼ 0 the dependence on depth 1992) and wetting-drying of expansive clay soils (Kirkby,
reduces back to the cube of depth as in the pure Newton- 1967; Fleming and Johnson, 1975). The central concept of
ian case in Equation (9.9). this process is that as the soil expands the surface moves
at right angles to the soil-saprolite interface (the slope of
which is implicitly assumed to be the same as the soil
surface). When it shrinks, the soil tends to fall vertically
9.2.4 Variable Depth
under the action of gravity (Figure 9.6). Only the upper
Many differences in the soil flow arise as a result of the portion of the soil will expand and contract (e.g. daily
soil depth variations in the downslope direction. The temperature fluctuations typically only penetrate a few
driver for soil thinning is that the soil-saprolite interface 10’s of centimetres), so the vertical displacements are
rises or falls relative to the soil surface (e.g. as a result of relative to the depth above which the process acts. This
spatial changes in soil production function; Saco et al., is the datum in Figure 9.6. If the expansion is the same all
2006). For steady state conditions without soil production the way down the profile to the datum, then the expansion
Qu ¼ Qd , where Q is the depth integrated downslope soil at any depth above the datum is linearly dependent on the
discharge. This ensures that the soil does not thicken or height above the datum. This leads to the piecewise linear
thin with time upslope of the soil depth change. The velocity profile
consequences of this for soil flow are the following: ( z zd
v0 for z > zd
v¼ D zd
• F u F d 6¼ 0: In Figure 9.3 the depths at the upslope 0 for z zd (9.14)
and downslope of the control volume are now differ-
v0 ¼ K 1 S
ent, so the net effect of the hydrostatic pressure on the
upslope and downslope interfaces of the soil control where v0 is the velocity at the surface and is a function of
volume is no longer balanced. If the depth is decreas- the vertical movement during each expansion-contraction
ing (i.e. travelling over a rise in the saprolite interface) event, how much of the vertical movement in each event
then F u F d > 0 and there is a net pushing force to is transformed into permanent downslope transport, the
soil movement, and vice versa if the soil depth is slope of the surface and how often the events occur. As in
increasing. the previous section this can be recast into Equation (9.2)
• The soil thins over the rise in the saprolite boundary and by noting that
the velocity of the soil increases to maintain uniform
v0 ðD zd ÞK 1 S
soil flow conditions downslope (i.e. Qu ¼ Qd where Q qs ¼ ðD zd Þ ¼ (9.15)
2 2
is the depth integrated soil discharge). This velocity
so that K ¼ ðDz2d ÞK 1 in Equation (9.2). This is similar to a
increase arises as a result of continuity.
formulation for solifluction used by Bovy et al. (2016)
• As the soil thins over the bump the streamlines of the
where they generalised the relationship so that
soil flow converge, the velocity gradient increases, the
shear stress/unit depth increases, but the bottom shear ðD zd Þe K 1 Sg
qs ¼ (9.16)
stress decreases because of the decrease in soil depth. 2
In the case of Bingham rheology we have one extra where e and g are parameters that they assumed to be 1.7
factor. The depth at which the shear stress equals the shear and 0.5, respectively, though no justification for these
stress threshold in the Bingham rheology also rises as the parameter values was provided.
soil passes over the rise. This needs to be accounted for in One extension to this simple model is to recognise that
the calculation of the discharge continuity. expansion and contraction events are not all the same size.
9.4 Creep as a Random Motion Process 159

downslope soil flow for one cycle
of expansion-contraction.
Long-term freezing events will penetrate deeper than short a result of Brownian motion. Culling recognised that soil
term freezing events. The frequent short events will only particles do not have free movement (as compared with
penetrate a small distance into the soil and result in small Brownian motion), and motion is thus limited by the
displacements only impacting on the near surface of the availability of pores in the soil that particles are physically
soil, while less frequent long events will penetrate deeper able to move into. By assuming a random distribution of
with larger displacements that penetrate further into the soil. available pores he was able to derive Equation (9.1) in
Elias et al. (2004) showed that the temperature range as a terms of the pore distribution and mean travel distance in
result of a sinusoidal temperature variation at the soil sur- the soil.
face declines exponentially with depth where the rate of An interesting observation that Culling made, but did
decline is a function of the thermal conductivity and specific not pursue, was that due to compaction (due to overlying
heat capacity of the soil (NB: This varies with soil moisture soil) the number of available pores would decrease with
content). If the velocity downslope is a linear function of the depth so that the diffusivity (and as a consequence the
temperature range, then a simple empirical relationship that downslope rate of transport) would decrease with depth
captures this trend without a threshold depth is the exponen- down the soil profile.
tial velocity profile adopted by Anderson (2015) Culling also noted that vegetation might influence
available pore distribution, and that the presence of roots
v ¼ v0 eðDzÞ=λ (9.17) may inhibit movement. Recent work has emphasized the
where λ is a parameter that determines the rate at which importance of macropore distribution for hydrology
the velocity declines with depth. More accurately the (Beven and Germann, 1982, 2013), so it is reasonable to
velocity will be a function of the number of times that assume that macropores might be an important, but
the temperature range crosses 0C, since the expansion unquantified, factor in movement downhill in the random
and contraction is primarily a function of the transition motion process, and the cavities probably act as an
from freeze to thaw and back again, rather than a function enhancement of downslope movement.
of the absolute temperature range. In this case if we assume Brownian motion for soil
Comparing these velocity profiles with data, the linear particles, then we can model creep by the Fickian diffu-
profile appears to be consistent with the velocity profiles sion equation
of Fleming and Johnson (1975) and Kirkby (1967), while
the exponential profile appears consistent with the data of dc d2 c d2 c d2 c
¼ K rx 2 þ K ry 2 þ K rz 2 (9.18)
Roering (2004) and West et al. (2014). dt dx dy dz
where c is the property of the soil that is being mixed by

9.4 Creep as a Random Motion Process the random process, and K rx , K ry and K rz are the diffu-
sivity in the x, y and z directions, respectively, and x, y are
Culling (1963) extended the Strahler (1952) idea that soil the two horizontal directions and z is in the direction of
flows downhill from random movement of soil particles as gravity. A natural orientation for x and y would be in the
steepest downslope direction and at right angles along the process with the same equations used for rainsplash (see
elevation contour, respectively. The diffusivity varies a derivation in Chapter 4)
with the mean travel distance per unit time, so it seems
∂z ∂2 z
likely that the diffusivity in the vertical direction will be ¼ Kr 2 (9.19)
different from those horizontally (i.e. K rz 6¼ K rx and ∂t ∂x
K rz 6¼ K ry ), and it is also conceivable that the diffusivity where K r is the diffusivity associated with rainsplash. If
in the across-slope and downslope directions will also be rainsplash is explicitly modelled in the landscape evolu-
different (i.e. K rx 6¼ K ry ). In this latter case macropores in tion model then Equation (9.19) does not need to be
some environments appear to be biased toward being modelled separately and can be lumped in with all of the
oriented in the direction of subsurface flow; i.e. down- other rainsplash effects (provided of course that the model
slope. Heimsath et al. (2005) implicitly assumed that all does simulate the deposition of the material on the surface
three diffusivities were the same and used cosmogenic by the ants).
data to estimate the vertical mixing rate and thus K rz : An unusual form of vertical soil mixing is gilgai soils,
They then used this estimate of K rz to estimate the down- which are common in Australia, the United States and
slope horizontal transport rate without justifying their Russia. Gilgai soils contain highly expansive soils. When
assumption of K rz ¼ K rx : the soils dry, they dry more rapidly at the surface and
large deep cracks appear at the surface. When rain occurs,
the water enters the soil through these cracks, and the soils
9.5 Creep and Bioturbation at the bottom of the cracks where the water collects
(typically metres below the surface) expand into the space
The main impact of bioturbation is to mix the soil available in the cracks, which results in deeper soils being
vertically. The net effect of this is that any differences forced to the surface through cycles of drying/cracking
in soil properties (e.g. viscosity) between the top and the driven at the surface and wetting/expansion driven from
bottom of the soil profile will be reduced. Whether the the base of the soil. Brimhall et al. (1992) discuss the
differences are removed completely or not will depend shrink-swell behaviour of soil profiles (in their case the
upon the relative rate of the mixing by bioturbation and shrink-swell behaviour arises from chemical weathering)
the processes producing the differences (e.g. physical and propose a simple model that might be useful for
and chemical weathering). This is no different from the modelling gilgai soils.
effect of bioturbation discussed in the physical and
chemical weathering chapters (Sections 7.3.5 and
8.3.4). If the soil profile weathering is modelled expli- 9.6 High Slopes
citly, then bioturbation can be handled at the same time
as for weathering and creep. If on the other hand It is convenient to now start talking about what happens at
weathering processes are not modelled, then some form higher slopes. While not strictly in the realms of soil slow
of explicit vertical mixing model should be considered. flow, there is a simple modification of the Culling creep
While the driver of the process is different, the math- equation that is commonly used to model hillslope evolu-
ematics is the same as for creep as a result of random tion at higher slopes. It is normally used to model the
movement. Most authors simply consider the mixing to elevations of the landform and has not been used for
be Fickian. However, it is worth noting that, as dis- modelling the evolution of the soil depth. When slopes
cussed previously, some forms of bioturbation such as are close to their angle of repose, then slope stability/
ants and termites are not simply Fickian mixing pro- failure becomes an issue, but this is not reflected in any
cesses, and explicitly move material from within the of the equations above. Andrews and Bucknam (1987)
profile directly to the surface. proposed a modification of Equation (9.2)
Bioturbation may also have a horizontal component. If
∂z
we consider for the moment the case of ants and termites,
qs ¼ K ∂x
then as the soil is brought to the surface there is a spread- ∂z
2 (9.20)
1 ∂x St
ing process at the surface, with the surface mound being
spread in all directions by the ants depositing the material where St is a threshold slope, considered equal to the
around the nest opening, and local erosion spreading it coefficient of friction of the soil surface. At this threshold
further afield. If the erosion process at the surface is rain- slope the rate of movement of soil downslope tends to
splash, then it is reasonable to model this horizontal infinity, and this is the angle of repose of the slope.
This equation was derived by moving a particle upslope/ 2015) since most of the processes yield different velocity
downslope with a given energy and determining the travel profiles, in addition to different relationships between the
distance of the particle sliding across the surface with a surface velocity and the profile average velocity.
coefficient of friction St : The distance travelled upslope is One tentative conclusion can be reached. The repre-
less than the travel distance downslope, and Equation sentation of soil flow using methods developed in the
(9.19) is the net sediment transport downstream. As the geomechanics field for viscoelastic fluids, and which
slope approaches the coefficient of friction, the downslope ignores field scale processes such as expansion and con-
particle will approach a travel distance of infinity. The traction, and bioturbation, appears to be a poor fit
power of 2 in the denominator arises naturally out of this to observed displacement profiles in the field. This
calculation and is not a fitting parameter. Andrews and still leaves an open question over what is the soil
Bucknam fitted these data to the rounding of old fault depth dependency of the diffusivity in the Culling
scarps and found a better fit than using Equation (9.2). Fickian model.
Roering et al. (1999, 2001, 2007) fit this equation to
hillslope profiles and also found that it gave a good fit to
hillslope profile, and that it fit better than assuming Equa- 9.8 Numerical Issues
tion (9.2) for slopes less than the threshold slope St , with
infinite transport for slopes greater than St : For the classic Culling depth-integrated creep there is the
Roering (2008) extended Equation (9.20) by consider- choice of whether to use Equations (9.1) or (9.2). Equa-
ing that the rate constant K was dependent on the depth tion (9.1) can be directly applied to the surface elevations
below the soil surface, and by assuming that the main but can be problematic if the planar boundaries of the
trigger for soil movement was the root density of trees, domain are not rectangular. In this case applying the
which decline exponentially with depth (Canadell et al., boundary condition (typically a zero slope, no flow, con-
1996). Equation (9.20) is then a depth-averaged equation dition) can be messy. On the other hand, the flux form in
after integration over the soil depth with depth-dependent Equation (9.2) when used in conjunction with a drainage
root density analysis algorithm (e.g. D8; see Chapter 4 for a detailed
discussion of drainage analysis algorithms) is easier to
K ¼ K 0 1 eβD cos θ (9.21)
apply at irregular boundary conditions. The downside of
where β is the length scale of the rate of decline of root a D8-based flux formulation is that normally diffusion
density with depth and K 0 is the rate constant for root dominated (i.e. creep-dominated) landforms tend to be
disturbance. Roering found that this equation gave a better convex down, and the flow directions are divergent. The
fit to their study site than assuming a constant value for K D8 algorithm does not model this divergence since it
with depth in Equation (9.20). allows flow only to one downslope node. The D∞ algo-
rithm goes some way to addressing this by allowing two
downslope flow directions. The main impact of the lack of
9.7 Conclusions divergence of D8 is that for short slopes (a few nodes
long) generated landforms tend to look a little rougher
The main conclusion to this chapter is that if all that needs than with Equation (9.1), and the catchment divides are
to be modelled is the mass flux to determine the soil sharper and less rounded. On the other hand, the total
surface elevation and the soil depth is the same every- mass flux is the same for both D8 and D∞ (Willgoose,
where, then most of the different processes can be reduced 2005a).
to the Culling Fickian diffusion model that has been When using TINs instead of grids, the flux formulation
traditionally used in landform evolution modelling. How- of diffusion is also much easier to use, and this is the
ever, if the soil depth is changing, as is likely to be the approach used in the CHILD LEM (Tucker, personal
case when modelling a coupled soilscape-landform evo- communication).
lution model, then the diffusivity in the Culling equation Perron (2011) proposed an implicit solver for Equation
is dependent on soil depth. Not only that but the func- (9.20), which we discussed in Section 4.7.
tional form of the dependence on the soil depth differs To date no researcher has published a creep model that
between the processes. Unfortunately, at this time there is models depth-dependent variations of velocity (Anderson
no consensus on what is the appropriate physics. The (2015) calculates analytical rather than numerical solu-
problem becomes worse if we are interested in the vel- tions), so we have no experience to share on numerical
ocity profile of the soil flow down the slope (Anderson, issues. We anticipate the main complexity here will be in
the coupling of the layering in the soil profile at each node vertical movement of soil streamlines when moving
with the horizontal movement of each layer when the soil downslope (and thus vertical movement and thickness
depth and velocity are varying downslope. As indicated change in the upslope layers as the profile moves down-
when we discussed viscous flow for a varying depth soil, slope). Thus explicit accounting for material exchange
the thickness of the layers will vary downslope as the between the layers (potentially defined differently at each
velocity downslope varies, so the depth boundaries of node on the hillslope) as the soil moves downslope is
the layer discretisation of the soil profile at the upstream required. In principle this is no different from the material
node will not match the layer discretisation of the soil exchange vertically within each individual profile that was
profile at the downslope node. Furthermore, Anderson performed in the physical weathering using mARM
(2015) shows that even with constant soil depth, the (Chapter 7), but it potentially adds an additional vertical
combination of soil production and creep results in numerical diffusion that may be an issue.
10 Soils: Colloids and Soil Organic Carbon
10.1 Introduction soil. Total vegetation biomass is about 50% of the total
biomass, and about 50% of vegetation biomass is above-
This chapter focuses on (1) clay transport within the soil ground (Larcher, 2003), so that about 75% of terrestrial
profile and (2) the dynamics of soil organic matter, of biomass is in the soil. This chapter is focussed on this
which soil carbon is the most important component. Clay 75% of terrestrial carbon. Figure 10.1 also makes clear
and soil organic matter combine in the soil profile and are that the difference between carbon release and carbon
a major driver of the soil structure evolution (the ‘clay- sequestration is quite small, so errors in the calculation
organic complexes’; Tisdall and Oades, 1982), and soil of the fluxes in the terrestrial carbon cycle will make large
structure is a major influence on the infiltration rate of relative errors in the net balance of the fluxes and thus
rainfall into the soil. There is a strong relationship rates of change in carbon storage.
between soil carbon content and soil water-holding cap- SOC is highly concentrated near the surface of the soil
acity (Rawls et al., 2003). Soil organic matter consists of and declines with depth, with 54% in the top metre
compounds containing carbon, phosphorus, nitrogen and (~1,500 Gt) and 26% (~615 Gt) in the top 20 cm. In
sulphur. Operational models for the evolution of soil water-limited ecosystems SOC storage is also highly cor-
organic matter typically model the cycles of these elem- related in space with soil moisture (Hobley et al., 2015;
ents (e.g. the CENTURY model; Metherell et al., 1994), Hobley and Wilson, 2016). This relationship is because
but in this chapter we will consider only soil organic plant growth and biomass are higher for higher soil mois-
matter (SOM) in total, or the carbon component of it, soil ture, so this source term for SOM is accordingly higher.
organic carbon (SOC). SOM contains 58% SOC in units SOC is important because it is the dominant sink for
of mass C/mass soil. SOM contains particulate organics, carbon in the terrestrial environment (e.g. forests soils store
humus, charcoal, living microbes, root exudates, fungi up to 90% of global SOC; Peterman and Bachelot, 2012),
and fine plant roots. The living biomass (the vast majority and recent work with climate models suggests that it is the
of which is microbes and fungi) is a major influence on most sensitive sink to climate change because of the rapid
the decomposition of dead organic matter and the storage change of the storage dynamics with changes in soil mois-
of carbon and nutrients (Wardle, 1992). ture (Friedlingstein et al., 2006). Stockmann et al. (2013)
Jobbágy and Jackson (2000), Houghton (2007) and noted that a 10% change in the stored SOC equals about
Stockmann et al. (2013) provide a summary of the fluxes, 30 years of current-day anthropogenic emissions.
storage and importance of the SOM and SOC. Figure 10.1
summarises the carbon cycle. While fluxes and storage
estimates are uncertain (Johnston et al., 2004) there is 10.2 Colloid Migration and Illuviation
general agreement that the amount of carbon in terrestrial
biomass (~550 Gt) is about equal to the amount of carbon Colloidal material moves down the profile through the
in the atmosphere (~800 Gt), and that the amount of SOC pores in the soil carried by infiltrating water. The colloidal
in the top three metres of soil is about 5 times the terres- materials we are interested here are clay and biochar from
trial biomass (~2,300 Gt). Soil carbon is the second big- wildfires. There is also evidence of microbial transport, but
gest store of carbon after the deep ocean. The vast we will not discuss this here. The principle of clay migra-
majority of the terrestrial biomass is in the soil with only tion is that clay is released at the surface of the soil profile
the aboveground biomass of vegetation not located in the due to rainsplash or as a result of chemical weathering 163
Atmosphere
597 + 165
120
0.2 119.6 2.6 Land 1.6 Land
GPP sink Use 6.4
Weathering Respiration Change
70.6 70 22.2 20
Vegetation, Fossil Fuels

Soil & Detritus 3700 – 244
2300 + 101 – 140
0.4 0.8
Rivers
Surface Ocean 50 Marine Biota
Weathering 900 + 18 39 3
0.2 90.2 1.6
101
11
Intermediate
& Deep Ocean
37,100 + 100
0.2
Reservoir sizes in GtC Surface sediment
Fluxes and Rates in GtC yr–1 150
FIGURE 10.1: The global carbon cycle. Fluxes and storages in black are natural; fluxes and storages in grey are anthropogenic. Units are
GtC or GtC per year. Houghton (2007) subdivides the vegetation, soil and detritus box into vegetation 550, detritus/litter 300 and soil
1,200 GtC so that soil carbon is about 60% of the terrestrial storage. For fluxes the soil litter flux into the soil approximately matches
decomposition (from IPCC, 2007).
within the profile, and then is transported down through the coefficient approximately equal to 0.7 and θ is the soil
profile by infiltration. It is then deposited within the profile moisture (volume m3/m3).
by a number of mechanisms: (1) flocculation, which occurs This equation models only mechanism (2) and does
if the salinity of the soil water increases down the profile, not account for the deposition of particles when infiltra-
(2) particles get caught in small pores within the profile, (3) tion stops, when water is transpired or evaporated from
deposition in pores, which occurs when infiltration has the profile, or flocculation.
reached its maximum depth in the profile for an individual Accompanying the transported clay is material trans-
rainfall event or (4) deposition on mineral surfaces (DeNo- ported adsorbed on the clay surfaces. In a typical soil
vio et al., 2004; Bradford and Torkzaban, 2008). Similar these components are organic molecules, phosphate and
processes drive biochar movement. metals. The amount adsorbed per unit surface area of clay
Rainsplash was discussed in detail in Chapter 4. The is strongly pH dependent with adsorbed amounts decreas-
key points are that the low salinity of rainwater disperses ing with decreased pH (Dijkstra et al., 2004). If pH
clay, and the energy of the rainsplash detaches the clay decreases, it is common to observe the release of these
particles from the surface, making them mobile. The adsorbed constituents into the soil water solution.
mathematics of chemical weathering leading to clay for-
mation was discussed in Chapter 8.
10.3 Single Soil Layer Soil Organic Carbon
Jarvis et al. (1999) provides an empirical equation for
Models
the capture of clay particles F (g/m3/hr) within the profile
F ¼ f ref vnref v1n cθ (10.1) Figure 10.2 shows how most soil organic carbon (SOC)
model conceptualisations are constructed. Generally the
where f ref is a media calibration coefficient measured at a soil profile is considered as one mixed layer, with no
reference flow velocity vref , v is the specific discharge, c is distinction being made with the storage at different depths
the particle concentration (mg/l), n is an empirical within the profile, so depth dependencies of SOC are
minerals (clay and others) is more resistant to

decomposition.
• Root exudates play a strong role in priming the fungal
decomposition of lignin.
• Leaf litter decomposition results from different pro-
cesses compared with SOC. Roots are significantly
more resistant to decomposition than shoots.
We will discuss the role of erosion (Chapter 4) in SOC

movement and SOC sequestration later in this chapter.
Two models of the soil carbon cycle are the basis of
many of the soil carbon models used worldwide (and are
themselves based on the model of Henin and Dupuis,
1945) and are the basis for the following discussion:
RothC (Jenkinson and Coleman, 2008; Coleman and
Jenkinson, 2014) and CENTURY (Metherell et al.,
1994. The discussion that follows focuses on the general
FIGURE 10.2: A schematic of the soil carbon cycle showing the principles in both of these models, but will use the termin-
surface leaf litter compartment and three SOC compartments. Note ology of CENTURY for consistency. While many SOC
that the third SOC compartment is called the ‘passive store’ in models follow the general structure of these two models,
CENTURY and ‘inert store’ in RothC (figure after Jenkinson and
they differ in detail. Plante and Parton (2007), Byrne and
Coleman, 2008). The sources of fresh carbon are primarily leaf litter
and fine roots, but only leaf litter is impacted by erosion.
Kiely (2008), and Stockmann et al. (2013) compare a
number of these models.
Within each of the compartments in the model
ignored. While it is not generally explicitly stated, the
(Figure 10.2) there is an input from some or all of the
single-layer models are typically applied to only the top
other compartments, and a decomposition that produces
30 cm of the soil, since this is where most of the SOC
(1) SOC products that are then inputs to other compart-
resides and where the temporal dynamics are strongest.
ments, (2) leaching of carbon in the form of dissolved
Figure 10.2 shows a number of conceptual compartments
organic carbon (DOC) and (3) release of carbon dioxide to
in the soil, each compartment being allocated a different
the soil atmosphere. The decomposition rate of SOC from
decomposition rate (Trumbore, 2009) and residence
all pools is by first order kinetics (Jenkinson and Raynor,
times. The surface SOM compartment is mostly leaf litter,
1977; van Veen and Paul, 1981) so that the change in the
and the three SOC compartments within the soil profile
SOC stored in any specific compartment is
have varying rates of decay (and thus residence time).
These compartments are not physically separated in the ∂C
¼ I KC (10.2)
soil but reflect the range of (bio)degradability of the soil ∂t
carbon in the profile. The residence times are determined
where C is the SOC concentration (mass SOC/mass soil),
by use of radiocarbon dating. The reasons for these differ-
I is the input to that compartment from all the other
ences in rates and residence times are an area of active
compartments and the rate parameter K is different for
research and arise from a range of sources (Schimel et al.,
each compartment. The rate parameter K is multiplied by
1994; Torn et al., 1997; Schmidt et al., 2011; Stockmann
a factor to account for the exponential increase in decom-
et al., 2013; Kaiser et al., 2016) including the following:
position rate with increase in temperature (Davidson and
• Chemical composition of the organic compounds, par- Janssens, 2006) and linear increase in decomposition rate
ticularly the carbon:nitrogen ratio and lignin content. with soil moisture (Davidson et al., 1998). The soil mois-
Higher lignin content SOM is more stable. ture link is because soil moisture deficiency reduces the
• Level of decomposition or SOC age. The least stable abundance of soil bacteria and fungi, in turn reducing both
organic molecules containing carbon are degraded first, SOM content directly but also the decomposition rate of
and the remaining decomposition products are more SOC (Maestre et al., 2015). Residence times for SOM for
decomposition resistant. temperate forest and grasslands are 20–90 years, tropical
• Soil grading and mineralogy influences soil aggregation grasslands 10 years, and for humic fractions and charcoal
and potential stability. SOM chemically associated with thousands of years (Horwath, 2007).
The decomposition is partitioned between the three Equation (10.2) yields a steady state solution C∗ ¼
destinations (i.e. the other compartments, DOC and I=K: The approach to this equilibrium is exponential:
CO2). The rationale for three compartments is that the
I
fast compartment (containing highly ‘labile’ SOC) C ¼ C0 eKt þ 1 eKt (10.3)
K
simulates the fast breakdown of fresh plant material,
the slow compartment the breakdown of the more The governing equations for the three-compartment
resistant material, and the passive compartment is system in Figure 10.2 are
poorly degradable material (low lability SOC) and ∂C1
charcoal (resulting from fire). It is generally recognised ¼ I K 1 C1
∂t
that there is a spectrum of decomposition rates from fast ∂C2
to slow (e.g. Gignoux et al., 2001). The state of current ¼ α12 K 1 C1 K 2 C2 (10.4)
∂t
knowledge of SOC chemistry, interactions with the soil ∂C3
mineral component (e.g. clay) and the role of environ- ¼ α13 K 1 C1 þ α23 K 2 C2 K 3 C 3
∂t
mental conditions in SOC decomposition rates is such
where αij is the proportion of the decomposition in com-
that more sophisticated representations are probably not
partment i that transitions to compartment j. If the chemical
warranted provided more than one SOC compartment,
composition of the compartment i remains the same and
with different decomposition rates in each compartment,
the decomposition process is also unchanged with time,
is modelled (Davidson and Janssens, 2006; Plante and
then it is reasonable to assume that αij remains the same
Parton, 2007). Bernoux et al. (1998) found little differ-
even if the rate of decomposition changes (e.g. with soil
ence in predictions of in-profile SOC when comparing a
moisture). This simplification allows us to solve this equa-
single-compartment model with a two-compartment
tion for the equilibrium concentrations (equilibrium values
model with different breakdown rates in each compart-
are denoted by the superscript *) if the carbon input I is
ment, but found different fluxes of carbon dioxide to the
constant and the decomposition rates are constant so that
atmosphere. Trumbore (2000) found that a one-
compartment model gave significantly different results I
C1 ∗ ¼
from a two-compartment model, when examining the K1
potential impact of climate change. CENTURY divides I
C2 ∗ ¼ α12 (10.5)
the leaf litter compartment into structural C (high lignin K2
concentration plant products) and metabolic C (low I
C3 ∗ ¼ ðα13 þ α23 α12 Þ
lignin concentration) to better model breakdown K3
processes before the plant materials are incorporated The time-varying solution can be derived using the stand-
into the soil, while RothC has a single leaf litter ard method for solving first order differential equations.
compartment. We also need to note that C1 equilibrates much faster than
One limitation of Equation (10.2) is that it doesn’t C2 and C 2 much faster than C3 so that at the time scales at
directly allow for the rate of decay to decrease with time, which Ciþ1 is varying Ci can be assumed to be constant
as the organic matter ‘ages’. To explain why this decrease with time (this assumption is required to determine that
occurs with aging, consider a single molecule of integration constant for C2 and C3 ) and with the equilib-
biological material, typically some long chain carbon rium value, so that
molecule with N, S atoms and OH molecules attached.
Initially the most reactive/unstable bonds within the mol- C 1 ¼ ½C 1 ð0Þ C1 ∗ eK 1 t þ C1 ∗
ecule will be decayed, leaving a molecule with propor- C 2 ¼ ½C 2 ð0Þ C2 ∗ eK 2 t þ C2 ∗ (10.6)
tionally less reactive components and more stable ∗ K 3 t ∗
C 3 ¼ ½C 3 ð0Þ C3 e þ C3
components. Over time the molecule becomes more stable
and less able to be decayed, and the K value decreases. and C i ð0Þ is the initial value at time equals 0 for the carbon
Some SOC models explicitly change the decay rate K with in compartment i. This shows that the solutions for carbon
time. In other models this is modelled by a flux of material change are simple exponentials with the decomposition
from compartments of highly labile SOC to less labile rate constant determining the rate of convergence to the
SOC, though the large difference in the decay rates equilibrium solution (Jenny, 1941). This exponential
between SOC compartments means that a smooth reduc- approach to equilibrium is qualitatively consistent with
tion in decay rate is only approximately captured (Plante time-varying simulations using CENTURY and RothC
and Parton, 2007). (e.g. Parton et al., 1988; Coleman et al., 1997) for
2 3 2 3
experiments where SOC inputs were constant (Figure 10.2). carbohydrates carbohydrates
However, in a comparison of nine SOM models, Smith 4 amino acids 5 ¼ eIAt 4 amino acids 5 (10.9)
et al. (1997) found that those that incorporated site-specific lignins t lignins t¼0
calibrated parameters (e.g. decomposition rates) performed
significantly better (both the time-averaged SOM and the where the exponentiation is a matrix exponentiation of the
temporal dynamics) than ones that used ‘book’ values, matrix I At.
suggesting that if these models are to be used in a land- It should be noted that the simplicity of the decoupling
scape evolution model, there is potential for significant of the decomposition rates and solutions for C1 , C 2 and
biases if the site changes significantly over time (e.g. soils, C3 in Equation (10.6) results from an approximation
vegetation). This is because while the model may start possible when there is a large difference in the rates of
calibrated to the field site through time, the landform and decay K i for the three compartments. If the rates of decay
soils evolve so that there is a need to ensure it remains in Equation (10.4) are comparable, then the approxima-
calibrated for the characteristics of the evolving landscape. tion leading to Equation (10.6) is not correct and the
Plante and Parton (2007) also note that SOC models need correct solution will be a matrix exponentiation as in
site-specific calibration to provide realistic results. Equation (10.9).
From Equation (10.5) we can determine the average Gignoux et al. (2001) describes an alternative model-
age of SOC. For compartment 1, since the age of the input ling methodology, SOMKO, where instead of the total
carbon I is approximately zero, then content of the SOM pool being lumped into three com-
partments, they explicitly model the evolution of separate
Dρb
average age ¼ (10.7) batches of SOM input at various times (they call these
SK 1
batches ‘cohorts’). The process is that SOM mass (the
where D is the depth of the soil and ρb is the soil bulk cohort) is input at a given time, and SOM, carbon and
density. nitrogen dynamics in this cohort are tracked over time,
Using a slightly different approach Kirby (1977) pro- independently of the other cohorts of SOM input at other
posed a state-space matrix equation for the evolution of times. A major advantage of this approach is that it allows
three components of SOC: carbohydrates, amino acids an explicit modelling of the aging process for SOM, so the
and lignins, each with their own transformation rate so decomposition rate K can be allowed to evolve, independ-
that ently for each cohort, smoothly over time. They also
explicitly modelled the microbial population, its impact
2 3 2 32 3
carbohydrates a11 a12 0 carbohydrates on the carbon:nitrogen ratio in the cohort and leaching of
4 amino acids 5 ¼ 4 a21 a22 0 54 amino acids 5 nitrogen, thus allowing for the possibility of nitrogen
lignins tþ1 0 0 a33 lignins limitation on microbial process and decomposition rate.
2 3 Their tracking of cohorts assumed cohorts were independ-
carbohydrates
¼ A4 amino acids 5 (10.8) ent, so they did not model soil mixing (e.g. as a result of
lignins bioturbation, so that one cohort mixed with another cohort
of a different age) and any interactions in the chemistry
where t is time and the matrix determines how one com- that might result from that mixing.
ponent transforms to another component in one timestep. The main deficiency of the models above is in their
The zeros indicate that lignin cannot be transformed into representation of erosion (Figure 10.2). Sheet erosion
either carbohydrate or amino acids, and vice versa. Even occurs at the surface and so only removes soil from the
though conceptually different from the compartment surface. Rill and gully erosion erodes soil down to the
models in Equations (10.4)(10.6), mathematically this depth of the rill/gully and in extreme cases can excavate
model is similar, even down to the three components/ from the entire soil profile; so eroded sediment reflects
compartments (though that is a historical coincidence). contributions from across the soil profile rather than just
The solution in time to Equation (10.8) is an exponential the surface. Erosion preferentially removes more mobile
decline with time at different rates for each component material, and because SOC is lower density than soil
that is mathematically very similar to Equation (10.6) with mineral matter, SOC is preferentially eroded relative to
the main difference being that the decline in carbohydrate the soil that contains it. This means that the SOC content
involves the decay rate of carbohydrate a11 and the trans- in the water flowing over the surface is higher than the
formation rate from amino acid a12 , and likewise for the concentration in the soil. The ratio of SOC in the water to
decline in amino acids which involves a21 and a22 , so that the SOC in the soil is called the ‘enrichment factor’ and
can be as much as 2–3, though more typically less than 2 whether it can be enhanced by modified land management
(Polyakov and Lal, 2004). Methods to estimate the enrich- practices (Lugato et al., 2014a, b, 2016). There is also
ment factor were discussed in Chapter 4, where erosion of increasing interest in the roles of lakes and oceans as sinks
sediment mixtures was discussed. for particulate SOC, and DOC (Kuehl et al., 2016; Leit-
The final problem with the models above is that they hold et al., 2016).
don’t capture the observed trend of declining SOC content Two properties of SOC are of particular interest:
with depth down the soil profile. If we are to capture the
movement of organic carbon by erosion, then we need a • The first is the amount of SOC in the soil, the concen-
tration, typically expressed in units of mass of SOC per
model for the depth distribution of soil carbon. One
unit mass of soil.
simple approximation is to take the models above and
distribute the SOC over depth using the exponential • The second property is the age of the SOC, defined as
the age from the time when the carbon stops living and
decline with depth that is commonly observed and/or
becomes dead. The reason for this definition is that
assumed (e.g. one empirical extension of CENTURY
radiocarbon dating is used to measure age. While the
uses this approach; Hilinski, 2001). However, given that
carbon sources (vegetation etc.) are alive, their ratio of
the time scales of the fastest SOC compartment are typic- 14
C (radioactive carbon) to 12C (stable carbon) is in
ally a few years, and that significant erosion (relative to
equilibrium with the atmosphere, once that biomass dies
the 20–30 cm soil depth that the most labile SOC is
it is no longer assimilating carbon and the ratio
preferentially concentrated in) can occur in that time, there 14 12
C: C declines as the 14C decays.
is some doubt that the dynamical interaction between
erosion and highly labile SOC can be correctly modelled
by simply assuming an exponential decline with depth.
Accordingly in the next section dynamical models of SOC 10.4.1 SOC Concentration
that explicitly model depth variation of SOC will be
Most published multilayer SOC models are variants of
discussed.
standard advection-diffusion-reaction (ADR) equations.
However, commonly, a number of approximations are
implicitly made (and for some scenarios these approxima-
10.4 Multilayer Soil Organic Carbon Models tions may be incorrect), so there is value in deriving the
SOC ADR equations from first principles to make those
The main motivation for a multilayer soil carbon model is
approximations explicit. SOC is moved down the soil
to be better able to simulate processes that depend on the
profile by either (1) bioturbation of the soil or (2) con-
depth distribution of the soil carbon, and where the
tained in water infiltrating down through the soil column.
dynamics of that process and the dynamics of the SOC
At each depth there is a first order decomposition term
process have similar time scales so that they coevolve
that is depleting the SOC. There may also be SOC
(Jenkinson and Coleman, 2008).
creation terms (e.g. dead microbes, fungi, plant roots) that
Our main agenda in this book is the interaction of SOC
will increase the SOC. For the moment, for simplicity, we
with erosion and deposition (Gregorich et al., 1998; Dia-
will ignore the different decomposition rates used in the
lynas et al., 2016). There is currently a debate about
compartment models of the previous section. We will
whether erosion and deposition are processes that produce
assume that all SOC degrades at the same rate, and we
a net depletion or accumulation of soil carbon on a land-
will return to generalise this simplification after the initial
scape wide average, and how significant the net impact is
derivation below. Figure 10.3 shows a single layer in a
on the global atmosphere (Lal, 2003; Doetterl et al.,
soil column of thickness Δz, with the z direction positive
2016). SOC is strongly correlated with regions of erosion
downwards. In plan the downslope and cross-slope
(SOC low) and deposition (SOC high) (Ritchie et al.,
dimensions of the column are Δx and Δy, respectively,
2007). If SOC is accumulating, then carbon from the
so the planar area of the soil column is ΔxΔy: We perform
atmosphere is being sequestered (typically in areas of
a mass balance on this layer of soil. The net mass of SOC
sediment deposition; VandenBygaart et al., 2015), while
entering the layer from above is Qc ΔxΔy and the net mass
if SOC is being depleted, it could be (1) undergoing
of SOC leaving the layer at the bottom is
release to the atmosphere or (2) lost to the ocean through
particulate and dissolved carbon transport in runoff. Con- Qc þ ∂Q∂z
c
Δz ΔxΔy, where the subscript c on Q indicates
sequently there is interest in determining terrestrial that it is the mass flux of carbon in the form of SOC.
sequestration of carbon by erosion/deposition and We note that SOC is normally expressed in units of

SOC transport down the soil
profile: (a) the profile and
(b) advection down the profile.
mass(C)/mass(soil) so to convert the SOC concentration with depth (Billings et al., 2010; Dialynas et al., 2016)
to a mass flux the SOC concentration must be multiplied with Dialynas adopting
by the soil bulk density ρb : If we define the decomposition
K ¼ K 0 ekz z (10.12)
rate of the SOC as before, then the mass of SOC degraded
is Kρb CΔxΔyΔz, and the source of SOC, which we will where k z is the rate of decline of decomposition rate with
assume is independent of the SOC in the soil, is depth, and K 0 is the decomposition rate at the soil surface.
Sρb ΔxΔyΔz where S is the rate of generation of SOC per A variety of constitutive equations are used for the flux
unit mass of soil. Writing the mass balance equation yields term in Equation (10.11). Bioturbation (Chapters 6 and 7)
can be modelled by Fickian diffusion so that
∂ðρb CÞ ∂Q
ΔxΔyΔz ¼ Qc Qc þ c Δz ΔxΔy ∂C
∂t ∂z Qc ¼ Dρb (10.13)
∂z
Kρb CΔxΔyΔz þ Sρb ΔxΔyΔz
(10.10) where the diffusivity D is a function of the rate of the
bioturbation and will vary (typically declining along with
where C is defined as the mass of SOC per unit mass of declining biological activity) with depth. The flux term in
soil so that Cρb ΔxΔyΔz is the mass of SOC in the layer. If Equation (10.11) then becomes
we expand the flux term inside the square brackets and
divide the equation by ΔxΔyΔz we obtain the mass balance ∂Q ∂ ∂C
c¼ Dρb
differential equation for SOC ∂z ∂z ∂z
∂ ðρ b C Þ ∂Q ∂2 C ∂D ∂C ∂ρ ∂C
¼ c Kρb C þ Sρb (10.11) ¼ Dρb þ ρb þD b (10.14)
∂t ∂z ∂z2 ∂z ∂z ∂z ∂z
where the first term on the right-hand side is the rate of In the case where SOC is transported down the profile
change of SOC as a result of a changing SOC flux down by infiltrating water, it can be modelled with an advection
the profile, the second term is the loss of SOC due to term where SOC moves down with velocity v
decomposition, and the third term is the increase in SOC (Figure 10.3b). The mass entering from above the layer is
from SOC production within the profile.
Qc ¼ vρb C (10.15)
The decomposition rate for SOC K can be of quite
general form, though it is common to assume that it is where v is the velocity at which the SOC is transported
constant with depth. Some data suggest that K can decline down the profile. Note that this is not the velocity of the
with depth as result of a decline in biological activity with infiltration water carrying the SOC but is the average
depth (Hobley et al., 2014) though the data are far from velocity at which particles of SOC move down the profile,
definitive, while some studies have assumed a decline which will be less than the infiltration velocity. During an
infiltration event some particles will move but many will in soil but two radioisotope tracers (137Cs and 210Pb).
not, and the SOC velocity is the average across all SOC They found that the velocity of transport down the profile
particles in the layer. The infiltration velocity may also for SOC was significantly slower than that for the radio-
decrease with depth as discussed in Chapter 3. The mass isotopes, but subject to this discrepancy they were able to
flux of SOC leaving the layer is obtain excellent fits for the depth profiles of SOC, 137Cs
and 210Pb. Wells et al. (2013) found a poor fit when using
∂Qc ∂ðvρb CÞ
Qc þ Δz ¼ vρb C þ Δz the same model at a site dominated by cracking soils,
∂z ∂z
where they believed that the vertical cracks created pref-
∂C ∂ρb ∂v
¼ vρb C þ vρb þv C þ ρb C Δz erential pathways for water and SOC. This preferential
∂z ∂z ∂z transport in cracks invalidates the one-dimensional verti-
(10.16) cal formulation of Equations (10.18) through (10.20).
so for advection the flux term in Equation (10.11) is There is one end member case for Equation (10.20) for
which an analytic solution is possible. At equilibrium
∂Q ∂C ∂ρ ∂v (∂C
c ¼ vρb þ v b C þ ρb C (10.17) ∂t ¼ 0) when bioturbation is small (D ¼ 0), transport
∂z ∂z ∂z ∂z velocity is the same throughout the profile (∂v ∂z ¼ 0), and
The main differences from the commonly used forms for all SOC is sourced from the surface and none is created
advection are the terms in (1) ∂v in-profile (i.e. S ¼ 0), then Equation (10.20) is
∂z which is required for
cases where infiltration doesn’t flow through the full soil ∂C
profile every time it rains (see Chapters 3 and 7) so the v KC ¼ 0 (10.21)
∂z
average velocity will decline with depth, and (2) ∂ρ∂zb which
is required for cases where changes in bulk density occur and solving this first order differential equation gives an
down profile (e.g. compaction). In the former case the top exponential distribution of SOC down the profile
of the soil profile is wet more frequently than deeper down
C ¼ Cð0ÞeKz=v (10.22)
the profile, so there are more opportunities for SOC to be
transported at the top of the profile than lower down, and where C ð0Þ is the SOC at the soil surface. This shows that
the average velocity of the SOC particles at the top of the as the decomposition rate K increases and/or the velocity v
soil profile will be higher than it is deeper in the profile. decreases, the SOC declines faster with depth. For
Combining these relationships yields the general equa- instance, Richter et al. (1999) attribute a rapid decline in
tion incorporating bioturbation and infiltration for a multi- SOC with depth to very porous soils that led to fast
layer SOC model with one SOC compartment decomposition of SOC. An exponential profile of similar
functional form to Equation (10.22) is often used to fit
∂ðρb CÞ ∂2 C ∂D ∂C ∂ρ ∂C SOC data with depth.
¼ Dρb 2 þ ρb þD b
∂t ∂z ∂z ∂z ∂z ∂z If, in addition, SOC is generated within the profile at a

∂C ∂ρ ∂v constant rate (i.e. S ¼ constant), then the solution is
vρb þ v b C þ ρb C Kρb C þ Sρb
∂z ∂z ∂z S
(10.18) C ¼ Cð0ÞeKz=v þ (10.23)
K
This equation simplifies if the soil bulk density is constant with the SOC declining to the constant value S=K rather
in both space and time so that than zero as in Equation (10.22).
2 O’Brien and Stout (1978) examined the case of steady
∂C ∂ C ∂D ∂C ∂C ∂v
¼ D 2 þ v þ C KC þ S state where D is constant with depth, and with no advec-
∂t ∂z ∂z ∂z ∂z ∂z
tion (v ¼ 0) and found a near exponential decline of SOC
(10.19)
with depth, and found that both experimentally and
The equation further simplifies if both the diffusivity and numerically the more stable SOC declined more slowly
velocity are constant with depth so that with depth as expected from Equation (10.22). Jobbágy
and Jackson (2000) fitted a number of regression equa-
∂C ∂2 C ∂C
¼D 2 v KC þ S (10.20) tions to SOC data with depth across a range of different
∂t ∂z ∂z
biomes (Figure 10.4) and found Equation (10.22) gave a
In general these equations are complicated enough that good fit; however, a slightly better fit, particularly for
they must be solved numerically. Wells et al. (2012) used depths greater than 1 metre, was obtained with the equa-
Equation (10.20) to model not just the migration of SOC tion C ¼ azb where a and b were constants. Hobley et al.
Grasslands Shrublands Forests

Proportional distribution (interval content/0–100 cm content)
0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8
0–20 0.42 b 0.33 a 0.50 c
20–40 0.23 a 0.23 a 0.21 b

Soil depth (cm)
40–60 0.15 b 0.18 c 0.13 a
60–80 0.12 b 0.14 c 0.09 a
80–100 0.08 b 0.12 c 0.07 a
100–200 0.30 a 0.38 b 0.29 a
200–300 0.13 a 0.39 c 0.27 b
FIGURE 10.4: Global estimates of SOC content with depth for the three main vegetation types: grasslands, shrublands and forests showing the
decline with depth. The error bars are one standard deviation. The horizontal scale is normalised so that they show the proportion of SOC with
depth relative to the SOC content of the top metre of the profile, so that the scale shows the relative proportions of SOC with depth (from
Jobbágy and Jackson, 2000).
(2013) found that an exponential curve best fit SOC Before we finish this section it should be noted that
associated with different particle size distribution frac- some portion of the passive/inert SOM in soils is charcoal/
tions, while Hobley and Wilson (2016) found it satisfac- biochar resulting from fire (Rumpel, 2015). We will dis-
tory for most soil profiles across 100 profiles across cuss fire in detail in Chapter 14, but we note here that
contrasting soils and climates. charcoal is chemically very different from other forms of
Returning to the single-layer SOC compartment SOM, and it is highly resistant to decomposition. For a
models of the previous section and the different decom- semiarid site Hobley et al. (2013, 2014) used spectrometry
position rates in each compartment we can easily extend to determine the chemical composition of SOC down their
the multilayer model to include the multi-compartment profiles and found that they could distinguish charcoal
nature of the single layer models in the previous section. from the other SOC, and that charcoal was up to 26% of
Using the terminology of the previous section a three- the SOC in their profiles. Hart and Luckai (2014) also
compartment version of Equation (10.19) is report significant components of charcoal in SOC in
boreal forest soils, and related spatial variation in charcoal
∂C1 ∂2 C1 ∂D ∂C1 ∂C1 ∂v stability to different wildfire intensities.
¼D 2 þ v C1 K 1 C1 þ S
∂t ∂z ∂z ∂z ∂z ∂z Zhang et al. (2010) extended Equation (10.20) to
∂C2 ∂2 C2 ∂D ∂C2 ∂C2 ∂v model the downward movement of biochar. Their equation
¼D 2 þ v C2 K 2 C2
∂t ∂z ∂z ∂z ∂z ∂z included terms for (1) linear reversible equilibrium depos-
þ α12 K 1 C1 þ S ition (of the same form commonly used for modelling first
order sorption reactions) K e ∂C∂t and (2) irreversible depos-
∂C3 ∂2 C3 ∂D ∂C3 ∂C3 ∂v ition K d C: They assumed that the in profile generation
¼D 2 þ v C3 K 3 C3
∂t ∂z ∂z ∂z ∂z ∂z term S in Equation (10.20) was zero, and that there was no
þ α13 K 1 C1 þ α23 K 2 C2 þ S (10.24) depth variation of infiltration and bioturbation, yielding

∂C ∂2 C ∂C ∂C
where the transfer from compartment 1 to compartment ¼D 2 v Ke þ KdC (10.25).
2 (α12 K 1 C 1 in Equation (10.4)) is equivalent to an ∂t ∂z ∂z ∂t
additional source term S in the equation for C2 , and Zhang et al. concluded that the parameters K e and K d
likewise for transfers from compartments 1 and 2 to were strongly dependent on the soil water pH and salinity,
compartment 3 in the equation for C3 : Equations with retention of biochar particles increasing with lower
(10.18) and (10.20) can likewise be extended to multiple pH and higher salinity, both of which are consistent with a
compartments. first order sorption-like process.
10.4.2 SOC Age where the only difference between the equations for 12 C
and 14 C is the addition of the radioactive decay term
Using the mass balance approach from the previous section λð14 CÞ for 14 C: This formulation ignores fractionation
and illustrated in Figure 10.3 we can write equations that effects, where the microbial decomposition slightly pre-
describe the age of SOC. We first note that when we mix fers to decay the lighter isotope 12 C over the heavier
two samples of soil with different masses and ages, the age isotope 14 C so that K for 12 C is slightly higher.
of the mixture is not the mass-weighted average of the ages. At any time the age of the carbon in the soil can then
The age A is a function of the relative masses of 12 C be calculated by
and 14 C where the time evolution of a packet of soil is !
14
C=12 C
14
C 14
C ð0Þ λt A ¼ λ ln 14 (10.32)
¼ 12 e (10.26) Cð0Þ=12 Cð0Þ
12 C C ð0Þ
where 14 C and 12 C are the concentrations of the radioactive Finally we can derive an analytic result for age with depth
and stable carbon isotopes, C ð0Þ is the concentration at for the same assumptions that we used for the SOC
time zero (if all SOC is being input at the surface, this is concentration Equation (10.22). The depth profile for the
also the concentration at the surface), λ is the decay rate and isotopes is
t is time. The decay rate is related to the half-life of 14 C as
12
C ¼ 12 C0 eKz=v
ln 0:5 (10.33)
λ¼ (10.27) 14
C ¼ 14 C0 eðKþλÞz=v
t 1=2
where t 1=2 is the half-life of 14 C which is 5,730 years. and when we substitute Equation (10.33) into Equation
The mixing model for two packets of soil with differ- (10.32) we obtain the age profile
ent masses of carbon and ratios of 12 C to 14 C (and thus ðKþλÞz=v λ2 z
e λz=v
different ages) is A ¼ λ ln ¼ λ ln e ¼
eKz=v v
M 1 14 Cð0ÞeλA1 þ M 2 14 C ð0ÞeλA2 (10.34)
14
Cave ¼ (10.28)
ðM 1 þ M 2 Þ
which is approximately consistent with measured age and
and this can used to determine the age of the mixed sample concentration profiles, and the derivation is consistent
with typical assumptions used in deriving the concentra-
1 M 1 14 Cð0ÞeλA1 þ M 2 14 Cð0ÞeλA2
Aave ¼ ln tion profile (Rethemeyer et al., 2005). Note that the
λ 14 C ð0ÞðM þ M Þ
1 2
decomposition rate K divides out of the solution so this
(10.29)
result is independent of the rate of decline with depth of
This is different from what would be obtained if ages were the decomposition rate (e.g. Dialynas et al., 2016), so
simply a mass-weighted average, as is done for the mean depth-related changes in K will not change the distribution
residence time (MRT) method (Schaetzl and Thompson, 2015) of age down the profile. Note that any source of SOC
M 1 A1 þ M 2 A2 within the profile (i.e. Equation 10.32) will result, on
Aave ¼ (10.30) substitution into Equation (10.32), in the age growing
ðM 1 þ M 2 Þ
more slowly with depth than Equation (10.34).
Accordingly, the best way to model age is to model the two
carbon isotopes with two sets of coupled equations. Using
Equation (10.19), then the appropriate equations are 10.5 Erosion
2 12
∂ð12 CÞ ∂ ð C Þ ∂D ∂ð12 C Þ Soil erosion and deposition is a component of the SOM cycle
¼ D þ
∂t ∂z2 ∂z ∂z (Figure 10.2). We talked in detail about soil erosion in
12
∂ð C Þ ∂v 12 Chapter 4 (of particular relevance are Sections 4.1, 4.2.1
v þ C K 12 C þ S
∂z ∂z and 4.2.4). However, as a link to these sections a short
2 14 discussion follows. Here we make a distinction between the
∂ð14 CÞ ∂ ð C Þ ∂D ∂ð14 C Þ
¼ D þ mineral matter in the soil (i.e. the clay, silt and sand fractions),
∂t ∂z2 ∂z ∂z
14 which typically has a density in the range 2,000–3,000 kg/
∂ð C Þ ∂v 14 m3, the organic fraction (i.e. SOM), which typically has a
v þ C K 14 C λ 14 C þ S
∂z ∂z density in the range 300–1,000 kg/m3 and soil aggregates
(10.31) (see Section 10.6.1), which are about 1,300 kg/m3.
10.6 SOM-Hydrology Interactions 173
The primary driver of erosion is the shear stress reflection of SOC enrichment, and thus preferential
applied by the overland flow on the soil surface. Two removal of SOC during erosion, or whether other seques-
factors that change the erodibility of the soil surface (it tration processes were at work. They concluded, using a
is good enough here to consider erodibility equivalent to mixture of modelling and fieldwork, that there appeared to
the amount of erosion for a given discharge and slope, a be a sequestration of SOC in the deposition regions and
more accurate definition is provided in Sections 4.2.1 and that the quantity of SOC sequestered was greater than that
4.2.4), and thus the erosion rate: provided by the deposited sediment. They concluded that
deposition enhanced the burial of vegetated material at the
• The grading of the soil eroded from the surface into the deposition sites (i.e. I in Equation (10.4) was increased by
flow. The finer the soil, the higher the erosion rate. This
the act of deposition). Though Yoo et al. did not examine
is not just the grading of the mineral matter in the soil
erosion regions, it might be inferred that in regions of erosion
but also the grading of soil aggregates (and specifically
the value of I would be reduced, so the question of whether
the water stable aggregates). The soil aggregates have a
there was an enhancement of sequestration of carbon over
larger diameter than the mineral matter and the SOM in
the entire landscape including all regions of erosion and
the aggregates.
burial depends on the relative value of I in erosion and
• The density of the particles (e.g. Ahad et al., 2015). deposition regions, and the question remains open.
Less dense particles are easier to entrain into the flow
Billings et al. (2010) and Dialynas et al. (2016)
and thus erode. They also settle more slowly and so
coupled a model of erosion/deposition with a model for
deposit at a lower rate. Again this is not due to only the
SOC decomposition to explore the role of deposition in
density of the mineral matter but also the density of the
carbon sequestration. Both used a single-compartment
aggregates. The soil aggregates tend to be less dense
SOC model and assumed that the decomposition coeffi-
then the mineral matter because they are bound together
cient K declined with depth exponentially. Both showed a
by SOM. Soil aggregates have a higher SOC concen-
strong increase in sequestration as the SOC was deposited
tration than the remainder of the soil, so erosion tends to
but that the results were strongly dependent on the exact
have a higher concentration of SOC than the soil surface
form of the decomposition rate change with depth and that
as a whole. This ratio of SOC concentration in the
there were insufficient data to accurately define this rate of
sediment in the flow relative to the SOC in the soil is
change with depth.
called the enrichment factor and is typically greater than
For burial of SOM to be a viable mechanism for seques-
1 (Figure 4.5) (Rumpel et al 2006).
tering carbon the decomposition rate after burial must be
Thus the SOM at the soil surface has two competing less than the decomposition rate at the surface, otherwise the
processes. The aggregates tend to be larger than the min- generation rate of carbon dioxide per kilogram of SOM after
eral matter on the surface (decreasing erosion), but they burial is unchanged. While the idea of decline with depth of
also tend to be less dense (typically around 1,200–1,500 the decomposition rate (e.g. driven by reduced biological
kg/m3, increasing erosion). activity itself limited by reduced food and oxygen) is
Ritchie et al. (2007) found patterns of SOC that were appealing, the field evidence for a decline in decomposition
related to the regions of erosion and deposition (using rates with depth (other than in permanently saturated
caesium isotope analysis), with the locations of highest regions, so that conditions are anaerobic and decomposition
erosion having the lowest SOC percentage content, which rates low) in the soil profile is ambiguous.
was consistent with previous more qualitative observa-
tions (e.g. Gregorich et al., 1998). 10.6 Soil Organic Matter-Hydrology
Stallard (1998) estimated the global erosion rate and Interactions
concluded that transport of SOC may be significant
enough to have influenced the global carbon cycle. If this There are many ways that SOM interacts with hydrology.
eroded SOC was sequestered by deposition, then it may High values of SOM are associated with better soil struc-
be an important factor in terrestrial sequestration of ture, and higher water-holding capacities and infiltration
carbon. However, for deposition to preferentially seques- rates. These relationships generate a positive feedback
ter SOM requires that the decomposition rate must decline because when the SOM increases, both the infiltration
with depth. Stallard had significant difficulties in estimat- rate and water holding capacity increase, so that the soil
ing the net sequestration of erosion and deposition. Yoo moisture content of the soil increases, and the SOM
et al. (2005, 2006) examined whether the trend in spatial content of the soil increases as soil moisture increases.
distribution in SOC and erosion/deposition was simply a We will concentrate here on how SOM changes the
hydrology relationships. Determining how the relation- particulate organic matter and biofilms. Macroaggregates
ships below subsequently drive the hydrology and soil involve physical aggregation of micro- and mesoaggre-
moisture (e.g. using soil moisture models such as gates by decaying plant roots and fungal hyphae
HYDRUS) has been discussed in Chapter 3. (Voroney, 2007).
As well as influencing the infiltration rate, the presence
10.6.1 Soil Aggregation of water stable aggregates increases the size of soil
particles and thus reduces raindrop erosion (see
Soil is not simply particles of mineral and SOM Chapter 4). The effect of the change in soil grading,
coexisting independently in the soils. A variety of bio- porosity and pore-size distribution due to soil aggregates
logical, chemical and physical processes result in the can be estimated by pedotransfer function. For instance,
particles combining to form aggregates. These aggregates the Kozeny-Carmen equation (Freeze and Cherry, 1979)
are crucial in carbon storage, and the development of the relates the saturated hydraulic conductivity to the soil
soil structure that enhances infiltration rates. Furthermore grading. Unfortunately, however, no pedotransfer func-
different types of aggregates are formed as a result of tions provide guidance on how the grading and porosity of
interactions between different fractions of the soil. The the aggregates changes with SOM (Nimmo and Perkins,
smaller aggregates tend to result from chemical processes, 2002) though a number of fractal models have been
while the larger aggregates tend to result from physical suggested.
and biological processes. Some of these aggregates once Because of the central role of SOM in forming soil
formed are long lived, while others are more sensitive to aggregates, SOM content in aggregates can be much
disturbance and/or changes in the fungal biology of the higher than in the finer, nonaggregate, fraction of the soil
soil. Six et al. (2004) reviews the processes that drive the (e.g. SOC was 38% in particles greater than 250 μm, and
formation of the aggregates. less than 5% in particles less than 2 μm; Tisdall, 1996).
Tisdall and Oades (1982) introduced a framework for Some preliminary work has linked SOM decomposition
how aggregates are formed by the clay and silt fractions and aggregation (Malamoud et al., 2009).
combining, and how these aggregates are important for
soil structure development. Aggregates tend to be 10.6.2 Infiltration Rate
unstable and disintegrate when wet unless they are also
bound with the polysaccharide components of the SOM in Barzegar et al. (2002) provides a direct link between SOM
the soil. The resulting clay-silt-SOM aggregates are then content and hydrology. They applied varying rates of
‘water stable’, so the large pores between the soil aggre- organic material to an agricultural soil and found that
gates (that enable high infiltration) do not collapse when the steady-state infiltration rate increased with increasing
the soil is wet. There is no single relationship for aggre- SOM and the infiltration rate was independent of the type
gate stability that has been found applicable for all soils, of SOM (e.g. mulch, manure). While the data set was
but Tisdall and Oakes discuss a number of the common small and restricted to relatively high SOM contents
relationships. For instance, Kemper and Koch (1966) (8% < SOM < 10%) they fit
(referenced in Tisdall and Oades, 1982) found
I ¼ 0:0064ðSOMÞ3:4 (10.36)
aggregate stability ¼ 40:8 þ 17:6log ð%SOCÞ þ 0:73ð%clayÞ
where I is the continuing infiltration rate (mm/hr).
0:0045 %clay2 þ 3:2ð%Fe2 O3 Þ
Arshad et al. (1999) present a comparative analysis of
(10.35)
no-tillage paddocks with conventional tillage paddocks.
where aggregate stability is defined by USDA-ARS Their no-tillage paddocks had a higher SOM content than
(1999) and characterises the percentage by mass of conventional tillage by between 10% and 30% (the actual
aggregates greater than 0.25 mm in diameter that remains SOC content was between 2.2% and 4.5%), for the no-
intact after soaking in distilled water for 5 minutes. tillage sites the mean diameter of the water stable aggre-
Aggregates are classified on size as 0.002–0.02 mm, gates was larger by 20–100%, and the steady-state infil-
0.02–0.25 mm and <0.25 mm, microaggregates, mesoag- tration rate was higher by about 60%. They attributed the
gregates and macroaggregates, respectively. Microaggre- differences to the SOM content. Azooz and Arshad
gates are formed by flocculation of clays and organic (1996) did not report SOC content but found that both
molecules (particularly polysaccharides produced by root saturated and unsaturated conductivities of no-tillage soils
and fungal exudates). Mesoaggregates involve decaying were 2–3 times higher than for a conventional tillage site.
10.7 Catchment and River Network-Scale Carbon Cycling 175
In both cases the differences were attributed to soil FC is not as strongly impacted by SOM (Minasny, per-
structure changes and not to physical disturbance by tillage. sonal communication).
These results are qualitatively consistent with Barzegar Hudson (1994) found when SOM increased from 0.5%
et al. (2002). to 3% by mass that the plant available water capacity was
Dunkerley (2002) measured infiltration rates for a site doubled across all soil types. Plant available water cap-
with patchy groves of vegetation and compared the infil- acity is the difference between field capacity and the
tration rates inside and outside the groves of trees. He permanent wilting point.
found that the infiltration rate outside the groves (7.4–30.5 Adams (1973) derived a relationship for predicting
mm/hr) was significantly less than inside the groves bulk density:
(as much as 292 mm/hr), and that there was a significant
100
correlation between infiltration rate and distance to ρbðsoilÞ ¼ ! ! (10.38)
%SOM 100
the nearest tree stem. He did not measure SOM/SOC þ
content and so was not able to distinguish whether this ρbðsomÞ ρbðmineralÞ
was a result of tree root macropores or SOM-soil
where ρb(soil), ρb(som) and ρb(mineral), are the bulk densities
structure effects, or surface sealing by rainsplash outside
of the soil, SOM and mineral respectively.
the groves (Wickens and Collier, 1971); however, others
have found significant correlations between SOM content
and distance from trees (Avirmed et al., 2014), suggesting 10.7 Catchment and River Network-Scale
that a link between SOM and infiltration rate was Carbon Cycling
plausible.
Improved soil structure results in an increase in pore This chapter has been focussed on soil and carbon pro-
sizes and a reduction in bulk density, and it is likely that cesses on the hillslopes. However, given the book’s over-
these are the fundamental drivers of the hydrology all focus on catchment and landscapes interactions, some
impacts of SOM. Thus models for the relationship discussion on the role of river networks in transporting
between SOM, aggregation and porosity are key to and transforming organic carbon on its route to the ocean
predicting the influence of SOM on hydrology (specif- is warranted. For many years the view has been that rivers
ically infiltration rates), but we do not have such rela- are simply the pipeline to transport eroded particulate
tionships at the current time (Minasny, personal organic carbon (POC) and dissolved organic carbon
communication). (DOC) to the ocean with some small decomposition of
organic carbon while in transit. Recent years have seen a
recognition that there is a significant mismatch between
10.6.3 Water-Holding Capacity and Bulk organic carbon creation on the hillslopes and the organic
Density carbon delivered to the ocean, so that (1) significant
transformation is likely occurring en route or (2) there is
Another impact of SOM is that it increases the soil water- storage with long time scales within the catchment. While
holding capacity and field capacity, and these are import- the data are by no means definitive, a recent review (Wohl
ant for vegetation because they determine the amount of et al., 2017) suggests that both processes are important.
water available between rainfall events. Saxton and Rawls This discussion follows the interpretation presented
(2006) present various equations for the Van Genuchten by Wohl.
unsaturated conductivity model that relate these param- In Chapter 1 and Figure 1.2 the strings and beads
eters to percentages of sand, silt, clay and SOC content. conceptualisation of rivers was discussed. This conceptu-
He found the strongest effect of SOM was on the field alisation divides the river into two types of reaches: (1)
capacity, FC, of the soil those with associated floodplains (the beads, typically
alluvial river cross sections) interspersed with (2) river
FC ¼ FCt þ 1:283FCt 2 0:374FCt 0:015 reaches without associated floodplains (the strings, typic-
FCt ¼ 0:251S þ 0:195C þ 0:011SOM þ 0:006ðS SOMÞ ally bedrock river cross sections). Wohl argues that little
0:027ðC SOMÞ þ 0:452ðS CÞ þ 0:299 carbon transformation and storage occurs in the bedrock
(10.37) river reaches because transit time are fast and there is
where S, C and SOM are the sand, clay and SOM little potential for storage of carbon within the reach.
proportion by mass, respectively. Other data suggest that Accordingly all the transformation and storage must

processes operating in the
floodplain determining exchange
and storage of organic carbon in
the channel and the floodplain.
occur in the alluvial river reaches, and that the exchange between the channel and the floodplain by bank erosion as
processes between the channel and the floodplain (which a result of meandering, and Cfp is the average concen-
includes the riparian or hyporheic zones) for both water tration of POC in the floodplain sediments, ζ is the aver-
and particulate matter are key to these processes. Because age rate of meander movement laterally, Dfp is the height
the floodplain is wet, it is an active zone for soil organic of the floodplain above the channel bed on the outside of
matter activity because vegetation growth and microbial the meander, and εL is the effective length of the channel
decomposition are both high. Since SOM values are within the floodplain where L is the straight-line distance
typically higher on floodplains than hillslopes, it seems between the upstream and downstream ends of the flood-
reasonable to infer that the enhanced vegetation activity plain so that ε captures the enhanced length of the channel
more than offsets increased decomposition. The from the tortuosity of the meandering. The second term on
exchange processes for POC are the same as they are the right-hand side is the deposition of organic matter on
for mineral matter (Figure 1.2; Sections 1.3.2 and 4.5) the floodplain during high flow, where we use the notation
though organic matter tends to be more mobile because of of Chapter 4 that deposition is the negative erosion E
its lower density. These processes are (1) deposition on (depth/time), and Afp is the planar area of the floodplain.
the floodplain surface during flood events, (2) erosion The third term on the right-hand side is the in-channel
on the outside of the meander bends and (3) deposition decomposition of the POC (assumed to be a first order
on the inside of meander bends. The DOC exchange is a decay process) and K ch is the in-channel decomposition
result of groundwater exchange between the river and rate (units 1/time), and v is the velocity of the flow so that
the groundwater table within the floodplain (the εL=v is the time spent in the floodplain reach. The fourth
hyporheic zone). term is the POC delivered to the channel by hillslopes
The discussion below will focus on the POC. A simple contributing to the floodplain, and efp is the efficiency of
model for the change of POC in the floodplain reach of the capture of POC from the hillslope by the floodplain
channel is a function of the exchange of sediment with the (efp ¼ 1 is 100% capture of the POC by the floodplain
floodplain so that (using the terminology in Section 4.5; with none reaching the channel).
Figure 10.5) This equation requires knowledge of the POC concen-
tration in the floodplain sediments. We thus need an
Cin þ C out
qs,out Cout qs,in Cin ¼ ζ Dfp ρb Cfp εL equation that predicts the dynamics of POC in the
2
floodplain:
Cin þ C out
þ Eρb Afp
2 ∂Cfp C in þ Cout
Afp Dfp ρb ¼ ζ Dfp ρb Cfp εL
q Cin þ qs,out x εL ∂t 2
K ch s,in
2 v
Cin þ Cout
þ 1 efp qs,hill Chill (10.39) Eρb Afp þ efp qs,hill Chill
2
where qs is the sediment transport in the channel (units þ SAfp K fp Cfp Afp Dfp ρb (10.40)
mass/time, the subscripts in and out indicating at the
upstream and downstream end of the floodplain, where the left-hand side is the rate of change of the total
respectively), C is the concentration of POC in the sedi- mass of POC in the floodplain sediments. The first two
ment (units mass POC/mass sediment) and ρb is the terms of the right-hand side come directly from Equation
average bulk density of the floodplain deposits. The first (10.39) and are the channel-floodplain exchange pro-
term on the right-hand side is the exchange of sediment cesses. The third term on the right-hand side is the
hillslope POC transport that is captured by the floodplain. 10.8 Conclusions

The fourth term is the source term for POC on the flood-
plain with S the rate parameter (mass/unit area of flood- Before we finish, we note that the main source for SOM/
plain). The fifth term is a simple first order decay term as SOC is vegetation leaf litter and dead roots, and the
discussed in the SOM compartment models discussed in microbes feeding on them. Thus the terrestrial component
previous sections where K fp is the decomposition rate for of the carbon cycle has vegetation at its core: i.e. SOM !
the floodplain. infiltration ! soil moisture ! vegetation growth ! SOM
These two equations are simple conceptualisations for source ! SOM. Vegetation will be discussed in detail in
the channel and floodplain processes, and it is likely they Chapter 14.
do not capture the full complexity of processes in flood-
plain reaches. However, given the extremely limited 10.9 Further Reading
understanding of floodplain organic matter processes rate
and storages (as discussed by Wohl et al., 2017), it is The organic content of soil is a complex topic, and this
likely that more complicated modelling is not warranted chapter has been able to only scratch the surface of it.
by the data at this time. Many processes are so complex as to have defeated quan-
Subject to that caveat some limitations might include tification to date (or even experimental study in some
that the decomposition terms in Equation (10.40) are likely cases), and so fall outside the scope of this book. Schaetzl
to require the multi-compartment models discussed in pre- and Thompson (2015) provide a good general overview.
vious sections. It may also be warranted to subdivide into Wardle (1992) provides an excellent review of the living
the soils above and below the floodplain groundwater table biomass in the soil and the factors that influence it. Plante
because rates within the anaerobic groundwater table will and Parton (2007) and Metherell et al. (1994) summarise
be significantly less than the aerobic conditions above the the state-of-the-art for single-layer compartment models.
water table (and this may be critical to long-term seques- Bondeau et al. (2007) present a modelling study using the
tration of POC within the floodplain sediments). The LPJ model (see Chapter 14) of the impacts of agriculture
source term will also be a function of the vegetation growth on the global carbon cycle, including a 10% reduction in
models to be discussed in Chapter 14. global soil carbon stocks, over the last century.
11 Soils: Constructing a Soilscape Evolution
Model – Details and Examples
11.1 Soilscape Modelling that this bulk density change may not form a distinct
boundary if the bulk density change is due to chemical
Before moving from soil evolution processes to landform weathering processes (e.g. Graham et al., 2010).
evolution processes we will pause to consider how a At the heart of the soil definition problem is that we
standalone soilscape evolution model can be constructed have four possible, but distinctly different, processes
from the components already discussed. Such a model can going on within the soil – (1) the soil production
be used to explore the dynamics of soils on a fixed function, (2) physical weathering, (3) chemical
landform. This is not to deny that landform evolution weathering and (4) downslope soil flow – and the different
may be important to the spatial distribution of soils but intellectual communities that have worked in these four
simply that soils may evolve more rapidly than the land- areas have each arrived at different definitions for ‘soil’.
form. In this case the soils will always be at or near Each of these processes contributes to the soil formation
equilibrium with the slowly evolving landform. Over the processes, but there is no consensus on their relative
long term the soils may still evolve, but only because the dominance. For example, Yoo and Mudd (2008, p. 249)
landform itself changes. This simplification should be state ‘colluvial soil production and colluvial transport
quite familiar because it is basically a restatement of the have not been considered in the geochemical evolution
principle underpinning ‘soil catena’, where the position of the soils and chemical weathering has been neglected in
on the hillslope, but not the history of the hillslope, understanding of the geomorphology of soil-covered
determines the soil properties. If the current-day soils hillslopes.’
were still responding to past landforms, then the link For the soil production function the recent evidence
between the position on the current-day landform and soil supporting its existence and form is the rate of bedrock
would be less apparent. conversion as a function of soil depth determined using
cosmogenic nuclides. Yet this same technique provides no
insight into what the processes are that drive this bedrock
11.2 Coupling Weathering Processes
conversion only that the rate of bedrock conversion
11.2.1 What Is Soil? decreases with soil thickness with a roughly exponential
decline. Burke et al. (2007) showed a relationship
Before outlining below how to couple all of the processes between pH, cumulative chemical weathering and the soil
in this chapter to develop a general coupled model for soil production function, suggesting an exponential decline
dynamics we need to consider ongoing debate over ‘what with depth for chemical weathering with a similar expo-
is soil’. At the beginning of the book we adopted a nent as the soil production function.
definition of soil as being all that material above the On the other hand, for chemical weathering Maher
saprolite interface, where there was distinct change in (2010) suggests that the equilibrium soil stops at the depth
the porosity and bulk density above and below the where the acid driving the weathering is completely
saprolite-soil interface. We discussed how this interface depleted. The argument is that if there is unused
might arise from bioturbation as a result of tree throw weathering potential (i.e. unused acid), then that acid will
where the tree roots rip rock out of the saprolite and mix further weather the bedrock and the soil will become
saprolite material into the soil (and in the process mix the deeper as a result. Soil deepening stops only when the
178 soil from top to bottom). However, other work suggests chemical weathering potential is depleted. This ignores
the role of SOM decomposition and respiration in provid- boundary as dispersivity normally scales linearly with
ing new acid potential in-profile. The philosophy behind groundwater velocity. Figure 11.1 shows a simulation that
Maher’s viewpoint is that soil is all the material that is is equivalent to Figure 8.16 but where PR ¼ 10 (e.g.
chemically weathered. On the other hand if there is a porosity above the tree throw boundary is 0.3 and below
distinct change in bulk density at the saprolite boundary, the boundary 0.03). The main observation is for the
it is hard to conceive of that being due to chemical weathering profile with multiplier ¼ 1:0. Where it previ-
dissolution, particularly if a humped chemical weathering ously peaked in the middle of the profile (Figure 8.16) it
profile has the maximum weathering rate in the mid- now peaks at the bottom of the profile. There are now two
regions of the soil profile. There is evidence that dissol- distinct weathering regimes that do not overlap: the first
ution is a significant cause of porosity development in soil where all the chemical weathering occurs above the por-
profile (e.g. Brimhall and Dietrich, 1987; Brimhall et al., osity contrast boundary (for high erosion rates) and the
1991; Anderson et al., 2002), but no credible mechanism second where it all occurs below (for low erosion rates).
for chemical dissolution to create an abrupt change in
porosity has been proposed. This is particularly the case
when the main indicator used for chemical weathering 11.2.2 What Soil Properties Should Be Replicated?
progress is the change in the ratio of mobile elements to
What we need to replicate will be different depending on
immobile elements in the saprolite and the soil, and they
the objective of the researcher. Data to test soilscape
typically do not show a step change at the saprolite
evolution models are rare. For instance, data from field
boundary (e.g. Anderson et al., 2002; Riebe et al.,
studies where soil properties are available and can be
2003), only a gradual change from the saprolite to the
dated, and where it is possible to definitively fingerprint
soil surface.
the pathway and rates from saprolite to soil is rare. Beal
There is thus the potential for conflict between the soil
et al. (2016) examines a volcanic field with a range of ages
production function and chemical weathering definitions
for lava flows and cinder cones up to about 12,000 years
of soil. It is possible to have deep soils that are not fully
and were able to compare soil properties between sites
chemically weathered to their full depth if the equilibrium
with a good assurance of similar initial conditions. The
depth from the soil production function is deeper than the
results from previous chapters (e.g. soils fine with age,
maximum depth of chemical weathering.
SOM increases with age) are consistent with their
There is also a geomorphic view of what defines soil.
findings.
Geomorphologists (Anderson and Anderson, 2010) define
However, beyond the pragmatic issue of prioritising
it as that part of the regolith that is mobile, which is any
the modelling of those characteristics for which field data
material above the saprolite that can potentially move
are available is the question of what characteristics of soil
downslope or that can be mixed vertically within the soil
behaviour actually need to be modelled for any given
profile. In this definition any material that is not physically
application. It is rare that we can model everything simul-
‘locked into’ the saprolite is soil. The movement down-
taneously, so we prioritise those characteristics that are
slope may be creep like processes, viscous or plastic flow
most important for our immediate needs.
of the soil, or mass movement.
Some geomorphology examples might include the
The extent of these difficulties can be seen with a
following:
simple example. Consider a soil where the top half of
the soil profile is subjected to the Gabet tree throw • Landslide researcher: Catchment wide soil depth so that
bioturbation mechanism: the underlying saprolite has not they can determine volumes of source material for
been excavated by tree roots, and the acid in the chemical potential landslides, and potential for saturation to trig-
weathering process has not been fully consumed at the ger a landslide.
bottom of the tree thrown layer. This can be modelled in • Landform evolution modeller: Grading of the soil sur-
the chemical weathering model discussed in previous face and its erosion potential so they can determine
chapters by considering that there is an abrupt increase future landform shape, and spatial distribution of rock
in the porosity of the soil above the tree throw–saprolite outcrops (i.e. lack of soil).
boundary. Let’s call the ratio of porosity above the bound- • Ecologist: Water-holding capacity of the soil (depth,
ary to the porosity below PR . The impact of the porosity texture and structure), which limits plant growth during
contrast is that the groundwater velocity below the bound- arid periods and nutrient availability from weathering.
ary will be PR times faster than above the boundary, and • Water quality modeller: Leachate properties from
the dispersivity will likewise be PR times higher below the weathering processes.
180 Soils: Constructing a Soilscape Evolution Model – Details and Examples
FIGURE 11.1: The chemical weathering model where there is a porosity change half way down the profile as a result of tree throw.
This result has the same parameters as Figure 8.16 except for a 90% reduction in porosity in the bottom half of the profile.
In an attempt to answer this question from the per- 1. Improve the ability of the models to respond to climate
spective of a soil scientist, Minasny et al. (2015) dynamics, climate change and landuse change
surveyed 29 case studies involving a range of soilscape 2. Balancing the desire for more process components
models. He listed 22 processes that potentially need to (and the potential for over-parameterisation) against
be included, of which only about 50% have ever been the need for model parsimony
studied in a soilscape model (Figure 11.2). While it is a 3. Explicit representation of temperature and moisture
laundry list of processes, it does indicate the range of dynamics
soil processes observed in the field. Taking a step back 4. Optimise codes for high performance and reduced
from the details of the processes, Minasny then sug- run times
gested four criteria against which soilscape models 5. Link models with global climate models and dynamic
should be evaluated: global vegetation models
6. Investigate the main drivers of soil spatial variability
1. Mass or solute fluxes along a soil catena (i.e. so fluxes and scaling properties of soils in space.
match those observed)
2. Effects of slope along toposequences (i.e. so soil Finally, from the soil science perspective McBratney
catena is correct) (personal communication) asserts that a key objective for
3. Effects of climate variations in climosequences (i.e. so soilscape models is the reproduction of the soil horizons,
the trend with changes in climate is correct) without having explicitly incorporated soil horizon infor-
4. Effects of the duration of soil formation in chronose- mation into the model in its formulation. Soilscape
quences of soils (so modelled soil ages match field soil models have not yet achieved this objective.
dating). Hydrologists are typically interested in the catchment
scale spatial trends in soil functional properties such as
Minasny concludes with six challenges for the hydraulic conductivity. However, even in a relatively
coupling of soils and landscape models: uniform soil, the grading and functional properties can
FIGURE 11.2: Presence (horizontal bars) of 22 soil processes in 29 case studies of profile models and distributed profile models (left two bar
charts, black), soil thickness (two-dimensional) models and three dimensional soilscape models (right two bar charts, grey) (from Minasny, 2015).
vary substantially within a few metres. If the underlying written, and interested readers are advised to download
source material is variable in space, then this may strongly model manuals off the web.
condition this spatial variability. For instance, soil prefer- It is generally considered that the modern era of quan-
entially forms in cracks in rock as a result of feedbacks titatively linking soil properties with pedogenic processes
between water collecting in the cracks and enhanced started with the empirical work of Jenny (1941, 1961),
weathering that results (e.g. Beal et al., 2016). Even with though it wasn’t until the 1970s that quantitative dynam-
uniform source rock geology it is also possible to postu- ical models based on process began to appear in the
late a spatial positive feedback in soil formation where literature. In parallel with the research motivated by
enhanced soil moisture enhances weathering, which Jenny, there was also the developing idea of soil catena,
enhances soil water-holding capacity, which enhances soil that a soil profile should not be viewed in isolation but as
moisture. Thus locations where soil moisture has slightly part of a (potentially evolving) hillslope system. For a
increased for some reason might exhibit runaway soil discussion of these historic developments see, for
weathering that results in spatially random, but persistent example, Huggett (1975) and Hoosbeek and Bryant
with time, patterns of soil moisture and weathering (e.g. (1992) and the subsequent discussion of the latter
the persistent but spatially random soil moisture patterns (Amundson, 1994; Hoosbeek, 1994).
observed by Western et al., 1999 might result from this Kirkby (1977, 1985) presents one of the first models to
mechanism). This random patterning, in addition to any attempt to model the evolution of saprolite weathering
mean trends in space, of soil moisture and soil functional over a hillslope. While he recognised the different
properties may then be an important property for hydrolo- weathering characteristics of different minerals, he
gists to be able to replicate. lumped all minerals into one component and modelled
the degree of weathering of the saprolite into soil by a
11.2.3 Previous Modelling single variable, what he called the weathering deficit,
ignoring the differences between the minerals. At any
This section is not intended as a comprehensive review of given time he considered that the solution and the soil
models, but rather as an outline of how previous model- have an equilibrium concentration as in Equation (8.15).
lers have constructed their models and place in context the However, he set the equilibrium leachate concentration as
components discussed in the previous chapters. Any a function of the unweathered proportion of saprolite S
review of a model will be out of date as soon as it is (figure 4 of Kirkby, 1977)
Substance remaining, r
p
T = 10
T = 20
T=5
T = 20
T=5
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.4r* 0.5 0.6 0.7 0.8 0.9 1.0
10
T = 16.8 mm/a
T = 2.4 mm/a
T = 4 mm/a
T = 7.2 mm/a
3
1 T
=2 0.5
T
=0
10
4
10
5
,1 1.0
2 04
10 4
a 10 5
a
T=
5 T=
Depth (m)
2 10 6
3 10 5 1.5
10 5
T=
0
(iii) (i) (ii)
10 5
4 2.0
depth (m)
2.5
5 3×10 5a
T=
2
3.0
6
3.5
7
3×105a
T=0
4-0
8
FIGURE 11.3: Equilibrium depletion profiles for a variety of 106a T=2

different parameter sets (from Kirkby, 1985). T=0
T = 2 mm/a
T = 5 mm/a
T = 10 mm/a
T = 20 mm/a
L
R ¼ k 1 k 5 ðS Smin Þ 1
Smin þ αðS Smin Þ (11.1) FIGURE 11.4: Transient depletion profile (from Kirkby, 1985).
A ¼ APCO2ðsoilÞ
where he assumed that the leachate concentration was at upper part of the soil profile, he found that the proportion
equilibrium with the saprolite. The term (S Smin ) meant of weathered saprolite in the top of the profile was fixed
that below Smin there was no weathering, so Smin was a by the minimum value for weathering over the entire
lower bound on how much saprolite could be weathered. depth. Like all subsequent workers in this area he
Unlike Equation (8.6), where the approach to zero reac- assumed the soil hydraulic conductivity was independent
tion rate is asymptotic, Kirkby’s was linear with an abrupt of the degree of weathering. Kirkby (1989) extended this
switching off of weathering at Smin : He also assumed that model to examine the effect of climate dynamics on soil
soil atmosphere was driven by microbial degradation of properties. Kirkby (1977) used the transition matrix for-
the organic matter so that the acid content was the same mulation presented in Chapter 7 to simulate the interactive
throughout the profile. His assumption that the system dynamics of four different types of soil organic matter
was at chemical equilibrium is the same as assuming the (Section 10.3).
reaction rate k 1 is very high in Equation (8.6). Kirkby Minasny and McBratney (1999) implemented a
coupled this chemical weathering model with a fixed rate coupled physical weathering and landform evolution
of erosion as was used in the sensitivity studies in Chap- model to predict the spatial distribution of the evolution
ter 8. He found depletion profiles (both the equilibrium of soil profiles on a hillslope. The physical weathering
profiles, Figure 11.3, and the transient approaching these was the exponential soil production function (Chapter 6).
equilibrium, Figure 11.4) that are very similar to those Minasny and McBratney (2001) extended this work to
shown in Chapter 8 with two differences. The first differ- implement both physical and chemical weathering
ence is that the saprolite is never 100% weathered because models that were depth averaged over the soil profile,
of the threshold Smin below which weathering does not and coupled them with a simple landform evolution
occur. The second, related, difference is that when he model to predict the evolution of the spatial distribution
included a bioturbation term that was concentrated in the of soil depth. The physical weathering they modelled
using a soil production function with an exponential the area draining to that location, the slope at that location
decline with soil thickness and when soil was produced and the particle size grading at that location, the area-
the bulk density was reduced so the volume increased. slope-diameter relationship (Figure 11.5). The trends of
Their chemical weathering model was quite simple. The surface grading for five hillslope profiles are illustrated in
chemical weathering model was a modification of the Figure 11.5b:
exponential function and as the soil chemically
1. This is a hillslope where the slope is increasing down
weathered the thickness of the soil decreased. The func-
the hillslope so is concave down in profile and looks
tion they adopted to calculate the rate of decrease in the
like a rounded hilltop. The d50 increases down the
soil thickness was a combination of (1) an exponential
hillslope (i.e. increasing area, moving from left to right
term that decreased with soil depth and (2) an exponen-
in Figure 11.5). The diameter contours always increase
tial term that decreased with both time and thickness so
from left to right and from bottom to top, so in general
that the rate of decrease decreased with time. The reduc-
concave hillslopes always coarsen downslope.
tion of weathering rate with soil depth reflected a con-
2. This hillslope has constant slope downslope and, as for
sumption of acid down the profile (and thus a weakening
slope 1, always coarsens downslope.
of weathering), while the reduction with time reflected a
3. This hillslope has slopes that are decreasing downslope
reduction of chemical weathering reactivity with time
and is concave up. The gradient of the hillslope in
(the most reactive material reacts first leaving behind less
Figure 11.5 is less than the gradient of the contours
reactive material).
so the hillslope coarsens downslope.
Probably the simplest implementation of chemical
4. This hillslope is similar to profile 3 except that the rate of
weathering was a dissolution model of Mudd and Furbish
decrease of slope downslope is more severe so the gra-
(2004). In a mass balance model for hillslope soil mass
dient of the hillslope in Figure 11.5 is steeper than the
loss they assumed that the loss by dissolution was con-
gradient of the contours. This hillslope fines downslope.
stant down the hillslope. By assuming that bioturbation
5. This hillslope is a classic catena profile with a rounded
disturbed the weathering soil sufficiently so that the bulk
hilltop and a concave profile downslope of the hilltop.
density of the soil did not change with dissolution (Brim-
Travelling down this hillslope it will initially coarsen.
hall et al., 1992) they assumed that this dissolution led to a
As the hillslope transitions to concave up, it will con-
proportionate loss of soil thickness in a similar fashion to
tinue to coarsen until the rate of reduction of the
Minasny and McBratney (2001).
hillslope slope is severe enough that is starts to fine
The matrix state-space approach for physical
downstream. Whether this latter region of fining occurs
weathering described in this Chapter 7 was first
will depend on the concavity of the hillslope and
described by Cohen et al. (2009, 2010) for his mARM
whether the concavity is strong enough relative to the
soilscape model, who in turn had extended the work of
gradient of the diameter contours in Figure 11.5.
Willgoose and Sharmeen (2006) that used their physic-
ally based ARMOUR fluvial erosion model (ARMOUR Figure 11.5 shows that the spatial distribution of soils,
is discussed in detail in Chapter 4). Cohen modelled the and any questions of downslope fining or coarsening of
soil by a large number (typically about 30) of size frac- those soils, depends on the interaction between the pedo-
tions from clay to large gravel, and used a variable number genesis processes that produce the soils (and thus drive
(typically about 20) of soil layers with depth. mARM was the slope-area dependence of soil grading) and landform
subsequently extended to examine the dynamics of soil evolution processes that generate those profiles (and the
profile adjustment due to climate variability (Cohen et al., slope-area relations for those slopes). Ultimately deeper
2013), and an aeolian deposition module was added to understanding of these links will come only from a coupled
simulate soil development and spatial patterns of erosion landscape-soilscape evolution model (see Chapter 16 for
at a field site in Israel (Cohen et al., 2015, 2016). Welivi- examples).
tiya et al. (2016) extended Cohen et al. (2010) to explore, Gasparini et al. (2004) used a landform evolution
using a parametric study, the generality, and process model that simulated sediment transport with a combin-
drivers, of scale-invariant behaviour found in the spatial ation of sand and gravel, and that could armour the catch-
organisation of soil grading generated by mARM across a ment surface by selective entrainment of the sand. She
landform. found for the landforms that she generated that the surface
One key outcome of Cohen and Welivitiya was the had a weak trend of fining downstream. Comparing her
discovery of a spatially organised relationship between results with Welivitiya et al. (2016), he found the slope of
FIGURE 11.5: Area-slope-d50 plots from the (a) mARM and (b) SSSPAM models showing how the surface soil grading changes downslope.
See the text for the explanation of the five lines in the SSSPAM plot (from Welivitiya et al., 2016).
the diameter contours was a function of the erosion expo- model following the principles outlined in Chapter 8 for
nents on discharge and slope. While the process formula- four saprolite components (anorthite, chlorite, microcline,
tions of Gasparini and Welivitiya are not directly albite) generating four leachate ions (calcium, magne-
comparable, the parameters of the erosion process used sium, potassium, sodium). It has a reaction rate that is a
by Gasparini result in the slope-area relationship for the generalisation of Equation (8.6) where the exponent on
topography (the dotted lines in Figure 11.5) being the acid concentration, α, is calibrated, and where the
approximately parallel to the contours for the d50 (the equilibrium leachate concentration is calculated based on
solid lines in Figure 11.5), so that if Welivitiya’s relation- an equilibrium sum of the four ion concentrations and the
ship is used, then the grading will not change greatly dissolved aluminium concentration, which is pH (i.e. acid
downslope, which is consistent with Gasparini’s results. concentration) dependent. The physical weathering model
Finke (2012) presents the SoilGen2 model, which is based on Salvador-Blanes et al. (2007), but instead of
couples physical and chemical weathering for a single soil having 1,000 size fractions different by just 1 μm, he has
profile, but which did not couple the processes spatially. size fractions that are powers of 2, so that a particle in one
The chemical weathering component builds upon the size fraction can be split into eight cubes and fall into the
LEACHC model. LEACHC couples Richard’s equation next size fraction smaller. This is similar to, but less
for unsaturated flow down through the profile (so v is soil general than, the size fraction approach of mARM, but
moisture dependent in Equation (8.6)) with a geochemical Salvador-Blanes modelled the fragmentation as stochastic
with time. The rate of fracturing is a function of the rate of

change of temperature, this being a surrogate for internal
stress. Finally SoilGen2 implements the clay translocation
model in Chapter 10. The capabilities for soil pedogenesis
of SoilGen2 have not been fully explored, but a number of
papers have shown promising comparisons with the field
(Finke, 2012; Sauer et al., 2012; Yu et al., 2013; Jagerci-
kova et al., 2014; Finke et al., 2015).
Vanwalleghem et al. (2013), and Temme and Vanwal-
leghem (2016) provide what is probably, to date, the most
complete implementation of a soilscape model that has
been coupled to a landform evolution model. Their soils- FIGURE 11.6: Conceptualisation of chemical weathering impacts
cape model includes an exponential soil production func- on particle shape and size: (a) equal dissolution around the surface of
tion for saprolite to soil conversion, and a physical the particle and (b) surface roughening and internal corrosion.
weathering model based on the mARM matrix formula-

tion described in Chapter 7. Vanwalleghem has only three Fickian diffusion, and that is linearly proportional to a
layers in this soil (conceptually simulating soil horizons), declining exponential function with depth.
while Temme and Vanwalleghem generalise that to an
infinite number of layers following the approach used by
Cohen et al. (2010) in mARM. The soil is represented by 11.2.4 Coupling Physical and Chemical
five size classes (fine clay, clay, silt, sand and gravel). Weathering
While they do not specifically have an armour surface
layer, they do implement an empirical selective entrain- This section discusses how we might couple the grading
ment model that is similar in concept to the Parker and evolution from the physical weathering model (Chapter 7),
Klingeman (1982) selective entrainment model used by with the chemical weathering model focussed on chemical
Sharmeen, Cohen, Welivitiya and co-authors. While transformation of the rock (Chapter 8). We conceptualise
acknowledging the principles of the chemical weathering how this could occur because chemical and physical
model discussed in Chapter 8, they chose to implement a weathering have, to date, not been coupled in soilscape
very simple chemical weathering model. This chemical models to the level of quantification discussed in the
weathering model (1) generates dissolved products, (2) previous chapters.
reduces the grain size as a result of dissolution by the Chemical weathering dissolution (Figure 11.6) may
process in Figure 11.6a and (3) generates clay particles. have two potential impacts on the grading.
Their chemical weathering rate is linearly proportional to The first impact is if all the dissolution is from around
the specific surface area of the soil, and has an exponential the outside of the particles, then all particles reduce in
decline in rate with depth below the surface. The sensitiv- size. If we can assume to first order dissolution on the
ity studies in the chemical weathering chapter (Chapter 8) surface of each particle is equal, then this mass of loss can
showed how under many conditions an exponential be easily converted into a volume loss, and therefore a
decline in weathering rate with depth was approximately diameter loss for the particles. All the particles will lose
valid. The main situation where the exponential decline is the same depth of the material around the outside rind of
not valid is when weathering results in all of the saprolite the particle (e.g. Vanwalleghem et al., 2013).
being weathered by the time it reaches the surface, in The second impact is if the dissolution occurs within
which the humped weathering function is the best the particle without loss of the material from the outside.
representation. Their clay particles are formed from the That this mechanism occurs is supported by the large
dissolved product, and this is a humped function with increase in specific area of weathered particles observed
the maximum clay generation just below the surface in laboratory experiments. This increase could come
(Barshad, 1959). They have a simple clay illuviation about only by the development of pores and cracks within
model based on the clay content of the surface layer and the particle. In this case the diameter of the particles does
where the travel distance is a declining exponential func- not change, mass is lost (the mass lost equals the mass
tion with depth. Finally they implement bioturbation that dissolved) and the bulk density changes in proportion so
mixes soil between layers in a process that approximates that the depth of the layer does not change.
If enough material is lost from the interior of the Varadachari et al., 1994). Biotite’s specific gravity is
particle, it may become ‘rotten’ and collapse under the about 2.7–3.1 while that of kaolinite is 2.2–2.7, suggest-
applied load of the soil above, or bioturbation may ing a chemical weathering strain of about 15–20%. Buss
physically break up the rotten particles. In this case the et al. (2008) identified the chemical breakdown of biotite
soil ‘collapses’. In this case the diameter of the particle as causing the physical weathering of corestones within a
will reduce (and it will simultaneously break into a soil profile. Over the long term there may be leaching of
number of smaller particles), and the bulk density will kaolinite, so that over the long term the strain associated
increase. The actual collapse will not change the mass in with chemical weathering biotite may be negative, but if
the layer so that means the layer will become thinner as a there exists a transition through a positive strain phase,
result of the change in the bulk density. Everything else then it is possible to induce physical weathering. Nor are
being equal, it seems reasonable to suggest that collapse these conclusions solely applicable to biotite-kaolinite
will become greater with greater applied vertical over- conversion. Parizek and Girty (2014) found that
burden load and/or greater bioturbation. Accordingly, weathering of biotite to vermiculite produced a chemical
the rate of collapse is likely to (1) linearly increase with weathering strain of about 10%, while Graham et al.
depth below the surface, just as the overburden pressure (2010) suggested a strain as high as 30%. Goodfellow
increases linearly with depth and (2) increase near the et al. (2016) found a strain of 5% when oxidising biotite,
surface where bioturbation is strongest exponentially and they found substantial physical changes (e.g. per-
declining with depth. meability increased by two orders of magnitude) in the
We now consider the impact of chemical transform- rock even in the early stages of the weathering process,
ation from primary to secondary minerals. This process is long before there was any substantial element leaching
complicated because it is unclear what the diameter char- (leaching is a common indicator of the progress of
acteristics of the chemically transformed products will be. chemical weathering).
In the chemical literature it is commonly assumed that the The role of chemical weathering in spalling fragmen-
secondary mineral is clay (Anderson and Anderson, 2010; tation also cannot be ignored, with Fletcher and Brantley
Vanwalleghem et al., 2013). In this case when mass is (2010) suggesting the volume increases by oxidation of
modified (in dissolution it is removed entirely from the iron silicate minerals results in spalling of the outer rind of
profile), that mass would then be placed into a size corestones. For a quartz diorite they suggested a spalling
grading reflecting a clay fraction (e.g. less than 2 μm). rate of 2 cm/200 years.
Because a unit mass of rock does not yield a unit mass of There is evidence to suggest that chemical weathering
clay (typically clays are hydrated so chemically may be rate limited by the rate at which fresh surfaces are
weathering a rock with no dissolution loss increases the created by physical weathering (Barman et al., 1992),
mass), so there would need to be a mass balancing reflect- suggesting that it is only by a coupling of chemical
ing the types of clays generated from the primary weathering with physical weathering that we will be able
minerals. to correctly model chemical weathering.
Finally there is the possibility that chemical trans- In conclusion, a proposed soilscape-modelling frame-
formation of minerals may contribute to physical break- work would look as follows. At each timestep:
down processes over the long term. Chemical weathering
may weaken rock so that it is more susceptible to phys- 1. The soil production function generates fresh soil at the
ical weathering processes (Migon and Thomas, 2002; base of the profile for one timestep.
Graham et al., 2010). When they weather, many primary 2. Erosion or deposition adds or removes soil at the
minerals generate secondary minerals that have a lower surface for one timestep.
density than the source mineral. This means that if the 3. For each of the physical weathering processes, modify
space that contains the weathered mineral within the rock the grading of the profile in the layers for one timestep.
matrix is fixed, the positive chemical weathering strain For a numerical explicit solver the rates of processes
may induce internal tensile stresses in the rock fragment that are a function of depth below the surface will use
and drive physical fragmentation of the rock. For the bulk density from the preceding timestep to deter-
instance, it is common to observe for weathering of mine the depth below the surface of the layers. When
biotite that there are weathering products (typically there is transport between the layers the model will
kaolinite) between the layers of biotite, and that these need to track not just the soil grading but also the
kaolinite deposits are associated with delamination of the mineralogy of the particles and the chemical
biotite (e.g. Bisdom et al., 1982; Fordham, 1990; weathering state of those particles.
4. The chemical weathering model would then be applied impacts on pedogenesis. When we discussed soil organic
for one timestep. There may be feedbacks between the carbon, anthropogenic impacts were central to soil carbon
grading and the infiltration velocity through the profile. dynamics. Veenstra and Burras (2015) observed that
Potentially track how the grading of particles is 50 years of agriculture have had significant impacts on
changed as a result of one timestep of chemical deep (1 m and more) soil chemistry and structure, with
weathering. implications for deep organic carbon storage, but left
5. Calculate the new bulk density for each layer and then many unanswered questions about the processes leading
the depth below the surface of all the layers. to the observations.
6. Rinse and repeat for as many timesteps as required. One question we have not directly addressed is the role
7. The rapidly changing dynamics of soil organic matter of erosion and in-profile soil processes on agricultural
evolution relative to the chemical and physical productivity. Van Oost and Bakker (2012) provide an
weathering processes suggest that the SOM model interesting conceptual model that involves many of the
would be best run separately with the output (e.g. acid processes discussed in the previous sections. They postu-
potential, soil gas carbon dioxide) being input into the late a case of newly developed agricultural land where
chemical weathering model, and the (slowing) evolv- crop productivity can be a function of either water
ing soil resulting from erosion and weathering would limitation and/or nutrient limitation. Initially the land is
be an input to the SOM model. Since the SOM model not limited by either water or nutrients. The first few
varies more rapidly than the soil, then a number of millimetres of erosion will have very little impact on the
timesteps with the SOM model would be run for each water-holding capacity of the soil so will not induce water
timestep of the pedogenesis model, and the average limitation. However, the nutrient status (SOM and associ-
over these timesteps would be the output from the ated nutrients) will be severely impacted because most
SOM model that is input into the weathering model. nutrients and SOM are highly concentrated near the sur-
face, leading to potential nutrient limitation. As the cumu-
One thing to note is that the physical weathering and
lative erosion depth increases, and as erosion increasingly
chemical weathering models are largely independent as
removes soil with declining nutrient storage, then the
they are calculated above. The interaction comes about
relative incremental impact of the erosion on nutrient
because the chemical weathering of the previous time-
limitation will decline. As erosion continues, the soil
steps influences the physical weathering in the next time-
profile will thin enough that the water-holding capacity
step. Also, depending on the rates of change of the various
of the profile is depleted sufficiently that water availability
processes, a process may not need to be recalculated at
becomes an issue. Roots tend to be concentrated near the
each timestep. For instance, chemical weathering is
soil surface (e.g. Jackson et al., 1996), so as the soil thins
slower than physical weathering, so physical weathering
further the water limitation effects become incrementally
may need to be calculated at every timestep while chem-
more significant as the higher root density fraction of the
ical weathering may need to be calculated only every X
soil profile becomes increasingly affected. The combin-
timesteps (though every time chemical weathering was
ation of these processes results in a crop productivity
calculated it would need to be calculated at X times the
decline versus cumulative erosion curve that (1) initially
rate it would be calculated at if it were calculated each
declines quickly (due to nutrient-limitation), (2) levels out
timestep).
and (3) for high cumulative erosion declines quickly again
to zero (due to water limitation) (Figure 11.7). This con-
11.2.5 Other Items ceptualisation was developed to understand the impact of
increased erosion after agricultural development and may
The soilscape discussion in this book has ignored any not be appropriate for natural systems. On the other hand
anthropogenic impact on in-soil pedogenic processes. it is possible to imagine this process at work after natural
There has been only a little discussion in the literature of disturbances of vegetation (e.g. fire, tree throw, climate
the role of known agricultural impacts on pedogenic pro- change).
cesses (e.g. acidification, salinization and soil wetness).
This may be because of a perception that the durations
over which western agriculture have been operational is 11.3 Numerical Issues
short relative to pedogenic time scales, organic carbon
being a notable exception. It is thus difficult to propose, Most models break the soil into a series of layers as
with any confidence, mathematical models of agricultural discussed above. Many of them fix the thickness of, or
In the existing models of physical weathering some

have defined the layers based on thickness of the layer,
while others have defined them on the basis of mass
within the layer. The mARM model defined the layers
on thickness, but we have subsequently found that this
definition makes formulation of the transition matrices
complex, particularly if the layers are different thickness
so SSSPAM (an enhancement of mARM) uses mass to
define a layer. This means that you need to have a model
for porosity so that this mass can be converted to soil
depth. This porosity dependence is implicit in the soil
thickness formulations.
A related issue is how to define the size fractions
FIGURE 11.7: Conceptualisation of how erosion might impact on within the layer. In mARM we defined them on the basis
crop yield, first by reducing the stored nutrients, and then by of the proportion by mass within each size fraction. This is
reducing the water-holding capacity (from Van Oost and Bakker, consistent with how particle size distributions are nor-
2012).
mally expressed in the soil science community. However,
subsequent experience has been that a numerical imple-
mentation based on the mass (rather than proportion) in
mass in, the layers. With the exception of mARM most do each size within each layer is computationally and con-
not do anything special with the uppermost layer. They ceptually easier, and consistent with the mass per layer
simply consider that erosion is a loss of material from the formulation in the previous paragraph.
top layer and do not model the change in the grading that Defining each layer by mass avoids another issue,
results in armouring. which is whether to define the layer boundaries by (1)
MILESD uses a different vertical discretisation for the elevations relative to the soil surface (which itself is
layers, modelling just three layers, corresponding to soil changing as a result of the erosion and landform evolu-
horizons, and models the evolution of the thickness and tion) or (2) absolute elevations (e.g. above sea level).
internal characteristics of these layers. This has the advan- Defining the layers with mass avoids the need to define
tage of allowing them to specialise the physics and chem- the elevation of the boundaries in the soil model except
istry on what happens in each of the horizons but has the when weathering rates need to be determined relative to
disadvantage of not allowing them to investigate what in the surface, in which case all layers are relative to the
the physics and chemistry quantitatively drives the devel- evolving soil surface.
opment of the horizons in the first place. Subsequently, If porosity is changing, then the thickness of each layer
Vanwalleghem et al. (2013) implemented a series of pro- will naturally change relative to how much mass is in each
cesses that have no regard for the horizons themselves, layer. It is then conceivable that the soil surface elevation
but only the depth of the numerical layers below the will rise if the right combination of changes in porosity
surface and their layer thickness can change with time. and mass within all the layers occurs. For instance, con-
The main numerical issue is the coupling of the surface sider the case of the dilation of a soil due to tree throw, but
layer in the physical model with the erosion mechanics. where erosion is low. The porosity will increase as a result
A key feature of the surface is that if there is a coarse of the tearing out of the saprolite of the rock, and this will
weathering-resistant component of the soil, then a surface inevitably lead to a soil thickening and a rise in elevation
armour will be formed. This armour consists of the mater- of the soil surface.
ial that cannot be moved by the overland flow, and its
existence reduces the erosion rate. Typical armour layer
depths observed in the field are of the order of 1–2 times 11.4 Further Reading
the diameter of the particles in the armour layer, so it is
quite thin relative to the depth of the soil. Our work has There is no single reference that covers all the material in
found that a thin surface layer is needed to correctly the soilscape chapters here, Chapters 5–11. Schaetzl and
capture the time-varying dynamics of armouring. Thicker Thompson (2015) provide an excellent, though less quan-
layers then underlie the armour layer, so that there aren’t titative, overview. At the risk of offending some research-
an unnecessarily large number of layers within the profile. ers by overgeneralisation, typically soil chemists discuss
soil chemistry in the absence of the geomorphic pro- the field interactions between chemical weathering, phys-
cesses, geomorphologists discuss transport processes in ical weathering and bioturbation. Maher (2010) discusses
the absence of pedogenic processes, and soil scientists do the geochemical weathering model presented here in
not quantify the long-term evolutionary processes leading greater detail, while Kump et al. (2000) provides a good
to the soils observed. The best modern overviews are overview of geochemical weathering in the field. There is
Anderson and Anderson (2010), and Minasny et al. no equivalent paper to Maher about physical weathering
(2015). Minasny and McBratney (2001) was the first other than the Sharmeen, Cohen, Wells and Welivitiya
(and very readable) paper to couple a pedogenesis model papers that were the basis for Chapter 7, though Wells
(albeit simple) to a landform evolution model. Brimhall et al. (2008) provides a quantitative interpretation of their
et al. (1992) uses field studies to discuss the complexity of experimental studies.
12 Tectonics and Geology
12.1 Introduction continental crust) comprise lower density rocks than the
plates that underlie the oceans (oceanic crust). Both crust
One of the most important drivers of the movement of types are less dense than the mantle, which is why they
landforms is movement of the geology underlying the both float on the mantle. This will be crucial later.
landform. Tectonic uplift provides the initial elevation Because of their lower density the terrestrial plates float
that is then eroded away by the land-forming processes higher in the mantle than oceanic crust. This is why the
discussed in following chapters. upper surface of the terrestrial plates is above sea level. At
These tectonic forcings observed at and near the land their horizontal boundaries the plates can be moving apart
surface are largely driven by interactions between the creating rift valleys or regions of normal faulting, on land,
crust and the underlying mantle. The flow processes in and spreading centres with undersea volcanic activity in
the mantle lift, tilt, warp and translate the overlying geol- the oceans. If the plates are converging, then mountains
ogy, and the land surface responds accordingly. It is these and subduction zones (where one plate slides under
latter land surface responses that are our primary interest another, typically oceanic crust underneath continental
in this book, but to understand them fully we do need to crust because of the density differences) are created.
understand how the mantle drives them (even if we don’t Finally there is the possibility that the plates are moving
model the mantle explicitly, that being the primary laterally relative to each other, creating active strike-slip
domain of the field of geodynamics; e.g. Turcotte and fault zones (e.g. the San Andreas fault in California).
Schubert, 2002). Our focus will not only be on how the The drivers of the plate movement are the convection
mantle drives the landforms, but how the mantle responds currents within the mantle where hot liquid rises
to landform changes, and how that in turn influences the (upwelling) and cool liquid falls (downwelling). Each pair
landform evolution into the future: i.e. the interaction of rising and falling currents is called a convection cell (or
between the landforms and the crust-mantle system. in other fields a Hadley or Rayleigh-Benard cell). The
A very brief overview of the crust-mantle system is schematic in Figure 12.1 suggests a convection schematic
warranted before we discuss details, and interested readers that is rather more neat and tidy than the current under-
can find a more detailed overview in any modern intro- standing of mantle dynamics which involves multiple
ductory undergraduate geology book. The crust-mantle internal layers (e.g. the mantle is demarcated into two
system consists of a thin layer (typically at most a few regions, upper and lower mantle, by a change in seismic
10’s of kms thick) of solid material (the crust) floating on velocity at a depth of 660 km), isolated plumes and
top of a liquid layer underlying it (the mantle) superplumes, and geochemical separation as a result of
(Figure 12.1). We will return to just how viscous the localised internal heat generation and density differences
mantle is in a moment, but for the moment it suffices to (Bercovici, 2003; Ogawa, 2008). It is believed that the
think of it as a very viscous fluid (e.g. like refrigerated driver for these convection currents is radioactive decay
honey on steroids rather than water). This crust is not one within the mantle and the core underlying the mantle. At
continuous piece covering the whole planet, but a large the top of the convection cell the flow direction is pre-
number of interconnecting pieces (i.e. the plates in plate dominantly horizontal, and it is the friction between the
tectonics) that are floating around and interacting with bottom of the crust and the horizontal mantle flow that
each other at their horizontal boundaries. The plates that drives the horizontal movement of the plates. Recent work
190 make up the land surface of the planet (called terrestrial or also suggests that downwelling actively drags the crust

mantle convection cells and crustal
response. Note (1) the potential
existence of mountain ranges at
the subduction zone, (2) the
increasing thickness of the oceanic
crust with distance away from the
spreading centre and (3) the
relative thicknesses of the oceanic
and continental crust.
down into the mantle in subduction zones (Conrad and

Lithgow-Bertelloni, 2002; Mitrovica and Forte, 2004).
These mantle convection velocities are typically less than
100 mm/year (Montelli et al., 2004). Some tectonic uplift
is generated over the regions of upwelling (e.g. Yellow-
stone National Park) to balance the forces resulting from
the change in the direction of the momentum of the
upwelling convection current at the top of the plume,
FIGURE 12.2: Schematic of isostatic compensation for continental
though whether this uplifting is ongoing (i.e. the topog- and oceanic crust. Note that (1) the oceanic crust is thinner than
raphy is continuing to rise or fall) will depend on whether continental crust and (2) the proportion of the oceanic crust above
there is a force balance between the extra weight of the the MZD is less than continental crust because the oceanic crust is
uplifted crust and mantle (acting downward), and the higher density.
momentum change (acting upward).
Finally there is a slight complexity which we will not to gravity, and h is the depth below the surface of the
address here. This is that when the mantle is loaded, two mantle), then by a force balance between the weight of the
processes happen. As mentioned above, over the long crust and the pressure on the bottom surface, the height of
term the mantle flows viscously until equilibrium occurs. the crust above the mantle surface is
However, in the first few years of the loading the domin-
ρ
ant cause of deflection is the elastic compression of the ðd hÞ ¼ d 1 crust (12.1)
ρmantle
mantle fluid. It’s only after a few years that the viscous
flow has become sufficient that it dominates the elastic so that the proportion of the crustal thickness that is above
response. Since we are interested in landform evolution the mantle ððd hÞ=d Þ is given by the difference between
over thousands to millions of years, we will ignore this the density of the mantle and crustal density. When the
elastic response. crust obeys Equation (12.1) it is referred to as being in
isostatic equilibrium or simply ‘equilibrated’, and h is
called the ‘compensation depth’. Landforms do not
12.2 Isostasy and Isostatic Compensation instantly achieve Equation (12.1) because of the high
viscosity of the mantle, and prior to equilibrium the land-
12.2.1 The Isostatic Compensation Principle forms will tend to evolve toward equilibration with some
response time, as will be discussed in following sections.
The most fundamental concept in tectonic-landform inter-
Figure 12.2 is simplistic in three ways:
actions is that of isostatic compensation. The basic idea of
this is that the crust (both marine and terrestrial) floats on • The figure shows a mantle free surface against which
the mantle below (Figure 12.2) much as an ice cube floats height is measured. Since the entire planet is covered by
in water. If we consider a unit area of crust of thickness d crust, then a free surface does not actually exist, so it is
with density ρcrust floating in a mantle with density ρmantle not possible to measure heights against it. However, as
and the pressure in the mantle is hydrostatic (so the a conceptualisation, this figure suffices. It is convenient
pressure is p ¼ ρmantle gh, where g is the acceleration due below to refer to an equivalent of the mantle free surface
TABLE 12.1 Typical crustal properties
Density (kg/m3) Typical thickness (km) Heights relative to MZD (km)
Upper mantle 3,2002

Terrestrial/continental crust 2,600–2,8002 15–751,2 7.1–14.1
Marine/oceanic crust 2,800–3,0002 5–101,2,3,4 0.6–2.4
isolated locations to 35 km4
1
Hammer et al. (2013); Klepeis et al. (2003), 2Rogers (2007). 3Canales et al. (2002), 4Sallares and Calahorrano (2007).
FIGURE 12.3: Schematic showing the net effect of the combination of erosion and isostatic compensation. (a) The original catchment
showing the base level provided by the ocean height, (b) the change in land elevations if the crust is eroded instantaneously and (c) the
adjustment provided by isostatic compensation over the long term. Note that in (c) not only is the net elevation lower by the amount shown,
but also if the ocean level has not changed, then the base level provided by the ocean has retreated inland.
height and here we will label it the ‘mantle zero datum’ on the oceanic crust (so the marine crust will not rise quite
(MZD) as an analogy to the zero pressure datum pro- as high as Table 12.1 suggests), but this calculation shows
vided by the free surface of a fluid (e.g. water in a lake). the difference in the maximum height of the terrestrial and
• The figure does not consider mass continuity within the marine crust. These depths are approximately consistent
mantle. If the crust mass sinks into the mantle due to with the difference in the maximum depth of the ocean
loading, then the displaced mantle material must move (Mariana Trench, 10,971 m) and the land (Mt Everest
elsewhere. At equilibrium the mantle everywhere out- 8,848 m), and the average depths of the ocean (4,000 m)
side the location of the load must rise to compensate and the height of the land (840 m). The table also explains
(and the MZD must rise) even if only by a small why marine crust underlies the ocean: it is more dense and
amount. thinner, and so forms the low spots on the surface of the
• Finally, as we will see below, it may take some time for crust, which is where the water collects.
this equilibrium to be attained, and in the meantime the Equation (12.1) is important in determining how the
displaced mantle will travel out in a slowly dissipating terrestrial crust responds to erosion and deposition of
wave from the load, which may create wavelike uplift material. From this equation and Table 12.1, 1.0 m of
transients in the MZD (Lambeck and Johnston, 1998). erosion (Figure 12.3) will yield an isostatic rise in the
crust of 0.76 m so that the net reduction in the elevation
As noted above, the entire surface of the planet is
about the MZD is 0.24 m. This calculation ignores many
covered by crust and thus there is no free surface, so the
important factors that will be discussed below but illus-
height in Equation (12.1) is the height above the MZD.
trates the hydrostatic balance that the terrestrial crust
Table 12.1 lists approximate densities for mantle and
exhibits with the underlying mantle. The important factors
crust, typical thicknesses for continental and marine crusts
that are missing from this calculation are the following:
and heights above the MZD to which the surface of crust
can rise. 1. The mantle is extremely viscous, and while in the
The heights above MZD in Table 12.1 are an approxi- example above the hydrostatic balance produces a
mation because they ignore the weight of the ocean water 0.76 m isostatic uplift, this uplift may take tens of
thousands of years to reach equilibrium. For instance, spatial patterns of uplift and subsidence around the planet
regions that were covered in ice kilometres thick at the and (3) data from satellite gravity missions (e.g. GRACE).
end of the last ice age (about 10,000 years ago) are still However, there is explanatory benefit from consider-
uplifting as a result of the isostatic adjustment to the ing a simple single-layer mantle model. To illustrate the
loss of ice (Section 12.2.3). effect of the lateral stiffness consider a two-dimensional
2. The stiffness of the terrestrial crust laterally (i.e. east problem with a local out-of-plane (i.e. at right angles to
and north) means that any localised reduction in the the crust) load (Figure 12.4a), but ignore for the moment
load of the crust due to erosion will be spread over a the buoyancy forces resulting from floating on the mantle.
larger area than the area of erosion so local adjustments This problem can be a model for a volcano supported by
will typically be less than predicted by Equation (12.1) the crust. If we assume that the crust is the same thickness
but spread over a larger area. and elasticity everywhere (Li et al., 2004), then
3. The lateral stiffness of the plates combined with the
PL3
buoyancy force means that there are some subtle far w¼ (12.2)
192 EI
field effects from the local isostatic readjustment.
where w is the deflection relative to the unloaded case, E
We will now discuss the details of these additional
is the elasticity of the plate, P is the point load, L is the
complicating factors. These factors form the core of
distance between supports and I is the moment of inertia
models collectively called Glacial Isostatic Adjustment
of the plate. For a unit width plate the moment of inertia is
(GIA) models (e.g. ICE-3G, -4G, -5G; Tushingham and
I ¼ T 3 =12 where T is the thickness of the crust. From
Peltier, 1991; Peltier, 1999, 2004). As their name sug-
Equation (12.2) it can be seen that the maximum deflec-
gests, they have been developed to understand crustal
tion is reduced by either a stiffer crust (E high) or a thicker
movements since the end of the last ice age and to
crust (I high). A doubling of the thickness of the crust
reconstruct past ice histories from existing crustal
reduces the deflection by a factor of 8.
movements. Insights from them provide a deeper under-
The incorporation of the buoyancy force to this problem
standing of crust-mantle interactions. At the simplest
means that instead of the point load being supported at the
level these models couple the motion of a thin elastic
edges of the plate as in Figure 12.4a the load is supported
crustal plate floating on a viscous mantle (Turcotte and
by the mantle nearer the load (Figure 12.4b). A good
Schubert, 2002). They model (1) the response of the
analogy for how the loads are distributed is obtained by
crust to the removal of ice (up to 4 km thick) from the
noting that the buoyancy force is linear with deflection
northern extremes of North American, Europe and
below the MZD so the buoyancy force can be replaced
Russia, (2) the extra loading on the crust of the resulting
by a linear spring. The load on the spring is linearly related
global rise in sea levels of about 130–135 m over the
to the deflection of the spring, so that a classic formulation
last 21,000 years (the end of the last glacial maximum,
for this looks like Figure 12.4c. In the case of an infinitely
Lambeck et al., 2014) and (3) the (mostly) coastal
flexible crust the entire load is taken by the spring directly
retreat due to the combination of isostatic adjustment
under the load (this is the simple isostatic problem dis-
and sea level rise.
cussed at the start of this section). As the crust becomes
stiffer, more of the load is carried by the springs to the left
12.2.2 Spatial Linkages and right. The general two-dimensional solution for the
deflection is (Ventsel and Krauthammer, 2001)
The first of the complicating factors we will discuss is the
lateral stiffness of the crust, which means that vertical ∂4 w ∂4 w ∂4 w 1
þ 2 2 2 þ 4 ¼ ½Pðx; yÞ qa (12.3)
deformations do not just occur locally near the load but ∂x 4 ∂x y ∂y D
also at a distance away from the load. A common repre-
where P is the vertical load per unit plan area, qa is the
sentation is to model the crust as a thin elastic plate
elastic reaction force of the mantle per unit plan area qa ¼ kw,
floating on a viscous fluid(s). The earliest models assumed
k ¼ ρmantle g and D is the flexural rigidity of the plate
that the mantle was uniform throughout with one viscos-
ity. Recent models use multiple layers with depth and
ET3
temperature-dependent variation in viscosity (e.g. Figure D¼ (12.4)
12 1 ν2
1.12 in Rogers, 2007). The layer structure has been
inferred from (1) seismic data, (2) better fits provided by and ν is Poisson’s ratio (typically in the range 0.2–0.3 for
the multilayer mantle models within GIA models to uncracked rock; Zhang and Bentley, 2005; Gercek, 2007).
FIGURE 12.5: Two lateral constraint conditions for the elastic plate
representing end members for the lateral movement of the crust
under loading. The unshaded region shows the plate before loading,
and the shaded region shows the plate after loading. (a) The case of an
infinite plate where the out-of-plane load induces both bending in the
plate as well as tensile forces that results from the plate being
constraining from moving inward. Note that that plate is slightly
thinner after loading because it is stretched over a slightly longer
length than before loading, (b) The case where the ends of the plate
can move laterally freely so that no tensile stresses are induced in the
plate, and consequentially the deflection of the plate is slightly larger
than the constrained case. In comparison with (a), in case (b) the plate
does not thin after loading.
Equation (12.3) is an infinite plate solution and is

FIGURE 12.4: Schematic showing how an out-of-plane load can derived assuming an elastic plate that is infinite in both
be supported by the crust in two end member conditions, and a third
horizontal directions and thus implicitly assumes that the
more realistic support condition: (a) if the crust is supported only
by the end of the crust and there is no buoyancy, (b) if the crust
elastic plate is horizontally constrained from shrinking
is very flexible, then all the load is supported by buoyancy in the laterally as the out-of-plane load is applied. This constraint
locality of the applied load and (c) a spring support condition for an means that a tensile load is induced in the plate with
elastic plate where the springs represent the support provided by increasing load (Figure 12.5). As an approximate analogy,
buoyancy.
consider a high-wire walker on a cable between two trees.
The load of the high-wire walker induces a tensile load in
The elastic reaction force qa is a good approximation to the cable. This tensile force is a direct response to the
the buoyancy force if the density of the upper mantle is deflection of the wire. However, if the lateral constraint
constant with depth. on cable fails (i.e. the wire breaks), then the tensile force
Turcotte and Schubert (2002) consider a number of and stress will be released, and the high-wire walker will
variants of the formulation of Equation (12.3). They are fall on the ground (rather embarrassingly). This horizontal
(1) where the deflected region is filled with water (e.g. the constraint results in the appearance of Poisson ratio (which
crust is underwater both before and after the load is describes the shrinkage of the elastic material at right
applied) and (2) where the deflected area is filled with angles to the tensile load in the plate) in the denominator
crustal material (e.g. where the deflected area is filled with of Equation (12.4), and the tensile forces induced by lateral
eroded sediment and any uplifted area is eroded back flat). constraint make the plate slightly more rigid than it would
In the first case the new loading is P0 ¼ P þ ρwater gw otherwise be (due to the high-wire walker effect). If this
while for the latter case the elastic reaction force is P0 ¼ lateral constraint is zero, then the rigidity is
P þ ρdeposited gw where ρdeposited is the density of the
ET3
deposited material, noting that sedimentary rocks typic- D¼ 12 (12.5)
ally have a lower density than igneous and metamorphic
rocks, so the deposited density is likely to be lower than which is the rigidity used to derive the deflection in
the crust as a whole. Equation (12.3). The key implication of this is that the
lateral constraint on movement of the plate will reduce the

magnitude of the deflection slightly (typical values for ν
of 0.2–0.3 suggest by about 5–10%). This lateral con-
straint is provided by (1) subduction zones at the plate
boundary and (2) frictional resistance to horizontal move-
ment of the crust provided by the mantle. The lateral
constraint is also less if the crust is vertically cracked.
Most studies that use Equation (12.3) assume perfect
lateral constraint and use Equation (12.4). The small dif-
ference between Equations (12.4) and (12.5) can reason-
ably be ignored in the field, but the differences may be
important if using Equation (12.3) to validate the accuracy
of computer models, where lateral constraints may be
modelled explicitly, e.g. through the drag force of the
mantle on the overlying crust, downwelling crust at FIGURE 12.6: The elastic deformation of the crust for a uniformly
applied distributed load between 200 km and 300 km. The curves are
subduction zones.
for different thicknesses of crust varying from 10 km to 100 km.
Finally it can be shown that the isostatic compensation
equation (Equation (12.1)) can be derived from Equation
(12.3). If the rigidity of the plate is zero (D ¼ 0), then Figure 12.6 shows the deflection of the crust under a
Equation (12.3) can be written distributed load. The distributed load is assumed to be
over 100 km wide in the x direction. The five different
Pðx; yÞ ¼ qa ¼ ðρmantle ρcrust Þgw (12.6) deflection curves are for a range of crustal thicknesses.
and if the load P is the weight of the crust P ¼ ρcrust gd The nominal thickness is 30 km, with the thinnest thick-
then w ¼ ðρ ρcrust d
ρ Þ which yields Equation (12.1).
ness being 10 km and the thickest 100 km, which covers
mantle crust
the range of crustal thicknesses observed. The elasticity of
the crust is 5 1010 Pa, which is a typical value for
12.2.2.1 One-Dimensional Example: Line Loading uncracked granite. The load is 10 km3/km length with
To demonstrate some of the characteristics of the physics specific gravity of 2.65, which is about the line loading
of Equation (12.3) we can solve some simple one- of the Hawaiian island chain (note: making this example a
dimensional problems. More complex spatially distrib- parallel to Hawaii assumes that the plate is continuous
uted two-dimensional solutions can be found in Turcotte under Hawaii, which recent studies suggest may not be
and Schubert (2002) and elsewhere, but the underlying the case; Klein, 2016). Several features in Figure 12.6 are
concepts and consequences are similar. For one dimen- of note. The first is the existence of close to constant
sion, Equation (12.3) simplifies to deflection under the load. This value of deflection (about
800 m in the figure) is the isostatic compensation depth for
∂4 w 1
¼ ½PðxÞ qa (12.7) this problem. Outside the loaded area there is a small
∂x4 D
upward deflection (about 25 m), which is called the
For the case where the load is independent of x (the forebulge. This forebulge is accompanied by an equal
crust thickness and density is the same everywhere), then and opposite increase in deflection just under the edge of
for an infinite plate (i.e. the boundary conditions are at the load. As the thickness of the crust increases, its rigid-
infinity), then the deflection w is also independent of x, ity increases, the deflection curve is smoother, and the
and P ¼ qa . This is the isostatic equilibrium condition forebulge moves further out from the edge of the loading.
where the buoyancy force qa equals the weight of the Also the absolute maximum deflection is not when the
crust P. Equations (12.3) and (12.7) are linear in deflec- rigidity of the crust is lowest (as might be expected), but at
tion and load, so superposition can be used to separate the a midrange rigidity where the distance to the forebulge is
various loads on the crust, calculate the deflections caused equal to the width of the loading (called the ‘flexural
by each of the loads, and then sum up the deflections for parameter’). Figure 12.7 shows the same loading and
the total deflection. For the examples below we will only crustal thickness cases when the same load is applied as
consider the effect of an additional load over and above a point in x (called a point load), not spread over 100 km.
the weight of the crust and ocean, and consider only the In this case the maximum deflection under the load is
incremental deflection as a result of this additional load. much higher than the distributed case (even though the
FIGURE 12.7: The elastic deformation of the crust for a point FIGURE 12.8: The deflection of the crust with an ice load of
load at 250 km. The total load applied is the same as applied in 1 km thick applied at 250 km and greater. The different curves are
Figure 12.6, only it is applied at a point rather than being distributed different thicknesses of crust varying from 10 km to 100 km, with the
over 100 km. The curves are for different thicknesses of crust nominal thickness of 30 km as for the other examples in this chapter.
varying from 10 km to 100 km.
of the deflection, determined by the load P and the plate

total loading is exactly the same) and as the rigidity
rigidity D. This is consistent with Figures 12.6 and 12.7,
increases the maximum deflection decreases (and is sig-
which show the forebulge moving outwards as the rigidity
nificantly higher than Figure 12.6), but the region over
D increases. The ratio of the height of the forebulge to the
which the deflections occur becomes much larger. As the
deflection under the load is fixed. For typical properties
rigidity increases, the height of the forebulge decreases,
(i.e. the nominal parameter set in Figure 12.7) α 33 and
the distance to the forebulge increases and the maximum
the distance to the forebulge is about 100 km from the
deflection decreases. An analytical solution for a point
point loading.
loading case follows below which explains the results in
While the height of the forebulge is quite small relative
Figure 12.7.
to the primary deflection under the load in situations
Equation (12.7) can be solved analytically, and the
where slopes in the region of the forebulge are very low,
analytic solution yields insights into the form of the
it is still possible for it to be important. Driscoll and
deflections in Figures 12.6 and 12.7. The deflection is
Kerner (1994) suggest that drainage patterns on the coast
(Turcotte and Schubert, 2002)
near the mouth of the Amazon result from the forebulge
Pα3 x=α x x from the loading of the Amazon delta. Coastal streams in
w¼ e cos þ sin the area drain inland rather than to the coast, and, given
8D α α
1=4 (12.8) the low slopes in the area, this is consistent with the
4D
α¼ forebulge forming a barrier to seaward drainage. Thus,
ρmantle g
even though small, it may be important to model this
where P is the line load at location x ¼ 0, and α is the forebulge feedback between deposition and tectonic
‘flexural parameter’ and is the length scale that determines response in landform evolution models.
the scaling in the x direction of the deflections. If α is
larger, then the deflections spread further. The forebulge
12.2.2.2 One-Dimensional Example: Ice Sheet
seen in Figure 12.7 is at location xfb ¼ πα and the height
Boundary
of the forebulge is wfb ¼ 0:0432w0 : The maximum
Another interesting example is that of the deflection of the
deflection that occurs directly under the load (setting x ¼ 0
crust as a result of loading from an ice sheet, and estimating
in Equation (12.8)) is
the deflections near the edge of the ice sheet. Figure 12.8
Pα3 shows the case of 1,000 m of ice on a 10–100 km thick
w0 ¼ (12.9)
8D crust with a Young’s modulus of 5 1010 Pa (the same
Equations (12.8) and (12.9) show that the shape of the parameters used in the previous example). The deflection
deflection is always the same and that the only change is of the crust is 300 m, and the height of the forebulge is
the scale in the x direction, given by α, and the magnitude about 11.2 m (or about 3% of the fully compensated
12.2.3 Time-Varying Processes
The second complicating factor impacting on isostatic

compensation is the rate at which the compensation
occurs. The elastic plate response in Equation (12.3)
above does not occur instantaneously, because for this to
occur the liquid supporting the crust would need to be
inviscid (i.e. zero viscosity) and for it to move instantan-
eously. Field data show a long-term post-earthquake
adjustment of the gravity field to single earthquakes.
These post-event changes are attributed to redistribution
of mantle fluid (Han et al., 2008).
The mantle has very high viscosity (~1020–1021 Pa∙s;
Peltier, 1999; Mitovica and Forte, 2004; Paulson et al.,
2007; Ivins and Wolf, 2008; Klemann et al., 2008) so that
FIGURE 12.9: The net force balance of the downward weight of the rate of adjustment of elevations to changes in crustal
the ice minus the upward buoyancy force, showing the net forces
loading is of the order of thousands of years. For instance,
applied by the elastic deformation of the crust. This is for the same
case as in Figure 12.8. areas of terrestrial crust near the poles are still rising at up
to 20 mm/year in response to the end of the last glacial
deflection under the ice). For a crust that is 100 km thick, max 21,000 years ago (Peltier, 1999; Klemann et al.,
the forebulge is about 60 km in front of the ice front. The 2008). This elevation adjustment varies markedly with
slight additional deflection just behind the front of the ice distance from past glaciation, with locations just outside
sheet is exactly symmetric with the forebulge, as was the the glaciation showing subsidence as the former ice-age
case of the distributed load in Figure 12.6. forebulge declines (e.g. Sella et al., 2007). Figure 12.10
Figure 12.9 shows the vertical force balance for the shows histories of relative sea level for the last 20 kyr for
crust relative to isostatic equilibrium at that location. The three locations in Canada and one in Sweden. In Canada
ice covers the right-hand side of the figure from 250 km to the relative sea level combines the effect of (1) sea level
500 km, and 250 km is the location of the ice front. In the rise due to ice melting and (2) crustal rise due to ice
50 km in front of the ice front (200 km to 250 km) there is removal, but since they are both driven by the same
net upward force, which reflects that the crust is deflected crustal response, they give a good indication of response
but does not have any load applied to it (Figure 12.8), so time (Peltier, 1998, Figure 36). The solid line is an expo-
there is a net buoyancy uplift that is balanced by the nential fit to the field data with response times (i.e. time
downward force of the elastic deflection of the crust. when the response is 1/e the full response, the e-folding
Moving further away from the front into forebulge, there time) of 3.3–4.7 kyr (Table 12.2) with an average of
is a net downward force, which indicates that the crust has 3.7–4 kyr. These numbers agree well with the relaxation
been uplifted by the elastic forces of the crust (i.e. the times derived by direct application of the ICE-4G GIA
forebulge in Figure 12.8). If on the other hand in the model for Angerman River, Sweden, of 4.2 kyr (Peltier,
region immediately underneath and behind the ice front, 1998), 1.7–3.3 (McConnell, 1968; presented in Peltier,
there is a large net downward force, because the crust just 1998) and other independent analyses of Mitrovica and
behind the ice front has not deflected enough to balance Forte (2004).
the ice load and is being held up by the crust immediately The relevance of the exponential time scale fitting is
adjacent and in front of the ice front. Moving further away that it potentially provides an alternative less complex
from the front under the ice, there is a negative equivalent model for viscous impacts on isostatic adjustments than
to the forebulge. Note that the net effect of these force direct use of GIA models. Moreover, there is considerable
imbalances is to create a clockwise moment, which results uncertainty in the parameterisation of GIA models, spe-
in the crust rotating clockwise at the ice front, which is cifically mantle physics (e.g. the type of viscosity
consistent with the deflections in Figure 12.8. In (Maxwell versus Newtonian), and the pattern of viscosity
Figure 12.9 for a vertical force balance the area above both with depth and globally within the mantle, and the
the zero line must exactly equal the area below, which is absolute values of mantle viscosity), and geodynamics
ensured by the symmetry in this example, and this is researchers have found that the parameters for the viscos-
consistent with the symmetric forebulges in Figure 12.8. ity parameterisations in GIA models are poorly
TABLE 12.2 The e-folding response times for GIA
Site Response time (kyr)
Peltier (1999) Peltier (2004) Mitrovica and Forte (2004)
Richmond Gulf, Canada 4.7 - 5.3 1.3

James Bay, Canada 3.3 3.4 2.4 0.4
Hudson Bay, Canada 3.9 3.4 -
Angerman River, Sweden 4.7 4.4 4.9 0.9
Average 4.1 (4.0 excl. Richmond) 3.7 4.2
FIGURE 12.10: Eustatic sea level

rise for four sites in Canada
showing the initially fast response
from the ice unloading, which
gradually decreases, approximately
exponentially, with time (from
Peltier, 1999). The time axis is
years before present.
identifiable (e.g. Peltier, 1998; Paulson et al., 2007). where zðt Þ is the amount of isostatic elevation adjustment
Finally, loads that cover a larger area mobilise deeper at a given time t after the load is applied to the crust, z∞ is
mantle material during the adjustment and have a different the isostatic elevation adjustment that occurs at equilib-
response time to more localised loads (e.g. Figures 2 and rium (or, equivalently, would occur instantly if the mantle
4 in Peltier, 2004). Thus even with state-of-the-art GIA was inviscid) and tr is the response time of the isostatic
models there are uncertainties about the physical mechan- response, discussed above. This response time is likely to
isms driving mantle viscosity, and the inferred isostatic vary only slightly globally because it is primarily driven
response times vary by a factor of 2 or more. Conse- by mantle viscosity and does not interact with the ocean
quently, in the absence of better information (or a lack and continental crust thickness so that it is independent of
of desire to couple the landform evolution model directly the elastic plate spatial response (Equation (12.3)).
to a GIA model), an alternative and conceptually simpler Accordingly, the response time t r and the equilibrium
approach would seem to be to assume an exponentially isostatic adjustment z∞ are independent. Comparing a
declining isostatic adjustment response with a specified coupled lithosphere-mantle model with uniform litho-
response time spheric thickness with one with realistic global variation
in thickness, Zhong et al. (2003) showed that the response
zðt Þ ¼ z∞ eð =tr Þ
t
(12.10) time t r to glacial rebound is relatively insensitive to
12.3 Mountain Building 199
lithospheric thickness if mantle viscosity is unchanged temperature increases with depth. This is the realm of
(though the magnitude of rebound z∞ was sensitive geodynamics and will not be discussed further here.
because of the changing stiffness of the elastic plate). The subject of this section is what happens with the
Moreover, while Peltier (1998, 1999) does not directly material that does not get subducted down into the mantle.
address the question of global variations in response Figure 12.11 shows a conceptualisation of two end-
times, his data, which include sites in both the northern member cases where the converging material piles up to
and southern hemispheres and sites inside and outside create a mountain range. In this conceptualisation the left-
glaciation in the last glacial maximum, are consistent with hand crust is moving toward a stationary crust on the
there being little variation (a maximum of a factor of 2) in right-hand side. A fixed proportion of the crust is sub-
response times globally. Thus to first order a starting point ducted and forms the root (not shown in the figure), and
for the response time in Equation (12.10) is the response the remainder piles up on the surface. It is how this
time in Table 12.2. material piles up that defines the differences between the
The final complicating factor is that horizontal flow of models. The main method is critical wedge/taper theory
mantle material is required to move mantle material from (Dahlen, 1990), where the material is assumed to be a
areas that are subsiding to areas that are uplifting. This Coulomb material where movement is instantaneous
flow induces friction stress on the bottom of the crustal when stresses exceed the yield stress (e.g. a granular
plates and results in horizontal motion in the plates. This material like sand).
motion is in addition to any mantle plume–induced Willet et al. (2001) summarise the concepts underpin-
motions (James and Morgan, 1990). Klemann et al. ning the competing approaches:
(2008) and Argus and Peltier (2010) suggest rates of
GIA-induced horizontal movement of the order of 1 mm/ • Frontal accretion is simply the piling up of material
year and that this rate of movement is typically less than against a resisting wall, which is the stationary crust on
10% of the total observed horizontal motion of plates the right-hand side of Figure 12.11. Over time as new
globally. Argus and Peltier (2010) highlight the difficulty material is accreted at the mountain range front, two
in using GPS data in GIA models to infer horizontal things occur: (1) the mountains get higher and steeper
movement rates. This suggests that, given the parameter as more material is contained within the horizontal
uncertainty bounds in the other parts of the GIA model, width of the mountain range (L in the x direction in
these induced horizontal flows are second order and can Figure 12.11) and (2) the width of the range gets wider
be ignored. so that L increases and the range front moves to the
left.
◦ In case (1) any given vertical cross section through
12.3 Mountain Building the mountain range (the vertical cross sections in the
figure) over time is (a) compressed (i.e. becomes
The previous section talked about the movement of the thinner in the x direction) horizontally, (b) to maintain
crust under loading but didn’t address the question of mass continuity the same cross section gets higher
evolution of the crust itself. The main topic of interest is and (c) the cross section itself moves from left
mountain building as a result of convergence of two to right.
sections (typically two plates) of crust toward each other ◦ In case (2) and if the prevailing surface slopes are
(Figure 12.11). As a result of continuity of mass the crust unchanged, then the (a) left-hand side of the moun-
must thicken and mountains are formed. Isostasy is still tain surface will rise parallel to existing surface, (b)
active, but because of the dynamics of the crustal the mountain divide will rise and move to the left and
thickening the crust may not be equilibrated. As a result (c) the point L will move to the left. As in case (1) the
of isostasy there is a thickening of the crust under the vertical cross section will be compressed, higher and
mountains (called a ‘root’), but the dynamics of this moved to the right, but the effect will be less dramatic
process cannot be described by the methods in the previ- than in case (1).
ous section. ◦ The dotted lines with arrows show typical paths for a
The main issue is determining the dynamics of the particle of crust. Once it reaches the surface, it is
crustal thickening that occurs at the collision point. For removed by erosion. At the land surface of the moun-
the case where one plate subducts under another the tain range the surface velocity has both horizontal and
motion of the underlying plate is normally modelled with vertical components, with the surface moving from
an elastic-plastic model where the plate softens as the left to right, as well as upwards.
FIGURE 12.11: The two mountain range building conceptualisations of Willet et al. (2001). (a) The initial condition. The dots are labelled
points, and the columns finite volumes used to show movements in the bottom two panels. The heavy arrow on the left-hand side is the
direction of movement of the crust. (b) Frontal accretion, where material is pushed from the left into mountain range equally from the top to
bottom of the crust (akin to a bulldozer blade). The black arrows show the movement of the labelled points from (a) to (b), and the light dotted
lines show trajectories from beginning to end of rocks passing through each labelled point. The light grey arrows at the surface show the
direction of the movement at the surface (note that for dynamic equilibrium where erosion equals uplift in the vertical direction, then the
vertical component of these grey arrows are all equal). Column width x 2 is less than x1 because the column is compressed in the horizontal
direction and the ratio x 2 : x 1 can be calculated from volume continuity. (c) Underplating, where the material being pushed in from the left is
preferentially pushed underneath the mountain range. All the arrows have the same meaning as in (b). Note that the columns are not
compressed in the horizontal direction, so the surface movements do not have a horizontal component.
• Underplating is conceptually simpler than frontal accre- horizontal motion. The dotted lines with arrows show
tion and is the case where the crust moves under the typical paths, and what is important is that the particle
right-hand side and simply lifts the existing mountain moves horizontally in the underplating region before it
wedge upwards. In this case the motion of the material finally moves vertically. The different paths for frontal
as it reaches the surface is vertical and the motion of the accretion and underplating may have an impact of ther-
surface of the mountain range is solely vertical with no mochronology dating techniques because the cooling
12.4 Extensional Settings 201
profile of the rock will be different from frontal accre- 12.4 Extensional Settings
tion where the pathway is always moving closer to the
surface. The discussion of the previous section considered settings
• The shape of wedges in the cross section is a direct where the crust was in compression. In this section we
result of the flat base below the horizontal crustal move- will consider settings in tension. The main way that ten-
ment and the assumption of a Coulomb material. If the sional strains can be accommodated is through faulting.
base is not flat, then the geometry is more complicated. Figure 12.12 shows how a single extensional fault and the
The general form of the wedges is also well supported crust interact. The subsidence and erosion in the region of
by sand box experiments, sand also being a Coulomb the fault are a result of the stiffness of the elastic plate,
material. which can be modelled with the thin plate techniques
previously discussed, and far from the fault, elevations
Note that in both cases there was no interaction are unchanged from that predicted by isostasy. The failure
between crust and the mantle, so isostatic buoyancy pro- surface is predicted by the Mohr-Coulomb failure surface
cesses play no part in this conceptualisation. Willet et al. assuming that the material is a Coulomb material (Olive
(2001) coupled both models to a landform evolution et al., 2014).
model to identify unique geomorphic signatures to these When looking at a large area under tension, it has been
processes so that they can be distinguished in the field. found that thick crust tends to extend with a number
Their main finding was the resulting mountain ranges of smaller offset faults, while thinner crust tends to
created were symmetric for underplating, but asymmetric break with a lesser number of higher offset faults (Buck,
for frontal accretion. In the latter case the divide of the 1993; Lavier et al., 2000; Lavier and Buck, 2002)
mountain range was pushed from left to right, so the cross (Figure 12.12).
section conceptually looked like their schematic for the The angle of the fault is determined by the requirement
process (i.e. Figure 12.11b). that the two sides of the fault do not separate but slide
Whipple and Meade (2004) explored the effect of the relative to each other. The failure criterion is that the
coupling between erosion and accretion and allowed the frictional shear strength of the failure plane per unit area
width of the orogen to adjust to these processes. They in the horizontal direction (which is a function of the
found for steady state that an erosion increase (e.g. by normal stress perpendicular to the fault) times the area
climate changes) reduced the width of the orogen and of the fault is equal to the tension force in the crust
increased the uplift rate. This increased uplift rate is (Figure 12.12). The angle that results in this strength
required by the balance, at steady state, between erosion being least is the angle of the fault. The shear strength
and uplift (see the discussions about slope-area analysis in of any surface is
Chapter 3). A consequence of this balance between ero- τ ¼ μσ n þ τ c (12.11)
sion and uplift is that the rock uplift rate is relatively
insensitive to the convergence rate and is most sensitive where μ is the coefficient of friction of the fault surface, σ n
to the erosion rate. Roe et al. (2008) extended the under- is the stress perpendicular to the fault surface and τ c is the
plating approach by postulating a precipitation variation cohesive shear strength of the fault surface. For a friction
with elevation, and leeward and windward sides of the coefficient of μ ¼ 0:6 the optimal angle to the horizontal
range. They performed a scaling analysis to show the of the fault is about 60 (Buck, 1993). Whether the fault
impact of the precipitation feedback on the equilibrium rises to the left (as shown in Figure 12.12a and b) or to the
height, width and shape of the mountain range. Recent right is a result of random chance and is likely driven by
work has focussed on collecting experimental evidence minor random variations in the rock mass that bias the
for this erosion driver (e.g. Whipple, 2009). fault in one direction or the other.
Roe and Brandon (2011) extended this work by It should be clear from Figure 12.12 that the faulting
looking at the effect of different forms of rheology leads to higher elevation on one side of the crust. Poten-
governing the way the crust flows in the convergence tially this could form a mountain range. Buck (1993)
zone where the mountain range is being formed. They calculated that the upwarping at the fault exposure could
focussed on the relationship of width to height of the be as much as 1–2 km. Sachau et al. (2013) propose an
range, and found the range characteristics to be relatively additional way that faulting may result in mountain build-
insensitive to the type of rheology used. This is consistent ing. If the faulting does not fully penetrate the crust from
with erosion, rather than convergence dynamics, being the top to bottom but only penetrates the top part of the crust
primary driver of orogen characteristics. while the hotter, softer, underlying crust undergoes
FIGURE 12.12: The form of elevation change resulting from tensional faulting in the crust as a function of thickness of the crust: (a) crustal
thickness 10 km, (b) crustal thickness 15 km, (c) crustal thickness 30 km. (a) and (b) generate a single fault, while (c) for the thicker crust
generates multiple faults and complex topography (from Lavier and Buck, 2002).
viscous flow, then the crustal plate will upwarp because of geologic time, then there must be a process that is rejuven-
the parasitic moments induced in the crust by the tensile ating crust. For landscape evolution studies of less than a
load that is no longer being applied through the centroid few million years this process is unlikely to be an import-
of the crust. Sachau proposes this as an explanation for ant variable to model, but over longer time scales it is a
crustal warping of 2,500 m and a gravity anomaly in the consideration, though one that is poorly understood
Alvertine rift system in Africa. (O’Reilly et al., 2001; Bishop, 2007). A number of pro-
cesses are discussed in the literature, and most evidence
comes from geochemical analysis of rocks brought to the
12.5 Renewal of Terrestrial Crust surface by igneous process, or by seismic studies. The
possible processes include the following:
If the terrestrial crust is always eroding and this eroded
sediment is deposited in the ocean, an important question 1. Buoyancy of mantle materials. Lighter melted rocks
is how the continents still exist. Perhaps they should have will float to the surface of the mantle as a part of a
all eroded away by now. Portenga et al. (2011) estimated natural differentiation of materials. The specific gravity
global erosion rates using 10Be. For drainage basins he differences required for this process may be due to
estimated the mean erosion rate to be 218 m per million temperature differences or differences in geochemistry.
years. This suggests that a 30-km-thick continental crust However, if it is geochemistry, then there must be
will be eroded completely away in just 137 million years, another parallel process that ensures that the lighter
which is significantly less than the measured age of materials collect preferentially under the continental
continental rocks. plates, otherwise there must be another process that
If erosion is thinning the continental crust and the rate maintains a difference in the densities between the
of thinning is high enough to destroy the crust within mantle underlying oceanic and continental plates.

how a triangulation deforms
across a strike-slip fault (a) the
initial triangulation, (b) the
deformed triangulation and (c) the
retriangulated nodes. The dotted
line shows the strike-slip fault. The
arrows in (b) show the direction of
movement on each side of
the fault.
2. Heat flow. If there is a heat flow balance in the crust, case a TIN discretisation has significant advantages over a
then the solidification of mantle rocks always occurs grid because at each time step the nodes at which the
when the crust reduces below an equilibrium thick- elevations are discretised can be moved with the lateral
ness. As material erodes on the surface an equal movement. For instance, points either side of a strike-slip
amount of mantle material solidifies on the base of fault can be moved relative to each other, and any features
the continental crust. This will be the lighter mantle (e.g. catchment divides) will be kept intact (Figure 12.13).
rocks as they will have floated to the mantle surface. If this strike-slip behaviour is modelled with a fixed grid,
This will not occur under the oceanic crust because then at each step there will be a smoothing of the terrain as
material is not being eroded, so the insulating blanket the divide moves between grid points and is interpolated
of the lithosphere is not being thinned under the ocean. between the grid points and smoothed as a result. The
In fact, under the ocean with the deposition of eroded reader might then legitimately ask why TINs are not
sediment, then the lithosphere will be thickening, and used more frequently in landform evolution models. The
material may melt and detach from the oceanic reason is that TINs add complexity in a number of differ-
crust base. ent ways, which may be unnecessary if lateral movement
3. Preferential recycling of crustal material. O’Reilly is not significant in the problem being studied.
et al. (2001) propose a mechanism of recycling where Specifically
continental lithosphere is delaminated from the litho-
1. The representation of the landform in TINs requires
sphere and recycled back into the mantle, and then
not just the easting and northing locations of the nodes
back into the lithosphere. Recent geochemical studies
in the DEM, but the geometry of each of the triangles
appear to support this hypothesis (Kelemen and Behn,
joining these nodes (called the triangulation), and the
2016).
geometry of how each of the triangles is joined to its
Along with many other questions about the behaviour adjacent triangles. As a result the calculation of flows
of the mantle the mechanisms driving crustal rejuvenation across a TIN is more complicated than across a grid.
appear to be unresolved. 2. How the triangulation changes from timestep to time-
step requires a degree of care, otherwise mass balance
12.6 Numerical Issues errors may enter the calculations. Normally the tri-
angulation is done using a Delaunay triangulation
The main feature of tectonic problems compared with algorithm (e.g. Sloan, 1987), which ensures that the
other processes in this book is the possibility for features triangles are close to equilateral in shape, which is
to move laterally (e.g. strike-slip faults, convergence, optimal for numerical reasons. The area associated
spreading) rather than movement being predominantly with each node is then determined using a Voronoi
up and down. In this case a discretisation of the elevations tessellation (e.g. Tucker et al., 2001b). When the tri-
that can move with these features is desirable to ensure angles are long and skinny, which over time is inevit-
that they are kept intact under lateral movement. In this able for triangles across a strike-slip, then periodic
retriangulation is required (Figure 12.13). But if the modelling (i.e. centennial to millennial time scales)
nodes move relative to each other and the nodes are requires one of the GIA models that simulate the viscous
simply retriangulated, then there is no guarantee that flow of the mantle.
the mass balance of the landform is maintained after Wickert (2015) provides a set of open source tools in
retriangulation. Python, gFlex, for solving elastic plate flexure problem
3. There is additional bookkeeping involved with the with isostasy. It uses the finite difference approximation to
triangles and retriangulation. If the triangle’s plan area the flexure equation (Equation (12.3)) and allows the
changes area, this indicates lateral compression of the crustal rigidity to change laterally. The codes can be used
crust, so that the height of the elevations at the triangle standalone or integrated into the GRASS GIS package.
corners must be increased to maintain the same mass
within the triangle. But this will also change the mass
in adjacent triangles because changing the elevation at 12.7 Further Reading
a corner changes the triangles that also have this node
For a review of the research questions around the impli-
as a corner. The Voronoi tessellation gives the area
cations of the coupling of tectonics, erosion and climate
associated with each node, and the expansion and
the reader is referred to Bishop (2007). The classic text
contraction of the nodal area in the Voronoi tessella-
discussing the details of isostasy is Watts (2001). For
tion that occurs with node movement can be used to do
greater depth on the elastic plate modelling of the equilib-
this efficiently.
rium crust-mantle interaction see Turcotte and Schubert
4. If new nodes are added, or old nodes removed, than
(2002). For the dynamics of the crust-mantle interaction
mass balance checks need to be done before and after
the reader is best served by reading some of the work on
the addition/removal.
glacial isostatic rebound modelling or GIA modelling.
None of these issues are insurmountable but they do The details of these GIA models can be quite complex,
add to the complexity of the computer code. Accordingly but the modelling is also an area of active interest outside
many computer codes are gridded, which is less flexible, of earth sciences because of its time-varying impacts on
but these bookkeeping issues do not arise. satellite trajectories and in the field of geodesy (e.g. Han
The REAR model (Melini et al., 2015) calculates the et al., 2008). Accordingly it is difficult to identify a single
elastic response of the crust and the mantle to time- and publication that neatly summaries the state-of-the-art that
space-varying ice load. This model does not model the won’t be out of date very quickly, but Lambert and
viscous response of the mantle but only the short-term Johnston (1998) provide a readable overview of the issues
elastic response of the mantle. This elastic response has of GIA model formulation and impacts of glacial rebound.
not been discussed in this chapter. Spada et al. (2012) Gerya (2010) has a specialist discussion of mantle
used this approach to examine the tectonic effect of rheology and how to model various aspects of mantle
changes in ice thickness over the last decade in Greenland geodynamics, and significantly extends the discussions
as measured by satellite instruments. To do longer term here on mantle dynamics.
13 High-Slope Gravity Processes
This chapter is about processes that are primarily driven them ill-suited for long-term evolution modelling
by gravity. These processes are typically also only active approaches, and they are better addressed with site-
in steeper regions because they involve some form of specific models that explicitly account for local geologic
strength failure in the hillslope. These processes are dom- effects. Rather the focus here will be on shallow mass
inant on the steep hillslopes in the upper reaches of movements, which tend to be driven by processes in the
catchments. They may also be active in regions of the top few metres of the soil-saprolite profile, and which are
landscape’s lowlands if localised geology is such that largely contained within the surface soil. It is thus unsur-
cliffs and scree slopes occur. prising that the soil-saprolite interface figures heavily in
There is some overlap with previous chapters, particu- the sections that follow. As noted in Chapter 6 a number
larly the chapter on soil flow and creep (Chapter 9) where of processes and properties change either abruptly or very
slow-moving processes without soil or saprolite failure rapidly at this interface (e.g. porosity, permeability,
were discussed. In this chapter the distinguishing features mobility) so that the soil-saprolite interface tends to be
of the processes are (1) the process begins only after either a plane of weakness, or a plane either side of which
failure of the strength of the material, the trigger mechan- there are abrupt changes in transport and hydrology
ism and (2) the subsequent transport of the material is processes.
fast. These two features are linked because if there is a
strength failure, then immediately after the failure there is
a significant force imbalance (typically the force con- 13.1 Trigger Criteria
straining the movement has disappeared as a result of
In this section we will discuss the mechanisms that start
the failure) and the soil/rock mass is rapidly accelerated
gravity-driven mass transport. The components of this are
by this force imbalance. While these trigger and transport
the processes that trigger the slope failure, and that deter-
processes are conceptually linked it is convenient below
mine how big the resulting slope failure will be.
to treat them separately. In contrast the processes dis-
cussed in Chapter 9 do not involve a failure trigger mech-
anism, and consequently the dramatic and sudden force 13.1.1 Empirical Relationships
imbalance does not exist so that transport rates are much
lower and less abrupt. Before we discuss physically based models of landslide
This chapter is organised into main parts. The first part triggering we will look at empirical relationships that have
discusses mechanisms that trigger failure of the hillslope been observed in the field or derived from remote sensing
(most of the processes in this chapter start on hillslopes studies. The typical methodology for these studies is that
rather than in the channels). The second part discusses the there has been some macroscale trigger process such as
movement of this material downslope (both on the hill- heavy rain, earthquake or snowmelt that has triggered a
slope and in the valleys/channels) once this failure has large number of microscale landslide events in the region
occurred. affected by that macroscale event. These studies provide
This chapter will not concern itself with deep-seated, insight into what has been observed in the field. More
multi-cubic-kilometre, landslides, which tend to be driven importantly they may also provide a simple basis for a
by very site-specific geology. This site-specificity makes stochastic model of landsliding where landslides occur
205
examined landslides in Taiwan and New Zealand, and,

while they fitted a double Pareto distribution on the entire
dataset, for large areas they also found a power law
relationship with landslide area with an exponent of
2.11 for Taiwan and 2.44 to 2.48 for New Zealand.
Using Equation (13.1) the average area of landslides
AL is
a
AL ¼ þs (13.3)
ρ1
These relationships give the distribution of landslide

areas for specific macroscale trigger events, but they lack
information about (1) how often the macroscale trigger
events occur, (2) the relationship for the absolute number
of landslides as a function of the magnitude of the macro-
scale trigger event and (3) the volume of sediment moved
given a landslide area. We will return to these power law
relationships with area in Section 13.2.2.
FIGURE 13.1: Dependence of landslide probability density function Turning now to the volume of the landslide, if the land-
p for landslide area AL (from Malamud et al., 2004).
slide removes all soil down to the soil-saprolite interface,
then the volume moved is V L ¼ DAL where D is the soil
depth. However, studies have shown that the volume scales
randomly in space and with random mass, where the nonlinearly with area, so that events with larger areas
landslide properties can be simulated using Monte Carlo remove a greater depth of material. Simonett (1967) and
sampling from these empirically derived probability Hovius et al. (1997) found relationships of the form
distributions.
V L ¼ εAL γ (13.4).
For instance, Stark and Hovius (2001) and Malamud
et al. (2004) examined landslide area data for a number of For landslides in Papua New Guinea Simonett (1967)
macroscale events worldwide. Malamud et al. (2004) fitted ε ¼ 0:024 and γ ¼ 1:37 while for landslides in
examined published studies of landslide trigger rates New Zealand and Taiwan Hovius et al. (1997) fitted
(probability per unit area) and the volume of the landslide ε ¼ 0:05 0:02 and γ ¼ 1:5. Larsen et al. (2010)
that resulted for three locations affected by three different report ε ¼ 0:15 and γ ¼ 1:11:3 for their entire
macroscale trigger processes (rain, snowmelt and earth- data set, but when they separated soil and bedrock
quake). Malamud found that the microscale data for three landslides they found γsoil ¼ 1:145 0:008 and
macroscale events were well fitted by a three-parameter γrock ¼ 1:35 0:01. The value near 1.0 for the soil
inverse gamma distribution (Figure 13.1) landslides is consistent with the idea that their depth is
ρþ1 independent of the area of the landslide, and that all soil
1 a a
down to a threshold layer (i.e. soil-saprolite interface) is
pðAL Þ ¼ e AL s (13.1)
aΓð ρÞ AL s removed regardless of the area of the landslide. For
instance, for a landslide area of 1 ha Simonett’s param-
where Γð ρÞ is the gamma function of ρ with parameters eters give an average depth of the landslide of 4.3 m,
ρ ¼ 1:4, a ¼ 1:28 102 km2 , and s ¼ 1:32 104 km2 . while Hovius’s parameters give 5 m. These depths are
For large areas this is well approximated by both larger than typical hillslope soil depths suggesting
ρþ1 that large landslides (see Figure 13.1) will remove some
1 a
pðAL Þ ¼ (13.2) of the saprolite as well as the soil, while smaller landslides
aΓð ρÞ AL
(e.g. an area of 0.01 ha yields average depth of 0.8 m and
Both of these equations are for large areas of more than a 0.5 m for Simonett and Hovius, respectively) might be
few hectares. Equation (13.2) shows that as the landslide confined within the soil. Larsen et al. also noted that the
area AL increases the likelihood of a landslide occurrence depths of their small bedrock landslides tended to be
decreases with a power law with exponent ð ρ þ 1Þ and indistinguishable from soil landslides, both being con-
ρ þ 1 ¼ 2:4 for Malamud’s data. Stark and Hovius (2001) strained by the soil availability.
FIGURE 13.2: Dependence of

landslide probability density
function p for landslide volume V L
(from Brunetti et al., 2009).
Equations (13.2) and (13.4) can be combined to create data together with a number of other published studies of
a relationship for the probability distribution of landslide similarly smallish landslide scars and found that the mean
volumes of the length:width ratio was about 1–2, but with a stand-
" ρþ1 # ard deviation in log-space of about 2–3 around that mean.
aρþ1 1 ð γ Þ They also found, by plotting depth against area, that γ in
V L ð γ Þ
ρþ1
pðV L Þ ¼ (13.5)
aΓðρÞ ε Equation (13.4) was about 1.4, consistent with the studies
of the larger scars discussed above.
which for Hovius’s parameters yields an exponent on the
volume V L of 1.6.
Brunetti et al. (2009) examined 19 datasets of land- 13.1.2 The Physically Based Infinite Slope Model
slide volumes to determine a probability distribution for
landslide volume and found (Figure 13.2) The most commonly used physically based criterion for
β shallow slope failure has been the ‘infinite slope’ model.
pðV L Þ / V L (13.6)
This model examines the stability of the soil for a hill-
where β ¼ 1:1 1:4 for rockfalls and rock slides, and slope that is infinite in both downslope and across slope
β ¼ 1:5 1:9 for soil slides, so that the scalings with directions, and where the hillslope is planar with no diver-
volume in Equations (13.5) and (13.6) are approximately gence or convergence (i.e. no ridges or valleys). The soil
consistent even though they were derived from different is assumed to be uniform in strength and depth every-
datasets. where. The slope is assumed to fail everywhere at the
Warburton et al. (2008) mapped widths and lengths of same time, so there are no lateral boundary effects influ-
landslide scars (their maximum size of landslide scar encing the failure scar. The rationale for ignoring the
about 1 ha, so their landslides were much smaller than lateral boundary, finite size, effect is that the area associ-
the studies discussed above) and found that length ated with vertical edges of a landslide is much smaller
increased with width but there was significant scatter than the basal area. For instance, for the example in the
around this trend. Milledge et al. (2014) used these same previous section, of a 1 ha scar with a depth of 5 m, the
cohesive forces are the cohesive strength of clays and the

existence of vegetation roots that penetrate through the
failure surface (e.g. Gorsevski et al., 2006; Meisina and
Scarabelli, 2007).
The frictional shear force is a function of the weight of
the overlying soil and the roughness of the interface
between the soil and the saprolite, is active both pre-
and post-failure of the soil, and is expressed as
T f ¼ τA ¼ μσA ¼ μW n (13.7)
where T f and τ are the shear force and shear stress acting
at the failure surface, respectively, A is the area of failure
surface that the shear stress is acting on, W n and σ are the
components of the vertical force and stress acting normal
FIGURE 13.3: Schematic of the force balance for a control volume to the failure surface and μ is the friction coefficient. Pre-
of failed soil on a hillslope where the cross-slope direction is
considered infinite, where the failure surface occurs within the dry
failure (when the soil is stationary) μ ¼ μs , the static
soil profile. friction coefficient, while post-failure (when the soil has
started moving relative to the underlying soil/saprolite)
μ ¼ μd , the dynamic or kinematic friction coefficient. For
area of the vertical edges for a square landslide is 20% of solid to solid sliding (e.g. rock fragment sliding against a
the basal area, so to first order the edge area is small. rock surface) it is widely accepted that the dynamic coef-
Including the edge effects will be discussed in the next ficient is less than the static friction coefficient (Meriam
section when we extend this model. et al., 2012). For granular material this also appears to be
Ignoring edge effects allows us to examine the force true (Géminard et al., 1999; Géminard and Losert, 2002)
balance stability of a small element of the hillslope, inde- though there are fewer experimental data for granular
pendently of surrounding elements (Figure 13.3). Note material. Thus once movement starts, the frictional shear
that in the figure we have drawn a failure plane that does force resisting movement is reduced, and Geminard and
not necessarily coincide with the soil-saprolite interface, co-authors found this reduction to be of the order of 20%.
so that the discussion below is also valid if the slope fails Post-failure only the frictional force is active.
along a plane that is contained within the soil (for Summing the pre-failure forces in the direction parallel
instance, as discussed in the previous section for the depth to the slope, positive in the downslope direction
of landslide scars where the equations can predict a depth X
less than the depth of the soil). F ¼ W sin θ T f T c
(13.8)
There are three main forces acting on the element. ¼ ρb gzΔx sin θ μs ρb gzΔx cos θ τ c Δx
They are the weight force W acting vertically downwards, where ρb is the bulk density of the overlying soil, z is the
and two shear forces at the failure plane acting to resist depth of the failure surface below the soil surface and τ c is
movement, collectively labelled T. The first resisting force the cohesive shear stress strength of the soil at the failure
T f is a frictional resistance force that is a function of the surface (the shear strength of clays is normally expressed
overburden pressure on the failure plane, and the second as a shear stress, while the shear strength due to roots will
T c is the cohesive shear strength of the soil at the failure be a function of the number of roots and/or the total area
plane surface and T ¼ T f þ T c . There are also two pres- P
of roots per unit area of the failure surface). If F is
sure forces at the upslope and downslope ends of the greater than zero then the slope fails because the destabil-
element, Pu and Pd , respectively, but they balance ising forces (the weight) exceeds the resisting forces. This
(because the soil and the failure plane is assumed the equation is commonly expressed as a factor of safety,
same in every direction) and so can be ignored until we FOS,
consider boundary effects in the next section.
μs ρb gz cos θ þ τ c
The cohesive shear strength force is active until it is FOS ¼ (13.9)
exceeded and then decreases to zero post-failure. It can be ρb gz sin θ
considered to be a force resulting from a ‘glue’ between where the destabilising forces are in the denominator and
the overlying soil and the underlying soil/saprolite. Once the stabilising forces are in the numerator. If FOS > 1
the glue is broken it is no longer active. The two main then the slope is stable, while if FOS < 1 then the slope
(b)
FIGURE 13.4: Schematic of the force balance for a control volume of failed soil, where the failure surface occurs within a soil profile with
a water table above the failure surface so that some of the failed soil is saturated with water: (a) the force balance, (b) the equipotentials
of the water table down the slope showing the pressure distribution down the slope.
fails. One end member is when the cohesive strength is everywhere, because once the soil is thicker than D∗
zero and the entire profile will be removed by landsliding
because the plane of failure is the soil-saprolite
μs cos θ μ
FOS ¼ ¼ s (13.10) interface, resetting the soil depth to zero at the failure
sin θ tan θ
location. Vanwalleghem et al. (2013) presents a simula-
so that at the failure threshold, when FOS ¼ 1, tion model that coupled a pedogenesis model with sto-
tan θ ¼ μs , where θ ¼ θr is the angle of repose and the chastic landslides and found that this combination of
internal friction angle for a granular material (i.e. a mater- mechanisms acted to limit the maximum depth of the
ial with no cohesive strength). For nonzero cohesion soil, and gave a better prediction of landslide initiation
τc locations than assuming a constant depth soil. Mont-
μs þ
ρb gz cos θ gomery and Dietrich (1994) found that for one steep
FOS ¼ (13.11)
tan θ study site that they classified as unconditionally
unstable (i.e. tan θ > μs and τ c ¼ 0) the site was dom-
which shows more explicitly that cohesion always
inated by exposed rock, consistent with Equation
improves the stability of the slope, but that this stabilisa-
(13.12), and the soil was being stripped off by land-
tion effect is reduced as z increases, so that if τ c is constant
sliding before being able to accumulate.
with depth, then the plane most at risk of failure is the
The main limitation of the infinite slope model as
deepest plane in the soil, the soil-saprolite interface. In
described above is that it is incapable of modelling hill-
fact, even if the cohesion strength increases with depth, if
slope response to a macroevent. In the model above, a
the cohesive strength varies as τ c zφ where φ < 1 then
hillslope is either stable or not, and the model cannot
the soil-saprolite interface (z ¼ D, where D is the depth
simulate time-varying instability and FOS (e.g. landslides
of the soil) is the weakest failure plane.
after a heavy rainfall event).
Finally in the situation where the soil depth is increas-
The main extension to the model is to simulate the
ing with time due to pedogenesis (see, for instance, the
existence of a time-varying water table within the soil.
soilscape evolution chapters), then for some depth D∗ the
Figure 13.4 extends Figure 13.3 to include a water table at
soil-saprolite interface will become unstable (provided
depth zw below the surface. As previously there are pres-
only that the hillslope slope is greater than the internal
sure forces at the upslope and downslope ends, Pu and Pd ,
friction angle), and from Equation (13.11)
respectively, which are now combinations of soil and
τc
D∗ ¼ (13.12) water pressures, but they cancel out as before. We also
ρb g cos θð tan θ μs Þ
note that the bulk density of the dry soil is ρb ¼ ρs ð1nÞ
This suggests a mechanism whereby a dynamic balance where ρs is the density of the soil particles and n is the
can develop between soil thickening by pedogenesis porosity of the soil. The total weight of the soil (noting
and soil thinning by landsliding. Note that this result that below the water table the pores in the soil will be
does not mean that the depth of the soil will be D∗ filled with water) is
W ¼ ρs ð1 nÞgzΔx þ ρw ngðz zw ÞΔx (13.13) this effect being the extra weight of the soil after it is
wet. Any reduction in the cohesive shear strength of the
where ρw is the density of water. However, the weight that
clays in the soil due to the increased water content is in
is effective on the failure plane to generate friction (the
addition to these changes. Thus we can see that a water
movement of the soil particles above the failure against
table in the soil always reduces the stability of the soil
the soil particles below the failure plane) is the weight of
relative to a dry soil. There are several slightly different
the wet soil minus the buoyancy effect, which is the
formulations of Equations (13.16) and (13.17), but they
vertical upwards force due to the pore pressures at the
differ only in the assumptions of what the pore pressure
failure plane. Figure 13.4b shows a flow net for the flow
distribution is at the failure surface, which depends on the
down the hillslope in the ground water. In particular it
assumed groundwater flow pattern (Iverson, 1990), and,
shows the equipotentials at right angles to the flow direc-
in any event, pore pressures are difficult to predict in the
tion. Along an equipotential the energy of the flow is the
field (Montgomery et al., 2009).
same so that ρpg þ z ¼ constant: Since the pressure is zero
w Montgomery and Dietrich (1994) called hillslopes that
at the groundwater table the pore pressure at the failure
were always unstable even when dry as ‘unconditionally
plane is p ¼ ρw gðz zw Þ cos θ. For those with an interest
unstable’, and those that are stable even when fully satur-
in fluid mechanics this is not the pressure resulting from
ated as ‘unconditionally stable’.
the hydrostatic pressure distribution from A to B (which
The final observation about the effect of the water
would be p ¼ ρw gðz zw Þ= cos θ) in Figure 13.4b. The
table is that this reduction in the FOS is a function of
difference is because there is a vertical component to the
how fast the water table varies. As we noted in Chapter 3
velocity from A to B, violating the hydrostatic assump-
the soil water table for saturation excess hydrology
tion. Thus the effective weight to determine the frictional
(Section 3.1.2.1) tends to vary slowly, yet it is common
resistance is
to see heavy rainfall trigger landslides very quickly. Sev-
W eff ¼ ½ðρs ρw Þð1 nÞz þ ρw ð1 nÞzw gΔx cos θ eral explanations for this apparent contradiction have been
(13.14). proposed including the following:
The force balance is then (1) Preferential flow paths (e.g. macropores) so that the
X soil-saprolite interface becomes very quickly satur-
F ¼ ðρs ð1 nÞz þ ρw nðz zw ÞÞgΔx sin θ ated even though the soil profile itself may only be
μs ½ðρs ρw Þð1 nÞz þ ρw ð1 nÞzw
partially saturated. This allows the buoyancy term to
gΔx cos 2 θ τ c Δx (13.15) be operative but without the increase in the soil dens-
and the FOS is ity due to wetting (i.e. the third term in the numerator
of Equation (13.17) changes, but the second term in
μs ½ðρs ρw Þð1 nÞz þ ρw ð1 nÞzw g cos 2 θ þ τ c the denominator does not change),
FOSwet ¼
ðρs ð1 nÞz þ ρw nðz zw ÞÞg sin θ (2) Hillslopes are at the edge of their stability threshold,
(13.16) and even a small change in the soil wetness is enough
which can be rearranged into a form that is more easily to trigger landslides. This does not appear to explain
compared with the dry soil FOS: observations that rapid changes in groundwater levels
trigger landslides, but equivalent, but slower, changes
μs ρb gz cos 2 θ þ τ c μs ρw ð1 nÞgðz zw Þ cos 2 θ do not.
FOSwet ¼
ρb gz sin θ þ ρw ngðz zw Þ sin θ
The concepts above form the basis of the SHALSTAB
(13.17)
(Dietrich and Montgomery, 1998), dSLAM (Wu and
The first two terms in the numerator and the first term in Sidle, 1995) and SINMAP (Pack et al., 1998) landslide
the denominator are almost the same as in the dry soil stability models that have been incorporated into GIS
FOS; the only difference is the cos 2 θ term instead of packages. Early work did not use the general solutions
cos θ which slightly reduces the FOS. The groundwater above, but assumed that only one of either friction or
table (1) subtracts a term (which is always positive) from cohesion was the stabilising force. Later work has recog-
the numerator that reduces the FOS, this effect being the nised that both may be important. Dietrich et al. (1992,
reduction of the weight of the solid component of the soil 1993), Montgomery and Dietrich (1994) and Borga et al.
acting on the failure surface as a result of buoyancy of the (2002) coupled this shallow landsliding model (using only
soil particles and (2) adds a term to the denominator frictional resistance, no cohesion) and a model for satur-
(which is always positive) that also reduces the FOS, ation excess runoff (to predict the depth of the water table
based on catchment geomorphology) to predict regions of pre-clearing), which stabilises back to pre-clearing levels
low stability, and then compared this spatial distribution after 10–20 years. The mechanism is that over the five
with field observations with a good qualitative match. years the strength of the older roots drops, and it is only
Dietrich et al. (1995) extended this work to include a after this drop that landsliding rates increase (Burroughs
spatially variable cohesion that was a function of root and Thomas, 1977; Sidle, 1992). We will return to this
density. Ho et al. (2012) were able to improve their topic of root strength when we discuss the impact of fire on
predictions of landslide scars by incorporating measured tree death and subsequent hillslope transport in Chapter 14.
soil depth data rather than assuming soil depth was the
same everywhere. Roering et al. (2003) found landslide
scars concentrated in areas with less root strength, con- 13.1.3 Finite Slope Modelling
firming the importance of the spatial distribution of root
strength in promoting cohesion in forested areas and the The major limitations of the infinite slope model are the
impact of forestry clear-cutting (Wu and Sidle, 1995; assumptions that (1) slope and soils are the same in all
Schmidt et al., 2001). Gorsevski et al. (2006) used a directions and (2) the landslides are infinite in extent.
continuous soil moisture accounting model to simulate Addressing these limitations results in formulations of
time-varying landslide susceptibility and coupled it with the shallow landslide model that consider a landslide of
a model for forest clear cut. They found a strong link finite extent. This means that in addition to the resisting
between landsliding and the dynamics of root strength in shear forces on the base of the landslide (as for the infinite
space and time. slope model, Tbasal , Figures 13.3 and 13.4), there are now
Griffiths et al. (2011) constructed a two-dimensional resisting shear forces due to friction and cohesion (Tleft ,
finite element hillslope model (the axes were downslope Tright ) along the sides of the landslide parallel to the
and down the soil profile with no cross-slope component) downslope direction, a tension force at the upper bound-
and used it to validate the results of the analytic infinite ary (Fupslope ) and a resistance force at the lower boundary
slope model discussed above. By varying the length of the (Fdownslope ) (Figure 13.5). The side normal forces (Fleft ,
modelled hillslope, they showed that the infinite slope Fright ) cancel as they are equal if the soil depth doesn’t
model gave good FOS estimates for length:depth ratios change. The triangular volume at the upslope end is the
more than 15. They extended this model to examine cases wedge of soil applying the destabilising force at the
where the soil strength increased with depth down the soil upslope end, while the triangular wedge at the downslope
profile. They were able to generate failure surfaces that end is the wedge of soil that is resisting the movement
were fully contained within the soil profile. This was downslope. As noted at the start of the previous section,
because the increasing strength of the soil with depth the main rationale for including boundary conditions is for
offset the decrease in stability with increasing depth in when the area of edges is comparable to the basal area so
Equation (13.12). the forces are potentially comparable, which occurs for
Most field studies have compared model predictions the low length to depth ratios typical of small landslides.
for the spatial distribution of landslides after a single Milledge et al. (2014) implemented a finite element
event, and relatively few studies have looked at the tem- (FEA) model for a long hillslope identical to that of
poral behaviour at the same site for a series of rainfall Griffiths et al. (2011) (see the previous section) and gen-
events. Shuin et al. (2012) examined 13 events using a eralised Griffith’s results using a parametric study.
model that coupled a two-dimensional groundwater Depending on the combination of parameters used, Mill-
model with an infinite slope stability model and found edge found that the infinite slope model was within 10%
that they could not satisfactorily predict landslide rate and 5% of the FEA model for length:depth ratios above
from event to event unless they allowed the soil cohesion 18 and 25, respectively.
to vary between storm events by an order of magnitude. Milledge et al. (2014) formulated a two-dimensional
The main effect of vegetation is to provide shear finite conceptual landslide model, called MD-STAB,
strength across the failure surface (typically the interface where the two dimensions are the downslope and cross-
between the bottom of the soil and top of the saprolite) slope direction (Figure 13.5) and which included a water
reducing the rate at which slopes fail. It has been observed table. An application at the Coos Bay landslide field site
that following clear cutting of forests for silvaculture that found that if the measured pore pressures were used, then
for the first five years after clear cutting there is no change the model gave better predictions of landslide occurrence
in rate of landsliding, but after that time there is marked than previous studies using the infinite slope model (e.g.
increase (up to a factor of five times higher than see Montgomery et al., 2009 for some introspection about

force balance on a three-
dimensional control volume of
failed soil.
the experimental and modelling work at this field site). stable collection of pixels in the landscape and then
One possible implication of Milledge’s model is that for assumes that these pixels fail as a single block. They
smaller landslides the FOS is increased relative to the calibrated the model to the Coos Bay field site and
infinite slope model, and the authors speculate that there found it was able, after calibration, to predict 65% of
may be a minimum area below which landsliding cannot the observed landslides with a relatively low false
occur. Milledge et al. speculated that this might explain positive rate.
the peak of the landsliding frequency curve in Figure 13.1
below which small events become less likely (Malamud
attributed it to undersampling of small landslides due to 13.1.4 Soil Destabilisation by Dynamics at the Soil
the limited spatial resolution of the data he used). Surface
The major difficulty with Milledge’s model (and any
finite area model for that matter) is the need to define the The sections above have dealt with the failure of an
spatial extent of the landslide. This spatial extent can initially stationary soil. Another case that is important is
change depending on the spatial variability of all the where soil, mud or water is flowing over the surface,
inputs into the model including topography, soil depth, applying a shear stress to the surface of the stationary
soil strength and pore pressure distribution. While it has soil, and potentially destabilising the underlying soil as a
been commonly assumed that pore pressure variations are result of this extra loading on the soil. This is an issue
the dominant unknown variable, it has been suggested that downslope of a failure along the path that the landslide
the spatial variation of soil depth and root strength distri- takes downslope, and is the mechanism whereby debris
bution are equally important variables (Montgomery and mudflows scour the hillslope downslope of a failure
et al., 2009; Bellugi et al., 2015a). In principle, defining source area upslope. Hereafter we refer to this case as a
the failure surface for a natural hillslope involves search- debris flow, but the discussion is equally applicable to any
ing through an infinite number of combinations of failure geophysical flow that applies weight and/or shear stress to
surface geometries and finding the least stable surface, the soil surface.
which is clearly computationally intensive. Bellugi et al. Figure 13.6 shows a schematic of this process, high-
(2015a,b) used Milledge’s MD-STAB model and lighting the differences with the discussion in the previous
developed a search algorithm that starts with some region section shown in Figure 13.4. The differences are the
(typically a single pixel) and adds to that region adjacent addition of a weight load W d and a shear force T d applied
pixels that are least stable and removes ones that are most at the interface between the debris flow and the underlying
stable. The coupling between the adjacent pixels occurs soil. The shear force T d is the component of the weight of
by the friction and cohesion forces on the vertical inter- the debris flow in the direction parallel to the surface. This
face between the pixels (e.g. the edge shear and normal has parallels to the shear force applied by the weight of
forces in Figure 13.5). In this way it searches for the least the soil on the failure surface. However, unlike for the soil
Thus the destabilisation force (the downslope component

of the weight of the debris flow) has increased, but the
resistance forces (cohesion and friction at the failure sur-
face) are unchanged so that the soil underlying the debris
flow becomes less stable then it was before. For the
extreme case where all of the extra weight is 100% sup-
ported by the undrained pore pressures and the flow is not
accelerating, then (from Equation (13.16))
μs ½ðρs ρw Þð1 nÞzw þ ρs ð1 nÞðz zw Þgcos θ þ τ c
FOSdb ¼
ðρs ð1 nÞz þ ρw nðz zw Þ þ ρbd Dd Þgsin θ
(13.19)
where the bolded term in the denominator (the third term
inside the parentheses) is the additional term due to the
debris flow, and since it is always positive, it reduces the
FIGURE 13.6: Schematic of the force balance for a control volume FOSdb . The determination of whether the underlying
of soil within the soil profile where a debris flow is travelling over the
surface of the soil applying a (potentially destabilising) shear force on
soil fails (and becomes part of the debris flow) is the
the surface of the underlying soil. same application of the FOS calculation as was done in
the previous sections.
Finally it should be noted that the assumption that in
underneath there is also a component involving the accel- the undrained case the water supports the entire load is an
eration of the debris flow. The force balance on the debris approximation. The load is actually proportioned between
flow in the downstream direction is the water and the soil in proportion to their stiffnesses (i.e.
X Young’s modulus). However, water is several orders of
F ¼ W d sin θ T d ¼ ρbd gDd Δx sin θ T d ¼ Ma
magnitude stiffer than most soils and/or gravels, so almost
¼ ρbd Dd Δxa
all load is transferred to the water and almost none to the
(13.18) soil, so to first order in the undrained case the water
where ρbd is the bulk density of the debris flow material, carries the entire load.
M is the mass of debris flow material in the control
volume defined by Δx and a is the downslope acceleration 13.2 Transport Principles, Momentum
of the debris flow. This equation shows that if the flow is and Energy
accelerating, then the shear force T d applied to the soil
surface is less than the downslope component of the The conceptual split into triggering processes and trans-
weight of the debris flow, while if the debris flow is port processes suggests that a convenient way of model-
decelerating, then the shear force is more than the down- ling gravity transport processes is with a Lagrangian
slope component of the weight. formulation (Section 2.2.3) rather than with a Eulerian
The other component to this force diagram is the pore formulation. Specifically the processes in this chapter are
pressure at the failure surface. The discussion leading to attractive targets for precipiton modelling (e.g. Chase,
Equation (13.16) assumed a ‘drained’ situation where the 1992; Crave and Davy, 2001), and the discussion that
pore pressures are in static equilibrium with the soil and follows is based around this approach where possible.
hydrostatic forces. If the soils are relatively impermeable We can use precipiton modelling for landslides while
and the debris flow is moving rapidly, then the weight of using traditional Eulerian approaches for the other land-
the debris flow may induce pore pressures that are higher form/landscape processes, by using the principle of oper-
than the drained case, leading to ‘undrained’ pore pres- ator splitting (Section 2.2.2).
sures at the failure surface. We will discuss pore pressures
shortly when we discuss debris flows in detail. In the end 13.2.1 Momentum and Energy Principles
member case where the soil is ‘fully undrained’, all of the
additional weight force of the debris flow in the direction The momentum conservation equation for a mass of
normal to the failure surface is supported by the water rock and debris moving downslope between two cross
and not by the soil matrix, so that the frictional resistance sections 1 and 2 (Figure 13.7) is derived from Newton’s
does not rise with the increased weight of the debris flow. second law
W ¼ Fs (13.23)
where W is the energy work of the force F acting over a

distance s. In the sections below this force will be the
frictional resistance working against the flow. Note that
the potential energy term in Equation (13.22) is the work
required to lift the mass from the zero elevation datum to
the elevation z (i.e. z ¼ 0).
If the mass between cross sections 1 and 2 does not
FIGURE 13.7: Schematic showing the components of the change, then we can rearrange the momentum equation
momentum and energy equations for two cross sections in the flow. into a conservation of energy equation along the direction
T is the bottom shear resistance to flow, W is the self-weight of of movement
the material, m is the mass of the volume of fluid and v is the
downslope velocity, with the subscripts 1 and 2 indicating the two X ∂v m ∂ðv2 Þ
cross sections. F ¼ mv ¼ (13.24)
∂x 2 ∂x
where the right-hand side is the rate of change with

X ∂M ∂ mv ∂ mv m2 v2 m1 v1 distance of the kinetic energy and is a function of the
F¼ ¼ ¼v ¼ lim
∂t ∂t ∂x Δt!0 Δt forces acting on the material. If this equation is integrated
(13.20) with respect to x between cross sections 1 and 2 in
where the summation is over all the forces (weight, fric- Figure 13.7, then
tion) acting on the mass of rock and debris, M is momen- X m 2
F s¼ v v21 (13.25)
tum, m is the mass of rock and debris, v is the velocity and 2 2
the underbar indicates a vector quantity. where the left-hand side is the work done by all the forces
The energy conservation equation along a flow line in acting on the material in Equation (13.23) and the right-
a fluid between cross sections 1 and 2 is normally hand side is the difference in kinetic energy between cross
expressed in terms of the energy head sections 1 and 2.
E1 ¼ E 2 energy losses The point of these derivations is to show how the
forces acting on the material change its momentum and
p v2 (13.21)
E¼ þ þz energy, and under what conditions the momentum and
ρg 2g
energy equations are equivalent. This latter point is
where E is the energy head, p is the fluid pressure, v is the important because much of what follows will use energy
velocity and z is the elevation. If we consider the energy conservation equations rather than momentum equations
equation for a given mass m of fluid, then by multiplying because cause and effect is more transparent.
this equation by mg we obtain
mp mv2
E mg ¼ þ þ mgz (13.22) 13.2.2 Momentum-Free Formulations
ρ 2
and this is a classical expression of energy, with the second Before presenting models based on energy and momen-
term on the right-hand side the kinetic energy of the mass tum principles, we will discuss transport models that do
and the third term the potential energy above the elevation not track momentum or energy. The classical model that
datum. The first term in pressure is important only when fits in this category is the sand pile model of Bak and
there is a change in pressure from cross sections 1 and 2 Wiesenfeld (1988). The idea in this model is that there is a
and this pressure term is the work done by the change in slope stability threshold at some angle of repose. Sand
pressure between the two cross sections. As per normal grains are dropped (without momentum) onto the pile of
practice for depth-averaged free surface flow we will sand, and if the location where the sand grain lands is no
ignore pressure changes (i.e. this is consistent with assum- longer stable (i.e. the slope is above the angle of repose),
ing hydrostatic pressure conditions, but does not require then the slope collapses and sand is transported down-
it). The energy losses term is the energy loss of the fluid stream. This transport downslope potentially destabilises
and is typically a transfer of energy into another form downslope locations leading to further collapse. The land-
outside the flow (e.g. noise, heat, vibration). For instance, slide stops when the sand grains that are transported
the work done by a force applied by the fluid is downstream arrive in a location where the angle of repose
is no longer exceeded as a result of the sand grains The key concept to note in both models is that neither
arriving from upslope. The landslide scar is the entire area simulates the velocity of the sand or the precipiton, so
that moved as a result of the destabilisation arising from accordingly they do not simulate momentum or energy
the single grain that was initially deposited on the slope. changes downslope. For Bak’s model the initial sand
There were two key insights from this work: (1) the grain does not have any momentum so that when it
introduction of the concept of self-organised criticality initially hits the surface it does not destabilise the sand
where the sand pile organises itself so that all locations pile as a result of the collision of the falling sand grain
are all just at the point of instability and (2) the probability with the sand grains already on the surface. Nor does the
distribution of the areas of the landslides follows a fat- resulting landslide transfer any momentum to the sand
tailed hyperbolic distribution so that the distribution of grains over which it flows, so the act of the landslide
landslide areas is fractal. This latter insight has triggered a flowing over the surface does not destabilise the under-
number of field projects (Hovius et al., 1997; Stark and lying material. Finally the landslide stops when the slope
Hovius, 2001; Malamud et al., 2004; Chien-Yuan et al., drops below the slope threshold so the landslide cannot
2007) and more realistic modelling studies (Hergarten, propagate up and over small undulations in the original
2003; Stark and Guzzetti, 2007) to examine the probabil- surface. To move up and over an undulation requires that
ity distribution of landslide scars and which broadly the landslide have energy and momentum that can be
confirm the hyperbolic distribution. The power law of used to propel the material over the obstruction. Chase’s
occurrence with area discussed in Section 13.1 and model shares many of the same characteristics. The only
Figures 13.1 and 13.2 is consistent with a hyperbolic difference is that he deposits sediment as a function of
probability distribution. the downstream slope. As soon as his model hits a node
The sand pile model has strong conceptual similarities that has a zero downstream slope, all the sediment is
to the landform evolution model of Chase (1992). In this deposited so again his model cannot propagate move-
model a package of sediment (he calls the package of ment up and over an obstruction nor model landslide
sediment a ‘precipiton’) is moved downslope through runout.
the gridded topography in the digital elevation model
(Section 2.2.3). At each point when the package arrives
at a new node downstream, the new slope to the down- 13.2.3 Including Momentum and Energy
stream nodes is evaluated. The transport capacity at that
There is a vast array of gravity processes all with different
point is considered to be proportional to slope, and a
names, and the vastness of this differentiation gives the
proportion of the sediment in the precipiton is deposited
appearance of many different and incoherent physical
based on the carrying capacity provided by the value of
processes at work. Here we try to focus on the common
slope. No discharge dependency is considered, and separ-
generic features. We will thus initially focus on the broad
ate precipitons are considered to operate independently,
principles of the processes at work.
so the process is linear with discharge (two separate pre-
The overall approach that will be taken in this section
cipitons are considered to have the same effect as a single
to explain the physics principles will be to track the
precipiton with twice the mass). The starting location for
energy and momentum of the material as it moves down
the precipiton is random in space, reflecting the random
the slope, because this approach links most transparently
nature of rainfall events. Landform evolution is a result of
to the cellular automata and precipiton-like models that
the application of a number of precipitons. In Section
appear to be the most promising approach to modelling
2.2.3 it is shown how this model (which is Lagrangian
these processes in LEMs. Thus we will explicitly focus on
because it tracks the path of the sediment down the slope)
energy balances and energy losses and transformations to
can be cast into a traditional differential equation for mass
model the movement of material downslope.
balance at a single location (which is a traditional Eulerian
If we consider the energy balance of a gravity event
formulation). Chase presented this model as a model of
from source to destinations, it is as follows:
fluvial erosion, but it is now accepted that the implicit
assumption of linearity in the discharge dependency is a 1. There is some initial energy input to dislodge the
fatally flawed assumption. However, the idea of precipi- material and start motion. The material may or may
ton modelling has merit as a Lagrangian method for not have some initial kinetic energy as a result of this
simulating landslides, and Section 2.2.3 shows how this failure mechanism.
approach is functionally equivalent to the Eulerian equa- 2. As the material travels downslope potential energy is
tions mostly used in the remainder of this book. converted into kinetic energy. This kinetic energy may
be longitudinal (i.e. straight-line motion) or rotational For a slide that is moving and assuming Coulomb
(i.e. rocks rolling). friction (i.e. where frictional shear resistance T on a sur-
3. Some of the kinetic energy is lost through various face is proportional to the load normal to the surface N so
energy dissipation processes including friction, frag- that T ¼ μd N), the energy dissipated per metre travel
mentation, inelastic and plastic impacts and failures, downslope
sound generation, atmospheric or water drag, and vis-
cous dissipation. W ¼ Ts1,2 ¼ ðμd mg cos θÞs1,2
z1,2 μ mgz1,2
4. When the increase in kinetic energy (due to conversion ¼ ðμd mg cos θÞ ¼ d (13.28)
from potential energy as the material drops in height sin θ tan θ
down the slope) is greater than the energy dissipation, where s1,2 is the distance along the slope between points
then the material accelerates, when the dissipation is 1 and 2, z1,2 is the drop in elevation from 1 to 2, μd is
greater than kinetic energy gain then the material slows. the dynamic coefficient of friction, θ is the angle of the
5. As the slope flattens dissipation losses continue and slope from horizontal and m is the mass of soil which is
kinetic energy gains decrease, resulting in deceleration. m ¼ ρb Ad where d is depth from the soil surface to the
Eventually motion stops. failure plane and defines the depth of moving soil, A is
The difficulties in this calculation are in determining the planar area of the slide and ρb is the bulk density of the
the criteria for the start of motion, and quantifying the soil. The dynamic coefficient of friction is the friction
energy dissipation processes (Bourrier et al., 2009). coefficient that applies once there is relative motion
Mathematically if we follow a particle of mass m along between the material above and below the interface. It is
its trajectory we can write an energy conservation equa- smaller than the static coefficient of friction, which is
tion for a downslope location 2, given a starting upslope applicable prior to motion (Meriam et al., 2012). In granu-
location 1 lar material this difference is attributed to the dilation that
must take place before particles above the interface can
PE1 þ KE1 dissipation ¼ PE2 þ KE2 (13.26) move up and over particles below the interface
where (Figure 13.8). Once this dilation has occurred, motion
occurs and the resistance to movement is reduced. At
PE ¼ mgz
some point downslope the motion stops, the particles
1 (13.27)
KE ¼ mv2 þ Iω2 settle so that the dilation reverses, and the static friction
2 again becomes applicable. As noted in Section 13.2.2 the
and z is the elevation above datum, g is the acceleration drop in friction coefficient for granular material upon
due to gravity, I is the moment of inertia of the particles
and ω is the angular velocity of the particle in radians/
second. For a spherical particle the moment of inertia is
2 (a)
I ¼ 2mr
5 : We will discuss rotational KE in more detail in
Section 13.5.
13.2.4 Slides
The primary difference between slides and flows is what

happens at the interface between the moving material and
the underlying soil/saprolite. By definition a flow has
(b)
a ‘no-slip’ condition at this interface, so the velocity of
the material in the flow at the interface is travelling at the
same velocity as the underlying material (typically the
underlying material is stationary). For a slide the material
directly above the interface is moving with some velocity
relative to the underlying material.
The significance of this distinction for us is that for a
FIGURE 13.8: Schematic how dilation and collapse occur as the
slide all the energy dissipation is at the interface, while for result of the geometry of movement relative to each other and the
a flow the energy dissipation occurs within the flow. impact on the water table within the soil: (a) dilation, (b) collapse.
motion can be of order of 20%, but the drop is highly single block with all energy dissipation occurring at the
sensitive to the grading of the material, which influences base of the block and no energy dissipation occurs intern-
how much dilation is required before motion can occur. ally within the slide.
The acceleration downslope can be calculated in two
ways. The first using Newton’s second law in a force
balance downslope to do a momentum balance, while 13.2.5 Debris Flows
the second uses the kinetic energy conservation.
Writing a force balance in the downslope direction The main type of flow we will talk about in this section is
(downslope positive) a debris flow. Classical flows have a no-slip condition at
X the base of the flow, and all the energy dissipation occurs
F ds ¼ ma ¼ mg sin θ μd mg cos θ (13.29)
internally within the flow. Debris flows are a little
and solving for the acceleration a different because they have aspects of flows and slides
simultaneously. They have some of the friction energy
a ¼ g cos θð tan θ μd Þ (13.30) losses at the flow-soil interface of the slide with the
We see that the acceleration is a function of the excess of energy losses within the flow as a result of turbulent
the slope over the dynamic coefficient of friction (i.e. and viscous losses. The discussion that follows is based
tan θ μd ). on the general principles outlined in Iverson (1997) and
The alternative derivation based on energy conserva- Iverson and George (2015).
tion can be obtained by writing the energy balance, which Debris flows are masses of rock, soil and water that
requires calculating energy conservation over a short travel downstream and deposit on flatter areas at the base
length of slope (Δx in the horizontal direction). The fric- of mountains. It is generally believed that a key condition
tional work is (from Equation (13.23)) is that the rock and soil mass must be saturated with water.
Their behaviour is somewhere intermediate between dry
W friction ¼ μd mg cos θ Δx=cos θ ¼ μd mgΔx (13.31) granular flow (e.g. dry sand) and water flow. Their vel-
ocity is fast, commonly greater than 10 m/s. A key aspect
and the change in kinetic energy over Δx (noting that for a of the debris in the flow is that the size of particles covers
short length of hillslope, so the kinetic energy change is a large range from large boulders down to clay, and the
small, v2 2 ¼ ðv1 þ ΔvÞ2 ¼ v1 2 þ 2v1 Δv þ Δv2 v1 2 þ shape of the particles is quite irregular. The fine particles
2v1 Δv when Δv is small) is are key because they form an important part of the ‘lubri-
m 2 cation’ of the flow. There is a strong size dependence to
ΔKE ¼ v2 v1 2 mv1 Δv ¼ 2mgΔxð tan θ μd Þ debris flows with larger flows travelling longer distances
2
(13.32) (Johnson et al., 2016).
so Δv ¼ gΔxð tan θ μd Þ=v1 . The acceleration is then It is common to treat the debris flow as having only
one component/phase (the mixture of rock, soil and
Δv gΔxð tan θ μd Þ v1 cos θ water), and to mathematically formulate the problem as
a¼
Δt v1 Δx a fluid flow with laminar viscosity where the viscosity is
¼ g cos θð tan θ μd Þ (13.33) described by a Bingham fluid (e.g. Whipple, 1997). We
discussed viscous Bingham flow in the context of soil
which is the same result as derived using the force balance creep in Chapter 9. Iverson and George (2015) indicate
method in Equation (13.30). that the calibration of a single-phase model results in
One point about the acceleration derived in Equations viscosities that are too low so that runout distances are
(13.30) and (13.33) is that it is independent of the depth too high. They proposed that the physics of the debris
below the surface and so is equally applicable to any flows is more clearly understood by differentiating two
interface at any depth below the surface, so there is no phases: (1) the granular rock phase and (2) the interstitial
relative acceleration between soil at different depths fluid between the rocky particles. The granular rock com-
below the surface. This is provided only that the dynamic ponent is typically about 40–80% of the volume of the
coefficient of friction does not change with depth. The material, and the fluid component is a mixture of water
implication of the lack of relative acceleration with depth and fine particles that form mud with about 10–20%
below the surface is that all the soil above the failure plane solids by volume.
slides as a single block. Thus, in the case of Coulomb There are three stages on the temporal evolution of a
friction, we can conclude that the material will flow as a debris flow: (1) the initiation of the debris flow movement

through some form of instability or slope failure, (2) the ∂h ∂ ∂h ∂2 h ∂K ∂h ∂2 h
flow of the rock, soil and water mass down the slope ¼ K ¼K 2þ K 2 (13.34)
∂t ∂z ∂z ∂z ∂z ∂z ∂z
and (3) the slowing down and eventual stopping of the
flow in what is normally a lobate-shaped deposit. In all
where the diffusivity for pressure is the hydraulic con-
three stages the fluid and its pore pressures in the inter-
ductivity of the flow in the vertical direction K, which is a
stices are key. Figure 13.8 shows how the rock and fluid
function of the void geometry between the solid particles
potentially interact at the moment when movement is
(e.g. the Kozeny-Carmen equation; Freeze and Cherry,
initiated.
1979). Pore pressure adjustment occurs primarily in the
As part of the debris movement process the material
vertical direction because the depth of the flow is much
can either dilate (expand) or collapse (contract). Dilation
less than the length or width of the flow, so the vertical
occurs when the large particles are initially tightly packed
hydraulic gradient is much greater than the horizontal
and cannot move downslope unless they move up and
gradient ∂h∂z
∂h ∂h
∂x , ∂y where x and y are the downstream
over the particles directly underneath. Thus to move the
and across stream directions. Note that the hydraulic
material must expand/dilate. One consequence of dilation
conductivity for the interstitial mud is not the same as
is that the volume of voids within the debris increases
the hydraulic conductivity of clear water, since the inter-
and the water level drops. If the material is initially fully
stitial mud is more viscous than clear water (by 102–104
saturated with no excess water, then upon movement the
times so the conductivity will be 102–104 times less for
top layers of the pile will become unsaturated as the pore
the mud).
water flows into the increased void space underneath. The
One consequence of Equation (13.34) is that the
opposite process occurs when the material collapses and
response time of the adjustment of pore pressures scales
excess water appears on the surface after it is expelled
with the ðdepth of the debris flowÞ2 so that deeper (and
from the decreased volume of voids. Water flow direction
bigger) flows will have excess pore pressures (and thus
is in the direction of reduction in head (pressure p is
reduced frictional resistance) for longer periods. Using a
sometimes used instead of head h but that is incorrect,
. similitude analysis Iverson and George (2015) suggest
h ¼ z þ p ρmg where h is head, z is elevation and ρm is the that large debris flows may take days to return to hydro-
density of the mud), so for dilation and water drop the static conditions, with the consequence that it seems rea-
pressure at the base of the debris flow must decrease sonable to suggest that excess pore pressures from debris
below hydrostatic pressure, while for collapse the pres- flow initiation may persist for the entire duration of the
sure at the base of the flow must exceed hydrostatic debris flow event. The increase in pore pressure relaxation
pressure to expel that water to the surface. An important times with size may also explain why larger debris flows
consequence of collapse is that if the collapse occurs appear to have lower frictional resistance and larger run-
because of movement and this movement was triggered outs than smaller debris flows (Iverson, 1997). This relax-
by rises in pore pressures (e.g. as a result of infiltration ation time scaling is consistent with field data that shows
during rainfall), then the pore pressures will rise further, excess pore pressures occur in the main body of a debris
further destabilising the material, leading to a positive flow deposit long after the event (Major and Iverson,
feedback. On the other hand dilation will lead to a reduc- 1999).
tion of the pore pressure with the possibility of a stabil- This scaling solution assumes that the only pathway
isation of the initial movement. Iverson and George for pore pressure relaxation is for water to flow from the
(2015) talk about dilation leading to a gradual stop-start base of the flow up through the layers of debris in the flow
destabilisation of the slope, while collapse leads to a to the surface. There are other potential mechanisms
rapid destabilisation. One consequence of this is that the (Figure 13.9), including (1) vertical mixing within the
initial condition of the soil on the slope is important for debris, (2) development of preferred flowpaths to the
how a debris flow commences since the initial consoli- surface of the debris flow such as cracks and (3) pressure
dation state of the soil determines whether it will collapse relief into the underlying soil material. All of these mech-
or dilate upon movement. anisms would significantly reduce the relaxation times,
The excess pore pressure at the base of the debris but they were not considered by Iverson and George
flow dissipates upwards and is driven by the vertical (2015).
head gradient. The rate of dissipation of pore pressure is For typical conditions, scaling analyses by Iverson and
akin to a Fickian diffusion process for head h as can be Denlinger (2001) and Iverson and George (2015) suggest
seen from the groundwater flow equation in the vertical that the dominant resistive force against flow is bottom
direction z: friction and not turbulent dissipation, so by that definition
FIGURE 13.9: Schematic showing pathways for pore pressure relief other than as a result of flow of pore water flow through pores.
The lighter hatched material is the moving debris flow material, and the darker hatching underneath indicates the natural surface the debris
flow is moving on top of. The arrows indicate the direction of pore pressure relief. (a) Vertical mixing where the higher pore pressure materials
are advected to the surface by vertical mixing, and the lower pore pressure materials from near the surface are transported downward,
(b) pore pressure relief by pore water flow into a vertical crack in the debris flow, (c) higher pore pressure water flow into the underlying
material (either into fully or partially saturated pores or cracks in the underlying material).
a ‘debris flow’ is not a flow but a slide, with most energy Iverson, 1999; Iverson and Denlinger, 2001; Figure 2 in
dissipation occurring at the bottom and sides of the debris Iverson and Vallance, 2001). The segregation that has also
flow. A simple model for debris flows is then the (1) slide been observed in these levees (and on the surface of
model of the previous section (Section 13.2.4) coupled flows) where the levees are richer in coarse materials will
with (2) a pore pressure dissipation model based on diffu- be a positive feedback, since coarse material will result in
sion (Equation (13.34)) and using debris flow depth, faster dissipation of the pore pressures (because of its
where the coupling occurs through (3) the reduction of higher hydraulic conductivity in Equation (13.34)),
the normal force in the friction equation as a result of enhancing stabilisation (Johnson et al., 2012).
excess pore pressure. The friction coefficient in part (3) Finally it should not be concluded from the discussion
should have an approximately log-log linear relationship above that pore pressures from water/mud in the sediment
with flow volume varying from about μ ¼ 0:5 for a is essential to debris flow behaviour. For instance,
volume of 106 m3 declining to about μ ¼ 0:1 for a volume Campbell et al. (1995) notes that the trend of reduced
of 1010 m3 (Campbell et al., 1995; Johnson et al., 2016). frictional resistance for large flows has also been observed
The derivation of Iverson and George does not directly on Mount St Helens and Mars. For Mount St Helens it is
address what happens in the termination stage of the flow unreasonable to expect that water played a role. Tradition-
when the flow is slowing to a stop and the lobate-shaped ally this was also the conclusion for Mars, but recent
deposits are being formed. However, it is quite simple, evidence of historical water processes may change that
and as the slope reduces at the bottom of the hillside, the interpretation. This suggests that there are additional, as
frictional resistance will be greater than the self-weight yet unresolved, mechanisms at work in large debris flows.
driving force and the flow will slow, eventually stopping.
Finally, one interesting consequence of this pore pres-
sure diffusion model is what happens around the edges 13.3 Rockfall Modelling
(i.e. the sides and front) of the debris flow. Because the
edges are closer to the atmosphere any excess pore pres- This section presents a modelling methodology for exam-
sures are more readily dissipated than deeper within the ining rockfalls, where the main transport process is indi-
flow. As a result the pore pressures are lower near the vidual rocks moving down a slope. Dorren (2003) reviews
edges so that the frictional resistance will be greater. It is the dominant physics of rockfall, noting that for low
then possible that the moving material at the edges will slopes (less than about 45 ) the dominant process is
fall below the undrained angle of repose and become rolling of rocks down the slope and that for steeper slopes
stable. Thus at the edges of the debris flow there may be rocks bounce. Sliding of rocks typically occurs only for
levees of deposited material formed, while at the front of the initial and final stages of movement.
the debris flow will be a pile of stable material potentially For rolling of rocks the main energy loss is rolling
being pushed by the ‘liquid’ debris flow behind the front resistance, and the stopping of motion typically occurs
and a surface layer on top of the debris flow (Major and quite suddenly as the rock is captured in the surface
roughness of the slope. Smaller rocks tend to be captured where KEbi is the KE of particle before the impact, KEai is
more easily than larger rocks, so the final positions of the total energy of all of the daughter particles after impact
rock fragments are sorted with the largest fragments and fragmentation, E l are the energy losses during colli-
being transported the furthest distance. Kirkby and Sta- sion and E f is the energy loss during fragmentation. If we
tham (1975) captured this sorting process with a simple first consider a particle that collides at right angles to the
model based on friction where the dynamic friction coef- surface without fragmentation (i.e. E f ¼ 0), then
ficient is 1
KEai ¼ mðCr vbi Þ2 ¼ C r 2 KEbi (13.37)
kd 2
μd ¼ tan ϕ þ (13.35)
2R where Cr is the coefficient of restitution of the particle.
where ϕ is the angle of internal friction or angle of repose For soil-covered slopes Cr is about 0.1–0.2, while for
(from 20 to 34 ), k is a constant (between 0.17 and 0.26), rock-covered slopes Cr is about 0.3–0.7 (Asteriou et al.,
d is the mean diameter of scree on the surface and R is the 2012). For particle-particle collisions without fragmenta-
radius of the moving rock. This coefficient of friction is tion it can be as high as 0.65 to 1 (Durda et al., 2011). If in
then used in the sliding model discussed in Section 13.2.3 addition the particle is travelling downslope so that there
to determine the travel distance. The larger the diameter of is a component of velocity parallel to the collision surface
the rock, the lower the friction coefficient and the longer as well as normal to the surface, then Cagnoli and Manga
the travel distance. This results in the aforementioned (2003) found that as the angle of impact decreased (i.e. the
downslope sorting on diameter. downslope velocity increased relative to the velocity per-
For higher slopes the approach that is described in pendicular to the surface), then C r increased (e.g. 0.15 for
more detail in the following section is based on Dorren an impact at 70 to the surface, up to 0.6 at 30 , both
et al. (2005, 2006) and Bourrier et al. (2009) who present figures for pumice particles). Asteriou et al. (2012) pro-
a feasible (in the sense of being able to be incorporated posed a relationship between Schmidt hammer hardness
into a LEM) model for rockfall, called RockyFor3D. of material and the coefficient of restitution for five dif-
Their model focuses on the transport process and does ferent rock types with Cr varying from 0.3 to 0.8
not address the triggering process. A review of this model, Cr ¼ 0:235e0:022R (13.38)
and other physically based models, is provided by
Volkwein et al. (2011). where R is the Schmidt hammer hardness value. They also
proposed a relationship for the coefficient of restitution
with impact angle, which was
13.3.1 Collision with Boundaries
θ
C r ,θ ¼ 1 α (13.39)
When a particle collides with another particle, (1) kinetic 90
energy (KE) is potentially transferred between particles where α defines the relationship between C r,θ and the
and (2) during this transfer of energy some of the KE is angle of impact θ (in degrees). The two relationships in
lost. If the particle hits a deformable surface (e.g. soil), Equation (13.38) and (13.39) are not necessarily consist-
there will be no transfer on KE (unless the soil starts to ent because Equation (13.38) is the result of lumping
move as a result of the collision), but some of the KE of together all of their results for different impact angles
the particle is lost and used in plastic deformation of the (ranging from 10 to 90 ).
soil. Finally if either of these collisions above results in Wyllie (2014) extended the definition of the coefficient
fragmentation of the particle, then some of the KE of the of restitution to consider separately the effects on the
particle(s) is used in the fragmentation process. In the component of the particle velocity normal to the surface
terminology of the fragmentation literature there is energy that is being impacted and the component parallel to the
associated with the creation of the new surfaces resulting surface so that
from the fragmentation, called ‘surface energy’. The more v N ,a
the particles fragment, the greater is the surface area C r,N ¼
v N ,a
generated and the greater is the transformation of the (13.40)
vT ,a
particle’s KE into surface energy. C r,T ¼
vT ,a
All the energy transformation processes above are
lumped into the ‘restitution coefficient’, which where the subscripts N and T indicate velocities normal
and tangential to the impacted surface, respectively. He
KEbi ¼ KEai þ E l þ E f (13.36) noted that the energy loss in the normal direction is
primarily a function of the plastic deformations of the momentum (in the x, y, z directions) for two particles
surface and the energy losses resulting from this plastic 1 and 2 are
deformation. The energy loss in the horizontal direction is
m1 v1b 2 þ m2 v2b 2 þ I 1 ω1b 2 þ I 2 ω2b 2 ¼ m1 v1a 2 þ m2 v2a 2
a function of the friction between the particle and surface,
which is a function of the friction coefficient between the þ I 1 ω1a 2 þ I 2 ω2a 2 þ losses
particle and the surface, and the normal force of the m1 v1xb þ m2 v2xb ¼ m1 v1xa þ m2 v2xa
impact (which in turn is a function of the normal impact m1 v1yb þ m2 v2yb ¼ m1 v1ya þ m2 v2ya
velocity and the plastic deformation of the surface). By m1 v1zb þ m2 v2zb ¼ m1 v1za þ m2 v2za (13.42)
examining various field rockfalls he found a tangential
where I is the moment of inertia of the particle. For a
restitution coefficient in the range 0.3–0.8, and a normal
sphere I ¼ 2=5 mr2 where r is the radius of the sphere. The
restitution coefficient
transfer of linear and rotational energy occurs as a result
C r,N ¼ 19:5θ1:03 (13.41) of tangential frictional force between the particles at the
time of collision, which in turn is a function of the normal
which gives a normal restitution coefficient greater than 1 force between the particles during the collision. Energy
for angles of impact less than about 18 . The values loss can occur as a result of these tangential friction forces
greater than 1 were attributed to interaction between the if the surfaces of the two particles slide against each other.
particle rotation and its translation upon impact, with This can occur when the inertial forces required to change
transfer between rotational energy and translational the rotation rate of the particle exceeds the tangential
energy (see Equation (13.27)). friction force. The energy lost is the work done by the
One complication of this discussion is that the defin- sliding surfaces W ¼ Fs where F is the frictional force
ition of the coefficient of restitution above considers only between the particle surfaces, which may vary during the
linear velocity and not rotation of the particle (Chau et al., collision as the normal contact forces change.
2002). At low angles of impact the rotational energy of This equation considers only collision without frag-
particles can interact with the rough surface being mentation. If multiple particles are generated by the colli-
impacted with, to result in impacts where the coefficient sion, then the energy and momentum of all the post-
of restitution as defined in Equation (13.37) is greater collision particles must be summed on the right-hand side.
than 1. This is a result of some of the particle rotational There will be additional energy losses as a result of
KE being transformed into translational KE by the geom- fragmentation, which will be discussed in the next section.
etry of impact. For instance, Buzzi et al. (2012) were able
to construct laboratory experiments using rotating par-
ticles that gave a coefficient of restitution that was greater 13.3.3 Fragmentation
than 1 and sometimes as high as 2 because the rotational
energy was converted to translational energy by the Returning to Equation (13.21) it remains only to charac-
impact, which is consistent with the findings of Wyllie terise the energy lost as a result of fragmentation.
(2014). Giacomini et al. (2009) performed a series of experiments
As a final note for cricket and tennis fans the inter- to define the energy loss due to fragmentation, E f , relative
action between rotational energy and translational energy to the impact energy (i.e. the ratio Ef =KEbi ) and found that
upon impact is how applying spin to a ball can result in a this ratio ranged from 0.1 to 0.6, and that it was independ-
ball ‘jumping’ with a higher speed after the impact with ent of the number of daughter particles generated by the
the ground than its speed before impact. impact so that they could express the energy lost in
fragmentation as
Ef ¼ Cf KEbi (13.43)
13.3.2 Particle-to-Particle Collisions
where C f varies from 0.1 to 0.6.
The discussion above considers collisions between a par- We are also interested in the grading of the daughter
ticle and a solid boundary. In the next section we will products from any fragmentation (e.g. see the discussion
discuss situations where many particles interact as a result about fragmentation modelling in Chapter 7). Using
of particle-to-particle collisions. In this case both kinetic experiment and discrete element modelling, Wang and
energy and momentum will be transferred between the Tonon (2011) tested the main distributions used to char-
particles. The pre- and post-collision (subscript b and a, acterise fragmentation products (exponential, Weibull and
respectively) energy (both linear and rotational) and Voronoi distributions) and found that the Weibull
distribution fitted the data and model results best. They the model digital elevation model. RockyFor3D does not
also found that mean particle size decreased as the strain deterministically calculate velocities (i.e. the downslope
rate (i.e. how rapidly the fragmentation occurred) and cross-slope components of the tangential velocity) but
increased. uses an empirical rule that is based on the probability
distribution and the steepest downslope direction derived
13.3.4 Modelling for the topography from the digital elevation model.
Potentially, the RockyFor3D approach could be
Bourrier et al. (2009) discusses an energy balance model extended by disaggregating the tangential velocity into
as outlined above, but doesn’t consider fragmentation of its downslope and cross-slope components (or its (x,y)
the rock. The general principle is that a rock bounces components if that is more convenient), and by disaggre-
down the slope, and at every bounce energy is lost and gating the rotational velocity into its three components.
this energy loss is parameterised by the restitution This would provide a greater physical basis for their post-
coefficient. However, instead of considering only the impact velocities so that the velocity vector would then be
energy loss from that component of the velocity at right 2 3
vx
angles to the surface, Bourrier also considers energy lost 6 vy 7
6 7
from the tangential (i.e. parallel to the surface) and rota- 6 vnormal 7
v¼6 7 (13.46)
tional energy. At each impact the velocities after impact 6 ωx 7
6 7
are determined before impact by the matrix equation 4 ωy 5
ωz
vai ¼ Avbi (13.44)
RockyFor3D did a good job of matching an experi-
where the velocity vector is the two translational veloci-
ment on a hillslope for (1) the spatial distribution of the
ties, and the rotational velocity and matrix A are
final location of the rocks (both distance downslope and
2 3 2 3
vtangential a11 a21 a31 lateral spreading) and (2) the trajectories of the bouncing
v ¼ 4 vnormal 5; A ¼ 4 a21 a22 a32 5 (13.45) rocks down the slope (compared with video imagery)
ω a31 a32 a33 except for two regions:
For the case where only velocities normal to the surface • The model underpredicted the number of particles close
are considered, as previously discussed, then aij ¼ 0 for to the original release point. This was because many of
all i and j except a22 ¼ Cr . This matrix equation charac- those particles did not roll but slid down the slope under
terises both the transformation of KE from one form to friction (because the experimental particles were not
another (the off-diagonal term) and the loss of KE for each perfectly round but blocky) and as a result stopped their
bounce (mostly the diagonal terms in the matrix, though motion closer to the release point due to frictional
the off-diagonal terms can include a loss component in the energy dissipation.
transformation process). For instance, a bounce at a low • The model overpredicted the number of particles at long
angle will tend to make the particle want to rotate, thus distances out in the runout zone. This was because a soil
transferring KE from the tangential component to rota- mantle at the surface dominated the runout zone in the
tional, while a particle that is rotating quickly prior to experiment, rather than a rock surface as assumed by
impact will tend to transfer rotational KE to tangential and the model. As noted above, the restitution coefficient for
normal velocities. Thus it should be clear that the idea that soil is much less than rock, so the model underestimated
there is a single restitution coefficient for each of the three the energy loss in these soil mantled regions, and the
velocity components is too simplistic (Bourrier et al., modelled particle moved further than the experiments.
2009). In fact, as was observed in the Buzzi et al. (2012)
Bourrier et al. (2009) does not document how they
experiments, the Bourrier model RockyFor3D predicts
derived the coefficients of the matrix A, but Bourrier
restitution coefficients greater than 1 for low impact angle.
(2008) indicates they were derived from laboratory experi-
One interesting aspect of the RockyFor3D model is
ments where particles were impacted on a rough surface.
that there is a random bounce component based on prob-
ability distributions presented by Dorren et al. (2005),
who measured impacts in a forested hillslope. In principle, 13.4 Discrete Element Models
a similar probability distribution could be derived or
assumed for rock-to-rock impact on rocky slopes, or on An emerging modelling technique, which is implicitly
the impacts between the rocks and the topography from Lagrangian and which follows directly from the
discussion of the previous section, is discrete element rotational and translational energy during impact with
modelling (DEM; not to be confused with Digital Eleva- the ground (Mitchell and Hungr, 2017). The trade-off of
tion Models). The concept of DEM is that each rock this approach is that irregular particles can be modelled,
fragment is modelled separately (i.e. mass, location, vel- but at the expense of using significantly more particles.
ocity etc.), and the interaction between the particles and Three-dimensional simulations are less common
between the particles and the boundaries are modelled. because of the computational requirements. Mead and
Typically many thousands of particles are modelled, and Cleary (2015) modelled a dry debris flow (comparing it
the average behaviour of the particles can provide a con- against experimental data in Iverson et al., 2004) in three
tinuum representation of the flow. dimensions and with irregularly shaped particles. The
The general physical principles of how the DEM par- particles were modelled to be superquadric in shape.
ticles move and interact (both with each other and with the Superquadrics are a generalised version of an ellipse. By
boundaries) are as discussed in Section 13.3. To model changing the properties of the particles they were able to
the motion of individual rocks from rockfalls requires the evaluate their effect on flow properties. They found par-
user only to simulate a small number of DEM particles ticle aspect ratio, angularity and interparticle friction to be
(e.g. Mitchell and Hungr, 2017). However, to model a important to the flow dynamics.
debris flow, a typical simulation involves tens of thou- The main limitation of DEM at the current time is that
sands up to millions of computational particles. Even with it is possible only to model the interactions between
these numbers of particles, the model is not simulating all particles, so it is unable to model the interaction between
of the particles in the flow, rather the idea is to model the particles and interstitial fluids like water and mud, and the
aggregate behaviour of the flow, which is dominated by resultant pore pressure effects. Thus they can model only
the interaction between the particles. Early DEM studies dry flows (i.e. the Iverson et al., 2014 data in the previous
were limited to two dimensions (e.g. the downslope dir- paragraph), while wet flows and snow avalanches (for the
ection and depth), circular particles and relatively simple latter air entrainment is important) cannot be modelled.
interaction dynamics.
Many DEM models follow the approach of Cundall
and Strack (1979). They modelled a two-dimensional, 13.5 Escarpment Retreat
slowly moving, geomechanics problem with a small
number of different size circular discs. The interparticle One recurring topic upon which LEMs have been applied
forces were modelled by a damped spring, as were the is escarpment and cliff retreat. The questions around this
tangential frictional forces. These interparticle spring topic are (1) what processes operate on or near the escarp-
forces were active only when the particle discs over- ment and how do they determine that an escarpment is
lapped. The timestep used for the solution must be small maintained and (2) what determines the rate of retreat of
enough to simulate the gradual overlap of the particles and an escarpment. The intent of this section is not to review
the subsequent development of the interparticle normal all the work in this area, but to discuss how the process
and frictional forces, which results in the method being formulation has been found to have profound implications
very compute intensive. Campbell et al. (1995) extended for whether cliffs are developed and/or maintained over
this approach to a highly dynamic situation by modelling the long term.
a debris flow. Johnson et al. (2016) used Campbell’s code Tucker (1994), Kooi and Beaumont (1994) and
to examine the enhanced runout with increased size Howard et al. (1994) carried out a series of one-
dependence of debris flows and was able to successfully dimensional simulations (a long profile that included a
model the mechanism. They attributed this behaviour to combination of plateau, cliff front and scree slope) using a
acoustic fluidisation, which results from the dynamic variety of river erosion formulations (see Chapter 4) to
interaction of the particles. model the evolution of the landform. The conclusion these
Computations of interparticle forces are significantly papers reached was that the stream power detachment-
simplified if only circular or spherical particles are limited model did a better job of maintaining the vertical
allowed. Accordingly many studies have adopted circular cliff face than the transport-limited model. Figures 13.10
or spherical particles. One way to model irregularly and 13.11 show similar simulations for similar starting
shaped particles is as aggregates of these spherical par- landforms for the range of parameters that are expected in
ticles. This is a common technique for rockfall analysis, as the field. All mechanisms show some flattening of the
rock shape has been shown to strongly influence trajec- initially (near) vertical cliff, but the detachment-limited
tory calculations as a result of the interaction between landforms show something that looks like parallel retreat,
FIGURE 13.10: Cliff retreat modelled with the stream power detachment limitation equation (Equation (4.29)) for the parameters indicated
and τ c ¼ 0. The heavy line is the initial hillslope profile, and the end time for all the simulations is when the average elevation of the landform is
6.0, a convenient number that allows comparison of the final hillslope profiles with Figures 13.11 and 13.12.
whereas the transport-limited results show that flattening decrease as the cliff approaches the catchment divide
of the cliff dominates. The detachment-limited cases for (i.e. x decreases).
n ¼ 1 are particularly interesting because they are of the Much of the smoothing of the cliff profile with time in
form of a pure advection equation with no diffusion, so Figure 13.10 is a result of numerical diffusion in the finite
that for our one-dimensional example differences used to solve the detachment-limited equation.
Lest the reader think that this smoothing in Figure 13.10 is
∂z ∂z
¼ βQm S ¼ βxm (13.47) a result of defective numerics, the algorithm used for the
∂t ∂x
calculations is identical to that used in most existing
where x is the distance from the divide (the left-hand landform evolution models, with the slope term being a
boundary of the figure). This is the classical advection forward finite difference (i.e. the slope is that slope from
equation where the velocity of movement of a disturbance the current node to the next node downstream as done in
in z (in this case the cliff line) is βxm where the sign D8) and the timestepping is explicit Euler. To further
indicates the movement is from right to left in confirm that this smoothing of the cliff face is numerical
Figure 13.10. Thus provided that the cliff remains vertical diffusion, Figure 13.12 shows a pure advection solution
(so that the top and the bottom of the cliff are both at equal with constant advection velocity to the equations (m ¼ 0,
distance from the divide), then the cliff will retreat at a n ¼ 1 in Equation (13.47)), and this smoothing of the cliff
velocity βxm and the cliff profile will NOT change with face is still evident and of about the same magnitude as
time, only retreat, and the parallel retreat speed will seen in Figure 13.10. Accordingly it is reasonable to
FIGURE 13.11: Cliff retreat modelled with the transport limitation equation (Equation (4.10)) for the parameters indicated. The heavy line is
the initial hillslope profile, and the end time for all the simulations is when the average elevation of the landform is 6.0, a convenient number that
allows the comparison of final hillslope profiles with Figures 13.10 and 13.12.
assume that at least part of the smoothing of the cliff

face that Tucker, Kooi and Howard observed in their
detachment-limited simulations was a result of numerical
diffusion.
Kooi and Beaumont (1994) also coupled their erosion
with an elastic plate to model isostatic rebound (see Chap-
ter 12). They found that the unloading of the plate as a
result of erosion led to the uplifting of the plateau upslope
of the cliff and warping so that flows upstream of the cliff
reversed direction and flowed away from the cliff face. As
a result, erosion of the cliff face was eliminated (since the
catchment divide had moved to the cliff face) and the cliff
face becomes stable. The cliff face only retreated as a result
of erosion and undercutting on the slope downstream of the
cliff face (for further discussion see Section 16.1.3).
FIGURE 13.12: Cliff retreat modelled with a pure advection equation The alternative simulation method for cliff retreat is to
(equivalent to detachment-limitation m ¼ 0, n ¼ 1). The heavy line is the
assume that the cliff face is more erosionally resistant than
initial hillslope profile, and the end time for all the simulations is when the
average elevation of the landform is 6.0, a convenient number that allows the geology upstream and downstream of the cliff face
the comparison of final hillslope profiles with Figures 13.10 and 13.11. (Weissel and Seidl, 1997). Figure 13.13 shows
FIGURE 13.13: Cliff retreat modelled with the detachment limitation equation (Equation (4.29)) for the parameters indicated and where the
cliff face has a detachment rate 0.1 times that of the material below and above the cliff face. The heavy line is the initial hillslope profile, and the
end time for all the simulations is when the average elevation of the landform is 6.0, a convenient number that allows comparison of the final
hillslope profiles with Figure 13.10.
simulations for a case where the cliff erodibility is 10% of

the material above and below the cliff face, and this
clearly shows the expected result of the maintenance of
the vertical cliff face, and that the numerical diffusion,
while still present, is also reduced.
The persistence of numerical diffusion in the simula-
tions results suggests that custom numerical solvers
might be advantageous for modelling cliffs. The main
problem is that cliffs are vertical or near vertical and that
the retreat occurs horizontally, and both of these proper-
ties are hard to resolve with a (fixed) horizontal grid of
land elevations.
One solution would be to model the cliff as an
explicit discontinuity in elevations and to move this FIGURE 13.14: A schematic showing the mass balance for
processes acting on a scree slope downslope of a stable retreating
discontinuity in elevations laterally across the grid with
cliff face.
the rate of movement determined by some retreat mech-
anism. The physics that control the retreat mechanism
would need to be parameterised. If the landform evolu- what stage does a cliff become steep and high enough
tion model uses a TIN with nodes that can move in time to require special treatment with a ‘cliff model’ as
(e.g. the CHILD landform evolution model), this discon- outlined above.
tinuity could be the boundary between the triangles of A simple conceptualisation of cliff retreat can be
the TIN so that the cliff face is modelled as occurring constructed for a cliff with a scree slope below. For cliff
along the boundary of the TINs, and as the cliff moves faces with scree slopes underneath them, the retreat rate
the triangle boundaries could also be moved. Such dis- is likely to be a balance between the clearance rate of
continuities along TIN boundaries are a common tech- scree at the base of the slope (e.g. by a river) and the
nique in structural and geomechanics finite element rate of collapse of the cliff face (Figure 13.14). If
codes. The author’s gridded SIBERIA model likewise the rate of removal of scree by the river is higher than
uses discontinuities in the grid to model contour banks the rate of retreat of the cliff, then eventually the river
(a discontinuity in flow directions, albeit stationary) and will remove all of the scree. If the rate of removal is
channels (a discontinuity in elevations based on where lower than the rate of cliff collapse, then the scree slope
channels occur) but is otherwise limited by its stationary will aggrade, push the river toward the other side of the
computational grid. The challenge is then determining at valley and eventually cover the cliff. Thus cliff retreat
rates may be a balance between the cliff source term and 13.6 Further Reading
the river removal term. Rather than being the result of a
lucky coincidence, this balance might be achieved by (1) For energy dissipation and fragmentation resulting from
if the river removal rate is too high, then the cliff face rockfall and collisions, there is a body of work in the asteroid
will be undercut increasing the rate of collapse due to and comet collision community that, while it may not be
instability of the cliff face and increasing the retreat rate directly relevant, may be of interest for this research com-
and (2) if the river removal rate is too low, then the munity’s different approaches (e.g. Durda et al., 2004,
scree aggrades covering part of the cliff, reducing the 2007). For greater detail on the theory and experimental
cliff face instability and weathering, stabilising the cliff evidence for debris flow processes, readers are referred to
and reducing the retreat rate. the extensive body of work by Iverson and his colleagues.
14 Vegetation and Wildfire
14.1 Introduction fungal/mycorrhizal processes on soil weathering and large

woody debris in rivers. The lack of coverage here is not to
This chapter is about the role of vegetation in the evolu- deny their importance. The lack of treatment simply
tion of landforms and soils, and the potential feedbacks acknowledges the lack of widely applicable models for
between the three. The chapter covers (1) vegetative them. See Viles et al. (2008) for a conceptual discussion
effects on fluvial erosion rates, (2) vegetation and climate of some of these processes.
feedbacks including the development of patterned vegeta-
tion and hillslopes in arid zones and (3) wildfire. We have
touched on other vegetative effects elsewhere in the book 14.2 Landforms, Soils and Vegetation
including (1) tree throw and its role in soil depth evolution
(Chapter 5) and soil mixing (Chapter 6) and (2) the role of Landforms, soils and vegetation are intimately linked, and
leaf litter as a source material for soil organic matter this linkage has significant impacts on other important
(Chapter 9). environmental processes. Models for exploring these link-
The chapter will touch on the broader issues of eco- ages and impacts are currently a work in progress, so this
systems dynamics and dynamic vegetation models. Typ- chapter focuses on the individual components of vegeta-
ically these models have been developed for other tion systems, with some examples of impacts from field
purposes such as climate change impact assessment and and usage in models where appropriate.
ecohydrology applications, and while they consider a As an example of these links, Peterman et al., (2014)
changing climate, they mostly assume fixed soils and examines the correlation between soil depth (Chapter 6)
landforms. This chapter will not provide a comprehensive and soil carbon (Chapter 10) and found a strong correl-
review of these models because that would require a book ation. The underlying driver of the correlation was that
in its own right (e.g. for ecohydrology see Eagleson, 2002 deeper soils provide greater water-holding capacity, and,
or Rodriguez-Iturbe and Porporato, 2005). Rather we will at least in water-limited regions, this increases the rate
examine these models to assess their suitability as models of production of vegetative biomass. As discussed in
in a coupled soilscape-landscape evolution model. Chapters 3 (Hydrology) and Chapter 11 (Soilscape
In any discussion of vegetation fire needs to be con- Modelling), the distribution of both water and soils is
sidered. In many places in the world fire is an integral part driven by the landform. Increased biomass is an
of the ecosystem function, including vegetation, and it is increased carbon source term in the models for soil
an important process impacting on long-term average soil carbon in Chapter 10 so soil carbon increases. Thus
erosion rates and soil organic matter mass balances. In the vegetation is the key link underpinning the empirical
context of soilscape and landform evolution research, it observation by Peterman and colleagues of a link
has often not been accorded the profile it deserves, and the between soil depth and soil carbon. They also observed
discussion here aims to emphasize the impact of fire, and a link between soil depth and fire, and as we will see
introduce techniques that might be suitable to study its later in the chapter, this also arises because of the vege-
impact. tation biomass growth. Finally, Peterman and Bachelet
We will not talk about some biotic factors that undoubt- (2012) note that soil nitrogen is a product of a biogeo-
edly influence landform and soilscape evolution such as chemical cycle driven by vegetation, so without vegeta-
228 animal effects (e.g. grazing, beaver dams), biocrusts, tion soil nitrogen has been found to decline sharply.
14.2 Landforms, Soils and Vegetation 229
14.2.1 A Simple Model for the Effect of Vegetation

on Erosion
A very simple way to allow for vegetation effects on

landform evolution is for the user to define the vegetation
and its characteristics. In this way the effect of vegetation
on erosion rate can be estimated using one of the many
ways that vegetation is accounted for in existing agricul-
tural erosion models. For instance, for the Universal Soil
Loss Equation (USLE) Table 10 from Wischmeier and
Smith (1978) (Figure 14.1) gives tabulated values for the
ratio of the hillslope fluvial erosion rate with and without
vegetation, called the cover factor, C. To allow for vege-
tation the erosion for unvegetated conditions is multiplied
by the cover factor for the vegetation, C, to calculate the
erosion rate with vegetation. This table explicitly
accounts for two factors: (1) the reduction of rainsplash
detachment as result of the canopy intercepting raindrops
(see Section 4.2) and (2) the reduction in the shear stress
applied to the soil surface, and thus erosion by overland
flow, as a result of the groundcover protection by grasses
(see Section 4.2.3). What is not clear from the table is that
it also implicitly includes the effect of vegetation on
infiltration and thus the reduced runoff that occurs for
highly vegetated areas. The RUSLE2 (Revised USLE,
Version 2; USDA-ARS, 2008) provides an explicit equa-
tion for calculating the cover factor that includes (1) the
canopy effects, as above, (2) the groundcover effect, as
above, (3) surface roughness, (4) ridge/furrow height, (5)
soil biomass, (6) soil consolidation, (7) water ponding
and (8) antecedent soil wetness. These factors can also
change over time, acknowledging, for instance, that it is
possible for peak rainfall/runoff to occur at times when
vegetation provides relatively poor cover. Vegetation
contributes directly to factors (1) and (2), and indirectly
to factors (5) and (8). The remainder are largely agricul- FIGURE 14.1: Cover factors for the relative impact of vegetation
cover on erosion rate. Note the distinction between the relatively
tural management factors. The cover factor is the multiple
small effects of canopy versus the strong effects of ground cover. The
of the eight factors above, but since this section is about differentiation between G and W ground covers reflects plants
vegetation, then if we lump all the non-vegetation factors where the canopy is strongly attached into the soil surface and thus
together, then the cover factor is provides strong protection against flowing water (G) versus plants
where the canopy simply lies on the surface and is easily moved or
C ¼ Ccanopy Cgroundcover Cremaining factors 3-8 detached from the soil surface by flowing water (W) (from
Wischmeier and Smith, 1978).
Ccanopy ¼ 1 f c e0:33hf (14.1)
bfg
Cgroundcover ¼ e
of the canopy (in metres), f c and f g are the fraction of the
where C is the cover factor where the canopy and ground- plan area covered by canopy and groundcover, respect-
cover effects are made explicit, and for the presentation ively, and b is a parameter in the range 0.025–0.06. Thus
here the remaining factors are simply lumped into one the reduction in erosion due to the canopy is linear with
factor (in RUSLE each of the remaining six factors also the proportion of cover, while the reduction due
have equations for their corresponding C factor, but they to groundcover is nonlinear with the maximum rate of
will not be discussed here). The parameter hf is the height reduction with increasing groundcover occurring for low
groundcovers. In the USLE (Figure 14.1) the cover and small term determined by taking the difference between
groundcover effects are lumped together in the C factor two large numbers (the total shear stress, and the resist-
for the erosion calculation. In our LEM calculations, if we ance to flow per unit area from the vegetation).
model rainsplash and rainflow separately from overland Foster (1982) and Foster et al. (1995) proposed a
flow fluvial erosion, then the Ccanopy term should be simpler version of Equation (14.2) that doesn’t explicitly
applied to the rainsplash transport term and Cgroundcover include the drag term, called the shear stress proportioning
to the fluvial erosion term. While the details of the vege- fraction, where
tation effects on erosion vary slightly between the various
fs
erosion models, the general principle of the cover factors τf ¼ τ0 (14.3)
ft
of the form in Equation (14.1) is commonly used world-
wide and is empirically well founded on field data. The where τ f is the shear stress acting on the soil surface, and
manuals for the USDA agricultural erosion models USLE fs and ft are the roughness of the bare soil and vegetated
(Wischmeier and Smith, 1978), RUSLE2 (USDA-ARS, surface, respectively.
2008) and WEPP (Laflen et al., 1991) provide more detail. While the formulations in Equations (14.1)–(14.3) are
It is important to note that these models have been different in detail, they all partition the shear stress
developed for application to agricultural areas subject to between that applied to the bare soil between the vegeta-
low gradient fluvial erosion, and so do not represent tion, and that applied to the vegetation. They all then
processes that primarily occur on steeper slopes including calculate the detachment from the soil by erosion as a
rilling, gullying and debris flows. function of the shear stress applied to the bare soil frac-
The form of the groundcover correction is exponential, tion. Moreover, all formulations have the steepest reduc-
and the reason for this nonlinearity requires a brief explan- tion in bare soil shear stress for low percentage cover by
ation. If the sole effect of groundcover was to protect the vegetation with the rate of reduction declining as the
surface by covering it, then the reduction with cover percentage cover increases.
would be linear from 1.0 at 0% cover to 0.0 at 100%. The main conclusions from this section is that if the
However, vegetation also reduces the shear stress action characteristics of the vegetation are known, then it is
on the soil surface by changing the soil surface roughness. possible to estimate the effect of vegetation on erosion
The bottom shear stress equation (i.e. τ 0 ¼ γyS for a wide by using the correction factors from agricultural erosion
channel) is derived by a force balance between that pro- models. Perhaps the main problem with using agricultural
portion of the self-weight of the water acting downslope erosion models is that they are heavily (though not com-
(the right-hand side of the bottom shear equation) and the pletely) focussed on agricultural applications, so the data
resistance to the flow provided by the bottom roughness for the cover factors are biased toward annual crops and
(the left-hand side of the bottom shear equation). In the agricultural management practices (e.g. ploughing).
case of vegetated surface there is the drag force from the Accordingly, given the focus on agriculture, much of the
flow acting on the stems of the groundcover, which pro- data are not applicable to problems involving the model-
vides a further resistance to flow so that the force balance ling of the pre-human agricultural past, steep mountainous
equation is landscapes and geomorphic time scales.
Willgoose (1995) used the table in Figure 14.1 to
drag
1 f g τ0 þ f gτv þ ¼ γyS (14.2) estimate the effect of vegetation (relative to an unvege-
area
tated condition) on a rehabilitated mine waste rock
where τ v is the shear stress acting on the groundcover, and dump for 1,000 year landform evolution simulations
the drag force per unit area is a function of the number of (Willgoose and Riley, 1998a, b). The main uncertainty
plant stems per unit area, the average area of a stem at in this approach was the ability to measure or predict the
right angles to the flow direction and the velocity of the vegetation canopy and groundcovers. Willgoose had
flow squared (Lopéz and Garcia, 1998). The drag on the measurements of these from adjacent natural field sites
stems results in a reduction in erosion over and above the and assumed that these were constant for the 1,000 year
groundcover effect (the second term in Equation (14.2)) design life of the containment structure (despite the land-
leading to the exponential decline. Nepf (2012) notes that form and soils being predicted to evolve significantly over
while Equation (14.2) is useful to explain the effect of the 1,000 years). Subsequent field experiments approxi-
vegetation, it is a poor tool for experimental determination mately validated the use of USLE cover factors for
of the soil bottom shear stress because the bottom shear current-day conditions (Evans et al., 1996; Evans and
stress term (the first term in Equation (14.2)) is typically a Willgoose, 2000) but was not able to test the assumption
of stationarity of the vegetation or the potential impact of to model not only the cycles of nutrients, climate and
fire over 1,000 years. water but also the evolution of the vegetative ecosystem
Temme and Veldkemp (2009) in their LAPSUS LEM that responds to those cycles so that the vegetation char-
also adopted a cover factor but where the erosion rate was acteristics (e.g. light use efficiency (LUE), water use
a linear function of vegetation cover. Relative to the efficiency (WUE), species structure, proportion of under-
USDA factors this approximation overestimates erosion storey to over-storey canopy) evolve in response to those
for midrange vegetation covers. They estimated vegeta- cycles. These cycles and the vegetative response are quite
tion cover as a function of soil thickness, and the rainfall complex and involve many processes (Figures 14.2 and
and temperature relative to averages over the last 50,000 14.3). In general they have not been fully characterised.
years. If soil was less than 0.5 m, then the vegetation was We will only be able to scratch the surface of these
reduced linearly by the soil thickness so that at 0.0 m soil processes in this section. There is considerable interest
thickness the vegetation cover was 0. In Section 14.4 we in dynamic vegetation models to model the terrestrial
will discuss models that exhibit a similar relationship with carbon cycle in global climate models (e.g. Cramer and
soil depth where the amount of water stored in the soil Field, 1999; Quillet et al., 2010), but the complexity of
profile limits the rate of growth, and therefore vegetation these models is beyond what will be discussed below,
density. even though the principles are similar.
Yetemen et al. (2015a) use a conceptually similar We start by outlining a simple mathematical frame-
approach to Equation (14.1), but they used it to predict work within which the DVMs discussed in the remainder
the effective shear stress applied to the soil surface rela- of the chapter can be viewed. Almost all DVMs and
tive to an unvegetated surface (functionally equivalent to DGVMs model biomass, and if they are interested in
a cover factor) as a function of LAI. They used an empir- factors that are related to biomass (e.g. vegetation cover),
ical function to relate the LAI to the percentage cover of they use empirically derived relationships or another
the surface. Like Equation (14.1) their function had the model to estimate them. Some models simulate carbon
greatest rate of erosion reduction for low vegetation dens- rather than biomass because they have a compelling inter-
ity and groundcover. est in the carbon cycle, though as indicated in Chapter 10
In cases with no ground-based data there has been a there is a good empirical relationship between carbon and
body of research, with mixed success, that has tried to biomass. The discussion below uses biomass per unit plan
estimate the canopy and groundcover directly from area as the property to be modelled. Conceptually the
remote sensing (Vrieling, 2006), or by correlating esti- biomass dynamics at a point can be formulated as a mass
mates of aboveground biomass with cover (e.g. Schön- balance equation
brodt et al., 2010). In the absence of defensible remote
∂B
sensing correlations it is very common to simply use ¼PRDHF (14.4)
∂t
tabulated values relating cover factor with cropping type
(e.g. Panagos et al., 2015). where B is the live biomass (mass per unit area), P is the
Finally it is important to note that the cover factor growth rate of live biomass, R is the respiration loss (mass
conflates the combined effect of vegetation on erosion per unit area per unit time), D is the death rate of live
resistance and hydrology. Thus the cover factor biomass and H and F are the biomass loss caused by
decreases with increasing vegetation cover because of herbivores and fire, respectively. We need to make a
(1) the protection of the vegetation, (2) the increased distinction here between live and dead biomass because
infiltration rate and (3) the reduced soil moisture at the while the total biomass is important for many processes
start of a rainfall event (Kinnell, 2010). In LEMS that (food for herbivores), only the live biomass grows, so D is
model erosion resistance and runoff generation separately, the conversion rate of live biomass (e.g. plant leaves and
in adjusting for vegetation it should be noted that the C flowers) to dead biomass (e.g. leaf litter and seeds). We
factor captures the combined effect of vegetation on both will return to modelling dead biomass shortly.
hydrology and erosion, not just the erosion. A simple biomass production equation is

B
P ¼ αB 1 ∗ (14.5)
B
14.3 Dynamic Vegetation Models
where B∗ is the maximum biomass possible, and α is the
Models of dynamic vegetation (DVM) (and their close rate parameter for biomass growth. For low biomass this
relatives, dynamic global vegetation models, DGVM) aim term is linear in B, and this reflects that for scattered trees
FIGURE 14.2: A summary of the energy, water, carbon and nutrients for a forested ecosystem. Note that the picture simplifies the typical
modelling of canopy and rhizosphere processes (see, e.g. Figure 14.3) since many models (1) discretise the processes with elevation above/
below the soil surface so that process rate varies with elevation, (2) allow a number of species to coexist and evolve as a result of competition
for resources and (3) differentiate the canopy and rhizosphere properties of these species (e.g. grasses and shrubs versus trees) (from Fatichi
et al., 2016).
(a) (b) (c) FIGURE 14.3: This is a

Ta Ta Ta schematic of some of the
approximations used to model
ra ra raC evapotranspiration in
ecohydrology and land surface
rsC models. T a is the total
rs raC rsC evapotranspiration, and the
raS
various r’s are the resistances to
S
ra flows across the atmospheric
gradient in water content: (a) bare
rs S rsS soil with no vegetation, (b) the
Trad Trad single layer vegetation canopy
which shades the soil, (c) where
One-source model Two-layer model Two-patch model the soil and vegetation are
independent, (d) a mixture of
(d) (e) (f) cases (b) and (c), (e) multiple
Ta Ta Ta independent patches of vegetation
(each patch potentially different
ra,a1 ra,a1 ra,a1 species) and (f ) a multilayer
canopy (from Zhang et al., 2016).
ra,C3
ra,C2
ra S ra,i ra,C1 rs,C3
S
rs,C2
ra
rs,C1
rs S rs S
Hybrid dual-source model Multi-patch model Multi-layer model

and grasses there is little interaction, synergy or competi- dV

¼ k g T V ð1 V Þ km V (14.6)
tion. For an individual leaf that has photosynthesis (i.e. dt
biomass production) that is light limited, then the rate of where V is the proportion of vegetation cover (in the range
biomass production is proportional to the leaf area (i.e. 0–1 as defined for Equation (14.1)), T is the transpiration
amount of biomass that is leaf area), and the constant of rate and kg and k m are the rates of growth and death for the
proportionality is the light use efficiency. vegetation cover. This formulation using T makes the
The term in parentheses in Equation (14.5) is an growth proportional to the water usage of the plant (in
empirical correction that reflects that as the biomass dens- water-limited conditions water use and carbon assimila-
ity increases competition, effects become more important, tion are very close to linearly related).
particularly competition for light (leaves higher in the Respiration is the process whereby the plant converts
canopy shade leaves lower in the canopy, the light extinc- the products of photosynthesis into energy for plant
tion curve) and water, and the biomass growth rate drops. growth and maintenance. Respiration typically consumes
Larcher (2003) indicates that once the leaf area index about 50% of the carbon assimilated by photosynthesis,
(LAI) reaches 4–6 m2/m2, competition for light inhibits though it can be as high as 75% for water-limited condi-
further growth. Ruimy et al. (1999) uses the light extinc- tions, slow-growing plants or plants toward the end of life
tion curve to estimate the amount of solar energy (Lambers et al., 2008). It is normally broken into two
absorbed, FPAR (the fraction of radiation absorbed by components: (1) the energy required to grow root and leaf
the plants in frequency band active in photosynthesis),
shoots (growth respiration) and (2) energy required to keep
as FPAR ¼ 0:95 1 ek:LAI where the light extinction the plant alive even in the absence of growth (maintenance
coefficient k was assumed to be k ¼ 0:5. This is consistent respiration). Typically the respiration is approximately
with Larcher as the maximum increase in photosynthesis evenly split between the two processes, but when growth
for an LAI greater than 6 is less than 5%. Empirical slows, maintenance respiration becomes a greater propor-
relationships between LAI and biomass are commonly tion of total respiration. Finally respiration operates (1) all
used in the interpretation of remote sensing of vegetation. the time while photosynthesis occurs only during the day,
The linear decline with B=B∗ in Equation (14.4) is com- so during the day carbon assimilation from photosynthesis
monly used, but other functional relationships are feas- dominates, while at night respiration and carbon release
ible. For instance, there might be a lower bound on the dominates and (2) in all parts of the plant, so that the roots
value of B below which the canopies of widely scattered are a net source of carbon dioxide to the soil atmosphere
plants do not overlap and thus don’t compete, so the term impacting on soil chemical weathering (Chapter 8).
inside the parentheses will be 1.0 below a small nonzero A simple equation that reflects the different mechanisms
B value. for the two processes is (after Sitch et al., 2003)
An alternative rationale for the form of Equation (14.5)
can be provided if biomass is considered equivalent to R ¼ r m B þ r g ðP rm BÞ (14.7)
surface coverage, so that if biomass doubles, so does
coverage. Consider a domain which is discretised into where r m and r g are rate parameters for maintenance and
cells where the cell is either vegetated or unvegetated, so growth respiration, respectively. Note that this equation
that B is proportional to the number of vegetated cells. As assumes that the first priority for the plant is to maintain
B increases, the number of cells that can be source for the existing plant biomass (i.e. maintenance respiration),
vegetation in other cells (e.g. seeds) increases, so that the and growth can occur only once maintenance require-
number of adjacent cells that can potentially switch from ments are met, so growth respiration uses whatever energy
unvegetated to vegetated increases. The term in paren- is left over after maintenance respiration. As the growth
theses characterises the number of cells that are available rate P decreases, growth respiration decreases. Mainten-
to switch. As the number of available unvegetated cells ance respiration is small when the live biomass is small.
decreases, then the actual rate of change drops as a func- Net Primary Production (NPP) is NPP ¼ P R and is
tion of that availability (e.g. seeds from a vegetated cell typically about 0.5P.
may fall into another vegetated cell, rather than an unve- The biomass death rate is generally modelled with a
getated cell) (Tilman, 1994). simple first order decay rate
A model based on this idea of vegetation cover in D ¼ δB (14.8)
pixels is possible. Collins et al. (2004) presents a model
for vegetation cover which has the same functional form where δ is the death rate, and the first order decay assumes
as Equations (14.4) and (14.5): that plant death rate is independent of the death rate of
surrounding plants. One complication is that the death rate needed to maintain themselves. Likewise if animals are
is a function of age and water stress, with older collections gaining weight, pregnant or lactating, then biomass
of trees (i.e. stands) having a lower death rate than requirements will be higher. A model including these
younger stands (van Mantgem et al., 2009). Water stress processes (Wainwright, 2008) is
will be discussed in Section 14.4; the death rate increases ε3
with stress. This may not be important in an undisturbed H ¼ ε2 þ þ ε4 þ ε5 pp þ ε6 pl N h (14.10)
B
area with a mixture of tree ages, but may have an impact
in an area recovering from fire or severe drought. More- where ε2 parameterises the maintenance energy require-
over, plants that have been historically subject to mild ment (ε0 in Equation (14.9)), ε3 the energy required to
water stress adapt and are better able to withstand a forage for food as the biomass density decreases, ε4 the
subsequent water stress event (called acclimation; energy required to gain weight (which will be related to
Lambers et al., 2008). Thus plant history may be import- the animal age), ε5 and pp the energy required for preg-
ant in determining plant death rate. Most models ignore nancy and the proportion of the animals that are pregnant,
these history effects. respectively, and ε6 and pl the energy required for lacta-
The herbivore consumption rate will be a function of tion and the proportion of the animals that are lactating,
the density of herbivores (both insects and animals), respectively.
which itself may be a function of the food (i.e. live and Finally fire removes both live and dead biomass by
dead biomass) availability, but may also be a function of burning it. Fire will be discussed in detail later in the
predation by carnivores. A simple equation ignoring car- chapter. If we confine ourselves for the moment to loss
nivores is (HilleRisLambers et al., 2001) of live biomass by fire, one way of representing it is as the
fraction of a pixel burnt in any one year so that
H ¼ ε1 B ðfood-limitedÞ
(14.9)
¼ ε0 N h ðdrinking water-limitedÞ F ¼ F B eB B (14.11)
where the linear relationship for food limitation assumes where F B is the fraction of the pixel burnt per year and eB
that if the food availability doubles so does the number is the efficiency with which the live biomass is burnt. In
of animals, ε0 is a rate term that encapsulates the amount fire-prone areas it is not uncommon for F B to be 5–10%
of biomass one animal consumes and N h is the number of (Price and Bradstock, 2011), and in areas with seasonal
animals and ε1 is the rate of consumption of biomass if dry seasons areas may burn every 2–4 years so that F B is
food is limiting on the herbivore population. Two cases 0.3–0.4 (Russell-Smith et al., 1997). These rates suggest
are listed: (1) food-limited where the number of animals is that loss of biomass by fire can be a significant component
limited by biomass availability and (2) drinking water- of the mass balance. In reality not all live biomass in the
limited where the number of animals is limited by the burnt area will be incinerated, so the efficiency eB will be
availability of water to drink. Note that the second case less than 1; it depends on the intensity of the fire. For low-
does not necessarily imply that the vegetation biomass intensity fires the trees and their canopy may be
growth is limited by water availability, only the drinkable untouched, and the only biomass burnt are shrubs and
water (ponds and creeks of moderate salinity water) for grass, while for high-intensity fires the canopy and a
the animals limits their numbers. If there are delays portion of the tree trunk and branches may also be
between biomass growth and herbivore population (e.g. removed. As will be seen later in this chapter, current fire
plants grow and in response the herbivore population models do not typically distinguish fire intensity but rather
grows but is delayed by the duration of pregnancy), the focus on burnt area, so it is difficult to quantify the
time-varying dynamics of this process can be complex, efficiency. Some models (e.g. LPJ, Sitch, 2003; CAT-
and herbivory and predation are often used as case studies GraSS, Zhou et al., 2013) are potentially better able to
of complex behaviour in nonlinear mathematical systems capture this intensity effect by breaking the modelling of
(e.g. chaos, fractals). biomass B into two components: structural biomass Bs
More sophisticated models of herbivory are possible, and nonstructural biomass Bl ; the former are the tree
since the rate of consumption of biomass by herbivores, trunk and big branches, the latter sometimes being called
ε1 , is not a simple reflection of the number of animals in green or live biomass, reflecting that it is the actively
Equation (14.9) and a fixed energy requirement per living part of the plant, and B ¼ Bs þ Bl . It is the non-
animal. For instance, as the biomass density decreases, structural biomass that will be most easily modified by
the animal will need to travel longer distances (and thus fire, with structural biomass being burnt only in the more
consume energy) to consume the amount of biomass severe fires.
Drawing on this discussion about live biomass we can biomass (HilleRisLambers et al., 2001), so Equation
write a mass balance equation for the dead biomass (leaf (14.3) becomes
litter, soil organic matter, char from fires and seeds) 2
∂B ∂ B ∂2 B
∂BD ¼ P R D H F þ DB þ (14.13)
¼ δB εD BD þ F B ðeBD B eD BD Þ E BD V B ∂t ∂x2 ∂y2
∂t
(14.12) where DB is the diffusivity with respect to biomass. While
straightforward, there are several conceptual limitations
where BD is the dead biomass (mass per unit area), and the
with this formulation. The first limitation is that the seeds
first three terms are the live biomass death rate, herbivory
form only a small portion of the biomass, and yet the
(assuming food limitation) and fire terms, respectively.
diffusive term applies to the entire live biomass, most of
Note that it is likely that ε1 > εD since herbivores have a
which is live plants which don’t spread or move. This is
preference to eat live biomass over dead biomass if live
easily fixed by setting DB ¼ es DS where DS is the diffu-
biomass is available. The fire term has a different effi-
sivity of the seeds (which can be derived from seed travel
ciency for burning dead biomass (since typically it has a
distances in the same way travel distances were used to
lower moisture content, see the fire section, so eD > eB ),
derive rainsplash diffusivity in Chapter 4) and eS is the
and there is also a term for live biomass that is killed by
proportion of the total living biomass that is seeds (5–20%
the fire and transferred to the dead biomass pool (e.g.
of NPP; Sitch et al., 2003). A more critical limitation
scorched leaves fall off trees and contribute to the post-
occurs when the main seed dispersal mechanism is ero-
fire leaf litter). The fourth term is the erosion rate by
sion. In this case seed movement is a function of surface
surface runoff and was discussed in Chapter 4. The bio-
erosion rates, and will be directional and downslope, with
mass removed by erosion is dominated by dead biomass
seeds continuing to move until they are deposited (Saco
(though there is a significant amount of live fungal matter
et al., 2007).
in the soil aggregates; see Chapter 10), so erosion of live
Implicitly included in the production term is reproduc-
biomass has been assumed to be zero in Equation (14.12).
tion and recruitment (i.e. seed germination). If recruitment
It is should be noted that dead biomass has a lower
for some reason is unsuccessful, then plant productivity
specific gravity than mineral matter, so it is easier for
will decline over time because new juvenile plants do not
erosion to transport biomass and the proportion of organic
grow to replace older plants that die. For biomass equilib-
matter in the eroded material is enriched relative to the
rium, then biomass death will be equal to biomass from
organic content of the soil surface (see the discussion of
juvenile plant growth. This suggests that biomass growth
the enrichment ratio in Chapter 4). The fifth term is the
as a result of recruitment R will be R ¼ δφB and that the
transport of dead biomass vertically down into the soil
biomass productivity term with an explicit recruitment
profile. This is the surface organic matter source/store in
term, at equilibrium, will be
the soil carbon models discussed in Chapter 10.
One process not included in Equation (14.12) is the
B
effect of sediment deposition on vegetation. Murray and P ¼ αB 1 ∗ δφB þ δφB (14.14)
B
Paola (2003) proposed a mechanism (they were examin-
ing braided rivers so the applicability to hillslopes is where the term in the square brackets is the biomass
unknown) that killed vegetation if the deposition or ero- productivity of existing vegetation, φ is the proportion
sion is high. They proposed a deposition rate of 0.5 mm/ of dead biomass that must be replaced by new plants
hour for 2 days (i.e. about 24 mm) killed vegetation, (e.g. leaf fall will naturally be replaced by the existing
which might be exceeded on hillslopes behind obstacles plant in due course, and doesn’t require a new plant to
during a severe rainfall event. replace the dead biomass; however, a plant that dies
The discussion above is by no means a definitive or a completely must be replaced by a new plant) and the
comprehensive representation of the processes, but it does right-hand term outside the brackets is the biomass prod-
provide a simple intellectual framework within which to uctivity of new vegetation (i.e. as resulting from
organise vegetation dynamics modelling at a single recruitment).
location. Another limitation of this simple biomass model arises
The most obvious missing component is the spatial if we return to the USDA USLE cover factors discussed in
coupling between two locations. Primarily this comes the previous section. The cover factors show that canopy
from the transport of seeds, either by erosion, wind, birds and groundcover operate in very different ways with
or animals. Some authors have used a diffusive term on respect to erosion protection. Canopy reduces the
rainsplash process by reducing raindrop energy, while is high enough (i.e. large N h ), then the equilibrium solu-
groundcover reduces shear stress applied to the soil sur- tion is negative, which is clearly unrealistic and an
face by overland flow and thus fluvial erosion rate. Thus undesirable mathematical feature if we are trying to
our biomass model needs to be able to distinguish bio- understand the behaviour of a landscape evolution model
mass that is groundcover from biomass that is canopy. that includes biomass. More sophisticated herbivore-
This opens up questions of competition between the biomass relationships and equations that include predation
species that form the canopy (e.g. trees) versus the species on the herbivores typically yield multiple equilibrium
that form the groundcover (e.g. grasses). This competition solutions and complicated dynamical behaviour for high
can occur in the form of competition for light (the trees herbivore loadings (e.g. Konno, 2016). This is a limitation
shade the grass and thus reduce grass photosynthesis and of the herbivore formulation provided above and which
thus grass growth), or water (tree roots penetrate deeper may be important if we desire to model the effect of
into the soil and can thus access deeper water not avail- vegetation on erosion in the presence of high herbivore
able to grasses, but the shallower grass roots can intercept loadings. Fortunately for most ecosystems the average
infiltration before it reaches these deeper depths). We will rate of herbivory is low, 1–20% of the biomass per year,
also see later that the water limitation and fire response of and Equation (14.8) is a good first approximation.
canopy and groundcover are different. However, while these equations are formulated for
If the biomass production term in Equation (14.5) is average conditions, this does not preclude the possibility
written explicitly in terms of the light or water availability, that the complicated temporal dynamics of the more
then the biomass is distributed between aboveground (i.e. sophisticated relationships may generate extremes (e.g.
canopy) and belowground (i.e. roots). Equation (14.4) still drought and overgrazing) that result in significant loss of
describes the dynamics of the total biomass, but some of vegetation cover and increased cumulative erosion. More-
the right-hand side terms are different for aboveground over in highly seasonal climates such as monsoonal, there
versus belowground biomass (e.g. some herbivores prefer may be strong annual cycles of biomass growth (e.g.
aboveground biomass, others dig belowground). Enquist during wet seasons) and destruction (e.g. fire during dry
and Niklas (2002) found that the ratio of aboveground seasons). Even outside these strongly seasonal climates it
biomass to belowground biomass is approximately con- is common to see different species of grasses being dom-
stant, and that BAGB ¼ 3:88BBGB 1:02 across a variety of inant at different times of the year (e.g. winter grasses
mature forest species. With a larger dataset Cheng and versus summer grasses). Seasonal sequencing of rainfall
Niklas (2007) found that the ratio varied and was in the and species dominance may be important in determining
range 2–4 depending on forest species. There is consider- the protection groundcover provides during high erosion
able scatter in this ratio because during drought plants potential periods. For instance, in monsoonal regions it is
typically allocate more new biomass belowground to common for vegetation to burn during the dry season so
create roots so that the plant can access more soil water. that there is no vegetation protection against erosion
While not explicitly stated above, these equations are during the early part of the wet season (until the grasses
formulated for temporally averaged behaviour. Thus the have grown as a result of the rainfall), and thus annual
effects of normal climate variability are not explicitly erosion is strongly influenced by the first few months of
included in this formulation, but are implicit in the par- the wet season when vegetation cover is regularly low.
ameters in the processes (i.e. the parameters are effective One remaining question about this formulation is
parameters at the time resolution of modelling). We could under what circumstances the various biomass destruction
introduce a time-varying biomass production term that processes are dominant and/or important and/or insignifi-
responds, for instance, to seasonal variations but the cant. As noted above natural undisturbed herbivory nor-
response of the system to changes in forcings are imme- mally removes 1–20% of the biomass per year. We will
diate, without any storage delay or lags in response. discuss fire in detail in a following section, but typically
The equilibrium solution for B using Equations for forest 5–10% is burnt per year (i.e. fires every 10–20
(14.4–14.11) is years), with up to 30% in some notably fire-prone regions
(e.g. seasonally dry monsoonal climates). If we adopt a
B∗
B¼ ðα r m Þ 1 r g δ ε1 F B e B (14.15) midrange value for herbivory and fire of 10% each, then
α 1 rg
to be of equal magnitude, the death rate needs to be 20%,
which indicates that there is one stable solution for bio- which corresponds to an average lifespan of biomass of
mass at equilibrium. This solution highlights one concep- five years. While this lifespan is long for grasses (which
tual problem of this formulation, which is that if herbivory may be as low as one growing season, that is, one year), it
is short for forests where average tree lifespans are meas- • They assume fixed soil properties: soil depth, water-
ured in decades or even centuries (van Mantgem et al., holding capacity, grading, SOC, chemistry and pH.
2009). We can thus reasonably assert that (1) for forests Changes in these properties may or may not be import-
tree death is not dominant and other processes likely ant depending on the time scales of interest. For
dominate the loss of live biomass, (2) for grasses seasonal instance, depth, water capacity and grading may not
growth and death factors such as water availability might change significantly at the decadal time scale unless
be dominant and (3) herbivory and fire are likely to be erosion is particularly high (e.g. the author has seen
important in many locations. Spanish olive groves on steep lands where erosion is
The discussion above simplifies processes consider- high enough that significant soil depth loss has
ably and does not include some of the complexities occurred), but SOC and chemistry and pH can change
built into existing DGVMs. Some of these excluded quite quickly.
processes include (Quillet et al., 2010) (1) changes in • The crop models are highly specialised, focussing on
plant functional type (i.e. ecosystem changes, such as crops of economic value, and so typically are not opti-
changes in the type of forests or grasses) in response to mised for rangeland, grazing and natural lands. In
evolving climate, soilscape and landform, (2) nutrient particular, by focussing on annual crops they do not
competition and nutrient cycling, (3) internal plant typically deal with recruitment and reproduction, nor
chemistry and (4) response to changes in temperature with competition. This is not true of all models, with
and atmospheric CO2. APSIM (Agricultural Production Systems sIMulator)
having a rangeland option that allows a mixture of
annual and perennial species (Johnson, 2008).
14.3.1 Agricultural Crop Models • In most cases fire is ignored.
There is an emerging interest in applying LEMs to agri- Other aspects of these crop models are more sophisti-
cultural settings either to look at the historical long-term cated than what has been discussed here including the
impact of agriculture on soils and landforms (Schrool and following:
Veldkamp, 2001; Baartman et al., 2012) or to investigate
• Nutrient cycles are modelled included N, P and
the future impacts of agriculture management practices C cycles. While the soil geochemistry is fixed, the
(Barreto et al., 2013; Fleskens et al., 2016). LEMs have nutrient cycles are dynamic responding to the geochem-
also been applied at mine sites with a range of different ical reactions, and sources and sinks of the nutrients.
surface conditions and surface treatments, though to date
• Very sophisticated crop physiology parameterisations
they have not been coupled to vegetation models (Han- might be difficult to predict for an evolving ecosystem.
cock et al., 2008). Competition, if it is modelled at all, is modelled very
LEMs provide a number of conceptual advantages simply.
over traditional erosion models including explicit gully/
• It is common to use daily climate data.
rill erosion (many traditional models have gully/rill cap-
abilities but typically the properties need to be input, One crop model, widely used internationally, is
rather than develop naturally out of the landform evolu- APSIM (Keating, 2003), which has been developed by
tion), soil depth changes due to erosion and the dynamics amalgamating a number of previous models customised to
of erosion and deposition for erosion protection measures specific crop types. In particular its pasture model is
(e.g. sediment deposition in the drainage line behind derived from the SGS model (Johnson, 2008), and the
contour/graded banks that result in the drainage line even- points above capture its structure. It is a complex model
tually choking and failing). It is common for traditional and includes extra complexity to do with farm manage-
erosion models to be coupled with a crop model to allow ment and irrigation. Many crop model users recognise that
the erosion model to assess erosion through the growing the complexity in most crop models leads to ‘low science
season. These crop models vary in complexity from (1) transparency’ (Wang et al., 2002), which is a shorthand
simply imposing an assumed growth cycle through to (2) for saying that it is difficult to understand how the model
fully coupled agricultural productivity models. In leads to the answers obtained.
principle there is no reason that these same crop models Ramankutty et al. (2013) recognised that a simpler
cannot be coupled with a landscape evolution model. model based on APSIM would elucidate first-order
However, various considerations may be important, dependencies in the model operation and would have
including the following: value in other applications. They fitted regression models
to the output of APSIM for a wheat crop in a water- respiration functions in Equations (14.5) and (14.7)
limited environment and found that a model that used are conceptually the same as LPJ, but the mechanics
only mean rainfall, number of rainfall events/month, mean of the chemistry, water limitation and canopy effects are
radiation and temperature provided a satisfactory fit much more explicit in LPJ.
(R2 ¼ 0:9). Thus one use for the agricultural models is • A soil organic matter model is included in LPJ. It is a
in the development of reduced-order models where much single-layer three-compartment model as discussed in
of the complexity of the reductionist approach in the Chapter 10, with an aboveground litter pool.
agricultural models is removed. The most important com- • LPJ explicitly models leaf phenology (e.g. evergreen
plexity removed by Ramankutty was the daily time series and deciduous) with leaves responsive to budburst and
of rainfall, temperature and net radiation, which we do temperature.
not typically have and cannot generate reliably (see • C3 and C4 photosynthesis are explicitly modelled
Chapter 3), for landscape evolution applications. Brooks in LPJ.
et al. (2001) presented a similar simplification of the • Respiration follows Equation (14.7), but LPJ has the
SIRUS crop model for a site in England and found that respiration rate also responsive to the carbon:nitrogen
the best predictors of crop yield were soil water capacity ratio in the plant.
and temperature, though they did not attempt to simplify • Tree mortality occurs as a result of light competition,
the leaf area, water balance and transpiration models. low growth (e.g. respiration greater than photosyn-
thesis) and heat stress,
• LPJ uses monthly climate data, with some of the
14.3.2 Ecosystem Models monthly data being disaggregated to daily resolution.
The ecosystem community has developed models of

vegetation evolution that have been adapted by the global 14.4 Soil Moisture and Water Limitation
climate modelling community, who have used them to
estimate the terrestrial interaction with the atmosphere. The dynamic vegetation models of the previous section
Originally this interest was simply to model latent heat did not explicitly discuss the impact of water limitation on
(i.e. evapotranspiration) exchange at the land surface as plant growth. About 50% of the planet is water-limited at
this is the major energy interaction between the land and some point during the growing season(s). To model this
the atmosphere. In recent years, as noted in Chapter 1, process explicitly we need to model the evolution of soil
there has developed an interest in the carbon assimilation water content (SWC, measured as depth of water in the
capacity of the terrestrial biomass. These models now soil profile) over time.
model the evolution of and competition between different In Section 14.2 we discussed a simple model used by
phenologies and ecosystems and are collectively known Temme and Veldkamp (2009) that limited vegetation cover
as Dynamic Global Vegetation Models (DGVM). It is as a result of shallow soil depth and rainfall. While not
generally considered that the BIOME model (Prentice explicitly derived for water-limited environments, the net
et al., 1992) was the start of this class of model. effect of the model was to impose a water limitation effect on
The LPJ (Lund-Potsdam-Jena) model is typical of vegetation cover. In this section we look at a more explicitly
these DGVMs and draws heavily from BIOME. The physically based model structure, albeit more complex,
discussion below follows Sitch et al. (2003) (Figure 14.4). for the effect of water limitation on vegetation cover.
The biomass model above generally follows the principles
in LPJ though with a number of differences:
14.4.1 The Physical Principles
• LPJ does not distinguish live and dead biomass, but
does distinguish between leaves and sapwood and roots If we model the soil as a single layer, then a mass balance
(i.e. nonstructural versus structural biomass). equation for the change of soil water content, SWC (mm
• LPJ divides vegetation into 10 ecosystem types (called depth of water), per unit time for a pixel with unit plan
plant functional types, PFTs), which are generally dif- area is
ferent types of forest (e.g. temperate versus tropical,
∂ðSWCÞ
evergreen versus deciduous), and each PFT has a separ- ¼ I T E DD L (14.16)
ate set of processes and parameters. ∂t
• The mechanics of photosynthesis in LPJ are modelled where I is infiltration of water from the surface (in mm
explicitly at the daily time step. The production and depth per unit time), T is transpiration consumption of
FIGURE 14.4: Flow chart of the calculation modules in LPJ (from Sitch et al., 2003).
water by the vegetation, E is bare soil evaporation of evapotranspiration, PET. The PET is calculated from an
water from the near surface, DD is deep drainage of water energy balance at the surface with the energy supplied
from the bottom of the soil profile and L is net lateral being the net incoming radiation, and this energy being
transfer of water down the slope within the soil profile consumed as the latent heat of vapourisation of water. PET
(see Chapter 3 for more detail). The key variables here are has a feedback with topography via the aspect of the
(1) the transpiration, which is the water use by the vege- hillslopes, with more incoming energy per square metre
tation, (2) soil water content and (3) the infiltration from for slopes facing the sun (i.e. equatorial facing; south
the surface. We will consider these in turn, and then return facing in the northern hemisphere, and north facing in the
to how the overall ecosystem responds to water limitation. southern hemisphere). For midlatitude (i.e. 45N or S) the
In dry climates transpiration is proportional to carbon increase in energy per square metre at midday between a
assimilation and biomass growth. This is a result of the 10% slopes facing the equator and a horizontal surface is
operation of stomata in leaves, which open and close as about 23%, and everything else being equal this means the
part of the photosynthesis process. Stomata are the open- PET will also be about 23% higher on the 10% slope
ings through which carbon dioxide diffuses from the (Yetemen et al., 2015a, b).
atmosphere into the plant. They are also the main source A key component of the PET process is that the moist
for water loss out of the leaves, with water diffusing air needs to be transported away from the bare soil and/
outward from the air in the stomata into the atmosphere or the leaves (otherwise the air will become saturated
surrounding the plant. The rate of diffusion (of both water near the leaf/soil so that no more water can be evapor-
and carbon dioxide) is governed by the size of the opening ated), and this vertical transport up away from the surface
in the stomata: the bigger the opening the higher the into the atmosphere is driven by turbulence generated by
diffusion rate. For small openings the increase in transpir- wind, so PET equations (mostly based on the Penman-
ation is linear with the total open area of stomata, but for Montieth equation) typically also involve a wind term. If,
large openings there is interference between the stomata, however, water is limited, then there is insufficient water
and the rate of increase in transpiration with area to allow this evapotranspiration process to act unre-
decreases for large openings (Ting and Loomis, 1965). stricted, so evapotranspiration is less than PET and is
Since this is a diffusion process, then the net transport rate called the actual evapotranspiration, AET. It is worth
of water out of the stomata is a function of the difference noting that the excess energy (from the net radiation) that
in the humidity of the air in the stomata (typically satur- is not used to vapourise water turns into heat energy and
ated) and the air around the leaf, which for water-limited temperature rises (see Kalma et al., 2008 and Chapter 3
environments is typically low (mainly because in water- for more detail). The key concept here is that as the soil
limited conditions the size of the stomata opening is becomes drier it becomes more difficult for the plants and
partially closed to reduce water loss). Some of the water soil to transpire at the maximum rate. A common
in the leaves is used in the transformation of carbon approximation that is used is (Rodriguez-Iturbe and Por-
dioxide into carbohydrates, but the water balance is dom- porato, 2005)
inated by the water loss out of the stomata. Since carbon
SWC SWCmin β
dioxide is the source material for biomass production, this AET ¼ PET (14.18)
SWCmax SWCmin
means that water use by the plant is proportional to
biomass production so that a good approximation to bio- where the subscripts min and max are the minimum and
mass production is maximum amount of water that can be held in the soil,
and β is a parameter that is commonly assumed to be 1 but
∂B
¼ μT (14.17) is sometimes larger (Walker et al., 2001). It is common
∂t
to assume that SWCmax ¼ ðsoil porosityÞðsoil depthÞ,
where μ is the coefficient relating biomass production per and SWCmin ¼ ðfield capacityÞðsoil depthÞ for bare soil
unit area to transpiration. evaporation and SWCmin ¼ ðwilting pointÞðsoil depthÞ
The soil water content constrains how much water the for transpiration. In Equation (14.17) we then substitute
plant can extract from the soil, and therefore how much T ¼ λ AET where λ is an empirical partitioning coeffi-
water the plant can transpire. If there is plentiful water, cient to determine how much of the AET is transpiration
then the plant can transpire at the maximum possible and ð1 λÞ for bare soil evaporation.
rate. In this unrestricted case bare soil evaporation and The drier the soil is at the start of a rainfall event, then
transpiration are lumped together as evapotranspiration, the faster the rate of water infiltration at the start of the
and the maximum rate is known as the potential rainfall event (Chapter 3). Infiltration during a storm is
divided into two parts: (1) the initial loss, which is the an estimate of infiltration rate directly from the biomass
water that wets up the soil to saturation at the start of equations in this chapter (Equation (14.4)). The more
the event, and is a function of the porosity and grading of complicated alternative would be to estimate biomass
the soil and the initial soil wetness and (2) the continuing from Equation (14.4) and use this as input to the soil
loss, which is the amount of water that can infiltrate carbon model in Chapter 10, and then link the SOM
through the profile under the action of gravity once the estimate with infiltration rate.
soil is saturated and is equal to the saturated hydraulic HillResLambers et al. (2001) use a simpler approxi-
conductivity of the soil. A commonly used expression for mation for the interaction between biomass productivity
these is the Philip equation (Philip, 1969) for the cumula- and aridity. While more conceptual, it still captures quali-
tive amount of water infiltrated I (or i the infiltration rate) tatively the feedback between reduced water availability
at time t: and reduced productivity

1=
I ¼ St 2 þ ct ∂B SWC
¼ gmax B (14.22)
1 1=
(14.19) ∂t SWC þ k 1
i ¼ St 2 þ c
2 where k 1 is a parameter and gmax is the maximum biomass
where S is the sorptivity (i.e. initial loss) and c is the productivity when water is not limiting. Conceptually this
continuing loss. In this form S is a (undefined) function relationship is similar to the term for SWC in Equation
of how wet the soil is before the rainfall event (note that (14.18), but is not normalised by the maximum water-
Philip derived this equation assuming that the soil was dry holding capacity of the soil.
at the start of the rainfall event). The Green-Ampt From the discussion above it is possible to construct a
equation (Section 3.1.4) yields similar equations with simple biomass dynamics model in water-limited environ-
appropriate soil property assumptions. This can be con- ments. What should be clear, however, is that this requires
verted into a more useful form by substituting the expres- modelling at the rainfall event scale, which may be prob-
sion for I into the expression for i to replace t (noting that lematic because of the data and computations required.
in the initial stages of the rainfall event the sorptivity term A considerable effort is currently being expended in the
dominates the continuing loss), which yields an equation ecohydrology research community to simplify this by
between the amount of water in the soil profile and the deriving temporally averaged equations. This averaging
infiltration rate is technically quite challenging and is well summarised
elsewhere (e.g. Eagleson, 2002; Rodriguez-Iturbe and
S =2
3
i¼ þc (14.20) Porporato, 2005).

2I =2
1
Eagleson (2002) provides a simple conceptualisation
and as the soil becomes wetter (i.e. I increases) the initial of the complicated feedbacks involved in water-limited
loss is reduced. The result of this transformation is that S ecosystems (Figure 14.5). Eagleson presented this graph
is now an intrinsic property of the soil rather than also for a single leaf with light intensity on the horizontal axis
being a function of the antecedent wetness. It is common and carbon assimilation (i.e. biomass production) on the
in engineering hydrology models to assume that SWC ¼ vertical axis, but it is easily extended to the case of an
fnðI Þ since I is a function of how wet the soil is at the soil ecosystem. Presenting Eagleson’s interpretation first, the
surface while SWC is the amount of the water in the soil left-hand side of the graph is the utilisation of light by the
profile (or at a minimum down to the base of the leaf, and as the light intensity (the horizontal axis)
root zone). increases, the carbon assimilation (the vertical axis)
As noted in Chapter 10 the infiltration rate is also a increases linearly until it reaches a threshold (labelled
function of SOM. Thompson et al. (2010) developed an ‘optimum bioclimatic state’) beyond which the leaf
empirical relationship between infiltration rate and dead cannot utilise additional light energy since it has reached
biomass for arid environments its maximum assimilation capacity. This is simulated by
Equation (14.5) though Eagleson has a discrete threshold
i ¼ 56 B0:43 (14.21) at B ¼ B∗ rather than the asymptotic approach to B∗ in
where i is in mm/hr and B is in kg/m2. The infiltration the equation. For increasing light intensity I0 above the
rates from this equation are quite high relative to infil- light threshold the biomass production rate remains the
tration rates that are typical of sandy loams (i.e. their same but SWC decreases (because of increasing evapor-
sites), so the results likely reflect macropore effects (see ation due to the increasing incoming energy that is not
Chapter 3). The advantage of this equation is that it allows used by the transpiration) until another threshold light
FIGURE 14.5: Schematic of light

and water limitation effects on the
production of biomass (from
Eagleson, 2002).
intensity is reached (and its corresponding SWC) when faster, penetrate deeper into the sandy soil reducing bare
there is insufficient water to allow the leaf to operate at soil evaporation, with the result that more water is avail-
maximum capacity. For increasing light intensity beyond able to plants. This is despite loam and clay soils having
this SWC threshold (and corresponding decreasing SWC), better water-holding characteristics so that they hold more
the stomata close to reduce water use and carbon assimi- of the infiltrated water between infiltration events. Sala
lation drops accordingly. This ‘water-limited’ part of the noted that the opposite occurs in humid areas, and high-
curve is Equation (14.18). water-capacity soils are preferred. The model structure
The ecosystem analogy to Eagleson’s curve is that the presented above is consistent with this behaviour (see, e.g.
light-limited part of the curve is much the same, but the sections 5.3 and 7.2 in Rodriguez-Iturbe and Porporato,
biomass generated (vertical axis) does not increase lin- 2005).
early because of shading of leaves as the leaf area index No discussion about water limitation and vegetation
and biomass increases (the horizontal axis). The ‘optimum would be complete without a brief discussion of C3 and
bioclimate state’ is then simply the biomass at which LAI C4 plants. The difference between C3 and C4 plants is in
saturates (i.e. LAI ¼ 4 6), and no extra incremental the chemistry of the photosynthesis process. All plants are
biomass can be generated with an increase of light. The C3, C4 or (less often) CAM. Most grasses are C4 and most
right-hand part of the axis is simply the response of the other plants are C3. The details of the chemical reactions
ecosystem to soil water limitation and AET. The horizon- are not important for this discussion, only that typically in
tal middle part of the curve may still exist and reflects an conditions without water limitation C4 plants use about
ecosystem that can’t grow any faster because it is already half the water of C3 for the same amount of carbon
using as much of the light as possible and SWC does not assimilation (and thus biomass production) (Larcher,
limit growth. 2003; Lambers et al., 2008), so that in Equation (14.17)
Noy-Meir (1973) and Sala et al. (1988) found that for the ratio μC4 : μC3 is about 2:1. However, their response to
water-limited ecosystems the precipitation and soil water- water limitation differs, and the water use efficiency
holding characteristics of the soil interacted, what has advantage of C4 diminishes with increasing water stress
subsequently become known as the ‘inverse soil texture (Nayyar and Gupta, 2006; Ghannoum, 2009; Taylor et al.,
hypothesis’. This hypothesis is that sandy soils have 2011). Modelling studies with varying water stress (Vico
higher biomass productivity than finer loam, or clay and Porporato, 2008) suggest that β in Equation (14.18) is
soils. The rationale for this is that water can infiltrate slightly smaller for C4 relative to C3, so that the water use
(a) (b) FIGURE 14.6: Soil moisture

100 versus normalised biomass
Assimilation rate (%) growth, where the growth rate is
normalised against the growth rate
for the non-water limited case: (a)
50 C3, (b) C4, for two different vapour
Helianthus annuus
VPD = 1 kPa pressure deficits (VPD) (from Vico
Panicum coloratum
VPD = 2.5 kPa and Porporato, 2008).
Bouteloua gracilis
0
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
s s
efficiency of C4 starts to decline more rapidly at a lower 4. Runoff from the unvegetated patches (and any sus-
water stress than C3 (Figure 14.6). Most studies involve pended sediment and nutrients) runs into the vegetated
experiments or models using a single species. In the field patches and infiltrates into the soil in the vegetated
C3 and C4 coexist (and probably compete), and there is patches. Thus the vegetated patches become sinks of
little information about these mixed cases. The approach water and nutrients, and unvegetated patches are
adopted in most DGVMs, which is to weight behaviour sources.
by the percentage cover within the pixel, seems a sensible 5. Destruction of the bands/patches has a disproportion-
approach in the absence of more definitive information. ate impact on water capture and net primary produc-
tion relative to the percentage of area covered by the
bands/patches.
14.4.2 Spatial Self-Organisation of Vegetation 6. Many hillslope profiles show a slight rise in topog-
Patches and Hillslope Profiles raphy (or at least a lessening of the hillslope gradient)
within the bands.
One apparent conclusion from the preceding section is
that as water becomes increasingly limiting, then the
Various models have been developed to capture these
biomass density should decrease uniformly until there is
principles (e.g. Dunkerley, 1997; Saco et al., 2007).
not enough SWC to support vegetation. However, one
Recent modelling work has suggested that coevolution
intriguing aspect of low-slope arid and semiarid land-
of the landforms and vegetation is essential to create some
scapes is the almost ubiquitous occurrence of banded
of the observed patterns (Saco et al., 2007; Saco and
and patchy vegetation where vegetation is concentrated
Moreno de Las Heras, 2013). Saco et al. coupled a land-
into high-productivity areas surrounded by areas of low or
form evolution model with a vegetation model and
no productivity (e.g. tiger bush). This suggests that arid
evolved the landform in response to the transfer of sedi-
ecosystems do not respond to increasing aridity by uni-
ment from the unvegetated to the vegetated zones. The
formly reducing biomass but by breaking up into a patch-
persistence of the spatial location of the patterns even after
work of high- and low-productivity zones (Dunkerley and
overgrazing and fire (Ludwig et al., 2005) suggests that a
Brown, 1995, 1999). The functioning of these ecosystems
long-lived process is essential to the formation and per-
and exact mathematical formulation of a set of processes
sistence of these bands whether it be topography, organic
that create this patterning are still subject to debate, but
matter changes in soil structure, nutrient storage or other
some general principles are generally accepted (Tongway
aspects of soil evolution.
et al., 1990; Ludwig et al., 1999, 2005; Dunkerley, 2000,
A more conceptual approach to the effect of banded
2002):
vegetation was adopted by HilleRisLambers et al. (2001)
1. In the downslope direction there are unvegetated and Rietkerk et al. (2002) where the erosion resulting
patches separated by vegetated patches/bands. from overland flow was not modelled, so it was not
2. The unvegetated patches are low in nutrients and possible to couple the vegetation and topography.
organic carbon, and have a low infiltration rate so A consistent finding with this model and by various
almost all rainfall on them runs off. subsequent authors (e.g. Dagbovie and Sheratt, 2014) is
3. The vegetated patches are high in nutrients and organic that the bands of vegetation generated move upslope for
carbon, and the infiltration rates within the band/patch almost all parameter values. The evidence for or against
are high. band movement in the field is weak, but a conclusion of
the modelling studies is that to fix the bands in space rates post-fire were large enough that the cumulative
requires a coupling between biomass and topography. impact on erosion was very important, so the contribution
Some intriguing questions about the pattern formation of fire to average erosion rates cannot be ignored.
remain open, including the following: There are three main types of wild fire. These are
ground fire (slowest moving and lowest temperature),
1. What drives the transition between uniform, patchy
surface fire and crown fire (fastest moving and highest
and banded vegetation? Some modelling suggests that
temperature). Ground fires largely burn and smoulder in
bands capture more overland flow because they infil-
the soil and consume peat, roots and other buried organic
trate all except the most extreme high flows, while
matter. Surface fires burn surface litter and grasses. Crown
patches allow short circuiting around the patches, par-
fires burn the canopy of forests. Another form of fire,
ticularly if the bands/patches have slightly higher top-
mainly in eucalyptus forests, is a ‘fire bomb’; this ignites
ography as a result of erosion and deposition (Moreno
organic vapours released into the atmosphere by heated
de Las Heras et al., 2012). This is consistent with the
vegetation and the fire front (in the atmosphere) can
field observation that patchy landscapes have slightly
propagate much faster than the ground and crown fires.
less improvement in biomass production relative to
Finally very high-intensity fires can induce convection
uniform vegetation than for banded landscapes
currents in the troposphere (resulting in a thunderstorm-
(Ludwig et al., 2005).
like cloud called a ‘fire storm’; Sharples et al., 2016), and
2. What is the governing process that determines the
the induced winds can be important in increasing both the
spacing of bands/patches? It seems reasonable to con-
fire intensity and the speed of propagation of the fire front.
clude that for a given aridity, the percentage of cover
Crown fires are further subdivided into (1) passive, where
on average will be determined by water availability
crowns of individual trees burn as a result of propagation
and the redistribution from unvegetated patches, but to
up from surface fire and (2) active/continuous, where fire
achieve that average biomass there is an undetermined
in the crowns propagates across the canopy from tree
degree of freedom between larger, more widely,
to tree.
spaced patches, versus smaller, more closely, spaced
Section 14.2 indicates the overwhelming influence of
patches.
groundcover on the erosion rate, with canopy cover being
3. Climate appears to control the percentage cover, but
secondary. This would indicate that the groundcover loss
how do the dynamics function in response to climate
by fire is more important than the canopy loss for erosion-
variability? As the climate dries, do patches get
driven landform evolution. However, crown fires also
smaller, or does the spacing increase (and vice versa)?
burn at higher temperatures and induce greater pyrogenic
For the latter case patches will need to move with
changes in the soil (e.g. destruction of soil carbon, thermal
climate variability. What are the implications for soil
transformation of clays, creation or destruction of soil
evolution if the patches are fixed in space?
hydrophobicity), though these pyrogenic changes have
been less well characterised and appear to be somewhat
14.5 Wildfire site and/or soil specific. Crown fires are important, how-
ever, in driving tree mortality, and they have longer-term
Wildfire is an integral part of natural vegetated systems by impact on catchment water yield as recruitment of tree
periodically removing vegetation, creating regions of new seedlings (and the accompanying water usage that accom-
vegetation growth and ecosystem patchiness. It is key for panies biomass production) occurs to replace the dead
soil carbon because fire is the main source of long-resi- trees (see the Kuczera curve, discussed later).
dence-time carbon in the soil, in the form of charcoal and The fire models that follow do not typically distinguish
other low-lability soil carbon. Fire is a significant process between these fire types and do not model atmospheric
in the return of carbon to the atmosphere from terrestrial feedbacks such as fire storms. The models focus on the
biomass (4.5 Gt/year, not listed separately in Figure 10.1) frequency and spatial extent of fire occurrence, and it is
(Stocker et al., 2013). Finally fire creates patches on the rare for models to simulate fire temperature. Most of the
landform where, in the aftermath of the fire and for some models that follow have a general structure with three
years afterwards, hydrology and erosion, and thus land- components (Venevsky et al., 2002): (1) fire ignition, a
form evolution, are changed. Istanbulluoglu et al. (2004) process for triggering the start of a fire, (2) fire spread,
modelled the impact of post-fire erosion using the pro- determining how fast a fire propagates once triggered and
cesses outlined below and found that even though fires at (3) fire termination, a process that suppresses a fire and
their field site were infrequent, the increases in erosion that is typically weather related (e.g. rain, change of wind
14.5 Wildfire 245
direction). For ecosystem modelling there is a fourth

component, the impact of the fire on ecosystem mortality,
which is related to fire type and temperature.
The discussion below will begin with physically
based models that simulate the three fire components
explicitly. These physically based models might be
too complex and computationally infeasible in a SLEM
but will highlight the physical processes. We will
then discuss simpler stochastic models that might be
adequate in other applications. There is a range of other
modelling approaches, such as cellular automata,
percolation networks and self-organised criticality, but
FIGURE 14.7: Elliptical spread of fire (from Finney, 1999).
to date the evidence is that, despite their conceptual
appeal, they perform poorly, and the reader is guided
to Sullivan (2009c) for a discussion of these (
approaches. 0:936e0:2566U þ 0:461e0:1548U 0:397 LB < 8
LB ¼
8
(14.23)
14.5.1 Physically Based Modelling where U is the velocity of wind at the height of the middle
of the flames. The ‘rate of spreading’ (interchangeably R
A number of deterministic fire models are available.
or ROS in the literature, the rate at which the ellipse
They are mostly used for modelling of fire propagation
increases in size with time) is given by
during a fire, that is, how a single fire, once started,
propagates. We will focus on one model that is typical I R ξ ð1 þ ϕ w þ ϕ s Þ
R¼ (14.24)
of these models, which will enable the highlighting of ρb εQig
generic physics that are required to model fire, and
where I R is the fire reaction intensity and is a function of
aspects of the input data and parameters that are required
the amount of energy released by the fire per unit time and
to run these models. It is the detail of these requirements
per unit length of the fire front, ρb is the bulk density of
that may make these models difficult to use in the con-
the fuel, Qig is the heat of pre-ignition and characterises
text of hundreds to thousands of years of landscape
how much energy per unit mass must be input into the
evolution simulation. The model discussed below is
fuel to ignite it, and ξ and ε are fitting parameters. The
FARSITE, developed and used by the USDA and US
effect of wind and slope is characterised by ϕw and ϕs ,
Forest Service (Finney, 1999, 2004).
respectively and are (Rothermel, 1972)
FARSITE models the propagation of a fire front, since
most of the dynamics of the fire occur at the fire front (the !E
B β
interface between the burnt area and the surrounding ϕw ¼ Cð3:281U Þ
βop (14.25)
unburnt material). The area burnt is then simply the region
traversed by the fire front during the simulation. The ϕs ¼ 5:275β0:3 tan 2 ϕ
methodology is that at every timestep the fire front is
where ϕ is the slope in the direction of the wind, and B, C,
advanced. The rate and direction of the fire front move-
E and β are fitting parameters (all typically positive). Note
ment are primarily a function of (1) wind speed and
that Finney (1999, 2004) incorrectly quotes the slope term
direction and (2) the intensity of the energy release by
in the slope equation as tan ϕ2 and fail to note that
the fire at the fire front. Many of the underlying physical
Rothermel derived this only for fire fronts travelling
principles in FARSITE, and a number of other models, are
upslope. For slopes less than 10%, the slope correction
based on Rothermel (1991).
is minor (Rothermel, 1991). The velocity of the front of
The spreading of a fire from a localised point is well
the ellipse is
approximated by an elongated ellipse. The initiation loca-
tion of the fire is at the rear foci of the ellipse, and the rate 3 R R
V front ¼ þ
at which the long axis of the ellipse extends is a function 2 HB 2
(14.26)
primarily of wind. The faster the wind, the more elongated
HB ¼ LBþðLB 1Þ LB LB2 1 0:5
2 0:5
the ellipse (Figure 14.7), with a length-to-breadth ratio of ð Þ

FIGURE 14.8: Rate of fire

spreading as a function of
wind speed (from Sullivan, 2009b).
Some key concepts embedded in these equations are front slows (this effect of wet fuel is in addition
common to many other models even if their mathematical to point 5).
formulations are different:
1. As the wind speed increases, the ellipse of burnt area Two broad conclusions can be reached here. The first
becomes increasingly elongated (Equation (14.23)), is the fire will interact with the topography because of the
ϕw becomes larger so that the rate of spreading of slope term in Equation (14.25). The second is that climate
the ellipse increases, so more area is burnt per unit data are needed, including (1) wind speed and direction
time both in the direction of the wind and at right and (2) inputs for a water balance model to determine how
angles to it (Equation (14.24)), and the velocity dry/wet the fuel is before the fire. The biomass model to
of advance of the fire front increases (Equation determine the quantity of available fuel from the previous
(14.26)). growing season(s) will also require pre-fire climate.
2. The parameter B has been found to be in the range 1–2 Point 5 highlights the important issue of the time sequen-
(Sullivan, 2009b), so the rate of spreading increases cing of climate variations. The worst fire conditions occur
nonlinearly, and more quickly than the wind speed as a result of a wet growing seasons in winter and spring
(Figure 14.8). followed by a dry hot summer because there is plentiful
3. Since wind speed increases with height above the biomass and it is dry (Bradstock, personal communica-
ground, the speed of the fire front will increase for tion; O’Donnell et al., 2011). Finally, FARSITE is a fire
higher flames because of the dependence on the mid- propagation model and does not simulate (1) when and
flame height U. where fires begin and (2) fire termination. While we won’t
4. As the slope of the topography increases then the discuss it further, FARSITE has criteria for surface fire
velocity of the fire front increases (Equation (14.25) and any transition to a crown fire, but the speed of propa-
and (14.24)). gation is assumed to be the same for both types.
5. As the amount of energy released increases (I R in FARSITE makes it clear that a considerable number of
Equation (14.24)), the velocity increases. Since wet spatially distributed data are required to describe the pre-
fuel releases less energy than dry fuel (because the fire conditions, including elevation, slope, aspect, the fuel
water has to be driven off before it can ignite), the fire (type, rates of accumulation, wetness), canopy cover,
front propagates less rapidly in wet conditions. If there canopy height, canopy base height and canopy bulk dens-
is more fuel (e.g. there has been a good vegetation ity, albeit the latter four are required only for crown fire
growth season leading into the fire season), then the modelling. These properties can be spatially variable, and
fire front will move more quickly. propagation of parts of the fire may be slower or faster as
6. As the fuel becomes harder to ignite (higher Qig in a result of local spatial variation. In addition the climate
Equation (14.24), typically because it is wet), the fire data (which may also be spatially variable though that
14.5 Wildfire 247
type of data is rarely available) required are (1) a wind hydrology applications using this type of data suggests
speed and direction time series and (2) minimum and that the RCM model’s ability to preserve relationships
maximum daily temperature, humidity and precipitation between climate variables is still patchy (Lockart et al.,
to determine fuel wetness. Note that the precipitation is 2016; Chowdhury, 2017; Parana Manage, 2017).
not used in FARSITE for fire termination.
The complexity of the modelling task using a physic-
These equations and their data requirements are typical
ally based model like FARSITE means that researchers
of deterministic physically based fire models. We can now
have tended to use simpler models that attempt to capture
consider how this model and others like it might be used
conceptually the processes above but without the detailed
in a landscape evolution simulation, which may be simu-
modelling of every process. All the approaches in the
lating thousands to millions of years. Clearly we will have
points above are suitable for inclusion in a LEM, the only
the topographic data required, and if we have a coupled
constraint being data availability. If these data are not
vegetation model, we will likely be able to simulate the
available or cannot be generated, then the approaches in
fuel properties required. The difficult issue is the simula-
the next two sections, though not as strongly physically
tion of the climate variables. There are three alternatives:
based, are preferred.
1. Use a historic climate record and repeat this record

multiple times until the total simulation time is com- 14.5.2 Conceptual Modelling
plete. This is a common solution (e.g. Willgoose and
Riley, 1998a) and ensures that all the different climate Venevsky et al. (2002) present a conceptual fire model,
series are correctly sequenced and the correlations Reg-FIRM, designed for incorporation into a global cli-
between the time series (e.g. relationships between mate model (GCM), and the model captures some of the
temperature and rainfall which will drive fuel wetness) physical processes from the previous section. Their model
are correct. This of course requires the climate series determines the average burnt area arising from a stochas-
data be available, and that they be consistent with the tic representation of the processes above, and from this
varying climate over the period being simulated (e.g. determines the average percentage of a GCM cell (typic-
does current-day data simulate what happened in an ice ally about 300 km 300 km) that is burnt. They do not
age?). One important consequence is, however, that recommend the model for grid cells less than 100 km2,
the input data cover only the extremes in the recorded which is larger than normally used in LEMs; however, we
data, and mega-events that might have occurred in the will discuss later how Reg-FIRM is the basis of another
pre-instrumental record cannot be simulated. model, PESERA (Fleskens et al., 2016), which was
2. Randomly simulate synthetic times series to use. This designed for typical LEM resolutions.
needs to be done with extreme caution. The generation The number of fires N fire in a grid cell per year is
of multiple randomly varying climate time series (i.e. estimated using two ignition potentials: (1) human-caused
temperatures, rainfall, evaporation) must be done so ignitions and (2) lightning strikes
that not only do they vary in time correctly (e.g. wet
N fire ¼ CN d δPθD þ l (14.27)
winters, dry summers) but also that the temporal rela-
tionships between the climate variables are also cor- where C is a Fire Danger Coefficient (FDC) averaged over
rectly maintained (e.g. wetter than average periods the daily values for the year, N d is the number of days
typically have lower than average evaporation, and with high fire danger per year, PD is the population
reduced temperature extremes). The dangers multiply density, l is the number of fires started by lightning strike,
considerably if you want to simulate random spatial θ is a coefficient that has been found to lie between 0.43
variability as well. Techniques to do this exist but and 0.57 and δ is a coefficient. Venesky defined FDC
typically rely on the variables being distributed with using the Nestorov Index, which is a function of the
a Gaussian distribution, and this can be problematic, temperature on preceding days and whether rainfall is less
particularly for rainfall, even after transforming the than 3 mm on those preceding days. They note that a
data (e.g. using the Box-Cox transformation; see range of fire danger indices are used worldwide that relate
Chapter 3). fire climate with the rate of fire ignitions that could also be
3. Generating high-resolution climates using regional cli- used. They also note that the number of human initiated
mate models (RCM) such as WRF (Weather Research fires is much greater than natural fires in recent decades
and Forecasting Model). It is early days for these (e.g. 97% in Spain, 86% in Russia and 87% in southeast
climate models, and the author’s experience with Australia of total fires; Venevsky et al., 2002; Collins
et al., 2015). For Russian boreal forest l is 0.015–0.028 exponential probability distribution for fire duration, and
fires per day per million hectares. The lightning ignition Agrid is the area of the grid cell. Spatial variations in PB
rate l increased with increased number of hot days, eleva- were then driven primarily by spatial variations in the rate
tion, terrain slope and percentage vegetation cover (Col- of spreading and population density, while temporal vari-
lins et al., 2015). Penman et al. (2013) found that the rate ations were a function of temporal variations in climate
at which a lightning strike caused ignition increased with (and thus the rate of spreading U).
number of years since the last bushfire, provided that the As noted above, Reg-FIRM has been used as the basis
last bushfire occurred more than 15 years previously, and of another model, PESERA, which is designed for use at
for less than 15 years the ignition rate was independent of high spatial resolutions (Fleskens et al., 2016). They cal-
the time since the last fire (and thus accumulated fuel/ culate a Fire Danger Index (FDI) for a given month that is
biomass). To yield l this ignition rate needs to be multi- a simplified version of Venevsky’s FDC as
plied by the lightning strike rate, which is largely a func-
tion of incidence of thunderstorms, and at the global scale FDI ¼ 1 1 eλN =λN
(14.29)
is strongly positively correlated with increases in eleva- N ¼ T ðT E =2 þ 4ÞD>3
tion and terrain slope and is higher over forest than bare
where T and T E are the mean monthly temperature and
soil (Kotroni and Lagaouvardos, 2008; Albrecht et al.,
mean monthly temperature range over a day and D>3 is
2016).
the number of days per month with more than 3 mm of
Lynch et al. (2007) and Verdon et al. (2004) used
rain. The mean number of fire starts per month is calcu-
paleoclimate data and ocean teleconnections to estimate
lated with Equation (14.27), and the number of wildfires
the temporal variability of forest fire risk. Verdon found
started in that month is FDI N fire . The area of fire spread
an increased risk of fire in eastern Australia during El
R (m/s) is calculated using a simplified form of the equa-
Niño episodes and IPO negative (i.e. dry hot periods).
tion used in Reg-FIRM
Proxy records based on ice cores and other paleo-data (see
Chapter 3) provide a possible means to develop millennial 0:9ð1 þ U Þ
R¼ (14.30)
scale paleo-fire records. B 16 þ 100AET2 PET
The area burnt by the fire once initiated is a function of
where AET and PET (mm/month) are actual and potential
the rate of spread of the fire and the duration of the fire.
evapotranspiration, respectively, B is the biomass (kg/m2),
Venevsky et al. (2002) used an exponential probability
U is the average wind speed (m/s), and fires are assumed
distribution for the duration of the fire (pðd Þ ¼ μeμd
to last one day and are circular in plan. A fire severity,
where d is the random duration) where the average dur-
based on biomass and dryness, is then used to estimate the
ation of the fires was one day, which was based on
loss of biomass and subsequent increase in erodibility
Russian fire statistics (Korovin, 1996 in Venevsky et al.,
(Kirkby, personal communication). To generate a time
2002). There is some rationale for the mean fire durations
series over a region, fires are assumed to randomly occur
being about one day because many fires burn during the
in time and space based on the expected number of fires
(hot) day but then extinguish in the cooler conditions
per month in a grid cell.
overnight (Penman et al., 2013). The rate of spread is a
The models above do not simulate fire intensity.
function primarily of the wind speed using the elliptical
A more advanced model that follows the approach of
fire burn area, as discussed above for FARSITE. The
Reg-FIRM is LPJ-SPITFIRE. In addition to modelling
effect of fuel load and soil moisture (a proxy for fuel
fire extents it also uses the biomass and climate models
moisture content) modifies the rate of fire spreading.
of LPJ to simulate the intensity of the fire (e.g. surface fire
Topography and any overlap between the burnt areas of
versus crown fire), and thus the proportion of the biomass
separate fires is ignored, so wind direction data are not
burnt by the fire (Thonicke et al., 2010).
required. This leads to their equation for the fraction of a
grid cell burnt in one year, PB,
2
N fire πU 14.5.3 Stochastic Modelling
PB ¼ (14.28)
8μ2 Agrid
Martin (2007) and Willgoose (2011) present simple sto-
where U is the annual average rate of fire spreading, chastic models with strong similarities that build upon
which is the average over the daily values, and the daily empirical observations of fire occurrence and spatial dis-
values are a function of average wind speed, fuel load and tribution. The simulation method is that a fire initiation
soil moisture on that day, μ is the exponent in the time and location is randomly simulated, the burnt area
14.5 Wildfire 249
did not show that this decadal variability resulted from

climate variability, other than changes in fire management
policies (e.g. Dennison et al., 2014), no alternative
explanation seems plausible. Verdon et al. (2004) found
a relationship for eastern Australia between fire occur-
rence and ocean teleconnections (El Niño and IPO). These
results are suggestive of a means of using their model to
simulate natural climate variability and anthropogenic
climate change: simply change the value of β so that fires
of a given area become more or less frequent, as was done
FIGURE 14.9: Cumulative size distribution of all detected fires in Willgoose (2011). Malamud did not link his results
for the two periods 1499–1650 and 1650–1880. Fires smaller than to climate data, so an empirical link between frequency
50 ha have been lumped together (from Niklasson and Granström, and climate needs to be developed for any given
2000). application site.
Willgoose (2011) calibrated his model to fire recur-
rence data of Russell-Smith et al. (1997) and found that a
from the fire is determined and the burn area placed at the range of area-scaling models, including that of Malamud,
fire initiation point. A key input is the size of the fire that and fire shapes (square, round, triangular) could be suc-
occurs. Malamud et al. (1998, 2005) and Nikalsson and cessfully calibrated to the Russell-Smith data, and that the
Granström (2000) analysed historical fire data and found only difference was that Malamud’s area scaling model
that the frequency of fire and the area of the fire were yields a spatial pattern of burning that had more fine detail
inversely related (Figure 14.9) by a relationship of the (because of the high frequency of small-area fires) and
form was less spatially clumped than models that had a more
uniform distribution of fire areas. Neither Martin nor Will-
p ¼ βAF α (14.31)
goose examined the spatial statistics of the burn patterns
where p is the probability of a fire larger than area AF (per that might be ecologically significant, for example, the
unit time and per unit area) and α was in the range connectivity of unburnt regions. Rather the focus of their
1.3–1.8. This general trend of decreasing frequency with studies was how a change in fire frequency (e.g. due to
increasing area has been widely found (Hantson et al., climate change) might change (1) landslide and sediment
2016). As a result, fires of small area are more frequent delivery impacts (Martin, 2007) and (2) runoff yield
than fires with large area. Malamud’s data were for a (Willgoose, 2011).
number of different locations worldwide and so are
applicable worldwide. O’Donnell et al. (2014) found a
similar relationship but with change in α for large areas, 14.5.4 Recovery from Fire
which they interpreted as being a size limitation based on
ecosystem fragmentation due to land use. Fire suppression One important variable for estimating the impacts of fire is
in the last 100 years has changed this relationship in some how long it takes for the ecosystem and soils to recover
areas (Steel et al., 2015). Nikalsson and Granström (2000) from the fire, and how these impacts decrease with time.
attributed a change in α pre- and post-1650 to human For instance, the vegetation models in previous sections
occupation. Malamud et al. (2005) calculated frequency can be used to model biomass recovery, but what about
rates for the United States and Europe and found that the other impacts such as on soils? Some impacts will have
number of fires per year with area greater than 10 km2 that longer recovery times (e.g. loss of soil carbon) than others
occurred in a 1,000 km2 area was about 1–10 for the (e.g. hydrophobicity in the soil), but in general it would be
western half of the United States and about 0.01 for the expected that there would be a gradual decline with time
eastern half, while in the Mediterranean it was about 2 per (e.g. exponential with time). This recovery process has
year. These rates were for recent years and so will likely rarely been quantified beyond a simple decline with time
have a significant human initiation component, as in lasting for a few years for small surface fires to decades
Equation (14.27). for larger more intense crown fires (e.g. stand-clearing
Malamud et al. (2005) also calibrated their recurrence fires that kill all or most trees). It is also important to
model to individual decades and found that there was a distinguish between biomass lost (mostly leaves and
decade-to-decade difference in β but not in α. While they ground litter) and tree mortality. Tree mortality effects
FIGURE 14.10: The Kuczera

curve showing the impact on
water yield from a catchment as a
function of years from fire (from
Kuczera, 1987).
will have longer recovery times because they involve the to separate it from the other effects discussed below. The
regrowth of fully grown trees (typically decades), while Kuczera curve mechanism will only occur in situations of
biomass recovery will be faster because it is recovery of water excess where water availability does not limit
vegetation on existing surviving trees (typically only years). growth.
Kuczera (1987) fit a curve (the Kuczera curve) to Much post-recovery work has used remote sensing and
catchment runoff yield after a stand-clearing (i.e. wide- NDVI to determine when the NDVI, reflecting photosyn-
spread tree mortality) fire in water supply catchments in thesis rates, has recovered to adjacent unburnt areas.
southeast Australia (Figure 14.10). Immediately post-fire Rodrigues et al. (2014) estimated the time for vegetation
the runoff yield was not significantly impacted (though in Spain to return to its pre-fire canopy height and canopy
some authors have noticed an immediate small increase in cover and found that grasses recover faster (two to five
water yield post-fire; Watson et al., 1999), then a short years) than shrubs (10–25 years) and trees (20–50) years,
time later the runoff yield of the catchment dropped as the and that if the vegetation recovered by sprouting on sur-
enhanced biomass growth, and consequent enhanced viving vegetation (the numbers quoted above) then recov-
water use, resulting from the forest recovery occurred. ery was about 2 times faster than if recovery was
Forest water use peaks at the time that the biomass predominantly from seed. These results are consistent
growth peaks and then asymptotically declines to the pre- with Polychronaki et al. (2014) who found in Greece
fire condition some decades later. Since water use is that 20 years post-fire the grass and shrub areas had
strongly linearly correlated with biomass growth, it is recovered their pre-fire NDVI, while forest areas were
reasonable to expect a yield reduction to occur in all still recovering.
catchments post-fire though the yield decline will vary Finally the question of how likely a previously burnt
with the amount of biomass lost (and thus the amount that area is to burn at some later stage is partly a question of
will be required to recover over and above normal plant recovery and particularly the recovery of the flammability
growth, which can be measured by NDVI remote sensing of the vegetation (which may not correlate exactly with
post-fire; e.g. Katagis et al., 2014) and the time scale of the amount of biomass; McCarthy et. al., 2001). Moreover
recovery. Nolan et al. (2015) estimated evapotranspiration it is also dependent on what percentage of the area is burnt
(ET) using sap flow measurements in a mixed species each year and how quickly it recovers. Low burnt area
burnt forest over time and found a result consistent with percentages mean that the impact of previously burnt
the Kuczera curve but with ET peaking at three to four areas on fire occurrence and area will be small (Price
years in her case rather than the 30–40 years found by and Bradstock, 2011; Price et al., 2012).
Kuczera. The fires studied by Nolan were for fires
involved biomass growth but low mortality, while the
fires studied by Kuczera involved widespread tree mortal- 14.5.5 Impacts of Fire
ity. Thus the recovery time for Nolan’s fires was much
less than Kuczera’s. This section can provide only a brief overview of the
If the loss of biomass is small, then the impact on impacts of fire, and accordingly the focus here is on those
runoff yield will likely be less, and thus it may be difficult with a first-order impact on soilscape and landscape
14.5 Wildfire 251
evolution. Unfortunately the literature in this area is three years. Decreased AET was driven by tree death,
incomplete with apparently contradictory conclusions for while increased AET was driven by seedling water
different case studies, and, at least in part, these differ- demand (Nolan et al., 2014b). A longer-term study
ences in responses are claimed to be a result of different suggested that AET of eucalyptus forests have
responses to different severity fires (Keeley, 2009). More- recovered by eight to 12 years post-fire (Nolan
over as the previous section indicates, there may be time- et al., 2015). Destruction of organic matter destroys
varying impacts during the post-fire recovery. However, soil structure and decreases the hydraulic conductiv-
in general, impacts are greatest for the highest intensity ity of the soil. Recovery times for vegetation can be
fires, which are generally correlated with those areas of variable from years to decades depending on the
highest biomass (Pierson et al., 2002). Shakesby and severity of the fire and percentage of tree mortality.
Doerr (2006), and Moody et al. (2013) discuss many of • Erosion: The main impact of post-fire erosion is the
the points below and note that fire effects on soils are loss of litter and groundcover protection of the soil and
relatively poorly understood compared to biomass and the resulting increased erosion (up to 25 times higher,
vegetation impacts. Nyman et al. (2013) review some of but more typically 5–10 times; Johansen et al., 2001;
the modelling approaches used for modelling fire impacts Llovet et al., 2009; Smith et al., 2011; Hosseni et al.,
on hydrology and geomorphology, while Langhans et al. 2016), and the recovery follows from reinstatement of
(2016) and Sidman et al. (2016) propose detailed model- the vegetation groundcover. The recovery to near pre-
ling approaches for hydrology and water quality impacts. fire response is highly variable (from 1 to 10 years) but
Some of the impacts of fire are as follows: typically takes about two years (Cerdà and Doerr,
2005; Lane et al., 2006; Sheridan et al., 2007; Ryan
• Hydrology: Hydrology impacts can be categorised as et al., 2011). There is evidence for dry ravel and debris
those that impact on event scale runoff that in turn flows post-fire, and, together with fluvial erosion, a
drives erosion processes, and longer-term-yield impacts significant amount of removed hillslope sediment can
that impact on the average runoff on the catchment over be stored elsewhere within the catchment in the imme-
the longer term. As discussed in the previous section, diate aftermath of the fire. This stored sediment is then
some impacts on runoff yield may last from years to removed from the catchment over the longer term
decades: (Shakesby and Doerr, 2006; Santi et al., 2008; Nyman
◦ Event-scale hydrology: Runoff response times tend to et al., 2011). The consumption of the organic-rich
decrease as a result of the loss of leaf litter, but with surface O layer in the soil can destabilise any estab-
smaller fires post-fire leaf fall offsets this. The ash lished surface armours (Shakesby, 2011), and provide
plays an important role in retarding runoff initiation loose sediment, ash and charcoal that is readily avail-
(Woods and Balfour, 2008). There is sometimes a able for transport by overland flow (Nyman et al.,
period of increased/decreased hydrophobicity due to 2013).
volatile by-products of burning organic matter • Water quality: The main impacts are an increase in
(lasting several years to decades; DeBano, 2000; suspended sediment (up 1,400 times higher), total
Cerdà and Doerr, 2005; Tessler et al., 2013; van N (specifically nitrate) increases in the first year, and
Eck et al., 2016), which must be seen relative to dissolved organic carbon being unchanged while par-
any pre-fire hydrophobicity, but which results in ticulate organic carbon increases markedly if rainfall
changed runoff rates post-fire. Higher runoff rates occurs soon after the fire (Smith et al., 2011). Signifi-
with increased fire severity can be attributed primarily cant impacts on the N and P cycles have been measured
to surface cover decreases, not hydrophobicity (Hos- up to 15 years post-fire (e.g. 50% increase in N fluxes
seini et al., 2016). 13 years after fire has been observed). The suspended
◦ Yield hydrology: Higher soil moistures at depth post- sediment is high in ash, and ash chemistry varies with
fire have been reported, and they appear to be related fire temperature. For temperatures < 450C the ash is
to the reduction of water use within the soil profile as rich in organic carbon, for temperatures > 450C most
a result of vegetation removal/damage. The higher organic carbon is volatised and carbon is in the form of
soil moisture has also been linked to increases in carbonates, and for temperatures > 580C ash is dom-
groundwater flow (Silva et al., 2006). Nolan et al. inated by oxides (Bodi et al., 2014) (see also ash effects
(2014a) found that in the first three years after forest under ‘Soils’ below). There have been some reports of
fires that AET decreased by 41% in areas of high elevated cyanide immediately post-fire as a result of
severity fire but then increased above pre-fire rates at aerial deposition.
• Mass movement: Dry ravel, debris flows and mini- • Ecosystem: Greene et al. (1990) found that cryptogam
landslides are common in the immediate aftermath of crusts were destroyed by fire resulting in increased
a fire (Ryan et al., 2011). This creates loose material that erosion, but cryptogam recolonisation occurred within
is then available for transport during subsequent runoff four years. A number of authors have noted that the
events (Cawson et al., 2012; Nyman et al., 2013). As balance of grass and trees in savannah ecosystems is
discussed in Section 13.1.2, rates of landsliding show a difficult to maintain without regular fire (Sankaran
marked increase after clear cutting of forest (up to a et al., 2005; Bowman et al., 2008; Lehsten et al.,
factor of 5), which is attributed to death of fine roots and 2016; Murphy et al., 2010). Patches of unburnt canopy
a reduction of strength in the older roots. Istanbulluoglu and wet gullies provide refuge for tree-dwelling
and Bras (2005) used clear cutting as an analogy for the mammals (Chia et al., 2015).
effect of fire on the rate of landsliding, though this is
probably valid only for stand-clearing fires. For smaller
fires the effect on landsliding rate is unknown. The 14.5.6 Fire and the Long-Term Erosion Rate
reduction in strength of roots will occur only for tree
death, so it is only in ecosystems with fire-sensitive This section is rather more speculative than the previous
trees that post-fire root weakening will be observed. sections and takes a long-term geomorphic perspective on
Another consideration is how quickly the forest the impact of fire on erosion and landform evolution.
recovers. If forest reestablishment is fast, then new roots Many of the ideas below have been triggered by conver-
may be established by the time the dead trees roots sations and debates with my colleague Gary Sheridan.
weaken (Sheridan, personal communication). In the previous section typical values quoted for
• Soils: Fire consumes soil carbon (SOC). SOC fluvial erosion are that it increases by at least 10–15
consumption starts when soil temperatures reach times the unburnt erosion rate (sometimes orders of
200–250C and is complete at 460C (Certini, 2005). magnitude more) and that impacts may last for two to
Low-severity fires may heat the soil only to 60C, 10 years. Let us consider a thought experiment. If the
while high-severity fires may heat the soil to increase is a factor of 10 for two years, both midrange
800–900C (Cawson et al., 2012). Consumption of values, then one fire will have the same amount of
SOC breaks down soil aggregates that are held together erosion as 20 years of unburnt erosion. Thus if an area
by SOC, destroys soil structure (down to about 10 mm burns every 21 years, then the post-fire erosion will be
below the surface; Nyman et al., 2013) and thus the same magnitude as the inter-fire unburnt erosion. In
reduces infiltration rates. Surface ash/char that may many fire-prone areas a return period for fires of 21 years
cover the surface post-fire is generally lost from the is not abnormally long, suggesting that post-fire erosion
surface within a year or two, either by erosion or by may be a significant, if not dominant, component of the
filling up soil macropores, with higher carbon loads in long-term erosion rate for a catchment. The thought
runoff post-fire primarily ash from the fire and not soil experiment shows that as the fire impact on erosion and
carbon (Reneau et al., 2007; Hosseini et al., 2016). the recovery time for that impact increases, and the
Rumpel et al. (2009) found that about half of the recurrence interval decreases then the relative importance
surface ash/char was eroded, while the remainder was of post-fire erosion in the long-term geomorphic erosion
transported down into the soil profile. Soil aggregate rate will increase.
stability was reduced by the thermal alteration of the Recent work has also highlighted other processes,
organic matter cementing particles at soil temperatures even though quantification has still not been performed.
around 380–460 (Greene et al., 1990; Garcia-Corona In the immediate aftermath of a fire dry ravel and small
et al., 2004), while Mataix-Solera et al. (2002) found debris flows move loose sediment (recall that fire can
greater SOM loss and soil structure breakdown after destroy soil structure and make soil more erodible by
surface fires than crown fires. Fires return nutrients volatising the SOM that binds soil into aggregates) from
from the aboveground biomass to the soil and enhance higher parts of the catchment to lower parts of the catch-
the cycling of nutrients (De Marco et al., 2005). Below- ment. This loose material is temporarily stored on the
ground vegetative biomass and microbial activity is hillslope and valleys, and is available for movement in
relatively unaffected by fire (for low fire severity), subsequent fluvial erosion after the catchment vegetation
and belowground productivity is more affected by has recovered. The evidence suggests that this mobile
grass species and drought than burning (Burnett et al., material is then eroded over the long term after the catch-
2012; Fontúrbel et al., 2012). ment has ‘recovered’. The quantities of sediment that may
14.6 Other Forms of Vegetation Disturbance 253
be temporarily stored in this way on the hillslope and in fire is likely to vary markedly with climate and dominant
the valleys are still unclear. ecosystem.
Let us return to our thought experiment. Let us assume The second implication is that the burn frequency in
50% (this is simply assumed for the thought experiment; populated areas is a function of population density, so that
I have no evidence to support this value) of the sediment it is possible to conceive that in areas of low burn fre-
eroded off the hillslope leaves the catchment during post- quency (where the catchment spends significant time in
fire runoff (and is thus the 10–15 times increased erosion Stage 3) that if the population density increases and the
that is observed in experiments at the catchment outlet) burn frequency also increases, then the average erosion
while the other 50% is moved from the hillslope into areas rate can increase because a greater proportion of time is
of temporary storage elsewhere downstream on the hill- spent in Stages 1 and 2. Thus it is possible for the geo-
slopes and in hollows. If this temporary storage is then morphically effective average erosion rate to increase as a
eroded over the long term after the catchment has result of human impacts without there being any obvious
‘recovered’, then this material may then be equal to direct link between human activities and erosion (e.g. a
10–15 years of unburnt erosion. We then speculate that direct impact would be farming, land management
there are three stages of erosion in a fire-prone catchment: changes or climate variability).
Much of the discussion above depends on the relative
1. Stage 1: In the year or two after the fire, the erosion
magnitude of burnt and unburnt impacts, most of which
rate (as observed at the catchment outlet) is elevated by
are poorly quantified at this time, so the discussion above
10–15 times. At the same time dry ravel and debris
should be considered somewhat speculative. However,
flows are mobilising soil, and this loose sediment is
typical numbers from the literature provide support for
stored lower down on the hillslopes and in the hillslope
the relative impact of burnt and unburnt erosion rates.
hollows.
The mechanisms of dry ravel and debris flows that are
2. Stage 2: For a period of a decade or more (recall there
the sources of the temporarily stored sediment are specific
are observed impacts on nutrient fluxes 15 years after-
to high-slope environments. However, the destruction of
wards, so a decade or more to recover does not seem
soil structure by the burning of SOM is equally applicable
unreasonable) fluvial erosion is removing the loose
to high, and low-slope environments, and it is possible
sediment temporarily stored on the hillslope and in
that there are also mechanisms that provide temporary
the valleys during Stage 1. Stage 2 ends when the
storage for mobile sediments on low-slope hillslopes.
stored sediment has all been removed. Nyman et al.
We thus conclude that it is possible that in fire-prone
(2013) suggest that the erosion rate post-fire declines
areas the average erosion rate may be determined by the
exponentially, so there may be a gradual transition
sourcing of sediment as a result of fire, and that the fluvial
from Stage 1 to Stage 2, and thence to Stage 3.
erosion between fires is simply the mechanism to deliver
3. Stage 3: After a decade or more of removal of loose
this sediment to the catchment outlet.
sediment created and stored on the hillslopes in Stage
2, the erosion drops to a source-limited condition
where the erosion is determined by the recovered
vegetation cover and the equilibrium unburnt soil con- 14.6 Other Forms of Vegetation
ditions. This might be called the true unburnt erosion Disturbance
rate. Stage 3 continues until the next fire.
Fire and herbivore grazing are only two forms of disturb-
There are two implications of this conceptualisation of ance of vegetation that might feed back to erosion rates.
erosion impacts into three stages. A number of other disturbances were briefly discussed in
The first implication is that if a catchment burns very Chapter 7 in the context of soil bioturbation in Section
frequently, then the catchment may never reach Stage 3. 7.3.5.
In this case the long-term erosion rate is solely determined Tree throw excavates a hole in the ground when the
by the post-fire transport processes. The burn frequency tree roots are pulled up. There are two aspects to the post-
for this case, while short, is not outside the bounds of tree throw behaviour. The soil in the root ball is now
observed burn frequencies. What should be clear is the above the ground and is loose and can potentially be more
impact of fire on erosion will be a function of the fire freely transported, leading to higher erosion rates. On the
frequency (how long the catchment remains in Stage 2), other hand, there is now a hole into which water and
and the severity of the fire and the recovery time (i.e. the sediment can flow and collect, potentially capturing ero-
duration of the Stage 2). Thus the geomorphic impact of sion. The post-tree throw erosion is the net effect of these
competing processes, and at least in the short term the can only scratch the surface of vegetation representation
roughening of the hillslope surface leads to a reduction in in a LEM in this chapter, even if the science actually
sediment delivery (Hancock et al., 2012). existed to attack all of these issues simultaneously. One
Fauna can also influence erosion by excavating around of the problems with the literature is that the treatment of
trees to feed on the roots. In Section 7.3.5 we discussed rates vegetation is rather fragmented, with little linkage
of soil turnover from wombats digging for roots to eat. Wild between the various approaches. Inevitably this chapter
pigs have also been observed to be an active soil turnover in part reflects this fragmentation.
mechanism. As for tree throw, the net effect of this digging is
to roughen the surface and enhance the capture of runoff
from the hillslope, reducing delivery of sediment to the 14.8 Further Reading
catchment outlet (Hancock et al., 2015b, 2017b).
For those with an interest in the range of agricultural crop
14.7 Conclusions simulation models used internationally, a good place to
start is the 2003 Special Issue of the European Journal of
Vegetation and fire impact on almost all the processes in Agronomy, 18(3–4). Research in Dynamic Global Vege-
the preceding chapters. Vegetation tation Models (DGVM) is extremely active and fast
moving because of global climate modelling imperatives,
• Increases the infiltration rate of the soil by the creation but the BIOME model (Prentice et al., 1992) and its
of macropores (Chapter 3), and influences soil moisture
descendants (BIOME2, BIOME3, BIOME4, LPJ) are
dynamics and water flow down the soil profile by tran-
seminal founding works. The 1999 Special Issue of
spiration (e.g. flow velocity down the soil profile in
Global Change Biology, 5(Supplement 1) compares
Chapter 8).
DGVM model characteristics (Cramer and Field, 1999)
• Protects hillslopes from the full effects of shear stress and while it is somewhat dated, it overviews many aspects
arising from overland flow, thus reducing sheet and
of DGVM construction and response at that time, and
gully erosion (Chapter 4).
provides a grounding for understanding more recent work.
• Appears to be the main driver of soil depth via tree Bondeau et al. (2007) used LPJ to examine the impact of
throw, with roots ripping rock fragments out of the
vegetation (and specifically agriculture) during the 20th
saprolite (Chapter 6).
century on the global carbon cycle and the reduction in
• Provides tree roots that strengthen the soil profile and soil carbon stocks. Eagleson (2002) provides a very
protect the soil against gravity processes such as creep
detailed discussion of vegetation adaptation in water-
(Chapter 9) and landslip (Chapter 13).
limited environments (Chapter 1 is a particularly good
• Is the main source of soil organic matter for the soil by overview of the principles). For those interested in a
providing leaf litter to the soil surface, and dead roots
broader range of forest fire models than discussed here,
and exudates within the soil profile (Chapter 10).
and the scientific principles underpinning them, see the
The complexity of a general vegetation model that review papers by Sullivan (2009a, b, c) and Hilton et al.
addresses all these issues simultaneously means that we (2015).
15 Constructing a Landscape Evolution
Model – Details
15.1 How to Couple the Processes A maximum step size can be used to obtain the desired
numerical accuracy for the elevation change, and that
The preceding chapters have focussed on how to solve the timestep size depends on the numerical solver used. It is
science of the various components of soilscape and land- straightforward to write a Courant number numerical con-
form evolution. In Chapter 2 we had a brief discussion of straint for this timestep size when an Euler solver is used.
some concepts of how to model these processes and how However, another factor is if the elevation change is large
to couple them. This coupling is key because many of the enough, then there is a possibility that the drainage direc-
interesting aspects of landscape evolution arise as a result tions in the code may be changed (i.e. the steepest slope
of the temporal or spatial coupling between processes. direction is changed) and the inflows and outflows into
Having covered the science details in the previous chap- that node are also changed (potentially dramatically)
ters, this chapter dives into the detail of how to couple because of the change in drainage area, leading to large
these processes in a model, calibrate it and test it against instabilities and model crashes at the next timestep. The
field data. experience of many LEM developers is that this latter
The core concept used to couple processes in time by drainage direction timestepping constraint is often the
most models is operator splitting, discussed in Section limiting constraint on timestep size when fluvial erosion
2.2.2. At each timestep the change due to each process is the dominant process. In this case the Courant number
over that timestep is calculated, and the impacts of all the constraint is not applicable, and using it typically leads to
processes are summed and applied for that timestep. At model crashes.
the next timestep the results of the previous timestep are These latter timestepping constraints have some inter-
used as the starting point. This simple idea hides some esting consequences.
subtle issues that may need consideration. The use of the The first consequence is that advanced solvers may not
results from the previous timestep as the starting point for yield the performance speedups conventionally seen in
each process means that during that timestep the processes other fields. A common method to speed up timestepping
are acting independently, and that the coupling occurs in traditional finite difference codes (e.g. a groundwater
through the summing of the processes at the end of the model) is to use an implicit solver instead of an explicit
timestep. This leads to a criterion on the maximum size of solver. Most LEMs use an explicit solver because it is
the timestep, which is that the timestep should be small significantly easier to implement. However, the solver
enough that the fastest changing interaction is captured. If instability that the Courant limit is designed to address
the timestep is too big, then (1) the interactions will not occurs only for explicit models, and implicit methods
occur as fast they would in the field system and the model allow a larger timestep. However, implicit methods are
will model the interaction as acting too slowly and (2) significantly more complex to code and require the inver-
instabilities may be created in the states that are interact- sion of a banded matrix that is N N where N is the
ing, leading to model mass balance errors or in the worst number of nodes in the domain. Domains of
case a model crash. 1,000 1,000 nodes are not uncommon, leading to the
A simple example may suffice to explain this principle need to invert a 106 106 banded matrix. If, however, for
in practice. To model erosion on a grid you solve the fluvial erosion the limit on the timestep size is the change
elevation change for a node; i.e. a mass balance is calcu- in drainage directions as a result of erosion and NOT the
lated on the inflows and outflows of sediment at that node. stability of the erosion solver at any node, then there may 255
256 Constructing a Landscape Evolution Model – Details
be no advantage in using a more sophisticated solver timestep reading the output files of both codes, extracting
because the problem that the solver is addressing is not out those data required as input for the next timestep and
the critical limit on the timestep size. constructing input files for the next timestep. While
The second consequence is how to write the computer writing/reading files to/from hard disks is slow, if RAM
code to couple the processes. It should be clear from the disks are used, then this slowness can be reduced to some
previous chapters that for some of the processes discussed extent. Solid state drive speeds are intermediate between
significant codes already exist to model them, and the hard drives and RAM and are an increasingly cost-
LEM developer is left with three alternatives as to how effective intermediate technology. This doesn’t require
to incorporate their capabilities in their LEMs. If we refer much software infrastructure other than scripts to run the
to the established process model as CodeA, then we can codes and set up the files for each timestep. However, the
(1) recode CodeA from scratch so that they fit within an two executable programs do need to be controlled by
existing landscape evolution framework, (2) refactor scripts. The scripting languages in Python or MATLAB
CodeA so it can be made to work within an existing both have the tools required to control executables that
landform evolution (typically by writing some wrapper use a command line interface and do the ancillary file
subroutines around CodeA) or (3) run the two codes (the modifications. However, the control by scripting generally
LEM and CodeA) in parallel and transfer the output requires that the executables not be run from a graphical
between codes using some intermediate computing infra- user interface (GUI) (e.g. controlled by a mouse) since the
structure (e.g. file transfers). current generation of Python and MATLAB scripting
The first option is a major waste of time. Recreating tools cannot control a GUI. This may be a problem if
CodeA from scratch, or translating CodeA from one lan- the GUI is heavily integrated into the computational
guage to another (if you are lucky enough to have access engine of the codes and so is difficult to isolate from the
to the source code for CodeA), and then debugging it, is science engine. Fortunately most science codes have fairly
potentially a major task. However, it is common practice. rudimentary GUIs, if they have a GUI at all, so this is
It also means that as developers update CodeA, then rarely a problem. Commonly science codes consist of
whether the LEM developer updates their LEM is a func- two separate programs, one which is the C or Fortran
tion of the availability of funds, time and energy, and it is computational engine, and then a separate graphical inter-
common not to have the most up-to-date science in face that may or may not be a separate program. In this
the LEM. latter case the graphical interface’s sole role is to provide
The second option is better but depends on (1) the the easy to use tools to create input files and interpret
CodeA having a structure that is amenable to having output files from the underlying computational engine.
wrappers written around it that can be used by the LEM We now return to the issue of what timestep to use.
and (2) the language of CodeA and/or the wrapper sub- The temptation is to use as large a timestep as possible
routines being compatible with the language of the LEM. (particularly for the third option) simply because the data
For instance, both Tellusim (Willgoose, 2009) and the transfer between codes is likely to be the limit on model
CSDMS BMI modelling framework (Peckham et al., performance. However, this may result in instability in the
2013) require four subroutines that can be called (a) interaction between processes. For instance, Temme et al.
program initialisation, (b) a computation loop, which (2011) coupled their LAPSUS SLEM with an agricultural
may be called repeatedly, (c) program finalisation and management model and used it on two case studies to
(d) data transfer routine(s) between the two programs. study the interaction between landform evolution and
For both Tellusim and BMI, if the model cannot be cast agricultural management options. They found that if the
into this four subroutines structure, then the coupling may timestep was less than one year, then the interaction
not be possible. The main difference between Tellusim between the processes in their model was satisfactorily
and BMI is that Tellusim is designed to couple with modelled, but for longer timesteps some of the inter-
models in the languages Fortran and Python/Cython, actions became unstable, even when the processes, when
while BMI provides support for a broader range of com- used on their own, were stable for timesteps of 50 years
puter languages. (water redistribution within the landscape) and 2,500
The third option is to simply run the LEM and CodeA years (tillage effects) (see their Figure 10). Thus the
in parallel and have some infrastructure to transfer data interactions forced the reduction of the timestep from
between the codes. At the simplest level this may involve 50 years to 1 year.
running both the LEM and CodeA for one timestep, Finally one advantage of having independent codes for
writing results to files, and then before running the next each process is that it addresses one major issue with
15.2 Model Testing 257
widely shared codes; the problem of code forking. Code Tellusim (Willgoose, 2009), SIGNUM (Refice et al.,
forking is where we encounter multiple versions of a 2012) and Landlab (Hobley et al., 2017) are all designed
‘single’ code, each of which has some unique capability to address this conflict between code forking and active
that a researcher has added to the code, but there is no code development by multiple research groups. The dif-
longer one single master code that contains all the cap- ferences are in the paths taken to achieve that objective.
abilities that all researchers have added. Thus when some- SIGNUM is a modelling system in MATLAB that mod-
body speaks about a model X, they may be referring to ularises the interacting processes but where each process
any number of multiple variants of a code, X1, X2, X3 is a conventional subroutine. In many respects SIGNUM
etc. Thus if a user needs the capabilities of X1 and X2, is akin to traditional programming of landform evolution
they need to manually merge those capabilities them- model, only it is done in MATLAB rather than Fortran or
selves. Nothing is stopping the ‘owner’ of the code from C. For example, both SIBERIA and CHILD are modu-
converging the various forked codes, but (1) this assumes larised in this way except they use Fortran and C, respect-
the ‘owner’ has access to all the variants of the code and ively, rather than MATLAB. On the other hand Landlab
that the developers are happy to share the source code for and Tellusim are designed around plug-in modules that
their variants, (2) the extensions in the variants may be have a standardised interface, and each module can be
incompatible and so cannot be converged into one master dynamically included or excluded without having to
code and (3) most research funding is to achieve insights modify the underlying landform evolution model. This
and results, and it is rare that funding support is available dynamic behaviour requires an amount of overarching
for code maintenance after a project is finished. However, infrastructure to support it as discussed above. Landlab
if a standardised interface/wrappers allows any extension is designed such that all modules are coded in Python,
to be used by the landscape evolution model without Cython or C, while for Tellusim they are coded in Python,
needing to modify the underlying landscape evolution Cython or Fortran. For both Landlab and Tellusim it is
code, then it is easier to manage the incipient ‘chaos’ that likely that other languages may be supported in the future
can occur when multiple researchers across multiple pro- because in both cases the overarching infrastructure is
jects and research groups are extending a code simultan- written in Python, and Python provides the capabilities
eously. This is a nontrivial management problem for any to link to many other languages.
successful open source computer code and is commonly
commented on by developers of open source environmen-
tal codes that are widely used (e.g. Holzworth et al., 15.2 Model Testing
2015). The plug-and-play and plug-in architectures in
The objective of model testing is to provide confidence
Tellusim, CSDMS/BMI and Landlab are designed to
that a model is working as expected and/or designed, and
begin to address this issue.
that it can be used to describe and/or predict behaviour
This discussion of how processes are coupled, and the
observed in the field. To provide this confidence the
various technologic ways that this may be done, may
model must be subjected to tests that are capable of
seem like overkill for many readers who are only inter-
identifying that the model is not working correctly when
ested in developing a small standalone landscape evolu-
it is wrong, and that does not have a high likelihood of
tion model to be used by themselves or at most a small
rejecting a good model (i.e. saying it is wrong when it is in
group of researchers. However, once the landscape evolu-
fact correct) or of accepting a bad model (i.e. saying it is
tion model escapes the lab environment and is used by
right when it is in fact wrong). The rest of this chapter
outsiders, the demands for project-specific extensions can
expands on these ideas and provides examples of methods
be overwhelming for the code developer (who no doubt
that have been used, or could conceivably be used, to test
has his or her own projects to worry about). Much of the
landscape evolution models. Along the way we will dis-
technology mentioned above is designed to address this
cuss aspects of the landscape evolution problem that limit
conflicting need to extend the model capabilities (by
model testing.
multiple groups) and the need to ensure that the basic
The first testing concepts are those of validation and
code is not subject to forking, while trying not to intro-
verification. The two concepts are often confused or con-
duce too much overhead into the code development pro-
flated in the literature:
cess. For instance, we don’t want graduate students, who
can be struggling to implement their own research code as 1. Verification is testing to ensure that the model solves
it is, to have to struggle with another layer of intellectual the problem as designed. Normally this is testing that
difficulty in the form of the coupling infrastructure. the mathematics of the model formulation is being
solved correctly by the numerical implementation. A second example is that calibrating the fluvial
Verification is normally done by running the model erosion model to data collected over a short period
against solutions for problems that are known to be (say a few years) may not accurately reproduce the
correct (e.g. analytic solutions to equations for simple erosion over the long term (hundreds to thousands
problems, other well-tested models, the same model of years), because aspects of erosion arise only after
with a different numerical solver) or convergence a significant time (e.g. incision, which is a function
testing (e.g. multiple runs with increasingly smaller of the feedback between erosion and landform evo-
timesteps and/or finer spatial discretisation to see that lution), and they are dependent on aspects of the
the results converge to a stable answer . . . though this erosion process not normally measured and cali-
approach can be fooled when the discretisation error of brated over the short term.
the solver is a function of the ratio of the timestep to
Another important aspect of modelling is determining
the grid size as is the case in advection problems; see
how reliable the model predictions are. If you make
Celia and Gray, 1991).
predictions (for a different site, or at some time in the
2. Validation is testing that the model correctly simulates
future), an estimate of the reliability of the projections
the practical problem being simulated (e.g. field or
may be required (e.g. best estimate standard devi-
laboratory experiments). This is a subtler task than
ation). As we will show below the validation process is
verification because it involves construction of a
key to determining the properties needed to make these
model. A ‘model’ is the combination of the computer
quantitative error estimates. For some engineering prob-
implementation of the mathematics, and the param-
lems where a statutory requirement might be a key
eters and boundary conditions used in the computer
design constraint (e.g. incision less than the thickness
implementation.
of a protection cap on hazardous waste) it may be neces-
a. Just how good this validation is, is then a question
sary to assess how likely the model projections are to
of how independent the data used in the construc-
meet this regulatory requirement. For example, if a waste
tion of the model is from the data used in the testing
containment structure must contain the waste against
of the fit. This latter point about independence is
release by erosion for 1,000 years, what is the probability
important because fitting a model to data (i.e. model
that this requirement will be breached during the design
calibration), and then asserting that this model is
lifetime of 1,000 years?
validated as a result, is incorrect. This just proves
that the model can be forced, with some choice of
parameters (realistic or otherwise), to fit a particular 15.3 Model Verification
situation so that the model is feasible. If, however,
the data were not used in determining the param- Model accuracy reflects various aspects of model per-
eters, then this is an independent test, and more formance. The most obvious is the numerical representa-
strongly validates that the underlying physics in tion of the processes accurately represents what is
the model and the process parameters leads to what happening in the field. Since LEMs are discretised in both
is observed in the field. Thus calibration of the space and time. an important question is what resolution is
model (i.e. determining the model parameters and required so that field processes are faithfully reproduced.
boundary conditions) is a separate task to validating All models are approximations of the real world, so this is
the model, and so validation requires data independ- a multistage question (1) how are the field processes
ent of those used in the calibration. This independ- converted into mathematics, (2) how is that mathematics
ence can be quite a severe constraint in earth spatially discretised and (3) how is that spatially discre-
sciences, and it is common for independent valid- tised model then discretised in time so that the temporal
ation to be only partly achieved. evolution can be modelled. The previous chapters of this
b. It is common to exercise only part of a model in one book have been focussed on the first of these stages,
or both of calibration and validation so that the generating the mathematics that reproduces field behav-
testing has validated only one aspect of the model iour. This section will focus on the latter two stages,
not all of it. For instance, if all the calibration and which is how accurately is the mathematic solved. How-
validation data were obtained in the diffusive part ever, the central purpose of this section is to describe the
of the landscape, you should not expect to be able methods that are used to assess the reliability of a model
to test and validate characteristics that are depend- as a whole and how this model can be validated against
ent on the nondiffusive components of the model. field data.
15.3.1 Model Spatial Resolution given elevation accuracy, the finer the DEM resolution,
the lower the slope accuracy. A simple example follows.
The main criterion for spatial resolution in an LEM is the Consider the calculation of slope between two adjacent
transition of the landform from concave down, divergent nodes A and B, separated by distance equal to the grid
flow (hilltops and ridges) to concave up convergent flow resolution Δx, where the variance of the elevation accur-
(valleys). In Chapter 3 we discussed how the slope-area acy of both A and B is σ e 2 so that the mean S and the
relationship and cumulative area diagram both allow variance σ S 2 of the slope estimate are
quantification of this transition area for catchments at
z A zB
dynamic equilibrium. In both cases we can calculate a S¼
Δx
catchment area at which this transition occurs. For the (15.1)
2
slope-area relationship, if we know the uplift (or the σS2 ¼ 2 σ e 2 σ AB 2
Δx
erosion rate . . . equivalent to a negative uplift), we can
estimate this area directly from the model parameters and where σ AB 2 is the covariance of the error between points
the analytic expressions for slope area. Zhang and Mont- A and B. If the error structure of the DEM is such that the
gomery (1994) analysed the saturation-excess hydrology errors in the elevations at A and B are independent, then
of a catchment that had a transition distance of about σ AB 2 ¼ 0. However, if the DEM is such that the error in
100 m and found no incremental benefit in a higher reso- the elevations is simply because the elevations have not
lution of 10 m, suggesting a grid size of about 10% the been georeferenced to ground control, then it is likely that
transition distance. Claessens et al. (2005) also found that all elevations at all nodes will be either biased high or low,
to model landslide hazard, a 10 m resolution DEM (the so that σ AB 2 ¼ σ 2 and the error in the slope is zero. For
highest they tested) was required, and this appeared to be example, Moreno de Las Heras et al. (2012) found that a
about 5% of the hillslope length (my visual estimate from major component of the errors in the ASTER DEM was
their DEM). Hancock (2005) found that a grid size of due to inaccurately ground-truthed stereo images, and
10 m was sufficient to capture hillslope shape and curva- found major jumps in elevations at image boundaries but
ture in natural catchments (i.e. no imposed drainage struc- little within one stereo image pair. Thus the correlation
tures such as contour banks), while Hancock et al. (2006) structure of the errors in the DEM is important.
demonstrated the potential errors in using coarse data such By comparing DEMs generated by three different
as the 90 m SRTM DEM. Wechsler (2007) reviewed methods (ground survey, cartographic photogrammetry
sources of errors in DEMs, including spatial resolution, and digital photogrammetry) Walker and Willgoose
and after reviewing the literature concluded that (1) higher (1999) showed that the derivation of the geomorphic
resolution is not necessarily better and (2) the optimal properties in Chapter 3 could be extremely error prone,
resolution ‘ultimately depends on characteristics of the and that the method of DEM creation changed the
study area such as topographic complexity.’ spatial correlation of the error structure and thus the
The limit of these works was that in many cases they slope errors.
did not have any higher resolution DEMs than 10 m, so all The derivation in Equation (15.1) is overly simple
that can be concluded is that coarser resolutions than 10 m because as the errors in the slopes become larger, another
were less satisfactory. This limitation may be resolved possible outcome is that the drainage direction may also
with the more recent availability of high spatial resolution change (the location of the next downstream node B for
LIDAR DEM data. For instance, Thomas et al. (2017) any particular node A). Gyasi-Agyei et al. (1995) found
found, after looking at LIDAR-derived DEMs of reso- that when the (elevation resolution or error)/(horizontal
lution ranging from 0.25 m to 5 m, that a 2 m DEM was resolution) approaches 1, it becomes difficult to reliably
optimal for capturing small microtopography effects on extract the drainage network. Other issues arise when the
drainage paths. only available data are those of coarse grid data. For
One important issue in determining the optimal grid example, the SRTM DEM allows unsurpassed global
resolution is the need to determine drainage directions and surface data; however, its coarse grid (at best 30 m) can
slope, and the need to use a field DEM as initial condi- provide poor catchment resolution (Hancock et al., 2006).
tions for the LEM or as a comparison with LEM predic- There are two conclusions:
tions. The error in the slope is a function of both the 1. That slope error is a function of the combination of the
elevation accuracy of the DEM and the horizontal reso- elevation error and the grid resolution, and neither
lution. For a given resolution, if the elevation accuracy is elevation error nor grid resolution should be con-
lower, then the slope accuracy will be lower, while for a sidered alone if slope is important.
(b)
(a)
FIGURE 15.1: The Ranger landform at times (a) 0 years, (b) 1,000 years. Note that the landform is vertically exaggerated with the horizontal length
being 1.1 km 1.5 km, and the elevation range is about 40 m. What appear to be gullies at 1,000 years are in fact valleys 7–8 m deep by 60 m wide.
2. DEM error can be significant so that when we discuss

model validation in a following section, it is important
to remember that the DEMs, which are the data against
which we compare our landform evolution models,
have their own errors which need to be considered at
the same time as model errors.
15.3.2 Model Timestepping Convergence Error
In this section we will demonstrate methods for quantita-

tively assessing model accuracy. For simplicity we will do
this by using an example of the SIBERIA landform evolution
model applied to a proposed rehabilitation of a mine site. This
DEM has been extensively discussed in papers by the author
and colleagues (e.g. Willgoose and Riley, 1998a, b), and it is
a convenient example for the discussions that follow because
in a single DEM it has an extensive range of features includ-
ing (1) areas of flow convergence and divergence in both FIGURE 15.2: Time evolution of the mean elevation of the Ranger
erosion and deposition regimes, (2) extensive areas of allu- landform over 1,000 years against timestep sizes ranging from 105
vial fan development on both flat and gently sloping terrain, year to 1 year.
(3) sudden changes in longitudinal slope that trigger gully
development, (4) a mixture of man-made and natural terrain does the mass change of the simulation match the exact
and (5) a defined objective for the solution of the problem, mass change. For most of our applications we do not
successful containment of the mining waste for 1,000 years. know what the exact change is because we don’t have
It is worth pointing out that this DEM is no longer the an exact solution. The major cause of inaccuracies in
proposed rehabilitation strategy, and the design lifetime has time-evolving models is the size of the timestep used, so
recently been extended to 10,000 years, so no conclusions to approximate the exact solution we use a model with a
should be reached by the reader about the performance of this very small timestep, assuming that this model is a good
landform for the currently proposed rehabilitation strategy. approximation to the exact solution. We then plot the
The parameters used in the simulations are those determined difference of the mass balance from the small timestep
in Willgoose and Riley (1998a, b). solution against solutions for various size timesteps.
One convenient way to characterise the accuracy of a Figure 15.1 shows the initial landform and the land-
model is the mass balance of the model, that is, how well form at 1,000 years. Figure 15.2 shows a plot of the mass
(a) (b)
(c) (d)
FIGURE 15.3: Plan of the Ranger landform at 1,000 years for four different times step sizes: (a) 1 year, (b) 0.1 year, (c) 0.001 year and (d)
105 year. Note the similarity of the locations of the valleys, even though the larger timestep results have less incision, which is consistent with
the lesser change in mean elevation for those times in Figure 15.2.
(i.e. (mean elevation) (area domain)) for the example • Large timesteps result in very large relative errors (mass
against timestep size (in years). Figure 15.3 shows the error/change in mass over 1,000 years) in the mass
initial DEM and the DEM for the different timestep balance, as much as 45% for the Euler solver.
sizes at 1,000 years. Figure 15.4 shows a classical error • The error declines as the timestep decreases, and this
analysis for time 1,000 years. The error shown in this decrease is roughly log-log linear (which is typical of
Figure 15.4 is the difference between (1) the DEM for convergence testing) with a slope of about 1.5 for
the given timestep size and (2) the DEM for the smaller timesteps, so the error decreases by about
smallest timestep (assumed to be close to the exact 30 times for a 10 times decrease in step size.
solution). Results for two different timestepping solvers • The nonlinear solver gives a better mass balance error
are shown (both discussed in Section 4.7.2), (1) a than the Euler solver for the same timestep size, by about
standard Euler explicit solver (solver = 1) and (2) the a factor 8 times for timesteps > 0.01 and 1.5 times for
nonlinear solver based on the approximate analytic timesteps < 0.01. Thus for the same mass balance the
solution to sediment transport (solver = 4; see Will- Euler solver needs a timestep about one-eighth that of the
goose et al., 1991a for details). A number of conclu- nonlinear solver when using large timesteps. This is
sions can be drawn from a cursory examination of these slightly offset by the longer CPU time for each
figures: timestep required by the nonlinear solver (about 25%).
It is still unfortunately true that when presenting a new

algorithm, no relative performance analysis is carried out
since it can be difficult to publish such papers. For
instance, Coulthard (2001) compared the performance of
models (among other things like breadth of physics imple-
mented, documentation etc.; but to his credit he did run all
the models himself rather than do a desktop comparison)
but did not ensure that the mass balance error was com-
parable, so it is not clear whether all the models generated
the same level of accuracy. Likewise other authors who
have claimed performance benefits from algorithm
improvements have not compared algorithms for the same
mass balance error, so it is difficult to independently and
quantitatively judge the relative performance of the algo-
rithms (e.g. Braun and Sambridge, 1997; Fagherazzi et al.,
2002; Perron, 2011). The only rigorous assessment
between two models has that been done comparing the
SIBERIA and CAESAR (and variants) by Hancock et al.
FIGURE 15.4: Convergence analysis of the absolute mass balance (2010, 2011, 2015a).
relative to the ‘correct value’ given by the simulation for timestep of
105 year.
15.4 Model Calibration

When using small timesteps, the differences between the
Model calibration is an important process, but even mod-
solvers decrease, since for small timesteps the non-linear
ellers can misunderstand it. None of the processes in the
solver converges to the Euler solver.
previous chapters have parameters that are universal con-
• For large timesteps both the solvers yield visually satis- stants. Even if you have to dig deeply into the process
factory results even if the relative mass balance errors
formulation, you will find that most parameters have been
are quite large (as much as 60% in this case). This
derived by fitting a process representation to either experi-
suggests that visual analysis is not a good way to judge
mental or field data. Thus all models have parameters that,
accuracy.
in some way or other, reflect a fitting process (i.e. calibra-
• While it does not occur in this example, it is common tion) and those data used in this fitting. On a naïve level,
for the explicit Euler solver to crash due to numerical
calibration is simply the process of fitting a model to some
instability for large timesteps, while the nonlinear solver
data. However, an important aspect of calibration is
is still able to provide a solution, albeit with a poor mass
developing an explicit estimate of the reliability of the
balance.
value for the parameter (e.g. standard error; parameter
value is mean standard deviation) for the data used in
The point of the example here is not to demonstrate calibration. Depending on the calibration method this
anything specific about one of the author’s models, standard error estimate not only reflects errors in the
SIBERIA, although the results here are consistent with parameter estimation process (and subtle questions of
unpublished performance assessments used in the incre- model parameter observability with respect to the data
mental improvements made to SIBERIA over the years collected), but also can also reflect measurement errors
since its original development. Rather, this section has in the calibration data and unknown randomness in the
exemplified, using a simple case study, how to use error data. An important use of these parameter standard error
analysis to assess the performance of competing algo- estimates is to determine how these errors propagate into
rithms and models. The error analysis here used mean the model predictions for (1) novel situations/applications
elevation because this is a direct measure of mass balance. at the same site (e.g. previous times, future projections,
It is also possible to use other measures of performance different climate) or (2) applications at other sites. Obser-
that might be relevant at a particular site (e.g. maximum/ vability is a technical measure of how well a parameter
mean depth of gullying, maximum/mean depth of depos- can be estimated if the data and model were perfect, and
ition) so as to determine the optimum timestep. characterises if a parameter can be calibrated from the
data. For example, measurements of groundcover might contributing area of the catchment increases and (3) the
help calibrate an erodibility parameter, but measurements parameter n determines how the erosion rate increases
of tree species might not, and this follows from the tree with increasing hillslope gradient.
canopy cover having only a small impact on erodibility, An important consideration when calibrating Equation
compared with an equivalent change in cover of ground- (15.2) is determining the ratio of m3 m to n. The ratio of
cover. Thus how well parameters can be calibrated reflects m3 m to n has a significant effect on the long-term topog-
not just the model but also the data. raphy and drainage network generated (Willgoose and
A model is ‘overparameterised’ if multiple combin- Gyasi-Agyei, 1995; Solyum and Tucker, 2004). Trad-
ations of parameters fit the calibration data set equally itional agricultural erosion models (e.g. USLE, RUSLE,
well. However, if a different data set is collected (e.g. the WEPP) have a fixed relationship between area and slope
same measurements but for a different climate, or a (i.e. ratio of m3 m : n), which may not be indicative of
completely different type of measurements), the differ- long-term spatial changes in erosion rate due to armouring
ences between the new data and the original data may and sediment sorting in space (Willgoose and Sharmeen,
mean the parameters might now be able to be uniquely 2006). This ratio of m3 m to n typically makes little
identified. A subset of this issue of overparameterisation difference to the short-term agricultural applications for
and poor observability is equifinality (Beven, 1996), which the traditional models have been optimised but is
where multiple different physics (different parameters crucial for long-term predictions when landform change is
or different equations) lead to the same long-term behav- significant enough to modify drainage patterns and there-
iour (e.g. Nicholas and Quine, 2010). In this case if, for fore the spatial distribution of discharge.
example, the only data you have is the long-term behav- The simplest method to calibrate Equation (15.2) is to
iour, you might not be able to uniquely calibrate the measure the long-term erosion from field erosion plots
parameters, and the model is overparameterised with with different areas and slopes. The different areas allow
respect to the long-term data (equifinality is normally the calibration of the power on area m3 m and the differ-
discussed with respect to long-term trends). However, if ent gradients allow the calibration of the power on slope,
other measurements are collected, it may be possible that n. Provided there are at least three plots (though more
the data may be able to distinguish the different param- plots are preferred to allow for experimental scatter), then
eters and/or models, and the model will no longer be the rate parameter β β3 can also be calibrated. This pro-
overparameterised (Hancock et al., 2010, 2011, 2014). cedure does not independently calibrate the discharge
As noted in Chapter 3 the classic case of overparameter- parameters β3 and m3 so that the model calibrated in this
isation in geomorphology is calibrating an erosion model way cannot model discharge independently of erosion.
solely to the slope-area diagram, which is the long-term The estimation of erosion could be done using field plot
form of the slope of the catchment. The slope-scaling data (e.g. Willgoose and Riley, 1998a, b used this method
exponent α can be easily determined, but this calibrates to calibrate the instantaneous erosion model used in the
only the ratio between m and n from the erosion equation next paragraph), or it could be done with synthetic desk-
(equally true for both transport or detachment limited), top studies using a traditional agricultural erosion model
not their absolute values. Other data are required to (e.g. USLE, RUSLE, WEPP) to predict the erosion on the
determine one or other of m and n. plots (Loch, personal communication). The latter method
For a discussion about model calibration we will ini- of calibrating to another model is subject to the aforemen-
tially focus on fluvial erosion processes, since these are tioned caveat about the ratio of m3 m to n assumed in
the main drivers of landscape form. As discussed in traditional erosion models.
Chapter 4 erosion processes (whether they are transport- A more complex approach that also calibrates the
limited or detachment-limited erosion) are of the form of discharge model was the approach of Willgoose and Riley
(1998a) used to calibrate the SIBERIA LEM (Figure 15.5)
Qs ðt Þ βQðt Þm Sn β β3 Am3 m Sn (15.2)
and more recently with other LEMs (Baartman et al.,
The key point here is that five parameters that need to be 2013; Coulthard and Skinner, 2016). Willgoose and Riley
fitted here: (1) β is the rate parameter and determines the calibrated a conceptual rainfall model to measured runoff
rate of erosion for a given discharge, while β3 determines from plot studies (both natural rainfall monitoring and
the runoff rate for the catchment so that the combined rainfall simulator trials). These parameters were then
parameter β β3 determines the rate of erosion per unit used with high temporal resolution pluviograph rainfall
area, (2) likewise the combined parameter m3 m deter- data (in this case 15-minute resolution for 20 years) to
mines the rate at which erosion increases as the predict the runoff for the catchments from the initial
FIGURE 15.5: Schematic of the calibration process used by Willgoose and Riley (1998a) (from Willgoose and Riley, 1998a).
landform for 20 years. In this way 15-minute runoff subcatchments) and discharges were calculated at the
records for 20 years for a large number of locations on outlet of these 30 subcatchments rather than for every
the landform were generated using the distributed Field DEM node. Despite the intensive calculations required
Williams hydrology model (Willgoose and Kuczera, by this approach (the runoff calculations took several
1995). Using the erosion plot data, an instantaneous sedi- weeks on a high-performance workstation at the time
ment transport rating curve was fit to the data so that the for the simplified subcatchment hydrology model), there
instantaneous sediment transport was related to slope and were a number of advantages of this approach, including
instantaneous discharge as discussed in the previous para- (1) both runoff and sediment transport models were
graph. This rating curve was then used with the 15-minute calibrated, (2) there was no need for a long continuous
resolution discharge time series to generate a 15-minute campaign of sediment transport monitoring on a number
resolution sediment transport time series for the same of plots to obtain the long-term erosion rate for different
locations at which the runoff series were generated. From areas and slopes and all that was needed was a number of
these data an annual average sediment transport rate was well-monitored events to which the hydrology model
determined, and the parameters β β3 , m3 m and n were and sediment transport rating curve could be calibrated
calibrated to the data. Likewise from the 15-minute runoff and (3) it avoided the need to use another erosion model
time series the parameters β3 and m3 were calculated. In to generate the erosion data for calibration of Equation
this way all five parameters were calibrated to the data. (15.2), with the aforementioned problems regarding the
These parameters were then used for the simulations as assumptions made in the model regarding the ratio
the landform evolved. m3 m : n in the erosion model. It is important to note
The computations required to run the rainfall-runoff that while the final result is an average erosion rate per
model at the resolution of the digital elevation model year (and its area and slope dependency), the erosion rate
(about 20,000 nodes) meant that an intermediate simpli- thus defined reflects the 15-minute variability of the
fication was required where the subcatchments were runoff process (see Section 4.3 for a discussion about
aggregated to make a smaller number of larger subcatch- the importance of temporal resolution and discharge
ments (20,000 nodes were aggregated into about 30 variability). This is our preferred calibration approach
for our industrial applications using SIBERIA and our only the fluvial erosion relationship is used, then over the
other LEM-based models. long term the landform will evolve to be concave up
A variant on this approach is when a significant pro- everywhere, without rounded hilltops. Thus to generate
portion of the erosion from plots is bed load. It is difficult realistic-looking landforms with rolling hills, a diffusive
to measure bed load during a flood event (as opposed to process must also be used. If we ignore what physical
suspended load, which can be easily measured during the process might be driving the diffusive process (see
event with a pumped sampler), so it is common to have a Chapter 9), we can calibrate the diffusivity. First, a fluvial
bed load trap that captures the cumulative bedload during erosion equation is calibrated. The LEM is run (typically
the runoff event (e.g. Evans et al., 2000; Hancock et al., to dynamic equilibrium) using the calibrated fluvial ero-
2017a). As a result, the bed load erosion data from a sion equation while setting the diffusivity K in Equation
single event is the cumulative sediment transport Qs,cum (9.1) to an assumed value. The value of K is then adjusted
during the runoff event manually until the inflection point of the slope area curve
ð (i.e. where the downslope landform concavity changes
Qs,cum ¼ βSn Qm dt (15.3) from concave down to concave up) is matched against
the analogue site being used for calibration (Hancock
Using a number of monitored runoff events, the calibra- et al., 2002; Wilson et al., 2006).
tion procedure for m is an iterative procedure of assuming The discussion above has focussed on how to cali-
a value of m, then estimating β and n that best fit the data brate the erosion model in an LEM. When dealing with
using that value of m, and then redoing these calculations paleo-reconstruction of a landform, there may be other
with another value of m. The best fit is then the value of m, processes that need to be calibrated, such as the tectonic
and the corresponding β and n, which best fits the data for history and spatial distribution of the tectonics over
the monitored runoff events. the landform. At the simplest level the slope-area
Another calibration method we have used is where we relationship provides an analytic solution to the average
are confident that the ratio m3 m : n has not changed tectonic uplift over a catchment, which can be evaluated
relative to erosion at another site, but where we know once the erosion and climate has been determined (Equa-
the long-term average erosion rate at the new site. We run tions (3.19) and (3.21)). However, if there is known
SIBERIA at the new site with the original m3 m and n spatial variability of uplift or erodibility, this spatial
parameters for a few years and then adjust β until the average may not provide enough insight. Willgoose
erosion estimated by SIBERIA for the site matches the (2001) used a DEM of a catchment in New Zealand near
estimate for erosion on the new site (Hancock et al., the Alpine Fault and mapped the spatial variation of
2017c). This is the approach we have used when we have U=K (Equation (3.19)) after fitting the other parameters
erosion estimates at the new site from (1) a single field of the slope-area relationship for the average behaviour
plot (so we have an estimate of the erosion rate but we of the entire catchment. The spatial pattern of U=K
don’t have the range of areas and slopes required to relative to its spatial average could be visualized and
calibrate the exponents), (2) erosion pins and (3) erosion patterns found consistent with the spatial distribution of
volumes estimated from landform differences obtained by valleys and ridges in the landform.
taking the differences between elevation obtained at dif- Finally a technique borrowed from hydrology called
ferent times (e.g. by ground surveying, photogrammetry, GLUE (Generalised Likelihood Unbiased Estimator;
terrestrial laser scanners or drones). We have found with Beven and Binley, 1992) has been used (Willgoose
(2) and (3) that while elevation changes are sufficient to et al., 2003; Braun and van der Beek, 2004; Braun and
estimate erosion rate, the amount of evolution of the Robert, 2005; Willgoose and Sharmeen, 2006; Nicholas
landform is generally insufficient to disentangle transient and Quine, 2010). GLUE involves randomly varying
landform evolution effects from the ratio of m3 m to n. parameters and/or forcings, and assessing the simula-
This method does not calibrate the hydrology model, as it tions using those parameters against the data. For this
only adjusts the erosion. example, we assume that simple least squares is the
This completes the discussion of how to calibrate the appropriate metric to assess the fit of the model to the
fluvial erosion equation. There are, however, other data. For each random parameter set θi where i indicates
calibration issues to note. If we remember back to the the ith set of randomly generated parameters, then the
discussion about the physical explanation of the slope- likelihood that the ith parameter set is the correct fit to
area relationship in Section 3.3.3, it should be clear that if the data is
1 measured in minutes to hours of CPU time. For example,

Li ¼ M
(15.4)
P 2 the GLUE simulations of Sharmeen and Willgoose (2007)
Eij Oj
j took nearly one month of CPU time on a desktop com-
puter, while Braun and van der Beek (2004) and Braun
where the summation in the denominator is the sum of the and Robert (2005) used a parallel supercomputer. That
squared differences between (1) the estimate Eij from the said, GLUE is easy to implement for parallel computing,
model using parameter set i and (2) the corresponding and modern multicore desktop processors are well suited
observation is Oj , where j denotes the jth observation and to GLUE simulation.
there are M observations. If the fit for parameter set i is
good, then the denominator is small and the likelihood Li
is large. There is a strict statistical definition for likelihood
15.5 Model Prediction and Validation
that has an additional term in the numerator that makes
this equivalent to the probability that this parameter set is Model testing and validation involve taking a calibrated
the correct parameter set. In practice GLUE takes an model and comparing it against data that were not used
empirical approach, and if there are N parameter sets with in the calibration and assessing the fit of the model
N corresponding values of likelihood, where Li corres- against these independent data. This independent data
ponds to parameter set i, then the likelihoods are normal- may take a variety of forms including (1) the same type
ised by calculating of data that was used in the calibration, but which was
Li not used in the calibration (e.g. data for a different time
pi ¼ (15.5) period than used in calibration, or if point spatial data
P
N
Li were used for calibration, then point data that were not
j
used in the calibration), (2) a different type of data that
where pi is then the probability that parameter set i is the was collected at the same or different time and/or loca-
P
correct parameter set and by definition Ni pi ¼ 1: Those tion as the calibration data or (3) the same or different
parameter sets that yield the best fits are the ones with the data for a different site.
largest pi and are most likely to fit the data. LEMs are sensitive to initial conditions, so slightly
Implicit in the GLUE methodology is the concept that varying initial conditions will lead to very different land-
all models are imperfect so there is no single ‘best-fit’ forms even if the physics is exactly the same (e.g. Lague
parameter set. Rather there is a range of parameter sets et al., 2003). There is evidence that this is also true for
that adequately fit the data to a greater or lesser extent. forcings (climate, etc.) and parameters, so that we have
Those parameter sets that fit the data best are more likely many cases where the details of the landform (i.e. exact
to represent the underlying processes than those parameter position of hills and valleys) vary quite wildly with only
sets that fit the data less well. GLUE allows the user to small changes in the inputs, yet the geomorphology stat-
eliminate parameters/forcings that don’t give feasible fits istics of the landform (see Chapter 3 for examples of these
to the data and prune the region of feasible parameters for statistics) are the same (Hancock et al., 2016; 2017c). This
the data (e.g. Nicholas and Quine, 2010) or prune the means, for example, that a direct comparison between a
types of physics that might be active at the site (e.g. model and a validation landform may be poor even
Temme et al., 2011), and it works well with large, dispar- though the model is good, simply because the valleys
ate data sets. The full benefit of GLUE will become are in different locations. Accordingly a test of a LEM
apparent in the next section when we discuss model should not be directly between the landform generated by
testing, but even at a pragmatic level this technique eluci- the LEM and the DEM in the field (i.e. not, for instance,
dates one aspect of parameter calibration: how unique is the sum of the differences in elevations at each node
the best fit parameter set? If two very different parameter location) but against statistical properties of the land-
sets fit the data well, then the observability of those form/landscape (i.e. some statistical property of the eleva-
parameters is poor using those data. The main downside tions such as the hypsometric curve).
of GLUE is the numerical intensity of the approach. Even The challenge then is to find statistics that are sensitive
for a few parameters it is common to need a few thousand to the physics being tested, but not sensitive to random
simulations, which may be practical for the relatively variability (i.e. that have a high signal-to-noise ratio).
simple hydrology models for which is was developed Ideally, these statistics would also capture some form of
(e.g. Chen et al., 2015), but can be difficult for LEM fundamental property of the landform. Some examples
applications where individual simulations may be follow:
15.5 Model Prediction and Validation 267
1. Willgoose (1989) showed using his SIBERIA much difference is acceptable? In hypothesis testing this
landform evolution model that Horton/Strahler/Toka- involves having a mean value for both the model and the
naga network topology statistics were very sensitive to field data, and an estimate of the expected variance of the
random variability in the channel network. While it difference between then so that we can calculate a prob-
was possible that they might have some signal of ability that the model and data are or are not statistically
underlying physics in the channel geometry, this significantly different. Thus being able to provide a confi-
random variability swamped any signal, so that they dence limit on both the data and the model is desirable so
were not a useful test of links between physics and that we can perform a hypothesis test on whether the
channel topology. model and the data are different.
2. Willgoose et al. (1991c) showed that the slope-area During the early development of LEMs it was
curve has a direct link to erosion physics in the channel common to see whether the model could generate the
and hillslopes. However, the scatter around the mean qualitative and quantitative characteristics of catchments
trend is so large that it has proven difficult to use the (e.g. Willgoose et al., 1991a, b; Hancock et al., 2002;
slope-area curve to test models against field data. That Lague et al., 2003). This involves adjusting parameters
said, if you know the channel network planar location, and/or governing physics to examine (typically visually)
then the channel elevations can be deterministically if the models can replicate the field behaviour. This is a
reconstructed from the channel network (this was the test that simply says that it is feasible that the model might
basis of the QUEL model for Quick Equilibrium Land- simulate observed field behaviour. What it does not say is
forms; Ibbitt at al., 1999), so the slope-area curve is a that the physics and parameters used in that modelling
key predictor of the elevations in catchments so has exercise are exactly the same as the physics and param-
been a focus of much subsequent work by the author eters applicable in the field. The match between model
and others. and field data may just be fortuitous. However, if, for
3. Perera and Willgoose (1998) showed the cumulative example, we refer back to the example in numerical
area diagram (Rodriguez-Iturbe et al., 1992) could be convergence testing used in Section 15.3.2, Figure 15.3
asymptotically derived from Strahler statistics but shows that even where there are significant numerical
without the aforementioned noise. It appears that the errors in the mass balance, the difference in the generated
cumulative area diagram might be insensitive to phys- valley network is visually small. This is just one example
ical processes (e.g. the slope of 0.5 seems to be of many that demonstrate that visual similarity is a weak
universal for all catchments), but that has not been test of model performance.
proven. The early testing of LEMs with experimental landform
4. Using the slope-area curve and cumulative area simulators involved either examining whether a model
diagram, and assuming spatially homogeneous soils, could be calibrated to simulator evolution (e.g. Hasbergen
Willgoose and Perera (2001) showed that you could and Paola, 2000; Lague et al., 2003) or calibrating the
predict the percentage of saturated area in a catchment, LEM to one-dimensional hillslope erosion experiments
and thus the runoff ratio for a catchment. and then comparing the model to independent two-
5. Willgoose and Hancock (1998) examined the hypso- dimensional landform evolution experiments (e.g. Han-
metric curve and integral, and found a link between the cock and Willgoose, 2001, 2002, 2004; Temme et al.,
slope-area curve and catchment shape. Both lack the 2011), using a range of geomorphology statistics and
scatter of the slope-area curve, and they were able to functions (see Chapter 3) and assessing visually whether
use them as statistics in comparisons between the these statistics were different between model and experi-
SIBERIA landform evolution models and laboratory ment. This same methodology was also used when exam-
landform simulator experiments (Hancock and Will- ining the evolution of degraded mine sites and
goose, 2001, 2002). See Section 2.3.3 about some undisturbed field sites (Hancock et al., 2000, 2002; Han-
caveats on the rescaling that is done to plot the hypso- cock, 2003). The limitation of this approach is that it does
metric curve, and which have implications for the not provide a statistical measure of how bad the fit
value of the hypsometric integral. between the statistics needs to be before we should
declare the model unsatisfactory.
Finally a key component of model validation and The main problem with estimating the variance of field
testing is being able to assess, with some objective data is the difficulty of replication of experiments. In
method, when the difference between the model and the classical field trials, as described in experimental design
test data is significant and when it is not. That is, how textbooks, a number of identical experiments are carried
out and the confidence limits determined from the Using GLUE in this way we can test a model to see
observed variability between the plots. However, in most how well it fits the independent validation data, and also
cases in the earth sciences and geomorphology, it is not make an objective assessment of how different the model
possible to do replicated experiments. In the field every and the validation data need to be before we can object-
catchment is different (different shape, geology, climate), ively conclude that the model is unsatisfactory.
so it is impossible to know whether the scatter in the Using GLUE Willgoose and Sharmeen (2006) cali-
relationships is a function of these differences or is just brated their armouring model ARMOUR to multiple plots,
the random sampling effect. and then used the parameter set likelihoods to calculate
GLUE can be used to circumvent these limitations. In error bands on predictions for other plots and other runoff/
the previous section we discussed how in GLUE erosion events. This validation testing showed that the
calibration a number N of parameter sets θi are generated model worked well for two different materials at two
where the subscript i is the ith parameter set, and that for different field sites. Furthermore they evaluated seven dif-
each parameter set a probability pi that it is the correct ferent potential armouring models (i.e. there were seven
parameter set for the model is calculated. The confidence different models, each was calibrated and validated against
limits of the model predictions are calculated as follows: the data) for armouring on hillslopes and were able to use
these prediction limits to decide that five of the proposed
1. For each parameter set θi the prediction of the model Pi
mechanisms could not simulate the field data. This shows
is calculated.
that GLUE can be used to eliminate infeasible physics (i.e.
2. At the end of step 1 you will have N predictions from
physics that could not be calibrated or validated satisfactor-
the model, and each prediction Pi has a probability that
ily against the data). While they did not validate their
it is correct pi .
model, Braun and van der Beek (2004) used this approach
3. The probability distribution for the predictions is then
to eliminate potential mechanisms for a crustal temperature
constructed by sorting the predictions from lowest to
model when they calibrated their thermal model against
highest. A cumulative probability function (cdf ) of the
crustal temperature data.
probability is derived where the cdf is the probability
Finally, Sharmeen and Willgoose (2007) used the like-
that the model prediction is greater than a given value.
lihood estimates for the parameter sets from GLUE to
To construct the cdf the first value in the sorted series
calculate confidence limits on predictions for erosion for
has a probability p1 that the prediction is less than that
600 years into the future. While these estimates could not
value where p1 is the probability of the first (lowest)
be validated, the value of using GLUE in this way was that
prediction in the sorted series. The second value of the
it provided an estimate of the confidence that could be
cdf has a probability that is the sum of the first and
placed in the predictions. In our industrial applications for
second probability in the sorted series, and so on.
mine sites and nuclear waste repositories this confidence
4. The confidence limits can then be calculated from this
estimate is a key outcome. Industry users of LEMs are
cdf. For instance, if we wish to plot the 90% confi-
unlikely to invest millions of dollars into remediation
dence limits (the values that 90% of model predictions
efforts unless they are confident the remediation will work.
fall between), then we simply need to look up the value
More specifically they reject the model results if (1) pre-
of the prediction for the probability of 0.05 and 0.95
dictions show a low degree of confidence or (2) where the
(5% of data below the lower limit and 5% above the
possible range of model performance (e.g. those results
higher limit).
inside the 90% confidence limits) encompasses a wide
When the confidence limits are calculated a compari- range of satisfactory and unsatisfactory performance.
son can be made between the observed and predicted The objective of the GLUE analysis above is to provide
values, and a conclusion reached about whether the model confidence limits on the model predictions so that when
and the data are consistent. For instance, for 90% confi- validating the model against independent data we can
dence limits it is expected that 5% of the observed data objectively assess if the model fit is too poor to be satisfac-
will lie above the upper limit and 5% below the lower tory. The main advantage of GLUE versus other tech-
limit. If more than 90% of the data lies between the niques is that it is conceptually simple even though it is
model’s 90% confidence limits, then the model is satis- computationally intensive. A limitation of GLUE is that it
factory. If more than 10% of the data lies outside the implicitly assumes that the sole reason for the misfit to the
limits, then it can be concluded that the model is not validation data is the uncertainty in the calibrated param-
satisfactory. In this way we can obtain a probability that eters; GLUE implicitly assumes that the model and the data
the data are consistent with the model. are perfect and not subject to their own errors. More
advanced techniques that aim to separately characterise the between model and observed topography, because they
impact on predictions of parameter uncertainty, model did not consider parameter or data errors.
error and measurement error are available. Examples We will finish this section to do with model calibration
include Kalman filter (e.g. Bras and Rodriguez-Iturbe, and testing with an overall observation. Traditional
1985) and BATEA (Kavetski et al., 2006). However, these hypothesis testing (and what is normally taught in experi-
methods are significantly more conceptually and computa- mental design courses) involves multiple replicates of field
tionally complex, and won’t be discussed here. tests so that the variability used in statistical tests (e.g.
We face other causes of uncertainty in model outputs t-test) is derived from the scatter from field plot to field
than just parameters, and these include model forcings plot. The null-hypothesis then used for the hypothesis
and initial conditions, such as the initial topography, testing is ‘Is the model different from the field data?’ and
paleoclimate and paleotectonics. Willgoose and Gyasi- this places the field data at the centre of the testing. As
Agyei (1995) and Willgoose and Riley (1998a, b) exam- noted previously, it is rare in the geosciences to have the
ined the impact of uncertain initial conditions on landform luxury of multiple replicates of data where everything other
evolution and found a significant effect on the determin- than the test property is kept constant. Accordingly the
istic planar geometry of the evolving hill/valley network discussion above has for the most part inverted this trad-
with markedly different locations for hills and valleys/ itional emphasis, since we can’t repeat field experiments
gullies. However, it was possible to define areas of the but we can repeat computer experiments. Provided we
landform that had a high probability of being subjected to know the sources of uncertainty and can quantify them,
excessive erosion rates for all the initial conditions, thus we can then use Monte Carlo simulation as discussed
allowing the mine owner to make an objective risk assess- above to generate the variability in the model predictions.
ment as to whether the probability of failure of their The hypothesis test then becomes ‘Is the field data different
encapsulation was too high, and, if so, how to modify from the range of behaviours predicted by the model?’
the design to reduce the probability of failure. Willgoose Some researchers have expressed dismay to the author that
et al. (2003) placed these insights into a formal statistical we are thus elevating the model to be truth, and we appear
framework and introduced the idea of using t-tests, to be testing whether the field data are correct. This is far
χ2-tests and Bayesian statistics to test whether the LEM from the truth, and in a rather difficult analysis in the
predictions for their Tin Camp Creek field site (Hancock second half of Willgoose et al. (2003) we showed that these
et al., 2002) fell outside the confidence limits of the two approaches are equivalent if we know and can quantify
modelling (and thus whether the landform was statistic- all sources of error. There are two subtleties to this equiva-
ally significant different from the model). It is important lence that arise if we underestimate the errors in the model
to note in the context of the discussion in this section that (e.g. we ignore some sources of uncertainty, or we under-
Willgoose et al. (2003) only varied the initial conditions. estimate the errors in the model or data by deciding the
The model erosion and hydrology parameters had been model/data is more reliable than it actually is). The first
estimated independently and were assumed perfectly subtlety is that if the errors are underestimated, then the
known in that study, so the parameters were not varied confidence limits used in the hypothesis testing will be too
as in GLUE. This meant that the confidence limits in tight and we are more likely to reject a good model. The
Willgoose et al. (2003) were too narrow because they second subtlety is that if these underestimated errors are
didn’t allow for uncertainty in model erosion parameters. used in independent predictions into the future (not for
Finally the DEM for Tin Camp Creek had been derived by hypothesis testing but, for instance, for assessment of the
digital photogrammetry and included errors in the eleva- safety of a waste encapsulation design), then the model will
tions from the stereo photo matching. However, Will- underestimate the amount of uncertainty in the predictions
goose et al. (2003) assumed that the topographic data and overestimate the confidence in the model predictions.
error was zero. At another field site Gyasi-Agyei et al.
(1995) and Walker and Willgoose (1999) showed that the
impact of digital photogrammetry errors on geomorph- 15.6 Further Reading
ology statistics (slope-area, cumulative area diagram
etc.) can be significant. These photogrammetry errors A recent publication that nicely summarises model testing
widen the confidence limits. Thus, while Willgoose in the field of hydrology is Pfister and Kirchner (2017).
et al. (2003) concluded that the LEM and field data were While it has a hydrology focus, many of the hypothesis
significantly different, they showed only that the initial techniques (and the errors in using them) are applicable
conditions could not be responsible for the differences across the geosciences.
16 Examples of Landscape Evolution
Modelling and Final Thoughts
16.1 Some Examples of Coupled Modelling evolution of soil depth and elevation (Equations (2.7)
and (2.8)) with time across a one-dimensional landform
This section will briefly highlight some of the applications (the landform being a cross section at right angles to the
to which landscape evolution models have been applied. ridge/valley axes and across a series of valleys) and
The previous technical chapters can seem a bit dry, and assessed what proportion of soil in hollows arose from
the motivation of this section is to provide some intuition hillslope diffusive transport versus in-situ residual
about how and why various processes have been coupled, weathering processes. Minasny and McBratney (2001)
what the compelling research questions were/are that the extended this analysis to a two-dimensional landscape
studies were designed to address, and the types of results and included a term for chemical weathering loss of soil
that were obtained. A comprehensive discussion of all mass that increased with time (empirically reflecting
work done in this area is too extensive to discuss here, increasing chemical weathering as the soil grading
so this section will give the reader a flavour of the types of becomes finer). The results were qualitatively consistent
applications and publications from these applications to with the earlier one-dimensional paper with hillslope
give him or her a place to start their online searching. soils being transported downslope and deposited into
The focus in the sections below is in the interaction the hollows.
(and potentially coevolution) of the processes. The Saco et al. (2006) extended the two-dimensional land-
behaviours are what are a result of these interactions that scape approach by making the soil production function a
are the ultimate objective of landscape evolution models. function of soil moisture and then predicting the soil depth
If we have two processes A and B, then we are not only and moisture using a variety of alternative formulations of
interested in how A impacts B, but also how the feedback the topographic wetness index. The soil depth model was
from B back to A changes both processes. There is the coupled to the SIBERIA landform evolution model that
possibility of (nonlinear) positive and negative feedbacks modelled both fluvial and diffusive erosion, rather than
between A and B that result in spatial and temporal only the diffusive transport used by Minasny. They were
organisation and/or complexity in the processes. able to replicate the localised regions of deeper soils and
valley-like forms in the soil-saprolite/bedrock interface
16.1.1 Coupled Soilscape and Landform (Figure 16.1) observed by Freer et al. (2002). The spatial
Evolution Models patterns of Freer’s soil depth and the elevations of the
soil-saprolite interface were different from that of the soil
Chapters 5 to 11 discussed standalone soilscape evolution surface, so the bedrock-lowering process (and thus the
models, with specified landforms and vegetation. A number dynamics of soil depth) appeared to be decoupled from
of published studies developing models couple a soilscape the dynamics of the soil surface. The key conclusion of
model (typically a one-dimensional pedon or profile Saco’s work was that if the soil moisture distribution is
model) with a landform evolution model. determined using the surface topography (the classic
The earliest work coupling soilscape and landform definition for the wetness index), then the bedrock
evolution models was Minasny and McBratney (1999, topography mirrored the soil surface topography, and it
2001). Minasny and McBratney (1999) coupled a diffu- was only if the soil wetness was distributed using the
sive surface transport model (Chapter 9) with an expo- bedrock topography that patterns similar to Freer’s were
270 nential soil production function (Chapter 6) to predict the simulated. In contrast, Minasny and McBratney’s earlier
between the layers and that declined exponentially with

depth below the soil surface (see Section 8.3.4). The
paper did not model vegetation. The landform evolution
model had a simple kinematic wave runoff model which
generated the discharge used in the detachment- and
transport-limited erosion model. They conceptually mod-
elled armouring at the surface as a result of selective
removal of fine soil fractions. The transport capacity of
the flow was independent of the grading of the sediment
either on the surface or in the flow, and the entrainment of
sediment into the flow during erosion was weighted to
entrain finer sediments preferentially. It deposited sedi-
ments equally irrespective of grading (Temme, personal
communication).
FIGURE 16.1: Spatial distribution of soil depth using a coupled Using LAPSUS Temme and Vanwalleghem (2016)
exponential soil production function and landform evolution model, carried out a sensitivity study to understand which of the
where the soil production rate increases with soil wetness and the model processes and parameters were most important in
soil wetness is determined using the bedrock topography (from Saco defining the soilscape and landform evolution. They
et al., 2006).
concluded that physical weathering and how it declined
down the profile was the most important process
work, which did not include soil wetness so that the soil (Figure 16.2). Physical weathering broke down the larger
production function did not vary in space, showed an less transportable particles into smaller more transportable
almost perfect match between (1) valley and deeper soils particles, thus being a key determinant of the erosion rate.
and (2) ridges and shallower soils, so that the spatial This was consistent with Sharmeen and Willgoose (2006)
distribution of the soil depths almost perfectly reflected who found that the weathering process limited the erosion
the surface topography. rate by determining the rate of generation of fine trans-
Recent coupled models have increasingly focussed portable material, producing a balance between armour
on the spatial variability of the soil profile characteristics development and armour breakdown. Temme and
(e.g. grading, organic carbon) rather than soil depth Vanwalleghem also found that the depth dependency of
alone. Temme and Vanwalleghem (2016) developed the the physical weathering was important because it deter-
LORICA model. LORICA is a soil pedon model (an mined the grading of the material just below the surface
enhanced version of MILESD, Vanwalleghem et al., armour layer.
2013) coupled with a landform evolution model Chemical weathering became important only toward
(LAPSUS, Schoorl and Veldkamp, 2001; Schoorl et al., the end of their simulations since its rate was determined
2014). The soils pedon model modelled the grading of by the specific surface area and was thus limited by the
the soil with five mineral fractions (coarse, sand, silt, clay small initial amount of clay in their profile. The clay in
and fine clay) and organic matter, and this profile was their simulations was initially low but increased over time
discretised with depth. A number of soil processes were (Figure 16.2). For clay formation they found that all
modelled as discussed in the preceding chapters includ- parameters of the clay model were important, but still
ing (1) physical weathering that fragments soil particles found the depth dependence of the physical weathering
as in Chapter 7, (2) a conceptual chemical weathering to be important. Physical weathering was important
process that was linearly proportional to specific surface because it generated the smaller particles (and increased
area of the soil particles and that declined exponentially the specific surface area) from which chemical weathering
with depth below the soil surface (see Chapter 8) and could then generate clays (see Figure 8.21).
chemical weathering dissolved the rock at the surface and Their conclusion of Temme and Vanwalleghem
precipitated it as clay, (3) clay eluviation as in Chapter 10, (2016) was that the most important parameters for the soil
(4) soil organic matter (SOM) in two compartments (see properties were the rates for clay formation and eluvia-
Chapter 10 for compartment modelling) and with all tion, and the rate of physical weathering. Figure 16.2
SOM being sourced within the profile at a rate that suggests that their soils and landforms had not reached
declined exponentially with depth and with no SOM equilibrium at the end of their simulations because the
source at the surface and (5) bioturbation that mixed soil states in Figure 16.2 were still changing at the end of the
272 Examples of Landscape Evolution Modelling and Final Thoughts
FIGURE 16.2: Soil properties over time as simulated by LORICA: (a) erosion and deposition, (b) the number of computational nodes eroding
or depositing, (c) the rate of physical weathering, (d) the rate of chemical weathering (note the log scale), (e) the rates of formation and
movement of clay and (f ) average soil thickness over the domain (from Temme and Vanwalleghem, 2016).
simulation. This may impact on their conclusion that above indicated that, at least for young materials, physical
chemical weathering issues were second order since weathering was a first-order process, and chemical
Figure 16.2e shows that the finest clay fraction was still weathering was second-order. Like Temme and Vanwal-
changing quite strongly with time, and since the fine clay leghem, Cohen and Welivitiya found that the soil profile
fraction was created by chemical weathering, it is likely grading, erosion properties of the surface and the temporal
that at times beyond those simulated the chemical evolution of the landform were critically dependent on the
weathering processes would be more important. depth dependency of the weathering process.
In complementary work Welivitiya (2017) presented Figure 16.3 shows three hillslope profiles (Welivitiya,
results using his SSSPAM coupled soilscape-landform 2017) that have (1) a declining weathering rate with depth
evolution model. The soilscape model was an improved below the surface (exponential), (2) a low rate at the
version of mARM3D (Cohen et al., 2010), while the land- surface increasing to a maximum within the profile and
form evolution model was based on ARMOUR (Willgoose then declining thereafter with depth (humped) and (3) the
and Sharmeen, 2006) and SIBERIA (Willgoose, 2005b) highest rate at the soil-saprolite interface declining with
but using only transport-limited erosion where the trans- distance above the interface (reversed exponential,
port capacity was a function of the grading in the flow. simulating a simple chemical weathering profile; see
Both the entrainment and deposition processes were a Figure 8.22). The three resulting soilscapes and landforms
function of the grading in the flow, so there was selective are markedly different. For the humped and reversed expo-
entrainment of fines during erosion and selective depos- nential profiles, once an armour is formed on the surface it
ition of coarse fractions based on settling velocity in is difficult to remove it because the weathering rate at the
depositional areas. The focus of the soilscape modelling surface is low. However, while there is a stable armour at
by Cohen and Welivitiya was on physical weathering the surface, the portion of the profile with the maximum
processes and this was justified because (1) physical weathering rate (i.e. the hump for the humped, the soil-
weathering is dominant for the mine rehabilitation and saprolite interface for the reversed exponential) below
nuclear waste problems motivating his work and (2) the the surface is weathering relatively rapidly, creating a
conclusions of Temme and Vanwalleghem discussed reservoir of highly erodible material just below the surface.
FIGURE 16.3A: Soil and landform profiles over time for a one-dimensional slope: (A) the exponential weathering profile, (B) the humped
weathering profile and (C) the reverse exponential. In all cases the upper panel is the long section through the profile showing the landform, the
soil profile d50 (see the colour bar) and original landform (the solid line), the middle panel is the d50 of the surface showing the armour grading
and the bottom panel is the soil profile where the top of the panel is the top of soil (from Welivitiya, 2017).
FIGURE 16.3B (cont.)

FIGURE 16.3C (cont.)

If the armour is destabilised, then the erosion rate acceler- obvious in Figure 16.3, this results in a banded soil profile
ates rapidly as the overland flow cuts through the reservoir where each band is formed by a cycle of armour destabil-
of fine material down to the coarser material underneath isation and downcutting, and oscillations in the erosion of
the band of fine material, and over time a stable armour the catchment which is most clearly seen in oscillations of
forms again and the cycle starts again. Though it is not the surface armour (Figure 16.4).
FIGURE 16.4: The evolution of the surface d50 with time for a planar hillslope for (a) exponential, (b) humped and (c) reversed exponential
weathering functions (from Welivitiya, 2017).
FIGURE 16.5: Planar view of the d50 of the surface grading of the
alluvial fan. Note the linear filaments of coarse material propagating
from the top of the fan at the centre top of the figure. The white area
does not have significant deposition, and the localised rectangular
areas at the top extreme left and top extreme right are artefacts of
the design of the experiment (from Welivitiya, 2017).
On a related topic Cohen et al. (2015, 2016) used his

mARM model to show how the deposition of a fine
aeolian fraction can destabilise an armour by reducing
the average particle size of entrained sediment and thus
FIGURE 16.6: The Langbein and Schumm relationship between
increase the sediment transport capacity of overland flow. erosion rate and climate: (a) the erosivity, which is a function of
Cohen’s conclusion was that even though the overland runoff and is proportional to erosion rate in the absence of vegetation
flow could not carry the coarse armour by itself, if the protection, (b) the erodibility of the landscape, which reflects the
armour particles are mixed with finer (aeolian) materials, increasing protection of the surface by vegetation cover and (c) the
actual sediment yield which is the multiple of erodibility and erosivity
then those coarse armour particles can be moved.
(from Moglen et al., 1998; after Langbein and Schumm, 1958).
Finally, Welivitiya simulated the deposition of an
alluvial fan (Figure 16.5) and found that his selective
deposition (on particle diameter) model created banding
vegetation protection) and a high erosion peak at an
in the particle size distribution down through the deposit,
intermediate rainfall (200–300 mm/year; Langbein and
some in-situ weathering within the deposits and a spatial
Schumm identified this rainfall as the transition between
pattern of grading across the fan, driven by avulsions, that
shrubland and grassland). They concluded, using their
was consistent with experimental and field data for allu-
LEM, that for deserts, drainage density (or valley density)
vial fans and modelling of submarine fans (Koltermann
increased with increasing rainfall, for humid/wet climates,
and Gorelick, 1992).
drainage density decreased with increasing rainfall, and
for intermediate rainfalls, the drainage density deceased
16.1.2 Coupled Vegetation and Landform with both increasing and decreasing rainfall, and that the
Evolution Models maximum drainage density occurred at the intermediate
rainfall when erosion peaked.
Moglen et al. (1998) examined the impact of climate Collins et al. (2004) and Collins and Bras (2010)
change on drainage density using an empirical relation- examined a coupled model of erosion and vegetation
ship between climate, vegetation and erosion that dates where the vegetation growth model simulated vegetation
back to Langbein and Schumm (1958). This relationship cover (not biomass) in Equation (14.6). Collins and Bras
(Figure 16.6) says that as the climate becomes wetter, the revisited the work of Moglen by applying their vegetation
vegetation cover increases, offsetting the increase in dynamics model and showed that the coupled model
erosion rate as a result of increased runoff. The result is could generate the coupling between erosion, vegetation
a curve (Figure 16.6c) that has a low erosion rate for a dry and valley density in Figure 16.6. Moreover, they high-
climate (i.e. no runoff ) and wet climate (i.e. heavy lighted that there is a third regime that occurs for even
the entire domain on average once every 200 years. The

vegetation model was coupled with the erosion model so
that erosion dropped with increased vegetation cover
(Section 14.2.1), and landsliding rate increased after fire
occurrence (see Sections 13.1.2 and 14.5.5). They per-
formed a number of landform evolution simulations to
explore the interactions between these processes. In the
first they simulated the difference between an unvegetated
and vegetated catchment for the case where the vegetation
had no impact on the landsliding rate. Unsurprisingly
(Figure 16.8), by reducing the erosion rate without
FIGURE 16.7: Schematic of the three regimes of erosion with the impacting on the diffusive rate of transport (due to land-
third region of increasing erosion rate with increasing rainfall on sliding), they generated a catchment with much longer
the far right of the figure (from Collins and Bras, 2010). hillslope length and greater distance between valleys. This
is consistent with the slope-area relationship (Chapter 3)
for two competing processes where the reduction of flu-
higher rainfalls than in Figure 16.6. In this third regime, as vial erosion rate results in the inflection point in the slope-
rainfall increases, the vegetation cover increases only area relationship shifting to larger areas (i.e. reduced
slowly because it (1) is already quite dense and (2) the valley density). Most of the catchment gradients in
marginal benefit of increased vegetation cover decreases Figure 16.8 are limited by the angle of repose, otherwise
markedly for high vegetation cover (Figure 16.7). Collins we would also expect the maximum gradients to increase
et al. (2004) examined the effect of the competing pro- with this shift of the inflection point. For the cases where
cesses of plant growth and death, and removal of vegeta- the vegetation is periodically removed by landsliding
tion by high shear stress during extreme runoff events. As and/or fire, the landform developed was intermediate
vegetation became denser, the shear stress threshold between the two landforms in Figure 16.8. Collins and Bras
increased. This had two consequences. The first conse- (2010) also found that when the erosion was less (as a result
quence was that as the vegetation cover became denser, of vegetation cover), the hillslope gradients increased.
the average erosion decreased and the slopes in the catch- Yetemen et al. (2015a, b) noted that in previous work
ment increased (this is consistent with the slope-area (e.g. Istanbulluoglu and Bras, 2005; Collins and Bras,
relationship and Equation (3.19)). The second conse- 2010; Saco et al., 2007) coupling vegetation to landform
quence was as the erosion threshold increased, the size evolution assumed a ‘flat earth’ assumption for vegetation
of the runoff event required to remove vegetation became dynamics. In the flat earth, the land surface is assumed
higher. The larger size of the erosion event meant that the horizontal and no account is taken of the hillslope gradient
relative impact of the extreme rainfall was higher, so and its orientation to the sun, so that all parts of the
sediment output from the catchment, while it decreased landscape are subjected to the same amount of incoming
on average, became more variable. Finally, Collins and solar energy per unit area irrespective of the evolving
Bras (2008) examined the impact of varying climate and landform and its aspect and gradient. Yetemen extended
found an asymmetry in vegetation recovery from climate the Istanbulluoglu and Bras model to include the effect of
variations. When the climate dried, the vegetation cover hillslope aspect. Figure 16.9 shows how aspect changes
declines as expected, and vice versa for a wetter climate. incoming solar radiation with latitude, showing the sig-
But the recovery time from a drier period was slower than nificance of the differences in aspect for incoming solar
for a wetter period because the vegetation growth rate was radiation, and how for hillslope gradient in the N-S direc-
smaller because it was limited by water availability. tion greater than 5 (about 9%) the effect of aspect is
Moreover, the increase in vegetation density is driven significant for midlatitudes. Yetemen also incorporated a
primarily by the growth rate, while the decrease in vege- soil water mass balance in his model so that as plants grew
tation density is driven primarily by the vegetation death they consumed water at a rate determined by their growth
rate, and these two rates are different. rate. Thus faster growing plants consumed more water.
Istanbulluoglu and Bras (2005) presented a landform For their study site they found that as a result of the soil
evolution that was coupled with a vegetation biomass water storage capacity that for the equatorial facing slopes
model (Equations (14.4), (14.5) and (14.8)), where the (EFSs, in the northern hemisphere these are the south
vegetation was subjected to a simple fire model that burnt facing slopes), the soil water limited growth during the
(a) (b)
400
300
40 200
20
100
0
0 700 0
0
100 600
100
200 500 700
200
300 400 600
300 500
400 300 400 400
500 200 500 300
200
600 100 600 100
700 0 700 0
FIGURE 16.8: Two landform simulations showing the impact of (a) no vegetation and (b) fully vegetated on the valley density and hillslope
gradients of the landform (from Istanbulluoglu and Bras, 2005).
FIGURE 16.9: The change in the incoming solar energy at the top of the atmosphere (i.e. cloud free) versus latitude, normalised by the annual
average solar energy at that latitude (from Yetemen et al., 2015a).
summer, while for the polar facing slopes (PFSs), growth vegetation cover on the PFSs relative to EFSs meant that
was less limited by water availability. The consequence of erosion was least on the PFSs, leading to an asymmetry in
this was that on average over the year the PFSs had higher the hillslope gradients where the PFSs were steeper and
biomass than the EFSs. Yetemen assumed that soils were shorter than the EFSs (Figure 16.10). The hillslope gradi-
the same for all hillslopes irrespective of aspect (i.e. flat ent dependency of erosion is because of the erodibility
earth for soil properties); however, the hydrology dependency in the slope-area relationship (Section 3.3.3).
(the runoff mechanism used was saturation excess; see This was consistent with observations at their field site.
Chapter 3) varied with aspect as a result of the coupling Examining seasonal dynamics Yetemen et al. (2015b)
with vegetation density and soil water storage. The higher highlights further important subtleties. The hillslope
FIGURE 16.10: Slope-area plots

for north (polar facing, PFS) and
south (equatorial facing, EFS)
facing slopes showing the steeper
and shorter hillslopes for the PFSs:
(a) 45N, (b) equator. Note that
this aspect effect on hillslope
gradient disappears for a site at the
equator (from Yetemen et al.,
2015a).
water-limited case the hillslope gradients will be greatest

on the PFSs. Thus the balance between water-limitation
and non-water-limitation on the EFSs is key to the devel-
opment of the hillslope gradient asymmetry. Figures 16.11
and 16.12 show the daily dynamics of the vegetation of
the simulations, and highlight the seasonal dynamics of
this process. Figures 16.11 and 16.12a show the spatially
averaged soil moisture and biomass highlighting two
annual trajectories (the times A, B and C in the two figures
are for different days and were selected to highlight the
three corners of the triangular trajectory; Yetemen, per-
sonal communication). As the soil moisture is increasing
(i.e. non-water limited since water in the soil is increasing,
A–B), then the vegetation growth is high, and as the soil
moisture is decreasing (B–C), the vegetation is initially
FIGURE 16.11: Plot of the trajectories of the catchment average maintained but then starts to decrease as the water
soil moisture over time (horizontal axis) and the variability within the becomes more limiting. The spatial pattern illustrates the
catchment of soil moisture (vertical axis). The dots are for the entire
varying influence of hillslope aspect. For the non-water-
period simulated, while the lines are for two selected years. Points A,
B and C, and the two years are for the same as in Figure 16.12 (from
limited part of the annual cycle (time B) the highest
Yetemen et al., 2015b). vegetation density is on the EFSs, showing the influence
of the higher incoming solar radiation. As the EFSs become
water limited (time C), the highest vegetation is now on the
asymmetry they calculated was a function of the water PFSs. Time A shows the persistence of this water limitation
limitation of the slopes. Consider the alternative case from the previous year with much lower vegetation dens-
where nowhere was the vegetation water-limited (e.g. ities on both the EFSs and PFSs (note the change in the
the soils were such that there was sufficient water to colour bar units between times A and C), but the relatively
sustain vegetation through the summer on the EFSs). In higher density of vegetation on the PFSs remains.
this case the highest vegetation growth will be on the The effect of aspect is greater at higher latitudes, but
EFSs because of the higher incoming energy per unit area, the effect levels off beyond about 30. The effect is
so that erosion will be least on the EFSs and the slopes directly related to the relationship between solar radiation
steepest. Thus for the non-water-limited case the hillslope and latitude, and Figure 16.9 shows this increasing mark-
gradients will be greatest on the EFSs, while for the edly up to about 30 and then increasing more slowly
FIGURE 16.12: Vegetation cover for the catchment over time: (a) the catchment average vegetation cover (horizontal axis) and the
variability of the vegetation cover over the catchment (vertical axis), (b)–(d) maps of the vegetation cover at the times A, B, C in plot (a).
Points A, B and C and the two years are for the same as in Figure 16.11 (from Yetemen et al., 2015b).
thereafter. Yetemen et al. (2015a) confirmed that the the soil profile and its organic matter content (Chapters 7,
simulated asymmetry in hillslope gradient increases with 8 and 10). This interaction between water limitation and
latitude up to about 30, and then declines at higher the soils raises the question of the landform response if
latitudes because of a general reduction in water limita- the model of Yetemen et al. is coupled to a soil evolution
tion at higher latitudes (Gerten, 2013), probably as a model as discussed in this book. Some possible inter-
result of decreased evaporation rates at high latitudes actions include the following:
(Yetemen, personal communication). Their modelling
varied only the latitude and aspect, and so did not con- • SOM increases infiltration and water-holding capacity
sider any latitudinal dependencies in altitude, soils, mean (Sections 10.6.2 and 10.6.3). If vegetation density is
annual precipitation and potential evaporation. At any lower (due to water limitation), then SOM will also be
specific site an important variable is how quickly the lower, and infiltration and soil water-holding capacity
EFS become water-limited, and this will be a function will both be lower, reducing the stored soil water and
of the plant available water capacity, which will be a providing a positive feedback enhancing the water
function of the depth of the soil (Chapter 6), the depth limitation.
distribution of the water in the profile (relative to the • Some researchers (e.g. Gabet and Mudd, 2010; Section
depth distribution of the plant roots) and the ability of 6.2) assert that the rate of soil production increases with
the soil to hold water between rainfall events (soil water- increased tree throw so that a lower tree density (hence
holding capacity), which is a function of the grading of lower rate of tree throw) will result in a lower rate of
soil production, shallower soils and thus a lower soil initial plateau
a
water storage capacity. This will lead to an enhanced
lf = 1 km
water limitation. Yetemen modelled only grass (since 19.2 My
Te = 10 km
erosion is primarily a function of groundcover rather
b
than tree cover), so it is unclear whether this will feed
lf = 1 km
back into his asymmetry. 19.2 My
Te = 30 km
• Clay is a transformation product from rock weathering c
(Chapters 7, 8 and 10). Physical weathering breaks up lf = 100 km
the rock into fine particles with high specific surface 240 My Te = 10 km
area (with a low specific surface area, chemical reac-

d
tions are very slow) and is primarily a function of 1 km
lf = 100 km
cycles of wetting/drying. It is likely to be highest on Te = 30 km
0 240 My
those slopes with greatest soil moisture variability 0 50 100 150 km
during the year. This variability is probably highest e ve = 15
on the water-limited hillslope, but all of Yetemen’s

FIGURE 16.13: Profiles across an evolving passive margin
slopes have a high variability over the year
escarpment for a variety of erosion rates and thickness of the crust
(Figure 16.11), it is just that the EFSs dry out earlier T e . The rigidity of the crust is linearly proportional to the thickness,
in the summer. Chemical weathering requires sufficient so the stiffer crust has T e ¼ 30 km. The erosion rate is proportional
water to facilitate the chemical reactions and is typic- to the travel distance for the sediment lf so the highest erosion rate is
ally highest in wetter soils. Thus physical weathering is for lf ¼ 100 km (from Kooi and Beaumont, 1994).
probably highest and chemical weathering lowest on
the water-limited hillslopes. The net feedback effect is
unclear and may be a function of other processes (e.g.
microbiology, fungi).
16.1.3 Coupled Geodynamics and Landform

Evolution Models
Coupled models of erosion and tectonics were one of the

first uses of coupled landform evolution models. The first
application was to couple a model for tectonics that
included isostatic rebound and flexural rigidity to explore
how the crust responded to flexural unloading as a result
of erosion. The post-breakup dynamics of passive margins
was an application for these types of coupled models of
isostasy, crustal rigidity (the combination of isostasy and FIGURE 16.14: The evolution of a passive margin escarpment
crustal rigidity is referred to as flexural isostasy) and when flexural isostacy is coupled with erosion with the evolution
proceeding from the initial condition at the back to the front
surface erosion (Gilchrist et al., 1994; Kooi and Beau-
with time (from Gilchrist et al., 1994).
mont, 1994). The idea tested is illustrated in Figure 16.13.
Initially after the breakup there is unloading on the coastal
region as a result of the erosion on the steep coastal
escarpment, which leads to isostatic rebound in the region Figure 16.14 shows the evolution of an escarpment
surrounding the unloading. The area behind the escarp- using a model involving a transport-limited erosion, dif-
ment up-warps, resulting in drainage on the other side of fusion and flexural isostasy. Though initially all the flow
the coastal escarpment draining away from the escarp- is from right to left very quickly a catchment divide is
ment rather than toward the escarpment. The importance created at the escarpment with the landform to the right of
of the crustal rigidity in distributing the uplift in the region the escarpment flowing away from the escarpment.
around the unloading site is shown in Figures 12.6 and One of the important insights from the works of Kooi,
12.7, both of which show the crustal deflection as a result Gilchrist and colleagues was that a relative stable and
of an imposed load (unloading is just the negative of ‘permanent’ escarpment was created even though a
loading). transport-limited erosion equation was used. It is
commonly accepted that the detachment-limited transport were able to identify constraints on the geologic settings
equation based on stream power (i.e. Equation (4.29)) within which fission track dating could be used to distin-
creates stable knickpoints and/or escarpments that travel guish competing hypotheses.
upstream without dissipating, while the transport-limited The other major application area for coupled models is
equation (i.e. Equation (4.10)) tends to smooth out any at plate boundaries where mountain ranges are being built.
knickpoints/escarpments. The signature research sites for this are Taiwan and the
The key insight is that the flexural isostasy creates a South Island of New Zealand. The attraction of these two
drainage divide that tends to stabilise the escarpment sites is the ability to couple an uplift model as a result of
against dissipation even though a transport-limited trans- converging crustal plates, erosion model and a climate
port equation was used. One consequence of this up- model where there is a time-invariant windward and lee-
warping is that the escarpment is generally perceived to ward side to the mountain range so that rainfall (and thus
be stable because there is no process other than diffusion runoff and erosion) was greater on the windward (western)
and mass movement to cause escarpment retreat, since no side of the mountain range than the leeward side, creating
transport process is able to flow over the escarpment and asymmetry in the mountain range profile. Willet (1999)
erode it (e.g. van der Beek et al., 2002). modelled the evolution of a one-dimensional cross section
Subsequent work on passive margin evolution has across a convergent orogen with a stream power–based
been focussed on what additional or modified processes detachment-limited fluvial erosion (Equation (4.29)).
are required to control the rate of continued cliff retreat Using a viscous/plastic flow model, he modelled the con-
while maintaining the form of the escarpment (called verging plates (where one plate was subducting under
‘parallel retreat’), including the initial topography and another, and where the viscosity of the plate was depend-
drainage network prior to passive margin creation. In the ent on temperature) and coupled this with the erosion
Kooi and Gilchrist modelling the parallel retreat observed model, which was used to estimate the unloading resulting
in the model simulations (Figure 16.14) was a result of the from erosion of the topography. He then explored the
diffusion process in the model, which increased transport spatial pattern of uplifting, the flow paths for the subsur-
at the escarpment as the escarpment steepened. No phys- face crustal material as the two plates converged, and the
ical mechanism was postulated by the authors for what the effect of orographic rainfall coupled with the evolving
diffusive process attempted to simulate, though a slope- elevation (Section 3.1.1). Figure 16.15 shows the equilib-
dependent mass movement (e.g. landslide) would be rea- rium result of experiments where the wind comes from
sonable. The representation of drainage divides in LEMs different directions, so that the windward high rainfall
may play a part in the stabilisation of the drainage divide occurs on different sides of the range. The grid shows
(Section 4.7.3). the flow paths of the viscous/plastic flow of the crust,
Braun and van der Beek (2004) use GLUE (see while the topographic surface is the result of the erosion
Section 15.4) to explore two alternative hypotheses of of the uplifted crust. It is obvious from this figure that there
evolution for the passive margin of southeast Australia: has been considerable erosion of material because the
(1) slow parallel retreat of the escarpment or (2) sudden height of the topographic divide (the highest point on the
creation of the escarpment and subsequent erosion across mountain range) is significantly lower than the maximum
the entire region rather than just at the escarpment. They height of the exhumed crust. It is also clear that the loca-
did this by coupling a model of heat flow through the crust tion of the topographic divide is a function of the direction
with a simplified version of their CASCADE landform of the wind with the divide being skewed away from the
evolution model and compared their model predictions side of the highest rainfall and toward the low rainfall side.
with measurements of temperature across the escarpment Finally, the flow processes in the crust are different for the
from apatite fission track dating. They searched for the two cases. This shows that the crustal processes have
best fit to the data over the feasible range of parameters for responded to changes in the rainfall and the spatial pattern
the two different competing hypotheses, with the two of rainfall, showing that a link between surface processes
parameters being the thickness of the crust and the heat and plate dynamics at convergent orogens exists. Subse-
flow through the crust. Unfortunately both models fit the quent work by Willett and others (Goren et al., 2014) has
data equally well, and the best fit parameters for both improved the physical representation, and compared
models gave thickness of the crust (~7 km) that were model predictions with a number of field sites, but the
inconsistent with estimates derived from other sources original conclusions remain broadly true.
(~17 km). While the GLUE analysis was unable to iden- The asymmetry resulting from the different rainfall on
tify the best explanation for escarpment evolution, they each side is consistent with the slope-area relationship in
(a) FIGURE 16.15: A coupled

crustal convergence-erosion
Ne=10 model with rainfall coming from
(a) the left so rainfall is highest on
WIND the left of the range, (b) the right
Topographic
Divide so rainfall is highest on the right of
t* = 8 the range. The grid shows the path
of the crustal material showing the
pattern of exhumation of crustal
material before erosion. The
Subducting Plate crustal plate from the left is
subducting under the plate from
(b) the right. The topographic surface
is the nearly flat line with a bump
at the ‘topographic divide’ (from
Willett, 1999).
Ne=10
WIND Topographic
Divide t* = 8
Subducting Plate
0.001 0.01 0.1 1

Strain Rate εh
vp
Equation (3.21) (Figure 16.16). If everything except rain- Table 14.1 shows that the erosion response to small
fall is the same on both sides of the divide, then on the changes in vegetation cover can be quite strong, so a
windward, high rainfall, side the runoff is also higher reverse in the asymmetry is possible. Equations (16.1)
(i.e. β3 ) so that the slope is relatively lower on the wind- and (16.2) are written using the detachment-limited trans-
ward side, and using Equation (3.21) port model, which will be dominant only in the river
m2nm3 network, but the vegetation will primarily determine the
Sw β3l 2
erosion from the hillslopes, which will be transport-
¼ (16.1)
Sl β3w limited (i.e. Equation (3.19)). The only difference is in
where the subscript w and l indicate the windward (high the parameters but there are likely to be interactions
rainfall) and leeward side, respectively. Figure 16.16a between the hillslopes and channels not accounted for in
shows Equation (16.1) where the erodibility is the same Equations (16.1) and (16.2).
on both sides of the catchment divide. However, if vege-
tation responds to the rainfall, then the vegetation on the
windward side will be denser than on the leeward side and 16.1.4 Coupled Soil Organic Matter, and Erosion
provide greater erosion protection so that Equation (16.1) and Deposition
is then
At the time scales that organic matter changes occur
n1 m2nm3 (years to decades), the landform doesn’t change greatly
Sw K 2l 2 β3l 2
¼ (16.2) unless it is degrading quite badly. Thus to couple ‘land-
Sl K 2w β3w
form evolution’ with organic matter, it is good enough to
Figure 16.16b shows that as the effect of vegetation examine erosion and deposition on a fixed landform, and
becomes increasingly dominant, the asymmetry due to treat landform evolution as a second-order issue.
rainfall disappears, and it is possible that the mountain This was the approach adopted by Lacoste et al.
range can become asymmetric in the opposite direction. (2015), who modelled soil organic carbon on a fixed
The extent of this reversal in symmetry will depend on the landform (with a 2 m resolution DEM) at monthly reso-
strength of the vegetation response to rainfall changes. lution (which means that the surface drainage pattern did
not change) but allowed the soil depth and elevation to looked at a number of different agricultural sites with
change in response to the erosion and deposition at different cropping and tillage patterns to examine the
decadal resolution. They coupled the RothC soil organic implications of land management strategies and climate
carbon (SOC) model (see Section 10.3) with the LandSoil change on SOC storage and loss. The soil was modelled
soil redistribution model (a simple transport-limited soil as eight layers down to 105 cm depth. Over their simula-
erosion and deposition as discussed in Section 4.2). tion period from 2010 to 2100, 22% of their 86 ha site had
Figure 16.17 shows a schematic of their model. They an elevation change due to erosion of more than 40 mm,
and 19% had an elevation change due to deposition of
more than 50 mm, while the net loss of soil (i.e. erosion
deposition) for the site was about 0.5 mm. This is consist-
ent with very slow landform evolution occurring during
the study period.
Roth-C is a single layer compartment model for the top
30 cm of the soil. Lacoste extended Roth-C to have five
SOC compartments (decomposable plant material, resist-
ant plant material, microbial biomass, humified organic
matter and inert organic matter), and the SOC dynamics of
each of the eight layers was modelled independently.
Organic matter was sourced from leaf litter at the surface
and from decomposing roots within the profile, with the
root depth distribution being specified for the agricultural
crops being grown. For want of data, decomposition rates
were assumed constant through the depth of the profile.
Only two methods of movement of SOC were considered:
(1) erosion and deposition that transported material
spatially and (2) tillage that mixed the soil down to the
depth of tillage (assumed to be 30 cm) but did not move
soil spatially. No coupling between SOC and soil
FIGURE 16.16: Schematic showing how asymmetry develops structure, and infiltration/runoff and SOC enrichment
across a mountain range due to orographic rainfall (a) where the high was considered.
rainfall has no impact on vegetation so erosion is higher on the left- Thus they ran two models in parallel; they described
hand side (i.e. K2 in Equation (3.21) is the same on both sides),
their models as ‘lightly coupled’. LandSoil calculated the
(b) where vegetation is stronger on the wetter side and thus reduces
the erosion on the wet side, with the arrows indicating how the
erosion and deposition on the DEM for every rainfall/
slopes on both sides would change for two cases with increasing erosion event. They summed up the erosion and depos-
strength of vegetation response to increased rainfall increases. ition from every event at every node, and evolved the
FIGURE 16.17: The model

coupling approach for soil organic
matter and erosion and deposition
(from Lacoste et al., 2015).
landform and the soil layer boundaries every decade. That said, there are a range of exciting new applica-
Between decadal updates of the landform the elevations tions for landscape evolution models, which are beginning
and the soil layer boundaries were considered unchanged. to emerge out of practical needs and new science, and
The erosion and deposition calculated for every event which I have made only passing reference to in this book.
were also summed up monthly, and the SOM movement No doubt these new applications will lead to new physical
and decomposition calculated for every node monthly. approximations and insights.
They used one notable simplification to split the erosion
from the SOC transport. They assumed that all soil that 16.2.1 Landscape Self-Organisation and
was deposited in each monthly timestep had the SOC Connectivity
content of the average of all the eroded areas in that same
timestep. Thus they did not track soil and SOC from its One of the challenges of understanding the landscape as a
erosion source to its deposition site. Rather, they simply functional system evolving in both space and time is the
assumed that all the eroded material had the same (on sheer range of active processes and the range of their
average) SOC content, irrespective of where it was spatial and temporal time scales and response times. This
sourced. It is unclear how important that assumption means to have a ‘complete’ model of the system, we
was because they did not plot any maps of the SOM over might need to model the system at very fine time and
the catchment, only presenting catchment wide averages space resolution, but then run these models over very long
in their analysis. They made this simplification because times and large areas to be able to capture the responses of
LandSoil was unable to track sediment and SOC the system. The models that we have discussed in the
trajectories. previous chapters all use considerable computer resources
In their analysis they then compared SOC evolution to simulate system response, so there is a compelling
for a number of climate change scenarios from IPCC desire to simplify these models if underlying organising
projections and analysed the implications of the different principles can be found.
climate projections for the different management practices At a personal level it was to understand the underlying
across their field site. They considered their study a pre- organising principles of channel networks, based on the
liminary first of its kind assessment, and so they didn’t physics of channel network formation, that we developed
provide definitive conclusions. However, they observed the SIBERIA landform evolution model. This insight has
that management practices and changes in them were been used to develop a deeper insight into why channel
more significant than projected changes in climate to networks have the mathematical characteristics they do
2100. The different management practices also had differ- (e.g. the slope-area diagram, the cumulative area diagram;
ent sensitivities to changes in climate, with the most see Chapter 3) and has allowed us to make observations
intensively farmed being the most sensitive. about channel response without needing to measure all the
Given the number of assumptions and simplifications details of every individual channel.
in the physics, it is unclear whether their conclusions are In the same way, we can measure in fine detail, from
generally applicable. Rather, the reason for presenting this remote sensing, the distribution of vegetation to include
study here is as a proof-of-concept approach, with all its this into hydrology models. However, if we can under-
simplifications, for coupling SOC and landform evolution. stand the organising principles of vegetation distribution
and function (and how they may coevolve with the land-
form and soils), we may be able to develop models of the
16.2 The ‘Future’ response of a catchment without having to measure all the
details of the vegetation. We have discussed several times
The scope of this book has been to present and explain the the objective of developing subgrid-scale parameterisa-
mathematical principles used in the modelling of soils- tions of processes, so as to allow us to model at coarser
cape, landform and landscape evolution. While a wider resolution without having to resolve all the small-scale
and more comprehensive discussion of applications in this detail. Likewise we would like to be able to develop
area was not possible, some discussion has been provided, models of the organisation of soils in space and time so
as much as anything, to motivate why certain processes that we can then use a subgrid parameterisation of the soil,
are modelled in the way they are, why certain approxima- to use this to develop models of the catchment-scale
tions are both made and are justifiable, and to provide, integrated response of the hydrology to the soils.
using examples, some concrete applications of what can One of the ways that these self-organisation principles
sometimes be rather esoteric abstractions. can be studied is by developing models of the spatial and
16.2 The ‘Future’ 287
temporal coupling, and examining the equilibrium But connectivity is about more than just water flow.
response and the evolutionary approach to that equilib- When discussing the impacts of fire, we discussed the idea
rium, based on the principles in this book. On the basis of that some of the sediment postfire flows directly off the
these simulations, we can then use these models to search hillslope into the river, while some of it moves down the
for simple models that capture the behaviour shown by the catchment only a short distance, to be moved at a later
complex models and in the field. This may seem to be an time by different postfire processes. Thus the fate of the
impossible task, but there is good reason to believe that latter sediment is a function of how well connected the
this is an achievable objective. Field workers have found transport trajectories are to the outlet of the catchment.
many empirical relationships in the field that seem to be A low connectivity to the catchment outlet will mean that
true across many landscapes and catchments, across cli- the sediment will be more strongly retained within the
mates and across ecosystems. It is highly likely that these catchment. An important question that can be asked is
relationships result from the coevolution and approach to how the coevolution of the processes and the resulting
equilibrium of the processes and fluxes that have been self-organisation of the landscape impacts on the connect-
discussed in this book. It is important to understand that ivity of the processes to the outlet of the catchment.
when we discuss equilibrium, we don’t mean a state This question of connectivity and the principles that
where the system no longer changes with time, because drive changes in it is a question that we have only the
there are many systems that oscillate in time, and this barest understanding of, and is a fertile area for future
oscillation in time is in fact the equilibrium behaviour research.
(this oscillation is the basis of the literature in the field
of nonlinear systems and chaos). 16.2.2 Sustainable Landscapes
One important aspect of self-organisation of these
systems in space is the question of how the processes Sustainable landscapes is an application area that has been
connect spatially. If we think of the patchy vegetation close to the author’s heart and has motivated much of his
discussed in Chapter 14, then not only are we interested work. It might be broadly described as ‘quantitative sus-
in the percentage cover of the hillslope by vegetation, but tainability assessment of landscapes’. If we wish to pre-
we are also interested in how those patches are arranged. dict the trajectory of an environmental system into the
Some arrangements of the patches are very effective at future, then we need to be able to model the (typically
trapping runoff on the hillslope, and this better retention nonlinear) transients, equilibria and feedbacks between
of water on the hillslope means that there is more water the different parts of the landscape. This may involve
available for vegetation growth, which in an arid climate complex coevolution problems that are difficult to untan-
is a competitive advantage. This behaviour can be pre- gle into their components, and where as a result of the
sented in terms of how well the flow paths of the hillslope feedbacks the trajectory may vary in nontrivial ways for
are connected (a patch of vegetation interrupts the flow even small changes in initial conditions and/or climate
path because the water in that flow path infiltrates in that inputs. Many environmental feedbacks also take many
vegetation patch). If flow paths are well connected, then years to exhibit themselves so that for a few years there
water is better able to flow off the hillslope and the may be little evidence of coevolution, but the final fate of
vegetation function (which is limited by water availability) the system may rely heavily upon them. One excellent
will be degraded. Thus flow connectivity is a measure of example of this is erosion. Over the short term, agricul-
the ecological health of the hillslope. In addition, the tural erosion models applied on a fixed landform provide
connectivity of the flow paths drives the hydrologic good estimates of the spatial distribution of soil erosion.
response of the hillslope, and if we model large catch- However, over the long term as the landform/paddock/
ments, we would prefer not to be required to simulate hillslope changes in response to erosion, these models
every patch and the hillslopes between these patches. An begin to break down; gullies and rills form, they concen-
‘effective’ representation of the hydrology of the hillslope trate flow and locally increase erosion, while at the same
and the patches on it (and their dynamics to changing time erosion is decreased outside the rills and gullies since
climate) that provides the integrated hydrology response the overland flow has concentrated into the gullies. The
of the hillslope would be a useful outcome. This effective discussions of the previous chapter are replete with pro-
model could be incorporated in the land surface schemes cesses that when coupled have the potential to produce
that are components of global climate models and that are this long-term nonlinear behaviour.
now being used to predict the impact of climate change on We could rely upon natural analogues to estimate
hydrology and ecology. behaviour, but it is better if we can model this, because
in many applications (1) we don’t have natural analogues the principle behind the QUEL model; Willgoose, 2001).
to these systems, (2) we wish to design an initial condition However, we know that when we do this that the eleva-
that provides the best long-term performance, and it is tions, and the drainage paths derived from them, are
only with models that we can compare design alternatives commonly inconsistent with the original drainage pattern
or (3) the field observations required operate at time scales used to construct the elevations (Ibbitt et al., 1999), and
that are too long, or the impacts are too subtle, to be an iteration is required to make the drainage lines, slope-
observable. area relationship and elevations consistent. The only
The author has pioneered applications of the LEMs in thing that has been approximated has been the scatter
the area of constructed landforms where landforms are around the mean slope-area relationship, indicating that
built to rehabilitate old mining sites, or as containment there is something in this scatter that is an essential
vessels for potentially hazardous wastes (mine tailings, characteristic of the landform, and that it is not simply
chemical wastes and low-level nuclear waste repositor- random scatter. It is likely that issues like this will arise
ies). In these cases a landform needs to be designed that in the use of the area-slope-d 50 relationship (discussed in
will be stable over some design lifetime so as to contain Chapter 11) for soil construction, and this is an area for
the waste (typically 100 to 10,000 years, and over this future investigation.
time significant landform evolution typically occurs), to
have minimal incremental impact on the surrounding
landscape, be consistent with social and mine operational 16.2.3 Extraterrestrial Geomorphology
constraints and be cost-effective. It is rare that all these
In the last two decades we have seen an explosion of
objectives can be met simultaneously, so a range of design
mapping and high-resolution photography missions to
alternatives need to be developed so that the positives and
other parts of the solar system. Baker (2008) makes the
negatives can be explicitly subject to trade-off analysis.
case that geomorphologists should be working on these
The Ranger Mine example discussed in Chapter 15 is just
applications, in addition to applications on Earth.
one published case where this tradeoff analysis has been
Other than sheer curiosity value there are several
ongoing for about 25 years. One final challenge is that the
reasons why extraterrestrial applications might be
owner of the rehabilitated mine site typically cannot be
important:
held responsible for the site forever (what company will
be around for the design lifetime of 10,000 years?), so at 1. The parameterisations of many of the processes in this
some stage government regulators must sign off on satis- book are empirical in nature. They are thus dependent
factory performance and assume long-term responsibility. on the conditions in which they have been derived:
This requires a suite of defensible tests against which the earth’s atmosphere, temperature and gravity as the
field site can be assessed, and these tests should, ideally, main factors. We can reasonably ask whether the dif-
be early indicators of longer-term performance, including ference in gravity results in fundamental changes in the
all the effects of coevolution over the design lifetime. We landform coevolution of the processes. No doubt many
would like these tests to be defensible in court because it unknowns exist with respect to geology on the ground,
is common for contentious designs to end up there under but can we infer these other dependencies? There are
challenge by opponents of the particular development. In the meandering rivers of liquid methane (i.e. different
many western countries unrefereed computer model viscosity to water) on Titan (Burr et al., 2013), paleo-
results are not admissible evidence in court proceedings. channels from meandering rivers of water on Mars
This remains a significant, and only partially resolved, (Howard, 2009) and glaciers of solid nitrogen (i.e.
challenge. different rheology) on Pluto (McKinnon et al., 2016).
Finally, there is an intriguing science question that is at The question is whether the landforms that they gener-
the heart of sustainable landscape design. What are the ate are fundamentally different in some measurable
essential statistics to describe a landscape? If we know way from those of Earth because of those differences.
these, then we know what features to design into our man- 2. Does the absence of biological activity on these planets
made landscape. If we confine our thinking to just the result in fundamentally different processes? For
landform for the moment, then if we know all the drainage instance, we know that biological activity is central
lines (and thus the area draining through every point in the to soil functioning, so the question is how does soil
landform), we should be able to use the average slope-area function in the absence of biological activity. We
relationship to construct the elevations of the landscape already have some information about this from colon-
deterministically starting at the catchment outlet (this is isation studies of barren sites (volcanic eruptions) and
rehabilitated mine sites, but what are the long-term tunnels between islands and oil/gas production platforms)
equilibria in the absence of biological activity (typic- that are regularly damaged by undersea transport events.
ally our barren sites recolonise long before they are We have not discussed undersea and estuarine processes
approaching geomorphic equilibria, so we cannot be in this book, but terrestrial fluvial erosion has its parallel
sure of what a nonbiological equilibrium looks like). If in the submarine environment: turbidity currents, flows
we can understand this question, we can then ask the downslope containing denser water (than the surrounding
question of whether we can remotely sense (i.e. either sea water) because of the entrained sediments. Gravity-
from satellite or from robots on the surface) the exist- driven mass movements driven by earthquakes also occur
ence of life (Dietrich and Perron, 2006). For instance, in the submarine environment, with the main difference
given the recent recognition of the role of vegetation in being that air is no longer a factor, and the sediment is
governing channel geometry and river meandering, always saturated so that liquefaction is almost always a
how did the meander scrolls on Mars and Pluto factor in both initiation and energy losses in the
develop in the absence of vegetation (Howard, 2009)? movements.
3. Can canyon formation on Earth (e.g. mega-floods at
the end of the last ice age; Baker, 2009) be compared 16.3 Conclusions
with canyon formation resulting from outburst floods
from craters on Mars (e.g. Lapotre et al., 2016)? Can This book started in Chapter 1 with a rather optimistic
canyon geometry relationships change as a result of the presentation of a schematic for landscape evolution: ‘The
difference in gravity, and can we infer Mars hydrology Model of Everything’ (Figures 1.1 to 1.3). This schematic
from the hydraulics of these canyons? of The Model provided a holistic overview of the agenda
4. Can we infer past climate on the planets from the of this book. The Model stressed the interconnected
topography (Moore et al., 2014)? nature of all the processes. I am reminded of the quote
from John Muir I put in the front of my PhD thesis in
16.2.4 Submarine Geomorphology 1989: ‘When we try and pick out anything by itself, we
find it hitched to everything else in the Universe.’ If
We conclude this discussion of new application areas a anything, I feel it to be even more true today than I did
little closer to home, but to a large extent almost as back then. One could be demoralised by this observation
unknown as other planets in the solar system: the submar- and feel that the problem is so vast, interconnected and
ine environment. Recent advances in multibeam sonar and complicated that there is no hope of achieving anything.
gravity satellites mean that we can collect data about the However, if the last 30 years of landform evolution mod-
sea floor and its changes, particularly before and after elling has proven anything, it is that by separating prob-
earthquakes, for example, the 2004 Sumatran earthquake lems with different time and space scales, by using
(Vince, 2005) and the 2011 Tōhoku/Fukashima earth- effective parameterisations for subgrid processes (so that
quake (Fujiwara et al., 2011). This also means that we the computer model runs in a feasible time) and, where
can now observe the undersea topography much better necessary, making pragmatic simplifications based on
than before. For example, the deep-sea search area in the field observation, we can chip away at the unknowns in
Indian Ocean for Malaysian Airlines flight MH370 has The Model and develop an understanding of the under-
now been imaged at 100 m resolution (Picard et al., 2017). lying self-organisations in the system. The emergence of
From these new sensors we can see channels from the the Critical Zone research agenda (NAS, 2010) means that
terrestrial rivers that travel considerable distances to the it is time to expand our focus away from solely landforms,
edge of the continental shelf (Gupta et al., 2007) and and extend the approaches we’ve successfully used for
sometimes beyond. While some of this geomorphology landform evolution research to understand soilscape and
no doubt results from terrestrial processes during low ecosystem evolution, i.e. landscape evolution. That seems
stands during ice ages when sea levels dropped (Section to me to be a worthwhile and achievable objective.
1.3.1), there are visual differences between terrestrial Having presented The Model, most of the book has
canyons and these submarine canyons that circumstanti- subsequently been about filling in the (sometimes
ally implicate other processes, for example, turbidity cur- daunting) details in the schematic of The Model. I have
rents. We can also see landslide scars at the edge of the tried not to avoid highlighting areas where our science or
continental shelf. The practical importance of these areas computational understanding is weak. Maybe a future
follows from the considerable essential infrastructure PhD student or two will take the hint and discover that
(mostly fibre optic cables, but in some cases power cables, therein lies an interesting PhD topic. The last chapters
have been more focussed on how to use a Landscape unwelcome. And don’t get me started on editing large
Evolution Model and presented some examples of recent text files that are too big or difficult to open with WordPad
uses of these models. and the like. Finally, there is the plethora of different
Before I finish, there are a number of topics that I think binary file formats (e.g. netCDF, HDF, geoTIFF, shape
are worth reiterating. files), each of which require a different set of tools to
Our current generation of models are computationally manipulate unless you are proficient in coding with
slow (a colleague calls them ‘slowscape models’ not scripting languages like MATLAB or Python, or
‘landscape models’). Sometimes this slowness means that advanced GIS packages. Short of actually running a land-
solution of problems at the scale that will provide science scape evolution model, many colleagues have found the
insight is impossible. For instance, for soilscape model- computer animations we regularly generate for research
ling our strictly physically based ARMOUR model and commercial clients useful for teaching and seminars.
(Willgoose and Sharmeen, 2006) was so slow that, while Unfortunately a book is not the place to include anima-
we could calibrate the model to field plot studies, we tions, but I cannot overemphasise the amount of positive
could barely simulate a hillslope over a few hundred feedback that we have received from educators and clients
years. However, once we had the insight that a state-state as a result of freely distributing our animations.
approximation was able to mimic the ARMOUR simula- Finally, I would like to return to a topic that has
tion results, the approach that eventually led to the motivated much of our research at Newcastle for the last
mARM (Cohen et al., 2009, 2010) and SSSPAM 30 years or so: the engineering application of landform
(Welivitiya et al., 2016; Welivitiya, 2017) suite of models, evolution modelling to the clean-up and sustainability
and the theory further extending the state-state approach assessment of rehabilitated mines and low-level nuclear
in Chapter 7, we were able to simulate (without a super- waste repositories. The first question we are always asked
computer) both spatially distributed soil profiles at the by new research partners and commercial clients is ‘How
catchment scale, soil spatial organisation and coupled do we know that the models are correct?’ Rarely do they
soilscape-landform evolution over periods up to a ask deep questions about the subtleties of the mathematics
million years. or the underlying science. Given that they are being asked
But there are other practical reasons that slow models or required to spend millions of dollars to (re)design their
are a problem. If we wish to quantitatively test our model rehabilitation approach on the basis of the predictions
predictions, then we need to become better at putting error from landscape evolution modelling studies, this is a
bands on both our model predictions and field measure- perfectly reasonable question. Yet it is a devilishly diffi-
ments. I see no other practical approach than Monte Carlo cult question to answer, and in ways that are not initially
simulation (see Chapter 15) to do that. This will either obvious. It is not simply that predictions are being made
require parallel computers that, at the moment, are out of hundreds to thousands of years into the future (the
the reach of all but the best-resourced research groups in facetious, but indefensible, answer is that we’ll all be dead
the world, or we need faster models. These faster models by then so who cares). It is, can we predict localised areas
will come from better solvers (e.g. can somebody please where enhanced protection might be required, against, for
find a way to remove drainage direction algorithms from instance, gully erosion that might penetrate through a
erosion solvers so we can take large timesteps), or protective cap. This requires predictions where not only
approximations to more sophisticated physics that are can we be sure that the gully/valley will occur in a specific
easier and/or faster to solve (e.g. like our mARM and location (see Chapter 15 for a deeper discussion of this),
SSSPAM approximations to the ARMOUR model). but also how deep the gully will be and how fast it will
One emerging area is the use of landscape evolution occur (e.g. some wastes become benign after a time). If
models for teaching and professional practice. My own we can do that (I’m not sure we can, as a general rule, do
experience and experience of colleagues is that our this everywhere), then can we give them some assessment
existing models are probably too complex and difficult (e.g. a probability) that the predictions are correct (e.g.
to use to be useful as routine tools for teaching geomorph- with 50% confidence that these protection measures will
ology and mine rehabilitation principles. Simpler and work for 1,000 years).
better graphical interfaces are required. The command line At a recent meeting about a new project we had a
is foreign to most undergraduates who have grown up discussion about how we believe that in general (due to
with GUI-based software like Windows, Word and Excel. all the randomness in the field) we cannot predict the
My experience is that even in the professional mine exact location of gully occurrence unless very specific
rehabilitation community a command line interface is (and quite restrictive) landscape design principles are
followed. One environmental manager asked the question the cultural differences between these two groups. Many
‘How about we let the constructed landform evolve and field workers come from a purer science background
see where gullies occur and then put protective measures whether that be geology, geography, agricultural and soil
(e.g. armouring) in that location. Will that work long term, science or ecology, while the modellers typically come
or will we push the gully somewhere else over the long from an engineering, geophysics or mathematics back-
term?’ This proposed management option opens a Pan- ground. Landscape evolution advances will come from
dora’s box of issues, including (1) is short-term behaviour being a broad church for all these workers, and for those
predictive of long-term behaviour, (2) the impact of local- workers of different backgrounds to be able to be deeply
ised armouring on the spatial distribution of long-term embedded together and understand each other’s view-
erosion and (3) do we have sufficient validation of our points, because it is unlikely in my view that any of us
landform evolution tools to be able to defend predictions can do it on our own.
of unusual and untested engineering solutions like this. The development of the Critical Zone Observatories
The answer we gave was that we don’t know. (CZOs) in the United States, is in my view, a step forward
We now conclude with a topic near to my heart, the in that direction, even though the current ones do not, yet,
collection of field data. I’m sure, coming from me, that in my view, fully encapsulate the broader range of topics
sentence surprised some of you. However, if we are to that have been discussed in this book. Likewise there are a
advance the science and practical applications of land- range of ongoing mini-CZOs around the world (but sadly
scape evolution models we need stronger, more robust, without dedicated CZO-like funding) that are studying
testing of the models. The models discussed in this book parts of the coupled landform-soilscape-ecosystem evolu-
(not just ours, but the models developed in other groups as tion system but not the whole (e.g. our SASMAS site,
well) make novel predictions, otherwise we would not Rüdiger et al., 2007; eriss’s Tin Camp Creek site,
develop them. Because these predictions are, well, novel, Hancock et al., 2002; U. Cordoba’s Santa Clotilde CZO,
unless we are extremely lucky it is unlikely that field Román-Sánchez et al., 2017). No group is holistically
workers would have been motivated/funded to collect addressing the whole. And the same is true of constructed
the data required to test those novel predictions. This landscapes, such as rehabilitated mines; there are no
new data collection task needs to be a collaboration CZOs for mined land. But, as they say, baby steps.
between the model builders/users and the field research- Hopefully this discussion, and the book in general,
ers. The model builders need to find predictions that are will stimulate some creative juices (and hopefully trigger
testable in the field or the laboratory, while the field robust debate) among the book’s readers, and generate
workers/experimentalists need to develop ways of collect- ideas for topics worthy of intellectual attention. If I have
ing the new data in a way that can be used to perform achieved that, then writing this book has been
these tests. This is a nontrivial task, not least because of worthwhile.
References
Abrahams, A. D. (1984), Channel networks: A geomorphological perspec- Amundson, R. (1994), Towards the quantitative modeling of pedogenesis –
tive, Water Resources Research, 20(2), 161–168. A review – Comment – Functional vs mechanistic theories: The paradox
Abrahams, A. D., and A. J. Parsons (1990), Determining the mean depth of paradigms, Geoderma, 63(3–4), 299–302.
of overland flow in field studies of flow hydraulics, Water Resources Amundson, R. (2001), The carbon budget in soils, Annual Review of Earth
Research, 26(3), 501–503. and Planetary Sciences, 29, 535–562, doi:10.1146/annurev.earth.29.1.535.
Abrahams, A. D., and A. J. Parsons (1991a), Relation between sediment Ancey, C. (2007), Plasticity and geophysical flows: A review,
yield and gradient on debris covered hillslopes, Walnut Gulch, Arizona, Journal of Non-Newtonian Fluid Mechanics, 142, 4–35, doi:10.1016/j.
Bulletin of the Geological Society of America, 103, 1109–1113. jnnfm.2006.05.005.
Abrahams, A. D., and A. J. Parsons (1991b), Resistance to overland flow Anders, A. M., and S. W. Nesbitt (2015), Altitudinal precipitation gradi-
on desert pavements and its implications for sediment transport modeling, ents in the tropics from Tropical Rainfall Measuring Mission (TRMM)
Water Resources Research, 27(8), 1827–1836. precipitation radar, Journal of Hydrometeorology, 16(1), 441–448,
Abrahams, A. D., and A. J. Parsons (1994), Hydraulics of interrill doi:10.1175/JHM-D-14-0178.1.
overland flow on stone-covered desert surfaces, Catena, 23(1–2), Anderson, R. S. (2002), Modeling the tor-dotted crests, bedrock edges,
111–140. and parabolic profiles of high alpine surfaces of the Wind River Range,
Adams, W. A. (1973), The effect of organic matter on the bulk and Wyoming, Geomorphology, 46, 35–58.
true densities of some uncultivated Podzolic soils, European Journal Anderson, R. S. (2015), Particle trajectories on hillslopes: Implications for
of Soil Science, 24(1), 10–17, doi:10.1111/j.1365-2389.1973.tb00737.x. particle age and 10Be structure, Journal of Geophysical Research (Earth
Ahad, T., T. A. Kanth, and S. Nabi (2015), Soil bulk density as related to Surface), 120, 1626–1644, doi:10.1002/2015JF003479.
texture, organic matter content and porosity in Kandi soils of District Anderson, R. S., and S. P. Anderson (2010), Geomorphology: The mech-
Kupwara (Kashmir Valley), India, International Journal of Scientific anics and chemistry of landscapes, Cambridge University Press,
Research, 4(1), 198–200. Cambridge.
Ahnert, F. (1976), Brief description of a comprehensive three-dimensional Anderson, S. P., W. E. Dietrich, and G. H. Brimhall (2002), Weathering
process-response model for landform development, Zeitschrift für Geo- profiles, mass-balance analysis, and rates of solute loss: Linkages between
morphologie N.F. Supplement, 25, 29–49. weathering and erosion in a small, steep catchment, Geological Society of
Ahnert, F. (1977), Some comments on the quantitative formulation of America Bulletin, 114(9), 1143–1158, doi:10.1130/0016-7606(2002)
geomorphological process in a theoretical model, Earth Surface Pro- 114<1143:WPMBAA>2.0.CO;2.
cesses, 2, 191–201. Andrews, D. J., and R. C. Bucknam (1987), Fitting degradation of shore-
Ahr, S. W., L. C. Nordt, and S. L. Forman (2013), Soil genesis, optical line scarps by a nonlinear diffusion model, Journal of Geophysical
dating, and geoarchaeological evaluation of two upland Alfisol pedons Research (Solid Earth), 92(B12), 12857–12867, doi:10.1029/
within the Tertiary Gulf Coastal Plain, Geoderma, 192, 211–226, JB092iB12p12857.
doi:10.1016/j.geoderma.2012.08.016. Angers, D. A., et al. (1997), Impact of tillage practices on organic
Albrecht, R. I., S. J. Goodman, D. E. Buechler, R. J. Blakeslee, and H. J. carbon and nitrogen storage in cool, humid soils of eastern Canada, Soil
Christian (2016), Where are the lightning hotspots on Earth?, Bulletin of and Tillage Research, 41(3–4), 191–201, doi:10.1016/S0167-1987(96)
the American Meteorological Society, 97(11), 2015–2068, doi:10.1175/ 01100-2.
BAMS-D-14-00193.1. Argus, D. F., and W. R. Peltier (2010), Constraining models of postglacial
Alley, R. B. (2014), The two-mile time machine: Ice cores, abrupt rebound using space geodesy: A detailed assessment of model ICE-5G
climate change, and our future, Princeton University Press, Princeton, (VM2) and its relatives, Geophysical Journal International, 181(2),
NJ. 697–723, doi:10.1111/j.1365-246X.2010.04562.x.
Alpert, P. (1986), Mesoscale indexing of the distribution of orographic Arshad, M. A., A. J. Franzluebbers, and R. H. Azooz (1999), Components
precipitation over high mountains, Journal of Climate and Applied of surface soil structure under conventional and no-tillage in northwestern
Meteorology, 25(4), 532–545, doi:10.1175/1520-0450(1986)025<0532: Canada, Soil Tillage Research, 53(1), 41–47, doi:10.1016/S0167-1987
MIOTDO>2.0.CO;2. (99)00075-6.
292
References 293
Asteriou, P., H. Saraglou, and G. Tsimbaos (2012), Geotechnical and Bazin, L., et al. (2013), An optimized multi-proxy, multi-site Antarctic ice
kinematic parameters affecting the coefficients of restitution for rock fall and gas orbital chronology (AICC2012): 120–800 ka, Climate of the Past,
analysis, International Journal of Rock Mechanics & Mining Sciences, 54, 9, 1715–1731, doi:10.5194/cp-9-1715-2013.
103–113, doi:10.1016/j.ijrmms.2012.05.029. Beal, L. K., D. P. Huber, S. E. Godsey, S. K. Nawotniak, and K. A. Lohse
Astete, C. E., W. D. Constant, L. J. Thibodeaux, R. K. Seals, and H. M. (2016), Controls on ecohydrologic properties in desert ecosystems: Differ-
Delim (2015), Bioturbation-driven particle transport in surface soil: The ences in soil age and volcanic morphology, Geoderma, 271, 32–41,
biodiffusion coefficient mobility parameter, Soil Science, 180(1), 2–9, doi:10.1016/j.geoderma.2016.01.030.
doi:10.1097/SS.0000000000000109. Bell, J. R. W., and G. R. Willgoose (1998), Monitoring of gully erosion at
Avirmed, O., I. C. Burke, M. L. Mobley, W. K. Lauenroth, and D. R. ERA Ranger Uranium Mine, Northern Territory, Australia, Internal Report
Schlaepfer (2014), Natural recovery of soil organic matter in 30–90-year- 274, Environmental Research Institute of the Supervising Scientist, Jabiru,
old abandoned oil and gas wells in sagebrush steppe, Ecosphere, 5(3), NT.
1–13, doi:10.1890/ES13-00272.1. Bellugi, D., D. G. Milledge, W. E. Dietrich, J. McKean, J. T. Perron,
Azooz, R. H., and M. A. Arshad (1996), Soil infiltration and hydraulic E. B. Sudderth, and B. Kazian (2015a), A spectral clustering search
conductivity under long-term no-tillage and conventional tillage systems, algorithm for predicting shallow landslide size and location, Journal of
Canadian Journal of Soil Science, 76(2), 143–152. Geophysical Research (Earth Surface), 120, 300–324, doi:10.1002/
Baartman, J. E. M., A. J. A. M. Temme, J. M. Schoorl, M. H. Braakhekke, 2014JF003137.
and T. Veldkamp (2012), Did tillage erosion play a role in millennial scale Bellugi, D., D. G. Milledge, W. E. Dietrich, J. T. Perron, and J. McKean
landscape development?, Earth Surface Processes and Landforms, 37(15), (2015b), Predicting shallow landslide size and location across a natural
1615–1626, doi:10.1002/esp.3262. landscape: Application of a spectral clustering search algorithm, Journal
Baartman, J. E. M., A. J. A. M. Temme, T. Veldkamp, V. G. Jetten, and J. of Geophysical Research (Earth Surface), 120, 2552–2585, doi:10.1002/
M. Schoorl (2013), Exploring the role of rainfall variability and extreme 2015JF003520.
events in long-term landscape development, Catena, 109, 25–38, Bercovici, D. (2003), The generation of plate tectonics from mantle
doi:10.1016/j.catena.2013.05.003. convection, Earth and Planetary Science Letters, 205(3–4), 107–121,
Bagnold, R. A. (1936), The movement of desert sand, Proceedings of the doi:10.1016/S0012-821X(02)01009-9.
Royal Society of London. Series A, Mathematical and Physical Sciences , Bergeron, T. (1961), Preliminary results of ‘Project Pluvius’, vol. 53,
157(892), 594–620. pp. 226–237. International Association of Hydrological Sciences Publica-
Bak, P., C. Tang, and K. Wiesenfeld (1988), Scale invariant spatial and tion, Gentbrugge.
temporal fluctuations in complex systems, in Random fluctuations and Bernoux, M., C. C. Cerri, C. Neill, and J. F. L. de Moraes (1998), The use
pattern growth: Experiments and models, edited by H. E. Stanley and N. of stable carbon isotopes for estimating soil organic matter turnover rates,
Ostrowsky, pp. 329–335, Kluwer, Berlin. Geoderma, 82(1–3), 43–58.
Baker, V. R. (2008), Planetary landscape systems: A limitless frontier, Earth Beven, K. J. (1996), Equifinality and uncertainty in geomorphological
Surface Processes and Landforms, 33, 1341–1353, doi:10.1002/esp.1713. modelling, in The scientific nature of geomorphology: Proceedings of
Baker, V. R. (2009), The Channeled Scabland: A retrospective, Annual the 27th Binghampton Symposium in geomorphology, 27–29 September,
Review of Earth and Planetary Sciences, 37, 393–411, doi:10.1146/ 1996, edited by B. L. Rhoads and C. E. Thorn, pp. 289–313, Wiley,
annurev.earth.061008.134726. Chichester, UK.
Baldwin, J. A., K. X. Whipple, and G. E. Tucker (2003), Implications of Beven, K. J. (2000), Uniqueness of place and process representations in
the shear stress river incision model for the timescale of postorogenic hydrological modelling, Hydrology and Earth System Sciences, 4(2),
decay of topography, Journal of Geophysical Research (Solid Earth), 203–213.
108(B3), art. no. 2158, doi:10.1029/2001JB000550. Beven, K. J. (2012), Rainfall-runoff modelling: The primer, 2nd ed.,
Balmforth, N. J., R. V. Craster, A. C. Rust, and R. Sassi (2007), Visco- Wiley-Blackwell, Chichester, UK.
plastic flow over an inclined surface, Journal of Non-Newtonian Fluid Beven, K. J., and A. M. Binley (1992), The future of distributed models:
Mechanics, 142, 219–243, doi:10.1016/j.jnnfm.2006.07.013. Model calibration and uncertainty prediction, Hydrological Processes, 6,
Barman, A. K., C. Varadachari, and K. Ghosh (1992), Weathering of 279–298.
silicate minerals by organic-acids. 1. Nature of cation solubilization, Beven, K. J., and K. Germann (1982), Macropores and water-flow
Geoderma, 53(1–2), 45–63. in soils, Water Resources Research, 18(5), 1311–1325, doi:10.1029/
Barreto, L., J. M. Schoorl, K. Kok, T. Veldkamp, and A. Hass (2013), WR018i005p01311.
Modelling potential landscape sediment delivery due to projected soybean Beven, K. J., and K. Germann (2013), Macropores and water flow in soils
expansion: A scenario study of the Balsas sub-basin, Cerrado, Maranhao revisited, Water Resources Research, 49(6), 3071–3092, doi:10.1002/
state, Brazil, Journal of Environmental Management, 115, 270–277, wrcr.20156.
doi:10.1016/j.jenvman.2012.11.017. Billings, S. A., R. W. Buddemeier, D. deB Richter, K. Van Oost, and G.
Barshad, I. (1959), Factors affecting clay formation, Clays and Clay Bohling (2010), A simple method for estimating the influence of eroding
Minerals, 6, 110–132. soil profiles on atmospheric CO2, Global Biogeochemal Cycles, 24,
Barzegar, A. R., A. Yousefi, and A. Daryashenas (2002), The effect of GB2001, doi:10.1029/2009GB003560.
addition of different amounts and types of organic materials on soil Birkeland, P. W. (1990), Soil-geomorphic research – A selective over-
physical properties and yield of wheat, Plant and Soil, 247(2), 295–301, view, Geomorphology, 3(3–4), 207–224, doi:10.1016/0169-555X(90)
doi:10.1023/A:1021561628045. 90004-A.
294 References
Bisdom, E. B. A., G. Stoops, J. Delvigne, P. Curmi, and H.-J. Altemuller Braun, J., and P. van der Beek (2004), Evolution of passive margin
(1982), Micromorphology of weathering Biotite and its secondary prod- escarpments: What can we learn from low-temperature thermochronol-
ucts, Pedologie, 32(2), 225–252. ogy?, Journal of Geophysical Research (Earth Surface), 109(F4), F04009.
Bishop, P. (2007), Long-term landscape evolution: Linking tectonics and Brimhall, G. H., O. A. Chadwick, C. J. Lewis, W. Compston, I. S.
surface processes, Earth Surface Processes and Landforms, 32, 329–365, Williams, K. J. Danti, W. E. Dietrich, M. E. Power, D. M. Hendricks,
doi:10.1002/esp.1493. and J. Bratt (1992), Deformational mass transport and invasive processes
Boardman, J., A. J. Parsons, R. Holland, P. J. Holmes, and R. Washington in soil evolution, Science, 255(5045), 695–702.
(2003), Development of badlands and gullies in the Sneeuberg, Great Brimhall, G. H., and W. E. Dietrich (1987), Constitutive mass balance
Karoo, South Africa, Catena, 50(2–4), 165–184. relations between chemical composition, volume, density, porosity, and
Bodí, M. B., D. A. Martin, V. N. Balfour, C. Santín, S. H. Doerr, P. strain in metasomatic hydrochemical systems: Results on weathering and
Pereira, A. Cerdà, and J. Mataix-Sorda (2014), Wildland fire ash: Produc- pedogenesis, Geochimica Cosmochimica Acta, 51, 567–587.
tion, composition and eco-hydro-geomorphic effects, Earth-Science Brimhall, G. H., C. J. Lewis, C. Ford, J. Bratt, G. Taylor, and O. Warin
Reviews, 130, 103–127, doi:10.1016/j.earscirev.2013.12.007. (1991), Quantitative geochemical approach to pedogenesis – Importance
Boillat, J. L., and W. H. Graf (1982), Settling velocity of spherical of parent material reduction, volumetric expansion, and Eolian influx in
particles in turbulent media, Journal of Hydraulic Research, 20, 395–413. Lateritization, Geoderma, 51(1–4), 51–91.
Bondeau, A., et al. (2007), Modelling the role of agriculture for the 20th Brocard, G. Y., J. K. Willenbring, T. E. Miller, and F. N. Scatena (2016),
century global terrestrial carbon balance, Global Change Biology, 13, Relict landscape resistance to dissection by upstream migrating knick-
679–706, doi:10.1111/j.1365-2486.2006.01305.x. points, Journal of Geophysical Research (Earth Surface), 121,
1182–1203, doi:10.1002/2015JF003678.
Borga, M., G. D. Fontana, and F. Cazorzi (2002), Analysis of topographic
and climatic control on rainfall-triggered shallow landsliding using a Brooks, R. J., M. A. Semenov, and P. D. Jamieson (2001), Simplifying
quasi-dynamic wetness index, Journal of Hydrology, 268, 56–71. Sirius: Sensitivity analysis and development of a meta-model for wheat
yield prediction, European Journal of Agronomics, 14, 43–60.
Bourrier, F. (2008), Modélisation de l’impact d’un bloc rocheux sur un
terrain naturel, application à la trajectographie des chutes de blocs, PhD Brunetti, M. T., F. Guzzetti, and M. Rossi (2009), Probability distributions
thesis, Institut National Polytechnique de Grenoble, Grenoble. of landslide volumes, Nonlinear Processes in Geophysics, 16(2), 179–188.
Bourrier, F., L. K. A. Dorren, F. Nicot, F. Berger, and F. Darve (2009), Buck, W. R. (1993), Effect of lithospheric thickness on the formation of
Toward objective rockfall trajectory simulation using a stochastic impact high- and low-angle normal faults, Geology, 21(10), 933–936,
model, Geomorphology, 110, 68–79, doi:10.1016/j.geomorph.2009.03.017. doi:10.1130/0091-7613(1993) 021<0933:EOLTOT> 2.3.CO;2.
Bovy, B., J. Braun, and A. Demoulin (2016), A new numerical framework Bufe, A., C. Paola, and D. W. Burbank (2016), Fluvial bevelling of
for simulating the control of weather and climate on the evolution of soil- topography controlled by lateral channel mobility and uplift rate, Nature
mantled hillslopes, Geomorphology, 263, 99–112, doi:10.1016/j. Geoscience, 9(9), 706–710, doi:10.1038/ngeo2773.
geomorph.2016.03.016. Bull, L. J., and M. J. Kirkby (1997), Gully processes and modelling,
Bowman, D. M. J. S., G. S. Boggs, and L. D. Prior (2008), Fire maintains an Progress in Physical Geography, 21(3), 354–374, doi:10.1177/
Acacia aneura shrubland – Triodia grassland mosaic in central Australia, 030913339702100302.
Journal of Arid Environments, 72, 34–47, doi:10.1016/j.jaridenv.2007.04.001. Bull, L. J., and M. J. Kirkby (2002), Channel heads and channel extension,
Bradford, S. A., and S. Torkzaban (2008), Colloid transport and retention in Dryland rivers: Hydrology and geomorphology of semi-arid channels,
in unsaturated porous media: A review of interface-, collector-, and pore- edited by L. J. Bull and M. J. Kirkby, pp. 263–298, Wiley, Chichester,
scale processes and models, Vadose Zone Journal, 7(2), 667–681, UK.
doi:10.2136/vzj2007.0092. Burke, B. C., A. M. Heimsath, and A. F. White (2007), Coupling chemical
Brantley, S. L., and M. I. Lebedeva (2011), Learning to read the chemistry weathering with soil production across soil-mantled landscapes, Earth
of regolith to understand the Critical Zone, Annual Review of Earth and Surface Processes and Landforms, 32, 853–873, doi:10.1002/esp.1443.
Planetary Sciences, 39, 387–416, doi:10.1146/annurev-earth-040809- Burnett, S. A., J. A. Hattey, J. E. Johnson, A. L. Swann, D. I. Moore, and
152321. S. L. Collins (2012), Effects of fire on belowground biomass in Chihua-
Bras, R. L., and I. Rodriguez-Iturbe (1985), Random functions and hydrol- huan desert grassland, Ecosphere, 3(11), 107, doi:10.1890/ES12-00248.1.
ogy, Addison-Wesley, New York. Burr, D. M., et al. (2013), Fluvial features on Titan: Insights from morph-
Braun, J., A. M. Heimsath, and J. Chappell (2001), Sediment transport ology and modeling, Geological Society of America Bulletin, 125(3–4),
mechanisms on soil-mantled hillslopes, Geology, 29(8), 683–686. 299–321, doi:10.1130/B30612.1.
Braun, J., J. Mercier, F. Guillocheau, and C. Robin (2016), A simple model Burroughs, E. R., and B. R. Thomas (1977), Declining root strength in
for regolith formation by chemical weathering, Journal of Geophysical Douglas-Fir after felling as a factor in slope stability, Rep. INT-190, US
Research (Earth Surface), 121, 2140–2171, doi:10.1002/2016JF003914. Department of Agriculture, Ogden, Utah.
Braun, J., and X. Robert (2005), Constraints on the rate of post-orogenic Buss, H. L., P. B. Sak, S. M. Webb, and S. L. Brantley (2008), Weathering
erosional decay from low-temperature thermochronological data: Applica- of the Rio Blanco quartz diorite, Luquillo Mountains, Puerto Rico:
tion to the Dabie Shan, China, Earth Surface Processes and Landforms, Coupling oxidation, dissolution, and fracturing, Geochimica Cosmochi-
30, 1203–1225, doi:10.1002/esp.1271. mica Acta, 72, 4488–4507, doi:10.1016/j.gca.2008.06.020.
Braun, J., and M. Sambridge (1997), Modelling landscape evolution on Buzzi, O., A. Giacomini, and M. Spadari (2012), Laboratory investigation
geological time scales: A new method based on irregular spatial discretiza- of high values of restitution coefficients, Rock Mechanics and Rock
tion, Basin Research, 9(1), 27–52. Engineering, 45, 35–43, doi:10.1007/s00603-011-0183-0.
References 295
Byrne, K. A., and G. Kiely (2008), Evaluation of models (PaSim, RothC, Chen, M., G. R. Willgoose, and P. M. Saco (2015), Evaluation of the
CENTURY and DNDC) for simulation of grassland carbon cycling at plot, hydrology of the IBIS land surface model in a semi-arid catchment,
field and regional scale, Environment Protection Agency, Wexford, Ireland. Hydrological Processes, 29, 653–670, doi:10.1002/hyp.10156.
Cagnoli, B., and M. Manga (2003), Pumice-pumice collisions and the Cheng, D.-L., and K. J. Niklas (2007), Above- and below-ground biomass
effect of the impact angle, Geophysical Research Letters, 30(12), 1636, relationships across 1534 forested communities, Annals of Botany, 99,
doi:10.1029/2003GL017421. 95–102, doi:10.1093/aob/mcl206.
Campbell, C. S., P. W. Cleary, and M. Hopkins (1995), Large-scale Chia, E. K., M. Bassett, D. G. Nimmo, S. W. J. Leonard, E. G. Ritchie, M.
landslide simulations: Global deformation, velocities and basal friction, F. Clarke, and A. F. Bennett (2015), Fire severity and fire-induced land-
Journal of Geophysical Research (Solid Earth), 100(B5), 8267–8283, scape heterogeneity affect arboreal mammals in fire-prone forests, Eco-
doi:10.1029/94JB00937. sphere, 6(10), 190.
Camporeale, C., E. Perucca, L. Ridolfi, and A. M. Gurnell (2013), Model- Chien-Yuan, C., Y. Fan-Chieh, L. Sheng-Chi, and C. Kei-Wai (2007),
ling the interactions between river morphodynamics and riparian vegeta- Discussion of landslide self-organized criticality and the initiation of
tion, Reviews of Geophysics, 51, 379–414, doi:10.1002/rog.20014. debris flow, Earth Surface Processes and Landforms, 32(2), 197–209,
Canadell, J. G., R. B. Jackson, J. R. Ehleringer, H. A. Mooney, O. E. Sala, doi:10.1002/esp.1400.
and E. D. Schulze (1996), Maximum rooting depth of vegetation types at Chigara, M., and T. Oyama (1999), Mechanism and effect of chemical
the global scale, Oecologia, 108(4), 583–595, doi:10.1007/BF00329030. weathering of sedimentary rocks, Engineering Geology, 55(1), 3–14,
Canales, J. B., G. Ito, R. S. Detrick, and J. Sinton (2002), Crustal thickness doi:10.1016/S0013-7952(99)00102-7.
along the western Galapagos Spreading Center and the compensation of the Chowdhury, A. F. M. K. (2017), Development and evaluation of stochas-
Galapagos hotspot swell, Earth and Planetary Science Letters, 203, 311–327. tic rainfall models for urban water security assessment, PhD Thesis,
Carson, M. A., and M. J. Kirkby (1972), Hillslope form and process, University of Newcastle, Callaghan, Australia.
Cambridge University Press, London. Chowdhury, A. F. M. K., N. Lockart, G. R. Willgoose, G. Kuczera, A. S.
Cawson, J. G., G. J. Sheridan, H. G. Smith, and P. N. J. Lane (2012), Kiem, and N. Parana Manage (2018), Development and evaluation of a
Surface runoff and erosion after prescribed burning and the effect of stochastic daily rainfall model with long term variability, Hydrology and
different fire regimes in forests and shrublands: A review, International Earth System Sciences, doi:10.5194/hess-2017-84.
Journal of Wildland Fire, 21, 857–872, doi:10.1071/WF11160. Claessens, L., G. B. M. Heuvelink, J. M. Schoorl, and A. Veldkamp
Celia, M. A., and W. G. Gray (1991), Numerical methods for differential (2005), DEM resolution effects on shallow landslide hazard and soil
equations: Fundamental concepts for scientific & engineering applica- redistribution modelling, Earth Surface Processes and Landforms, 30,
tions, Prentice-Hall, New York. 461–477, doi:10.1002/esp.1155.
Cerdà, A., and S. H. Doerr (2005), Influence of vegetation recovery on soil Clarke, L. E. (2015), Experimental alluvial fans: Advances in understand-
hydrology and erodibility following fire: An 11-year investigation, Inter- ing of fan dynamics and processes, Geomorphology, 244, 135–145,
national Journal of Wildland Fire, 14, 423–437, doi:10.1071/WF05044. doi:10.1016/j.geomorph.2015.04.013.
Certini, G. (2005), Effects of fire on properties of forest soils: A review, Clarke, R. H. (1979), Reservoir properties of conglomerates and conglom-
Oecologia, 143, 1–10, doi:10.1007/S00442-0041788-8. erate sandstones, American Association of Petroleum Geologists Bulletin,
63, 799–809.
Certini, G., R. Scalenghe, and W. I. Woods (2013), The impact of warfare
on the soil environment, Earth-Science Reviews, 127, 1–15, doi:10.1016/j. Cohen, S. (2010), Spatial description of soil properties through landscape-
earscirev.2013.08.009. pedogenesis modelling, PhD thesis, University of Newcastle, Callaghan,
Australia.
Chadwick, O. A., G. H. Brimhall, and D. M. Hendricks (1990), From a
black to a gray box: A mass balance interpretation of pedogenesis, Cohen, S., G. R. Willgoose, and G. R. Hancock (2008), A methodology
Geomorphology, 3, 369–390. for calculating the spatial distribution of the area-slope equation and the
hypsometric integral within a catchment, Journal of Geophysical Research
Chase, C. G. (1992), Fluvial landsculpting and the fractal dimension of
(Earth Surface), 113, F03027, doi:10.1029/2007JF000820.
topography, Geomorphology, 5(1/2), 39–57, doi:10.1016/0169-555X(92)
90057-U. Cohen, S., G. R. Willgoose, and G. R. Hancock (2009), The mARM
spatially distributed soil evolution model: A computationally efficient
Chatanantavet, P., and G. Parker (2009), Physically based modeling of
modeling framework and analysis of hillslope soil surface organization,
bedrock incision by abrasion, plucking, and macroabrasion, Journal of
Journal of Geophysical Research (Earth Surface), 114, F03001,
Geophysical Research (Earth Surface), 114, F04018, doi:10.1029/
doi:10.1029/2008JF001214.
2008JF001044.
Cohen, S., G. R. Willgoose, and G. R. Hancock (2010), The mARM3D
Chatanantavet, P., and G. Parker (2011), Quantitative testing of model of
spatially distributed soil evolution model: Three-dimensional model
bedrock channel incision by plucking and macroabrasion, Journal of
framework and analysis of hillslope and landform responses, Journal of
Hydraulic Division – ASCE, 137(11), 1311–1317, doi:10.1061/(ASCE)
Geophysical Research (Earth Surface), 115, F04013, doi:10.1029/
HY.1943-7900.0000421.
2009JF001536.
Chau, K. T., R. H. C. Wong, and J. J. Wu (2002), Coefficient of restitution and
Cohen, S., G. R. Willgoose, and G. R. Hancock (2013), Soil response
rotational motions of rockfall impacts, International Journal of Rock Mechan-
to late-Quaternary climatic oscillations, new insights based on numerical
ics & Mining Sciences, 39, 69–77, doi:10.1016/S1365-1609(02)00016–3.
simulations, Quaternary Research, 79(3), 452–457, doi:10.1016/j.
Chen, A., J. Darbon, and J.-M. Morel (2014), Landscape evolution yqres.2013.01.001.
models: A review of their fundamental equations, Geomorphology, 219,
Cohen, S., T. Svoray, S. Sela, G. R. Hancock, and G. R. Willgoose (2015),
68–86, doi:10.1016/j.geomorph.2014.04.037.
The effect of sediment-transport, weathering and aeolian mechanisms on
296 References
soil evolution, Journal of Geophysical Research (Earth Surface), 120(2), Culling, W. E. H. (1960), Analytical theory of erosion, Journal of Geol-
260–274, doi:10.1002/2014JF003186. ogy, 68(3), 336–344.
Cohen, S., T. Svoray, S. Sela, G. R. Hancock, and G. R. Willgoose Culling, W. E. H. (1963), Soil creep and the development of hillside
(2016), Soilscape evolution of aeolian-dominated hillslopes during the slopes, Journal of Geology, 71(2), 127–161.
Holocene: Investigation of sediment transport mechanisms and climatic- Cundall, P. A., and O. D. L. Strack (1979), A discrete numerical model for
anthropogenic drivers, Earth Surface Dynamics, 5, 101–112, granular assemblies, Geotechnique, 29(1), 47–65, doi:10.1680/
doi:10.5194/esurf-5-101-2017. geot.1979.29.1.47.
Coleman, K., and D. S. Jenkinson (2014), RothC – A model for the Dagbovie, A. S., and J. A. Sheratt (2014), Pattern selection and hysteresis
turnover of carbon in soil. Model description and users guide, Rothamsted in the Rietkerk model for banded vegetation in semi-arid environments,
Research, UK. Journal of the Royal Society Interface, 11, 20140465, doi:10.1098/
Coleman, K., D. S. Jenkinson, G. J. Crocker, P. R. Grace, J. Klir, M. rsif.2014.0465.
Korschens, P. R. Poulton, and D. D. Richter (1997), Simulating trends in Dahlen, F. A. (1990), Critical taper model of fold-and-thrust belts and
soil organic carbon in long-term experiments using RothC-26.3, Geo- accretionary wedges, Annual Review of Earth and Planetary Sciences, 18,
derma, 81(1–2), 29–44. 55–99, doi:10.1146/annurev.earth.18.1.55.
Collins, D. B. G., and R. L. Bras (2008), Climatic control of sediment Daily, J. W., and D. R. F. Harleman (1966), Fluid dynamics, Addison-
yield in dry lands following climate and land cover change, Water Wesley, Reading, MA.
Resources Research, 44, W10405, doi:10.1029/2007WR006474.
Davidson, E. A., E. Belk, and R. D. Boone (1998), Soil water content and
Collins, D. B. G., and R. L. Bras (2010), Climatic and ecological controls temperature as independent or confounded factors controlling soil respir-
of equilibrium drainage density, relief, and channel concavity in dry lands, ation in a temperate mixed hardwood forest, Global Change Biology, 4(2),
Water Resources Research, 46, W04508, doi:10.1029/2009WR008615. 217–227, doi:10.1046/j.1365-2486.1998.00128.x.
Collins, D. B. G., R. L. Bras, and G. E. Tucker (2004), Modeling the effects of Davidson, E. A., and I. A. Janssens (2006), Temperature sensitivity of soil
vegetation-erosion coupling on landscape evolution, Journal of Geophysical carbon decomposition and feedbacks to climate change, Nature, 440
Research (Earth Surface), 109(F3), F03004. (7081), 165–173, doi:10.1038/nature04514.
Collins, K. M., O. F. Price, and T. D. Penman (2015), Spatial patterns of Davis, W. M. (1899), The geographical cycle, Geographical Journal, 14,
wildfire ignitions in south-eastern Australia, International Journal of Wild- 481–504.
land Fire, 24, 1098–1108, doi:10.1071/WF15054.
Davy, P., and D. Lague (2009), Fluvial erosion/transport equation of
Conrad, C. P., and C. Lithgow-Bertelloni (2002), How mantle slabs drive landscape evolution models revisited, Journal of Geophysical Research
plate tectonics, Science, 298(5591), 207–209. (Earth Surface), 114, F03007, doi:10.1029/2008JF001146.
Corti, G., A. Agnelli, G. Certini, and F. C. Ugolini (2001), The soil DeBano, L. F. (2000), The role of fire and soil heating on water repellency in
skeleton as a tool for disentangling pedogenetic history: A case study in wildland environments: A review, Journal of Hydrology, 231–232, 195–206.
Tuscany, central Italy, Quaternary International, 78, 33–44, doi:10.1016/
De Marco, A., A. E. Gentile, C. Arena, and A. V. De Santo (2005),
S1040-6182(00)00113-0.
Organic matter, nutrient content, and biological activity in burned and
Coulthard, T. J. (2001), Landscape evolution models: A software review, unburned soils of a Mediterranean maquis area of southern Italy, Inter-
Hydrological Processes, 15(1), 165–173. national Journal of Wildland Fire, 14, 365–377, doi:10.1071/WF05030.
Coulthard, T. J., D. M. Hicks, and M. J. Van De Wiel (2007), Cellular Dennison, P. E., S. C. Brewer, J. D. Arnold, and M. A. Moritz (2014),
modelling of river catchments and reaches: Advantages, limitations and pro- Large wildfire trends in the western United States, 1984–2011, Geophys-
spects, Geomorphology, 90, 192–207, doi:10.1016/j.geomorph.2006.10.030. ical Research Letters, 41, 2928–2933, doi:10.1002/2014GL059576.
Coulthard, T. J., M. G. Macklin, and M. J. Kirkby (2002), A cellular DeNovio, N. M., J. E. Saiers, and J. N. Ryan (2004), Colloid movement in
model of Holocene upland river basin and alluvial fan evolution, Earth unsaturated porous media: Recent advances and future directions, Vadose
Surface Processes and Landforms, 27(3), 269–288. Zone Journal, 3(2), 338–251, doi:10.2113/3.2.338.
Coulthard, T. J., and C. J. Skinner (2016), The sensitivity of landscape Dessert, C., B. Dupre, J. Gaillardet, L. M. François, and C. J. Allegre
evolution models to spatial and temporal rainfall resolution, Earth Surface (2003), Basalt weathering laws and the impact of basalt weathering on the
Dynamics, 4, 757–771, doi:10.5194/esurf-4-757-2016. global carbon cycle, Chemical Geology, 202, 257–273, doi:10.1016/j.
Coventry, R. J., A. J. Moss, and E. Verster (1988), Thin surface soil layers chemgeo.2002.10.001.
attributable to rain-flow transportation on low-angle slopes: An example de Vries, H., T. Becker, and B. Eckhardt (1994), Power law distribution of
from semi-arid tropical Queensland, Australia, Earth Surface Processes discharge in ideal networks, Water Resources Research, 30(12),
and Landforms, 13, 421–430. 3541–3543.
Cox, N. R. (1980), On the relationship between bedrock lowering and de Vries, W., J. J. M. van Grinsven, N. van Breemen, E. E. J. M. Leeters,
regolith thickness, Earth Surface Processes, 5(3), 271–274, doi:10.1002/ and P. C. Jansen (1995), Impacts of acid deposition on concentrations and
esp.3760050305. fluxes of solutes in acid sandy forest soils in the Netherlands, Geoderma,
Cramer, W., and C. B. Field (1999), Comparing global models of terres- 67(1–2), 17–43, doi:10.1016/0016-7061(94)00056-G.
trial net primary productivity (NPP): Introduction, Global Change Biol- Dewitte, O., M. Daoudi, C. Bosco, and M. Van Der Eeckhaut (2015),
ogy, 5(Suppl. 1), 3–4. Predicting the susceptibility to gully initiation in data-poor regions, Geo-
Crave, A., and P. Davy (2001), A stochastic ‘‘precipiton’’ model for morphology, 228, 101–115, doi:10.1016/j.geomorph.2014.08.010.
simulating erosion/sedimentation dynamics, Computers & Geosciences, Dialynas, Y. G., S. Bastola, R. L. Bras, S. A. Billings, D. Markewitz, and
27(7), 815–827, doi:10.1016/S0098-3004(00)00167-9. D. D. Richter (2016), Topographic variability and the influence of soil
References 297
erosion on the carbon cycle, Global Biogeochemical Cycle, 30, Dunkerley, D. L. (2002), Infiltration rates and soil moisture in a groved
doi:10.1002/2015GB005302. mulga community near Alice Springs, arid central Australia: Evidence for
Dietrich, W. E. (1982), Settling velocity of natural particles, Water complex internal rainwater redistribution in a runoff–runon landscape,
Resources Research, 18(6), 1615–1626, doi:10.1029/WR018i006p01615. Journal of Arid Environments, 51(2), 199–219, doi:10.1006/
jare.2001.0941.
Dietrich, W. E., and D. R. Montgomery (1998), SHALSTAB: A digital
terrain model for mapping shallow landslide potential, National Council Dunkerley, D. L., and K. J. Brown (1995), Runoff and runon areas in a
for Air and Stream Improvement. patterned chenopod shrubland, arid western New South Wales, Australia:
Characteristics and origin, Journal of Arid Environments, 30(1), 41–55.
Dietrich, W. E., and J. T. Perron (2006), The search for a topographic
signature of life, Nature, 439(7075), 411–418. Dunkerley, D. L., and K. J. Brown (1999), Banded vegetation near Broken
Hill, Australia: Significance of surface roughness and soil physical prop-
Dietrich, W. E., R. Reiss, M. L. Hsu, and D. R. Montgomery (1995),
erties, Catena, 37(1–2), 75–88.
A process based model for colluvial soil depth and shallow landsliding
using digital elevation data, Hydrological Processes, 9(3/4), 383–400, Dunne, T., D. V. Malmon, and K. B. J. Dunne (2016), Limits on the
doi:10.1002/hyp.3360090311. morphogenetic role of rain splash transport in hillslope evolution, Journal
of Geophysical Research (Earth Surface), 121, doi:10.1002/
Dietrich, W. E., and P. Whiting (1989), Boundary shear stress and sedi-
2015JF003737.
ment transport in river meanders of sand and gravel, in River Meandering,
edited by S. Ikeda and G. Parker, pp. 1–40, American Geophysical Union, Dunne, T., D. V. Malmon, and S. M. Mudd (2010), A rain splash transport
Washington, DC. equation assimilating field and laboratory measurements, Journal of Geo-
physical Research (Earth Surface), 115, F01001, doi:10.1029/
Dietrich, W. E., C. J. Wilson, D. R. Montgomery, J. McKean, and R. A.
2009JF001302.
Bauer (1992), Erosion thresholds and land surface morphology, Geology,
20, 675–679. Durda, D. D., W. F. Bottke, B. L. Enke, W. J. Merline, E. Asphaug, D. C.
Richardson, and Z. M. Leinhardt (2004), The formation of asteroid satel-
Dietrich, W. E., C. J. Wilson, D. R. Montgomery, and J. McKean
lites in large impacts: Results from numerical simulations, Icarus, 170(1),
(1993), Analysis of erosion thresholds, channel networks, and land-
243–257.
scape morphology using a digital terrain model, Journal of Geology,
101(2), 259–278. Durda, D. D., W. F. Bottke, D. Nesvorny, B. L. Enke, W. J. Merline, E.
Asphaug, and D. C. Richardson (2007), Size–frequency distributions of
Dijkstra, J. J., J. C. L. Meeussen, and R. N. J. Comans (2004), Leaching of
fragments from SPH/N-body simulations of asteroid impacts: Comparison
heavy metals from contaminated soils: An experimental and modeling
with observed asteroid families, Icarus, 186(2), 498–516.
study, Environmental Science & Technology, 38(16), 4390–4395,
doi:10.1021/es049885v. Durda, D. D., N. Movshovitz, D. C. Richardson, E. Asphaug, A. Morgan,
A. R. Rawlings, and C. Vest (2011), Experimental determination of the
Doetterl, S., A. A. Berhe, E. Nadeu, Z. Wang, M. Sommer, and P. Fiener
coefficient of restitution for meter-scale granite spheres, Icarus, 211(1),
(2016), Erosion, deposition and soil carbon: A review of process-level
849–855.
controls, experimental tools and models to address C cycling in dynamic
landscapes, Earth-Science Reviews, 154, 102–122, doi:10.1016/j. Eagleson, P. S. (2002), Ecohydrology: Darwinian expression of vegetation form
earscirev.2015.12.005. and function, Cambridge University Press, Cambridge.
Dorren, L. K. A. (2003), A review of rockfall mechanics and modelling Einstein, H. A. (1950), The bed-load function for sediment transportation
approaches, Progress in Physical Geography, 27(1), 69–87. in open channel flows, Technical Bulletin 1026, US Department of Agri-
culture, Washington, DC.
Dorren, L. K. A., F. Berger, C. le Hir, E. Mermin, and P. Tardif (2005),
Mechanisms, effects and management implications of rockfall in forests, Elias, E. A., R. Cichota, H. H. Torriani, and Q. de Jong van Lier (2004),
Forest Ecology and Management, 215, 183–195, doi:10.1016/j. Analytical soil-temperature model: Correction for temporal variation of
foreco.2005.05.012. daily amplitude, Soil Science Society of America Journal, 68, 784–788,
doi:10.2136/sssaj2004.7840.
Dorren, L. K. A., F. Berger, and U. S. Putters (2006), Real-size experi-
ments and 3-D simulation of rockfall on forested and non-forested slopes, Enquist, B. J., and K. J. Niklas (2002), Global allocation rules for patterns
Natural Hazards and Earth System Sciences, 6, 145–153. of biomass partitioning in seed plants, Science, 295(5559), 1517–1520.
Driscoll, N. W., and G. D. Karner (1994), Flexural deformation due to Eppes, M. C., and D. Griffing (2010), Granular disintegration of marble in
Amazon fan loading: A feedback mechanism affecting sediment delivery nature: A thermal-mechanical origin for a grus and corestone landscape,
to margins, Geology, 22, 1015–1018, doi:10.1130/0091-7613(1994) Geomorphology, 117, 170–180, doi:10.1016/j.geomorph.2009.11.028.
022<1015:FDDTAF> 2.3.CO;2. Espinoza, J. C., S. Chavez, J. Ronchail, C. Junquas, K. Takahashi, and W.
Dunai, T. J. (2010), Cosmogenic nuclides: Principles, concepts and appli- Lavado (2015), Rainfall hotspots over the southern tropical Andes: Spatial
cations in the earth surface sciences, Cambridge University Press, distribution, rainfall intensity, and relations with large-scale atmospheric
Cambridge. circulation, Water Resources Research, 51, 3459–3475, doi:10.1002/
2014WR016273.
Dunkerley, D. L. (1997), Banded vegetation: Development under uniform
rainfall from a simple cellular automaton model, Plant Ecology, 129, Espirito-Santo, F. D. B., et al. (2014), Size and frequency of natural forest
103–111. disturbances and the Amazon forest carbon balance, Nature Communi-
cations, 5, 1–6, doi:10.1038/ncomms4434.
Dunkerley, D. L. (2000), Hydrologic effects of dryland shrubs: Defining
the spatial extent of modified soil water uptake rates at an Australian Evans, K. G., and R. J. Loch (1996), Using the RUSLE to identify factors
desert site, Journal of Arid Environments, 45, 159–172, doi:10.1006/ controlling erosion rates of mine spoils, Land Degradation and Develop-
jare.2000.0636. ment, 7, 267–277.
298 References
Evans, K. G., M. J. Saynor, and G. R. Willgoose (1996), The effect of Fletcher, R. C., H. L. Buss, and S. L. Brantley (2006), A spheroidal
vegetation on waste rock erosion, Ranger Uranium Mine, Northern Terri- weathering model coupling porewater chemistry to soil thicknesses during
tory, Bulletin of the Australian Institute of Mining and Metallurgy, 6, steady-state denudation, Earth and Planetary Science Letters, 244,
21–23. 444–457, doi:10.1016/j.epsl.2006.01.055.
Evans, K. G., M. J. Saynor, G. R. Willgoose, and S. J. Riley (2000), Post- Fontúrbel, M. T., A. Barreiro, J. A. Vega, A. Martín, E. Jiménez, T. Carballas,
mining landform evolution modelling. I. Derivation of sediment transport C. Fernández, and M. Díaz-Raviña (2012), Effects of an experimental fire and
model and rainfall-runoff model parameters, Earth Surface Processes and post-fire stabilization treatments on soil microbial communities, Geoderma,
Landforms, 25(7), 743–763. 191, 51–60, doi:10.1016/j.geoderma.2012.01.037.
Evans, K. G., and G. R. Willgoose (2000), Post-mining landform evolu- Fordham, A. W. (1990), Weathering of biotite into dioctahedral clay
tion modelling. II. Effects of vegetation and surface ripping, Earth Surface minerals, Clay Minerals, 25, 51–63.
Processes and Landforms, 25(8), 803–823. Foster, G. R. (1982), Modelling the erosion process, in Hydrologic Mod-
Fagherazzi, S., A. D. Howard, and P. L. Wiberg (2002), An implicit finite elling of Small Watersheds, edited by C. T. Haan, pp. 295–380, American
difference method for drainage basin evolution, Water Resources Society of Agricultural Engineers, St Joseph, Missouri.
Research, 38(7), art. no.-1116. Foster, G. R., D. C. Flanagan, M. A. Nearing, L. J. Lane, L. M. Risse,
FAO (2009), Harmonized World Soil Database (version 1.1), FAO, and S. C. Finkner (1995), Chapter 11: Hillslope erosion component, in
Rome, Italy, and IIASA, Laxenburg, Austria. USDA-Water Erosion Prediction Project Hillslope Profile and Water-
Fatichi, S., C. Pappas, and V. Y. Ivanov (2016), Modeling plant–water shed Model Documentation, NSERL Report #10, edited by D. C. Flana-
interactions: an ecohydrological overview from the cell to the global scale, gan and M. A. Nearing, US Department of Agriculture ARS, West
Wiley Interdisciplinary Reviews – Water, 3(3), 327–368, doi:10.1002/ Lafayette, Indiana.
wat2.1125. Fox, M., L. Goren, D. A. May, and S. D. Willett (2014), Inversion of fluvial
Field, J. B., and G. R. Anderson (2003), Biological agents in regolith channels for paleorock uplift rates in Taiwan, Journal of Geophysical
processes: Case study on the Southern Highlands, NSW, paper presented Research (Earth Surface), 119, 1853–1875, doi:10.1002/2014JF003196.
at Advances in Regolith: Proceedings of the CRC LEME Regional Regolith Fox, R. W., and A. T. McDonald (1998), Introduction to Fluid Mechanics,
Symposia, 2003, Cooperative Research Centre for Landscape Environments 5th ed., Wiley, Chichester, UK.
and Mineral Exploration (CRC LEME), Canberra, www.crcleme.org.au/ Fraser, H. J. (1935), Experimental study of the porosity and permeability
Pubs/Advancesinregolith/Field_Anderson.pdf. of clastic sediments, Journal of Geology, 83(8), 910–1010.
Finke, P. A. (2012), Modeling the genesis of luvisols as a function of Freer, J., J. J. McDonnell, K. J. Beven, N. E. Peters, D. A. Burns, R. P.
topographic position in loess parent material, Quaternary International, Hooper, B. Aulenbach, and C. Kendall (2002), The role of bedrock topog-
265, 3–17, doi:10.1016/j.quaint.2011.10.016. raphy on subsurface storm flow, Water Resources Research, 38(12), art. no.-
Finke, P. A., A. Samouelian, M. Sourez-Bonnet, B. Laroche, and S. S. 1269.
Cornu (2015), Assessing the usage potential of SoilGen2 to predict clay Freeze, R. A., and J. A. Cherry (1979), Groundwater, Prentice Hall,
translocation under forest and agricultural land uses, European Journal of Englewood Cliffs, NJ.
Soil Science, 66, 194–205, doi:10.1111/ejss.12190.
Friedlingstein, P., et al. (2006), Climate-carbon cycling feedback analysis
Finke, P. A., T. Vanwalleghem, E. Opolot, J. Poesen, and J. Deckers results from the C4MIP model intercomparison, Journal of Climate, 19
(2013), Estimating the effect of tree uprooting on variation of soil horizon (14), 3337–3353.
depth by confronting pedogenetic simulations to measurements in a Bel-
Fujiwara, T., S. Kodaira, T. No, Y. Kaiho, N. Takahashi, and Y. Kaneda
gian loess area, Journal of Geophysical Research (Earth Surface), 118,
(2011), The 2011 Tohoku-Oki earthquake: Displacement reaching the
2124–2139, doi:10.1002/jgrf.20153.
trench axis, Science, 334(6060), 1240, doi:10. 1126/science. 1211554.
Finney, M. A. (1999), Mechanistic modeling of landscape fire patterns, in
Furbish, D. J., A. E. Ball, and M. Schmeeckle (2012a), A probabilistic
Spatial modeling of forest landscape change: Approaches and applica-
description of the bed load sediment flux: 4. Fickian diffusion at low transport
tions, edited by D. J. Mladenoff and W. L. Baker, pp. 186–209, Cambridge
rates, Journal of Geophysical Research (Earth Surface), 117, F03034,
University Press, Cambridge.
doi:10.1029/2012JF002356.
Finney, M. A. (2004), FARSITE: Fire Area Simulator – Model Develop-
Furbish, D. J., E. M. Childs, P. K. Haff, and M. W. Schmeeckle (2009),
ment and Evaluation, Rep. RMRS-RP-4, USDA Forest Service Rocky
Rain splash of soil grains as a stochastic advection-dispersion process,
Mountain Research Station.
with implications for desert plant-soil interactions and land-surface evolu-
Fleming, R. W., and A. M. Johnson (1975), Rates of seasonal creep of silty tion, Journal of Geophysical Research (Earth Surface), 114, F00A03,
clay soil, Quarterly Journal of Engineering Geology and Hydrogeology, 8 doi:10.1029/2009JF001265.
(1), 1–29, doi:10.1144/GSL.QJEG.1975.008.01.01.
Furbish, D. J., and S. Fagherazzi (2001), Stability of creeping soil and
Fleskens, L., M. J. Kirkby, and B. J. Irvine (2016), The PESERA- implications for hillslope evolution, Water Resources Research, 37(10),
DESMICE modeling framework for spatial assessment of the 2607–2618.
physical impact and economic viability of land degradation mitigation
Furbish, D. J., P. K. Haff, J. C. Roseberry, and M. Schmeeckle (2012b),
technologies, Frontiers in Environmental Science, 4, 31, doi:10.3389/
A probabilistic description of the bed load sediment flux: 1. Theory,
fenvs.2016.00031.
Journal of Geophysical Research (Earth Surface), 117, F03031,
Fletcher, R. C., and S. L. Brantley (2010), Reduction of bedrock blocks as doi:10.1029/2012JF002352.
corestones in the weathering profile: Observations and models, American
Furbish, D. J., K. K. Hamner, M. Schmeeckle, M. N. Borosund, and S. M.
Journal of Science, 310, 131–164, doi:10.2475/03.2010.01].
Mudd (2007), Rain splash of dry sand revealed by high-speed imaging and
References 299
sticky paper splash targets, Journal of Geophysical Research (Earth Sur- Giacomini, A., O. Buzzi, B. Renard, and G. P. Giani (2009), Experimental
face), 112, F01001, doi:10.1029/2006JF000498. studies on fragmentation of rockfalls on impact with rock surfaces, Inter-
Furbish, D. J., J. C. Roseberry, and M. Schmeeckle (2012c), A probabilistic national Journal of Rock Mechanics & Mining Sciences, 46, 708–715,
description of the bed load sediment flux: 3. The particle velocity distribu- doi:10.1016/j.ijrmms.2008.09.007.
tion and the diffusive flux, Journal of Geophysical Research (Earth Sur- Gignoux, J., J. I. House, D. Hall, D. Masse, H. B. Nacro, and L. Abbadie
face), 117, F03033, doi:10.1029/2012JF002355. (2001), Design and test of a generic cohort model of soil organic matter
Gabet, E. J. (2000), Gopher bioturbation: Field evidence for nonlinear decomposition: The SOMKO model, Global Ecology and Biogeography,
hillslope diffusion, Earth Surface Processes and Landforms, 25(13), 10(6), 639–660.
1419–1428. Gilbert, G. (1909), The convexity of hillslopes, Journal of Geology, 17,
Gabet, E. J., and T. Dunne (2003), Sediment detachment by rain power, 344–350.
Water Resources Research, 39(1), ESG1, doi:10.1029/2001WR000656. Gilchrist, A. R., H. Kooi, and C. Beaumont (1994), Post-Gondwana
Gabet, E. J., and S. M. Mudd (2010), Bedrock erosion by root fracture and geomorphic evolution of south-western Africa: Implications for the con-
tree throw: A coupled biogeomorphic model to explore the humped soil trols on landscape development from observations and numerical
production function and the persistence of hillslope soils, Journal of experiments, Journal of Geophysical Research (Solid Earth), 99(B6),
Geophysical Research (Earth Surface), 115, F04005, doi:10.1029/ 12211–12228.
2009JF001526. Gläser, G., H. Wernli, A. Kerkweg, and F. Teubler (2015), The transatlan-
Gabet, E. J., O. J. Reichman, and E. W. Seabloom (2003), The effects of tic dust transport from North Africa to the Americas – Its characteristics
bioturbation on soil processes and sediment transport, Annual Review of and source regions, Journal of Geophysical Research (Atmospheres), 120
Earth and Planetary Sciences, 31, 249–273. (121), 11231–11252, doi:10.1002/2015JD023792.
Garcia-Corona, R., E. Benito, E. de Blas, and M. E. Varela (2004), Effects Gómez-Villar, A., and J. M. García-Ruiz (2000), Surface sediment char-
of heating on some soil physical properties related to its hydrological acteristics and present dynamics in alluvial fans of the central Spanish
behaviour in two north-western Spanish soils, International Journal of Pyrenees, Geomorphology, 34(3–4), 127–144, doi:10.1016/S0169-555X
Wildland Fire, 13, 195–199, doi:10.1071/WF03068. (99)00116-6.
Garreaud, R. D., M. Vuille, R. Compagnucci, and J. Marengo (2009), Goodfellow, B. W., G. E. Hilley, S. M. Webb, L. S. Sklar, S. Moon, and C.
Present-day South American climate, Paleogeography Paleoclimatology A. Olson (2016), The chemical, mechanical, and hydrological evolution of
Paleoecology, 281, 180–195, doi:10.1016/j.palaeo.2007.10.032. weathering granitoid, Journal of Geophysical Research (Earth Surface),
121, doi:10.1002/2016JF003822.
Gasparini, N. M., G. E. Tucker, and R. L. Bras (1999), Downstream fining
through selective particle sorting in an equilibrium drainage network, Goodwin, I. D., T. D. van Ommen, M. A. J. Curran, and P. A. Mayewski
Geology, 27(12), 1079–1082. (2004), Mid latitude winter climate variability in the South Indian and
southwest Pacific regions since 1300 AD, Climate Dynamics, 22,
Gasparini, N. M., G. E. Tucker, and R. L. Bras (2004), Network-scale
783–794, doi:10.1007/s00382-004-0403-3.
dynamics of grain-size sorting: Implications for downstream fining,
stream-profile concavity, and drainage basin morphology, Earth Surface Goren, L., S. D. Willett, F. Herman, and J. Braun (2014), Coupled
Processes and Landforms, 29(4), 401–421. numerical–analytical approach to landscape evolution modeling, Earth
Surface Processes and Landforms, 39(4), 522–545, doi:10.1002/esp.3514.
Gasparini, N. M., K. X. Whipple, and R. L. Bras (2007), Predictions of
steady state and transient landscape morphology using sediment-flux- Gorsevski, P. V., P. E. Gessler, J. Boll, W. J. Elliot, and R. B. Foltz (2006),
dependent river incision models, Journal of Geophysical Research (Earth Spatially and temporally distributed modeling of landslide susceptibility,
Surface), 112, F03S09, doi:10.1029/2006JF000567. Geomorphology, 80, 178–198, doi:10.1016/j.geomorph.2006.02.011.
Géminard, J.-C., and W. Losert (2002), Frictional properties of biodisperse Graf, W. H. (1984), Hydraulics of sediment transport, Water Resources
granular matter: Effect of mixing ratio, Physical Review E, 65(4), Publications, Highlands Ranch, CO.
D041301, doi:10.1103/PhysRevE.65.041301. Graham, R. C., A. M. Rossi, and K. R. Hubbert (2010), Rock to regolith
Géminard, J.-C., W. Losert, and J. P. Gollub (1999), Frictional mechanics conversion: Producing hospitable substrates for terrestrial ecosystems, GSA
of wet granular material, Physical Review E, 59(5), D5881, doi:10.1103/ Today, 20 (2), 4–9, doi:10.1130/GSAT57A.1.
PhysRevE.59.5881. Greene, R. S. B., C. J. Chartres, and K. A. Hodgkinson (1990), The effects
Gercek, H. (2007), Poisson’s ratio values for rocks, International Journal of fire on the soil in a degraded semi-arid woodland. I. Cryptogam cover
of Rock Mechanics & Mining Sciences, 44(1), 1–13, doi:10.1016/j. and physical and micromorphological properties, Australian Journal of
ijrmms.2006.04.011. Soil Research, 28, 755–777.
Gerten, D. (2013), A vital link: Water and vegetation in the Anthropocene, Gregorich, E. G., K. J. Greer, D. W. Anderson, and B. C. Liang (1998),
Hydrology and Earth System Sciences, 17, 3841–3852, doi:10.5194/hess- Carbon distribution and losses: Erosion and deposition effects, Soil Tillage
17-3841-2013. Research, 47(3–4), 291–302, doi:10.1016/S0167-1987(98)00117-2.
Gerya, T. (2010), Introduction to numerical geodynamic modelling, Cam- Griffiths, D. V., J. Huang, and G. F. deWolfe (2011), Numerical and
bridge University Press, Cambridge. analytical observations on long and infinite slopes, International Journal
for Numerical and Analytical Methods in Geomechanics, 35, 569–585,
Ghannoum, O. (2009), C4 photosynthesis and water stress, Annals of
doi:10.1002/nag.909.
Botany, 103, 6350644, doi:10.1093/aob/mcn093.
Grimm, V., et al. (2006), A standard protocol for describing individual-
Ghezzehei, T. A., and D. Or (2001), Rheological properties of wet soils
based and agent-based models, Ecological Modelling, 198, 115–126,
and clays under steady and oscillatory stresses, Soil Science Society of
doi:10.1016/j.ecolmodel.2006.04.023.
America, 65, 624–637.
300 References
Guerit, L., F. Metiver, O. Devauchelle, E. Lajeunesse, and L. Barrier Hancock, G. R., K. G. Evans, G. R. Willgoose, D. R. Moliere, M. J.
(2014), Laboratory alluvial fans in one dimension, Physical Review E, Saynor, and R. J. Loch (2000), Medium term erosion simulation of an
90, 022203, doi:10.1103/PhysRevE.90.022203. abandoned mine site using the SIBERIA landscape evolution model,
Güneralp, I., and R. A. Marston (2012), Process–form linkages in meander Australian Journal of Soil Research, 38, 249–263.
morphodynamics: Bridging theoretical modeling and real world complex- Hancock, G. R., J. Hugo, A. Webb, and L. Turner (2017a), Sediment
ity, Progress in Physical Geography, 36(6), 718–746, doi:10.1177/ transport in steep forested catchments – An assessment of scale and
0309133312451989. disturbance, Journal of Hydrology, 547, 613–622, doi:10.1016/j.
Gupta, S., J. S. Collier, A. Palmer-Fengate, and G. Potter (2007), Cata- jhydrol.2017.02.022.
strophic flooding origin of shelf valley systems in the English Channel, Hancock, G. R., J. B. C. Lowry, and T. J. Coulthard (2015a), Catchment
Nature, 448(19 July), 342–346, doi:doi:10.1038/nature06018. reconstruction – Erosional stability at millennial time scales using land-
Gyasi-Agyei, Y., G. R. Willgoose, and F. P. de Troch (1995), Effects of scape evolution models, Geomorphology, 231, 15–27, doi:10.1016/j.
vertical resolution and map scale of digital elevation maps on geomorpho- geomorph.2014.10.034.
logic parameters used in hydrology, Hydrological Processes, 9(3/4), Hancock, G. R., J. B. C. Lowry, T. J. Coulthard, K. G. Evans, and D. R.
121–140. Moliere (2010), A catchment scale evaluation of the SIBERIA and
Hack, J. T. (1957), Studies of longitudinal stream profiles in Virginia and CAESAR landscape evolution models, Earth Surface Processes and
Maryland, USGS Professional Papers 294-B, USGS. Landforms, 35(8), 863–875, doi:10.1002/esp.1863.
Hairsine, P. B., and C. W. Rose (1992a), Modelling water erosion due to Hancock, G. R., J. B. C. Lowry, and C. Dever (2017b), Surface disturb-
overland flow using physical principles: 1 Sheet flow, Water Resources ance and erosion by pigs: A medium term assessment for the Monsoonal
Research, 28(1), 237–243. tropics, Land Degradation and Development, 28(1), 255–264,
doi:10.1002/ldr.2636.
Hairsine, P. B., and C. W. Rose (1992b), Modelling water erosion due to
overland flow using physical principles: 2 Rill flow, Water Resources Hancock, G. R., J. B. C. Lowry, C. Dever, and M. Braggins (2015b), Does
Research, 28(1), 245–250. introduced fauna influence soil erosion? A field and modelling assessment,
Science of the Total Environment, 518–519, 189–200, doi:10.1016/j.
Hairsine, P. B., and G. C. Sander (2009), Comment on ‘A transport-
scitotenv.2015.02.086.
distance based approach to scaling erosion rates’: Parts 1, 2 and 3 by
Wainwright et al., Earth Surface Processes and Landforms, 34, 882–885, Hancock, G. R., J. B. C. Lowry, D. R. Moliere, and K. G. Evans (2008),
doi:10.1002/esp.1782. An evaluation of an enhanced soil erosion and landscape evolution model:
A case study assessment of the former Nabarlek uranium mine, Northern
Hallet, P. D., S. Caul, T. J. Daniell, P. Barre, and E. Paterson (2010), The
Territory, Australia, Earth Surface Processes and Landforms, 33(13),
rheology of rhizosphere formation by root exudates and soil microbes,
2045–2063, doi:10.1002/esp.1653.
paper presented at 19th World Congress of Soil Science, 1–6 August 2010,
Brisbane, Australia. Hancock, G. R., J. B. C. Lowry, and M. J. Saynor (2017c), Surface armour
and erosion – Impacts on long-term landscape evolution, Land Degrad-
Hammer, P. T. C., R. M. Clowes, F. A. Cook, K. Vasudevan, and A. J. van
ation and Development, doi:10.1002/ldr.2738.
der Velden (2013), The big picture: A lithospheric cross section of the
North American continent, GSA Today, 21(6), 4–10, doi:10.1130/ Hancock G. R., C. Martinez, K. G. Evans, and D. R. Moliere. (2006).
GSATG95A.1. A comparison of SRTM and high-resolution digital elevation models
and their use in catchment geomorphology and hydrology – Australian
Han, S. C., J. Sauber, S. B. Luthcke, C. Ji, and F. F. Pollitz (2008),
examples, Earth Surface Processes and Landforms, 31, 809–824.
Implications of postseismic gravity change following the great 2004
Sumatra-Andaman earthquake from the regional harmonic analysis of Hancock, G. R., and G. R. Willgoose (2001), The use of a landscape simula-
GRACE intersatellite tracking data, Journal of Geophysical Research tor in the validation of the SIBERIA catchment evolution model: Declining
(Solid Earth), 113(B11), B11413, doi:10.1029/2008JB005705. equilibrium landforms, Water Resources Research, 37(7), 1981–1992.
Hancock, G. R. (2003), Effect of catchment aspect ratio on geomorpho- Hancock, G. R., and G. R. Willgoose (2002), The use of a landscape
logical descriptors, in Prediction in Geomorphology, edited by P. R. Wil- simulator in the validation of the SIBERIA landscape evolution model:
cock and R. M. Iverson, pp. 217–230, American Geophysical Union, Transient landforms, Earth Surface Processes and Landforms, 27(12),
Washington, DC. 1321–1334.
Hancock, G. R. (2005), The use of digital elevation models in the identifi- Hancock, G. R., and G. R. Willgoose (2004), An experimental and
cation and characterisation of catchments, Hydrological Processes, 19, computer simulation study of erosion on a mine tailings dam wall, Earth
1727–1749. Surface Processes and Landforms, 29(4), 457–475.
Hancock, G. R., T. J. Coulthard, and J. B. C. Lowry (2016), Long-term Hancock, G. R., G. R. Willgoose, and K. G. Evans (2002), Testing of the
landscape trajectory – Can we make predictions about landscape form and SIBERIA landscape evolution model using the Tin Camp Creek, Northern
function for post-mining landforms?, Geomorphology, 266, 121–132, Territory, Australia, field catchment, Earth Surface Processes and Land-
doi:10.1016/j.geomorph.2016.05.014. forms, 27(2), 125–143.
Hancock, G. R., T. J. Coulthard, C. Martinez, and J. D. Kalma (2011), An Hantson, S., S. Pueyo, and E. Chuvieco (2016), Global fire size distribu-
evaluation of landscape evolution models to simulate decadal and centen- tion: From power law to log-normal, International Journal of Wildland
nial scale soil erosion in grassland catchments, Journal of Hydrology, 398 Fire, 25, 403–412, doi:10.1071/WF15108.
(3–4), 171–183, doi:10.1016/j.jhydrol.2010.12.002. Hart, S. A., and N. J. Luckai (2014), Charcoal carbon pool in North American
Hancock, G. R., K. G. Evans, J. J. McDonnell, and L. Hopp (2012), boreal forests, Ecosphere, 5(8), 99, doi:10.1890/ES13-00086.1.
Ecohydrological controls on soil erosion and landscape evolution, Ecohy- Hasbargen, L. E., and C. Paola (2000), Landscape instability in an experi-
drology, 5(4), 478–490, doi:10.1002/eco.241. mental drainage basin, Geology, 28(12), 1067–1070.
References 301
Hasegawa, K. (1977), Computer simulation of the gradual migration of spectra, SEM and radiocarbon dating, European Journal of Soil Science,
meandering channels, in Proceedings of the Hokkaido Branch, Japan 65(7), 751–762, doi:10.1111/ejss.12171.
Society of Civil Engineering, pp. 197–202 (in Japanese). Hobley, E. U., and B. Wilson (2016), The depth distribution of organic
Heimsath, A. M., J. Chappell, N. A. Spooner, and D. G. Questiaux (2003), carbon in the soils of eastern Australia, Ecosphere, 7(1), 1–21, doi:e01214.
Creeping soil, Geology, 30(2), 111–114. 10.1002/ecs2.1214.
Heimsath, A. M., W. E. Dietrich, K. Nishiizummi, and R. C. Finkel Hobley, E., B. Wilson, A. Wilkie, J. Gray, and T. Koen (2015), Drivers of
(1997), The soil production function and landscape equilibrium, Nature, soil organic carbon storage and vertical distribution in Eastern Australia,
388(6640), 358–361. Plant and Soil, 390(1–2), 111–127, doi:10.1007/s11104-015-2380-1.
Heimsath, A. M., W. E. Dietrich, K. Nishiizummi, and R. C. Finkel Hodson, M. E. (2006), Does reactive surface area depend on grain size?
(1999), Cosmogenic nuclides, topography, and the spatial variation of soil Results from pH 3, 25 C far-from-equilibrium flow-through dissolution
depth, Geomorphology, 27(1–2), 151–172. experiments on anorthite and biotite, Geochimica et Cosmochimica Acta,
Heimsath, A. M., D. J. Furbish, and W. E. Dietrich (2005), The illusion of 70(7), 1655–1667, doi:10.1016/j.gca.2006.01.001.
diffusion: Field evidence for depth-dependent sediment transport, Geology, Hole, F. D. (1981), Effects of animals on soils, Geoderma, 25(1–2),
33(12), 949–952, doi:10.1130/G21868.1. 75–112, doi:10.1016/0016-7061(81)90008-2.
Heimsath, A. M., G. R. Hancock, and D. Fink (2009), The ‘humped’ soil Holmes, K. W., R. D. A, S. Sweeney, I. Numata, E. Matricardi, T. W.
production function: Eroding Arnhem Land, Australia, Earth Surface Biggs, G. Batista, and O. A. Chadwick (2004), Soil databases and the
Processes and Landforms, 34(12), 1674–1684. problem of establishing regional biogeochemical trends, Global Change
Heister, K. (2016), How accessible is the specific surface area of minerals? Biology, 10(5), 796–814, doi:10.1111/j.1529-8817.2003.00753.x.
A comparative study with Al-containing minerals as model substances, Holzworth, D. P., V. Snow, S. Janssen, I. N. Athanasiadis, M. Donatelli,
Geoderma, 263, 8–15, doi:10.1016/j.geoderma.2015.09.001. G. Hoogenboom, J. W. White, and P. Thorburn (2015), Agricultural
Henderson, F. M. (1966), Open channel flow, Macmillan, New York. production systems modelling and software: Current status and future
prospects, Environmental Modelling Software, 72, 276–286,
Hénin, S., and M. Dupuis (1945), Essai de bilan de la matière organique
doi:10.1016/j.envsoft.2014.12.013.
des sols, Annales agronomiques, 15, 161–172.
Hoosbeek, M. R. (1994), Towards the quantitative modeling of pedogen-
Hergarten, S. (2003), Landslides, sandpiles, and self-organized criticality,
esis: A review – Reply – Pedology beyond the soil landscape paradigm:
Natural Hazards and Earth System Sciences, 3, 505–514.
Pedodynamics and the connection to other sciences, Geoderma, 63(3–4),
Hilinski, T. E. (2001), Implementation of exponential depth distribution of 303–307.
organic carbon in the CENTURY model, Colorado State University,
Hoosbeek, M. R., and R. B. Bryant (1992), Towards the quantitative
Fort Collins. www.nrel.colostate.edu/projects/century5/reference/html/Cen
modeling of pedogenesis – A review, Geoderma, 55(3–4), 183–210.
tury/exp-c-distrib.htm.
Horwath, W. (2007), Carbon cycling and formation of soil organic matter,
HilleRisLambers, R., M. Rietkerk, F. van Den Bosch, H. H. T. Prins, and
in Soil Microbiology, Ecology, and Biochemistry, edited by E. A. Paul,
H. de Kroon (2001), Vegetation pattern formation in semi-arid grazing
pp. 303–340, Academic Press, Amsterdam.
systems, Ecology, 82(1), 50–61, doi:10.1890/0012-9658(2001)082[0050:
VPFISA]2.0.CO;2. Hosseini, M., J. J. Keizer, O. G. Pelayo, S. A. Prats, C. J. Ritsema, and V.
Geissen (2016), Effect of fire frequency on runoff, soil erosion, and loss of
Hilton, J. E., C. Miller, A. L. Sullivan, and C. Rucinski (2015), Effects of
organic matter at the micro-plot scale in north-central Portugal, Geoderma,
spatial and temporal variation in environmental conditions on simulation
269, 126–137, doi:10.1016/j.geoderma.2016.02.004.
of wildfire spread, Environmental Modelling Software, 67, 118–127,
doi:10.1016/j.envsoft.2015.01.015. Houghton, R. A. (2007), Balancing the global carbon budget, Annual
Review of Earth and Planetary Sciences, 35, 313–347, doi:10.1146/
Ho, J.-Y., K. T. Lee, T.-C. Chang, Z.-Y. Wang, and Y.-H. Liao (2012),
annurev.earth.35.031306.140057.
Influences of spatial distribution of soil thickness on shallow landslide
prediction, Engineering Geology, 124, 38–46, doi:10.1016/j.enggeo.2011 Hovius, N., C. P. Stark, and P. A. Allen (1997), Sediment flux from a
.09.013. mountain belt derived by landslide mapping, Geology, 25(3), 231–234,
doi:10.1130/0091-7613(1997)025<0231:SFFAMB>2.3.CO;2.
Ho, M., D. C. Verdon-Kidd, A. S. Kiem, and R. N. Drysdale (2014),
Broadening the spatial applicability of paleoclimate information – A case Howard, A. D. (1971), Simulation of stream networks by headward
study for the Murray-Darling Basin, Australia, Journal of Climate, 27(7), growth and branching, Geographical Analysis, 3, 29–50.
2477–2495, doi:10.1175/JCLI-D-13-00071.1. Howard, A. D. (1980), Thresholds in river regimes, in Thresholds in
Hobley, D. E. J., J. M. Adams, S. S. Nudurupati, E. W. H. Hutton, N. M. Geomorphology, 9th Binghampton Geomorphology Symposium, 1978,
Gasparini, E. Istanbulluoglu, and G. E. Tucker (2017), Creative computing edited by D. R. Coates and J. D. Vitek, pp. 227–258, Allen and Unwin,
with Landlab: An open-source toolkit for building, coupling, and explor- Boston.
ing two-dimensional numerical models of Earth-surface dynamics, Earth Howard, A. D. (1992), Modelling channel migration and floodplain
Surface Dynamics, 5, 21–46, doi:10.5194/esurf-5-21-2017. sedimentation in meandering streams, in Lowland floodplain rivers, edited
Hobley, E. U., G. R. Willgoose, S. Frisia, and G. E. Jacobsen (2013), by P. Carling and G. E. Petts, pp. 1–42, Wiley, Chichester, UK.
Environmental and site factors controlling the vertical distribution and Howard, A. D. (1994), A detachment-limited model of drainage-basin
radiocarbon ages of organic carbon in a sandy soil, Biology and Fertility evolution, Water Resources Research, 30(7), 2261–2285.
of Soils, 49(8), 1015–1026, doi:10.1007/s00374-013–0800-z. Howard, A. D. (1996), Modelling channel evolution and floodplain
Hobley, E. U., G. R. Willgoose, S. Frisia, and G. E. Jacobsen (2014), morphology, in Floodplain processes, edited by M. G. Anderson, D. E.
Vertical distribution of charcoal in a sandy soil: Evidence from DRIFT Walling, and P. D. Bates, pp. 15–62, Wiley, Chichester, UK.
302 References
Howard, A. D. (1998), Long profile development of bedrock channels: Istanbulluoglu, E., D. G. Tarboton, R. T. Pack, and C. H. Luce (2004),
Interaction of weathering, mass wasting, bed erosion, and sediment trans- Modeling of the interactions between forest vegetation, disturbances, and
port, in Rivers over rocks: Fluvial processes in bedrock channels, edited sediment yields, Journal of Geophysical Research (Earth Surface), 109,
by K. J. Tinkler and E. E. Wohl, pp. 297–319, America Geophysical F01009, doi:10.1029/2003JF000041.
Union, Washington, DC. Iverson, R. M. (1990), Groundwater flow fields in infinite slopes, Geo-
Howard, A. D. (2009), How to make a meandering river, Proceedings of technique, 40(1), 139–143.
the National Academy of Sciences USA, 106(41), 17246–17246. Iverson, R. M. (1997), The physics of debris flows, Reviews of Geophys-
Howard, A. D., W. E. Dietrich, and M. A. Seidl (1994), Modeling fluvial ics, 35(3), 245–296.
erosion on regional to continental scales, Journal of Geophysical Research Iverson, R. M., and R. P. Denlinger (2001), Flow of variably fluidized
(Solid Earth), 99(B7), 13971–13986. granular masses across three-dimensional terrain 1. Coulomb mixture
Howard, A. D., and G. Kerby (1983), Channel changes in badlands, theory, Journal of Geophysical Research (Solid Earth), 106(B1),
Geological Society of America Bulletin, 94(6), 739–752. 537–552.
Howard, A. D., and T. R. Knutson (1984), Sufficient conditions for river Iverson, R. M., and D. L. George (2015), A depth-averaged debris-flow
meandering – A simulation approach, Water Resources Research, 20(11), model that includes the effects of evolving dilatancy. I. Physical basis,
1659–1667. Proceedings of the Royal Society of London. Series A, Mathematical and
Huang, H. Q., and G. R. Willgoose (1993), Some scale dependent proper- Physical Sciences, 470, 20130819, doi:10.1098/rspa.2013.0819.
ties of distributed rainfall-runoff models, paper presented at Towards the Iverson, R. M., M. Logan, and R. P. Denlinger (2004), Granular ava-
21st Century, Hydrology and Water Resources Symposium, Institution of lanches across irregular three-dimensional terrain: 2. Experimental tests,
Engineers (Aust.), Newcastle. Journal of Geophysical Research (Earth Surface), 109(F1), F01015,
Huang, X., and J. D. Niemann (2006), An evaluation of the geomorphi- doi:10.1029/2003JF000084.
cally effective event for fluvial processes over long periods, Journal of Iverson, R. M., and J. W. Vallance (2001), New views of granular mass
Geophysical Research (Earth Surface), 111, doi:10.1029/2006JF000477. flows, Geology, 29(2), 115–118, doi:10.1130/0091-7613(2001)029<0115:
Huang, X., and J. D. Niemann (2008), How do streamflow generation NVOGMF>2.0.CO;2.
mechanisms affect watershed hypsometry?, Earth Surface Processes and Ivins, E. R., and D. Wolf (2008), Glacial isostatic adjustment: New
Landforms, 33, 751–772, doi:10.1002/esp.1573. developments from advanced observing systems and modeling, Journal
Hudson, B. D. (1994), Soil organic matter and available water capacity, of Geodynamics, 46, 69–77, doi:10.1016/j.jog.2008.06.002.
Journal of Soil and Water Conservation, 49(2), 189–194. Ijjasz-Vasquez, E. J., R. L. Bras, I. Rodriguez-Iturbe, R. Rigon, and A.
Huggett, R. J. (1975), Soil landscape systems: A model for soil genesis, Rinaldo (1993), Are river basins Optimal Channel Networks, Advances in
Geoderma, 13, 1–22. Water Resources, 16(1), 69–79.
Huntly, N., and R. Inouye (1988), Pocket gophers in ecosystems: Patterns Jackson, R. B., J. G. Canadell, J. R. Ehleringer, H. A. Mooney, O. E.
and mechanisms, Bioscience, 38(1), 786–793. Sala, and E. D. Schulze (1996), A global analysis of root distributions
for terrestrial biomes, Oecologia, 108(3), 389–411, doi:10.1007/
Hupy, J. P., and R. J. Schaetzl (2006), Introducing ‘Bombturbation’, a
BF00333714.
singular type of soil disturbance and mixing, Soil Science, 171(11),
823–836, doi:10.1097/01.ss.0000228053.08087.19. Jagercikova, M., O. Evrard, J. Balesdent, I. Lefevre, and S. S. Cornu
(2014), Modeling the migration of fallout radionuclides to quantify the
Hupy, J. P., and R. J. Schaetzl (2008), Soil development on the WWI
contemporary transfer of fine particles in Luvisol profiles under different
battlefield of Verdun, France, Geoderma, 145, 37–49, doi:10.1016/j.
land uses and farming practices, Soil Tillage Research, 140, 82–97,
geoderma.2008.01.024.
doi:10.1016/j.still.2014.02.013.
Hutchinson, M. F. (1995), Interpolating mean rainfall using thin plate
James, A. L., J. J. McDonnell, I. Tromp-van Meerveld, and N. E. Peters
smoothing splines, International Journal of Geographical Information
(2010), Gypsies in the palace: Experimentalist’s view on the use of 3-D
Systems, 9(4), 385–403, doi:10.1080/02693799508902045.
physics-based simulation of hillslope hydrological response, Hydrological
Hutchinson, M. F. (1998), Interpolation of rainfall data with thin plate Processes, 24(26), 3878–3893, doi:10.1002/hyp.7819.
smoothing splines – Part II: Analysis of Topographic Dependence, Jour-
James, T. S., and W. J. Morgan (1990), Horizontal motions due to post-
nal of Geographic Information and Decision Analysis, 2(2), 152–167.
glacial rebound, Geophysical Research Letters, 17(7), 957–960.
Ibbitt, R. P., G. R. Willgoose, and M. J. Duncan (1999), Channel network
Jarvis, N. J., K. G. Villholth, and B. Ulen (1999), Modelling particle
simulation models compared with data from the Ashley River, New
mobilization and leaching in macroporous soil, European Journal of Soil
Zealand, Water Resources Research, 35(12), 3875–3890.
Science, 50(4), 621–632.
Ikeda, H., G. Parker, and K. Sawai (1981), Bend theory of river meanders:
Jenkinson, D. S., and K. Coleman (2008), The turnover of organic carbon
Part I, Linear development, Journal of Fluid Mechanics, 112, 363–377.
in subsoils. Part 2. Modelling carbon turnover, European Journal of Soil
IPCC (2007), Climate change 2007: The physical science basis. Contri- Science, 59, 400–413, doi:10.1111/j.1365-2389.2008.01026.x.
bution of Working Group 1 to the Fourth Assessment report of the
Jenkinson, D. S., and J. H. Rayner (1977), The turnover of soil organic
Intergovernmental Panel on Climate Change, edited by S. Solomon, D.
matter in some of the Rothamsted classical experiments, Soil Science, 123
Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H.
(5), 298–305, doi:10.1097/00010694-197705000-00005.
L. Miller, p. 996, Cambridge University Press, Cambridge.
Jenny, H. (1941), Factors of soil formation – A system of quantitative
Istanbulluoglu, E., and R. L. Bras (2005), Vegetation-modulated landscape
pedology, McGraw-Hill, New York.
evolution: Effects of vegetation on landscape processes, drainage density,
and topography, Journal of Geophysical Research (Earth Surface), 110, Jenny, H. (1961), Derivation of state factor equations of soils and ecosys-
F02012, doi:10.1029/2004JF000249. tems, Proceedings of the Soil Science Society of America, 25, 385–388.
References 303
Jobbágy, E. G., and R. B. Jackson (2000), The vertical distribution of soil Kinnell, P. I. A. (2010), Event soil loss, runoff and the Universal Soil Loss
organic carbon and its relation to climate and vegetation, Ecological Appli- Equation family of models: A review, Journal of Hydrology, 385,
cations, 10(2), 423–436, doi:10.2307/2641104. 384–397, doi:10.1016/j.jhydrol.2010.01.024.
Johansen, M. P., T. E. Hakonson, and D. D. Breshears (2001), Post-fire Kirchner, J. W. (1993), Statistical inevitability of Horton’s laws and the
runoff and erosion from rainfall simulation: Contrasting forests with apparent randomness of stream channel networks, Geology, 21, 591–594.
shrublands and grasslands, Hydrological Processes, 15(15), 2953–2965. Kirkby, M. J. (1967), Measurement and theory of soil creep, Journal of
Johnson, B. C., C. S. Campbell, and H. J. Melosh (2016), The reduction of Geology, 75(4), 359–378.
friction in long runout landslides as an emergent phenomenon, Journal of Kirkby, M. J. (1971), Hillslope process-response models based on the
Geophysical Research (Earth Surface), 121, 881–889, doi:10.1002/ continuity equation, in Slopes: Form and process, pp. 15–30, Institute of
2015JF003751. British Geographers, London.
Johnson, C. G., B. P. Kokelaar, R. M. Iverson, M. Logan, R. G. LaHusen, Kirkby, M. J. (1976), Tests of the random network model and its applica-
and J. M. N. T. Gray (2012), Grain-size segregation and levee formation in tion to basin hydrology, Earth Surface Processes, 1, 197–212,
geophysical mass flows, Journal of Geophysical Research (Earth Sur- doi:10.1002/esp.3290010302.
face), 117, F01032, doi:10.1029/2011JF002185.
Kirkby, M. J. (1977), Soil development models as a component of slope
Johnson, I. R. (2008), Biophysical pasture model documentation: Model models, Earth Surface Processes, 2, 203–230, doi:10.1002/esp.3290020212.
documentation for DairyMod. EcoMod and the SGS Pasture Model, http://
Kirkby, M. J. (1985), A basis for soil profile modelling in a geomorphic
imj.com.au/wp-content/uploads/2014/08/GrazeMod.pdf.
context, European Journal of Soil Science, 36(1), 97–121, doi:10.1111/
Johnson, J. P., and K. X. Whipple (2007), Feedbacks between erosion and j.1365-2389.1985.tb00316.x.
sediment transport in experimental bedrock channels, Earth Surface Pro-
Kirkby, M. J. (1989), A model to estimate the impact of climatic-change
cesses and Landforms, 32, 1048–1062, doi:10.1002/esp.1471.
on hillslope and regolith form, Catena, 16(4–5), 321–341, doi:10.1016/
Johnston, C. A., et al. (2004), Carbon cycling in soil, Frontiers in Ecology 0341-8162(89)90018-0.
and the Environment, 2(10), 522–528, doi:10.1890/1540-9295(2004)002
Kirkby, M. J. (2000), Limits to modelling in the Earth and environmental
[0522:CCIS]2.0.CO;2.
sciences, in Geocomputation, edited by R. J. Abrahart, S. Openshaw, and
Johnstone, S. A., and G. E. Hilley (2015), Lithologic control on the form of L. M. See, pp. 374–386, Taylor and Francis, London.
soil-mantled hillslopes, Geology, 43(1), 83–86, doi:10.1130/G36052.1.
Kirkby, M. J., L. J. Bull, J. Poesen, J. Nachtergaele, and L. Vandekerc-
Julien, P. Y. (2014), Downstream hydraulic geometry of alluvial rivers, in khove (2003), Observed and modelled distributions of channel and gully
Sediment dynamics from the summit to the sea, IAHS Publ. 367, IAHS, heads – with examples from SE Spain and Belgium, Catena, 50(2–4),
New Orleans, doi:10.5194/piahs-367-3-2015. 415–434.
Julien, P. Y., and J. Wargadalam (1995), Alluvial channel geometry – Kirkby, M. J., and I. Statham (1975), Surface stone movement and scree
Theory and applications, Journal of Hydraulic Division – ASCE, 121(4), formation, Journal of Geology, 83(3), 349–362.
312–325, doi:10.1061/(ASCE)0733–9429(1995)121:4(312).
Kirkby, M. J., and G. R. Willgoose (2005), Weathering limited or trans-
Kaiser, M., D. P. Zederer, R. H. Ellerbrook, M. Sommer, and B. Ludwig port limited removal in bedrock channel systems, paper presented at IAG,
(2016), Effects of mineral characteristics on content, composition, 7–11 September, Zaragoza, Spain.
and stability of organic matter fractions separated from seven forest
Klein, F. W. (2016), Lithospheric flexure under the Hawaiian volcanic
topsoils of different pedogenesis, Geoderma, 263, 1–7, doi:10.1016/j.
load: Internal stresses and a broken plate revealed by earthquakes, Journal
geoderma.2015.08.029.
of Geophysical Research (Solid Earth), 121, 2400–2428, doi:10.1002/
Kalma, J. D., T. R. McVicar, and M. F. McCabe (2008), Estimating land 2015JB012746.
surface evaporation: A review of methods using remotely sensed surface
Klemann, V., Z. Martinec, and E. R. Ivins (2008), Glacial isostasy and
temperature data, Surveys in Geophysics, 29, 421–469, doi:10.1007/
plate motion, Journal of Geodynamics, 46, 95–103, doi:10.1016/j.
s10712-008-9037-z.
jog.2008.04.005.
Katagis, T., I. Z. Gitas, P. Toukiloglou, S. Veraverbeke, and R. Goosens Klepeis, K. A., G. L. Clarke, and T. Rushmer (2003), Magma transport
(2014), Trend analysis of medium- and coarse-resolution time series image and coupling between deformation and magmatism in the continental
data for burned area mapping in a Mediterranean ecosystem, International lithosphere, GSA Today.
Journal of Wildland Fire, 23, 668–677, doi:10.1071/WF12055.
Knisel, W. (1980), CREAMS: A field-scale model for chemicals, runoff,
Kavetski, D., G. Kuczera, and S. W. Franks (2006), Bayesian analysis of and erosion from Agricultural Management Systems, Conservation
input uncertainty in hydrological modeling: 1. Theory, Water Resources Research Report No. 26, US Department of Agriculture, Washington, DC.
Research, 42(3), W03407.
Knorr, W., and V. Lakshmi (2001), Assimilation of fAPAR and surface
Keating, B. A., et al. (2003), An overview of APSIM, a model designed temperature into a land surface and vegetation model, in Land surface
for farming systems simulation, European Journal of Agronomy, 18, hydrology, meteorology and climate: Observations and modeling, edited
267–288. by V. Lakshmi, J. Albertson, and J. Schaake, pp. 177–200, American
Keeley, J. E. (2009), Fire intensity, fire severity and burn severity: A brief Geophysical Union, Washington, DC.
review and suggested usage, International Journal of Wildland Fire, 18, Koltermann, C. E., and S. M. Gorelick (1992), Paleoclimatic signature in
116–126, doi:10.1071/WF07049. terrestrial flood deposits, Science, 256, 1775–1782.
Kelemen, P. B., and M. D. Behn (2016), Formation of lower continental Koltermann, C. E., and S. M. Gorelick (1995), Fractional packing model
crust by relamination of buoyant arc lavas and plutons, Nature Geo- for hydraulic conductivity derived from sediment mixtures, Water
science, 9(3), 197–205, doi:10.1038/ngeo2662. Resources Research, 31(12), 3283–3297.
304 References
Konno, K. (2016), A general parameterized mathematical food web model Lambeck, K., H. Rouby, A. Purcell, Y. Sun, and M. Sambridge (2014),
that predicts a stable green world in the terrestrial ecosystem, Ecological Sea level and global ice volumes from the Last Glacial Maximum to the
Monographs, 86(2), 190–214, doi:10.1890/15-1420. Holocene, Proceedings of the National Academy of Sciences U.S.A, 111
Kooi, H., and C. Beaumont (1994), Escarpment evolution on high- (43), 15296–15303, doi:10.1073/pnas.1411762111.
elevation rifted margins: Insights derived from a surface process model Lambers, H., F. S. Chapin, and T. L. Pons (2008), Plant physiological
that combines diffusion, advection and reaction, Journal of Geophysical ecology, 2nd ed., Springer, New York.
Research (Solid Earth), 99(B6), 12191–12209. Lancaster, S. T., and R. L. Bras (2002), A simple model of river meander-
Kotroni, V., and K. Lagouvardos (2008), Lightning occurrence in relation ing and its comparison to natural channels, Hydrological Processes, 16(1),
with elevation, terrain slope, and vegetation cover in the Mediterranean, 1–26.
Journal of Geophysical Research (Atmospheres), D21118, doi:10.1029/ Lane, E. W. (1955), Design of stable channels, Transactions of the
2008JD010605. American Society of Civil Engineers, 120(1), 1234–1260.
Kuczera, G. (1987), Prediction of water yield reductions following a Lane, L. J., E. D. Shirley, and V. P. Singh (1988), Modelling erosion on
bushfire in ash-mixed species eucalypt forest, Journal of Hydrology, 94 hillslopes, in Modelling geomorphological systems, edited by M. G.
(3–4), 215–236. Anderson, pp. 287–308, John Wiley & Sons, New York.
Kuczera, G. (1989), An application of Bayesian nonlinear-regression to Lane, P. N. J., G. J. Sheridan, and P. J. Noske (2006), Changes in sediment
hydrologic models, Advances in Engineering Software and Workstations, loads and discharge from small mountain catchments following wildfire in
11(3), 149–155. south eastern Australia, Journal of Hydrology, 331, 495–510, doi:10.1016/
Kuehl, S. A., et al. (2016), A source-to-sink perspective of the Waipaoa j.jhydrol.2006.05.035.
River margin, Earth-Science Reviews, 153(301–334), doi:10.1016/j. Langbein, W. B., and S. A. Schumm (1958), Yield of sediment in relation
earscirev.2015.10.001. to mean annual precipitation, Transactions of the American Geophysical
Kump, L. R., S. L. Brantley, and M. A. Arthur (2000), Chemical Union, 30(6), 1076–1084, doi:10.1029/ TR039i006p01076.
weathering, atmospheric CO2, and climate, Annual Review of Earth and Langhans, C., H. G. Smith, D. M. O. Chong, P. Nyman, P. N. J. Lane, and
Planetary Sciences , 28, 611–667, doi:10.1146/annurev.earth.28.1.611. G. J. Sheridan (2016), A model for assessing water quality risk in catch-
Kyriakidis, P. C., J. Kim, and N. L. Miller (2001), Geostatistical mapping ments prone to wildfire, Journal of Hydrology, 534, 407–426,
of precipitation from rain gauge data using atmospheric and terrain char- doi:10.1016/j.jhydrol.2015.12.048.
acteristics, Journal of Applied Meteorology, 40, 1855–1877. Lapotre, M. G. A., M. P. Lamb, and R. M. E. Williams (2016), Canyon
Lacoste, M., V. Viaud, D. Michot, and C. Walter (2015), Landscape-scale formation constraints on the discharge of catastrophic outburst floods of
modelling of erosion processes and soil carbon dynamics under land-use Earth and Mars, Journal of Geophysical Research (Planets), 121,
and climate change in agroecosystems, European Journal of Soil Science, doi:10.1002/2016JE005061.
66(4), 780–791, doi:10.1111/ejss.12267. Larcher, W. (2003), Physiological plant ecology: Ecophysiology and
Laflen, J. M., L. J. Lane, and G. R. Foster (1991), WEPP: A new gener- stress physiology of functional groups, Springer-Verlag, Berlin.
ation of erosion prediction technology, Journal of Soil and Water Conser- Larsen, I. J., D. R. Montgomery, and O. Korup (2010), Landslide erosion
vation, 46(1), 34–38. controlled by hillslope material, Nature Geoscience, 3, 247–251,
Lague, D. (2014), The stream power river incision model: Evidence, doi:10.1038/NGEO776.
theory and beyond, Earth Surface Processes and Landforms, 39, 38–61, Larsen, L. G., M. B. Eppinga, P. Passalacqua, W. M. Getz, K. A. Rose,
doi:10.1002/esp.3462. and M. Liang (2016), Appropriate complexity landscape modeling, Earth-
Lague, D., A. Crave, and P. Davy (2003), Laboratory experiments simulat- Science Reviews, 160, 111–130, doi:10.1016/j.earscirev.2016.06.016.
ing the geomorphic response to tectonic uplift, Journal of Geophysical Lasaga, A. C. (1984), Chemical kinetics of water-rock interactions, Jour-
Research (Solid Earth), 108(B1), 2008, doi:10.1029/2002JB001785. nal of Geophysical Research, 89(B6), 4009–4025, doi:10.1029/
Lague, D., N. Hovius, and P. Davy (2005), Discharge, discharge variabil- JB089iB06p04009.
ity, and the bedrock channel profile, Journal of Geophysical Research Lauer, J. W., and G. Parker (2004), Modeling channel-floodplain co-
(Earth Surface), 110, F04006, doi:10.1029/2004JF000259. evolution in sand-bed streams, in ASCE World Water and Environ-
Lal, R. (2003), Soil erosion and the global carbon budget, Environment mental Resources 2004 Congress, 27 June–1 July, p. 10, ASCE, Salt
International, 29, 437–450, doi:10.1016/S0160-4120(02)00192-7. Lake City.
Lamb, M. P., W. E. Dietrich, and L. S. Sklar (2008), A model for fluvial Lavier, L. L., and W. R. Buck (2002), Half graben versus large-offset low-
bedrock incision by impacting suspended and bed load sediment, Journal angle normal fault: Importance of keeping cool during normal faulting,
of Geophysical Research (Earth Surface), 113, F03025, doi:10.1029/ Journal of Geophysical Research (Solid Earth), 107(B6), ETG 8–1,
2007JF000915. doi:10.1029/2001JB000513.
Lamb, M. P., N. J. Finnegan, J. S. Scheingross, and L. S. Sklar (2016), Lavier, L. L., W. R. Buck, and A. N. B. Poliakov (2000),
New insights into the mechanics of fluvial bedrock erosion through flume Factors controlling normal fault offset in an ideal brittle layer, Journal of
experiments and theory, Geomorphology, 244, 33–55, doi:10.1016/j. Geophysical Research (Solid Earth), 105(B10), 23431–23442.
geomorph.2015.03.003. LeB. Hooke, R. (2005), Principles of glacier mechanics, Cambridge
Lambeck, K., and P. Johnston (1998), The viscosity of the Mantle: University Press, Cambridge.
Evidence from analyses of glacial-rebound phenomena, in The Earth’s Lebedeva, M. I., R. C. Fletcher, V. N. Balashov, and S. L. Brantley (2007),
Mantle: Composition, structure and evolution, edited by I. Jackson, A reactive diffusion model describing transformation of bedrock to saprolite,
pp. 461–502, Cambridge University Press, Cambridge. Chemical Geology, 244, 624–645, doi:10.1016/j.chemgeo.2007.07.008.
References 305
Legates, D. R., R. Mahmood, D. F. Levia, T. L. DeLiberty, S. M. Quiring, Ludwig, J. A., D. J. Tongway, and S. G. Marsden (1999), Stripes, strands
C. Houser, and F. E. Nelson (2010), Soil moisture: A central and unifying or stipples: Modelling the influence of three landscape banding patterns on
theme in physical geography, Progress in Physical Geography, 35(1), resource capture and productivity in semi-arid woodlands, Australia,
65–86, doi:10.1177/0309133310386514. Catena, 37(1–2), 257–273.
Legros, J. P., and G. Pedro (1985), The causes of particle-size distribution Ludwig, J. A., B. P. Wilcox, D. D. Breshears, D. J. Tongway, and A. C.
in soil profiles derived from crystalline rocks, France, Geoderma, 36(1), Imeson (2005), Vegetation patches and runoff-erosion as interacting eco-
15–25. hydrological processes in semiarid landscapes, Ecology, 86(2), 288–297.
Lehsten, V., A. Arneth, A. Spessa, K. Thonicke, and A. Moustakas (2016), Lugato, E., F. Bampa, P. Panagos, L. Montanarella, and A. Jones (2015),
The effect of fire on tree–grass coexistence in savannas: A simulation Potential carbon sequestration of European arable soils estimated by
study, International Journal of Wildland Fire, 25, 137–146, doi:10.1071/ modelling a comprehensive set of management practices, Global Change
WF14205. Biology, 20, 3557–3567, doi:10.1111/gcb.12551.
Leithold, E. L., N. E. Blair, and K. W. Wegmann (2016), Source-to-sink Lugato, E., P. Panagos, F. Bampa, A. Jones, and L. Montanarella (2014),
sedimentary systems and global carbon burial: A river runs through it, A new baseline of organic carbon stock in European agricultural soils
Earth-Science Reviews, 153, 30–42, doi:10.1016/j.earscirev.2015.10.011. using a modelling approach, Global Change Biology, 20, 313–316,
Leopold, L. B., M. G. Wolman, and J. P. Miller (1964), Fluvial processes doi:10.1111/gcb.12292.
in geomorphology, Freeman, London. Lugato, E., K. Paustian, P. Panagos, A. Jones, and P. Borrelli (2016),
Li, F., C. Dyt, and C. Griffiths (2004), 3D modelling of flexural isostatic Quantifying the erosion effect on current carbon budget of European
deformation, Computers & Geosciences, 30, 1105–1115, doi:10.1016/j. agricultural soils at high spatial resolution, Global Change Biology, 22,
cageo.2004.08.005. 1976–1984, doi:10.1111/gcb.13198.
Li, Z., L. Liu, J. Chen, and H. H. Teng (2016), Cellular dissolution at Lynch, A. H., J. Beringer, P. Kershaw, A. Marshall, S. Mooney, N.
hypha- and spore-mineral interfaces revealing unrecognized mechanisms Tapper, C. Turney, and S. Van Der Kaars (2007), Using the paleorecord
and scales of fungal weathering, Geology, 44(4), 319–322, doi:10.1130/ to evaluate climate and fire interactions in Australia, Annual Review of
G37561.1. Earth and Planetary Sciences, 35, 215–239, doi:10.1146/annurev.
earth.35.092006.145055.
Little, D. A., J. B. Field, and S. A. Welch (2005), Metal Dissolution from
Rhizosphere and non-Rhizosphere soils using low molecular weight McBratney, A. B., I. O. A. Odeh, T. F. A. Bishop, M. S. Dunbar, and T.
organic acids, paper presented at Regolith 2005: Ten Years of CRC M. Shatar (2000), An overview of pedometric techniques for use in soil
LEME, Cooperative Research Centre for Landscape Environments and survey, Geoderma, 97(3–4), 293–327.
Mineral Exploration (CRC LEME), Canberra. McBratney, A. B., M. L. M. Santos, and B. Minasny (2003), On digital
Llovet, J., M. Ruiz-Valera, R. Josa, and V. R. Vallejo (2009), Soil soil mapping, Geoderma, 117(1–2), 3–52, doi:10.1016/S0016-7061(03)
responses to fire in Mediterranean forest landscapes in relation to the 00223-4.
previous stage of land abandonment, International Journal of Wildland McCarthy, M. A., A. M. Gill, and R. A. Bradstock (2001), Theoretical fire-
Fire, 18, 222–232, doi:10.1071/WF07089. interval distributions, International Journal of Wildland Fire, 10(1),
Lobry de Bruyn, L. A., and A. J. Conacher (1990), The role of termites and 73–77, doi:10.1071/wf01013.
ants in soil modification: A review, Australian Journal of Soil Research, McFadden, L. D., and P. L. K. Knuepfer (1990), Soil geomorphology: The
28(1), 55–93. linkage of pedology and surficial processes, Geomorphology, 3(3/4),
Lobry de Bruyn, L. A., and A. J. Conacher (1994), The bioturbation 197–205, doi:10.1016/0169–555X(90)90003-9.
activity of ants in agricultural and naturally vegetated habitats in semi- McGuire, L. A., J. D. Pelletier, J. A. Gomez, and M. A. Nearing (2013),
arid environments, Australian Journal of Soil Research, 32, 555–570, Controls on the spacing and geometry of rill networks on hillslopes: Rain
doi:10.1071/SR9940555. splash detachment, initial hillslope roughness, and the competition
Lockart, N., G. R. Willgoose, G. Kuczera, A. S. Kiem, A. F. M. K. between fluvial and colluvial transport, Journal of Geophysical Research
Chowdhury, N. P. Manage, L. Zhang, and C. Twomey (2016), Case study (Earth Surface), 118, 241–256, doi:10.1002/jgrf.20028.
on the use of dynamically downscaled GCM data for assessing water McKenzie, B. M., and A. R. Dexter (1993), Size and orientation of
security on coastal NSW, Journal of Southern Hemisphere Earth Systems burrows made by the earthworms Aporrectodea-Rosea and a-Caliginosa,
Science, 66(2), 177–202. Geoderma, 56(1–4), 233–241.
Loewenherz-Lawrence, D. S. (1994), Theoretical constraints on the develop- McKinnon, W. B., et al. (2016), Convection in a volatile nitrogen-ice-rich
ment of surface rills: Mode shapes, amplitude limitations and implications for layer drives Pluto’s geological vigour, Nature, 534(7605), 82–85,
nonlinear evolution, in Process Models and Theoretical Geomorphology, doi:10.1038/nature18289.
edited by M. J. Kirkby, pp. 315–334, Wiley, Chichester, UK. McVicar, T. R., T. G. Van Niel, L. T. Li, M. F. Hutchinson, X. M. Mu, and
Lopéz, F., and M. García (1998), Open-channel flow through simulated Z. H. Liu (2007), Spatially distributing monthly reference evapotranspira-
vegetation: Suspended sediment transport modeling, Water Resources tion and pan evaporation considering topographic influences, Journal of
Research, 34(9), 2341–2352, doi:10.1029/98WR01922. Hydrology, 338, 196–220, doi:10.1016/j.jhydrol.2007.02.018.
Lucas, Y. (2001), The role of plants in controlling rates and products of Maestre, F. T., et al. (2015), Increasing aridity reduces soil microbial diversity
weathering: Importance of biological pumping, Annual Review of Earth and abundance in global drylands, Proceedings of the National Academy of
and Planetary Sciences , 29, 135–163. Sciences U.S.A, 112(51), 15684–15689, doi:10.1073/pnas.1516684112.
Luchi, R., J. M. Hooke, G. Zolezzi, and W. Bertoldi (2010), Width variations Maher, K. (2010), The dependence of chemical weathering rates on fluid
and mid-channel bar inception in meanders: River Bollin (UK), Geomorph- residence time, Earth and Planetary Science Letters, 294(1–2), 101–110.
ology, 119, 1–8, doi:10.1016/j.geomorph.2010.01.010. doi:10.1016/j.epsl.2010.03.010.
306 References
Maher, K., C. I. Steefel, A. F. White, and D. A. Stonestrom (2009), The Mesa, O. J. (1986), Analysis of channel networks parameterized by eleva-
role of reaction affinity and secondary minerals in regulating chemical tion, PhD thesis, University of Mississippi.
weathering rates at the Santa Cruz Soil Chronosequence, California, Geo- Metherell, A. K., L. A. Harding, C. V. Cole, and W. J. Parton (1994),
chimica et Cosmochimica Acta, 73, 2804–2831, doi:10.1016/j. CENTURY Soil organic matter environment, Technical Documentation
gca.2009.01.030. Agrosystem, Version 4.0, Great Plains System Research Unit Technical
Major, J. J., and R. M. Iverson (1999), Debris-flow deposition: Effects of Report No. 4, USDA-ARS, Fort Collins, CO.
pore-fluid pressure and friction concentrated at flow margins, Geological Michaelides, K., and G. J. Martin (2012), Sediment transport by runoff on
Society of America Bulletin, 111(10), 1424–1434. debris-mantled dryland hillslopes, Journal of Geophysical Research
Malamoud, K., A. B. McBratney, B. Minasny, and D. J. Field (2009), (Earth Surface), 117, F03014, doi:10.1029/2012JF002415.
Modelling how carbon affects soil structure, Geoderma, 149(1–2), 19–26, Michaelides, K., and M. B. Singer (2014), Impact of coarse sediment
doi:10.1016/j.geoderma.2008.10.018. supply from hillslopes to the channel in runoff-dominated, dryland fluvial
Malamud, B. D., J. D. A. Millington, and G. L. W. Perry (2005), Charac- systems, Journal of Geophysical Research (Earth Surface), 119,
terizing wildfire regimes in the Unites States, Proceedings of the National 1205–1221, doi:10.1002/2013JF002959.
Academy of Sciences U.S.A, 102(13), 4694–4699, doi:10.1073/ Migon, P., and M. F. Thomas (2002), Grus weathering mantles – Problems
pnas.0500880102. of interpretation, Catena, 49(1–2), 5–24.
Malamud, B. D., G. Morein, and D. L. Turcotte (1998), Forest fires: An Millar, R. G. (2005), Theoretical regime equations for mobile gravel-bed
example of self-organized critical behavior, Science, 281(5384), 1840–1842. rivers with stable banks, Geomorphology, 64(3–4), 207–220, doi:10.1016/
Malamud, B. D., D. L. Turcotte, F. Guzzetti, and P. Reichenback (2004), j.geomorph.2004.07.001.
Landslide inventories and their statistical properties, Earth Surface Pro- Milledge, D. G., D. Bellugi, J. A. McKean, A. L. Densmore, and W. E.
cesses and Landforms, 29, 687–711, doi:10.1002/esp.1064. Dietrich (2014), A multidimensional stability model for predicting shallow
Mann, M. E., R. S. Bradley, and M. K. Hughes (1999), Northern hemi- landslide size and shape across landscapes, Journal of Geophysical
sphere temperatures during the past millennium: Inferences, uncertainties, Research (Earth Surface), 119, 2481–2504, doi:10.1002/2014JF003135.
and limitations, Geophysical Research Letters, 26(6), 759–762, Minasny, B., P. Finke, U. Stockmann, T. Vanwalleghem, and A. B.
doi:10.1029/1999GL900070. McBratney (2015), Resolving the integral connection between pedogen-
Mao, L., J. R. Cooper, and L. E. Frostick (2011), Grain size and topo- esis and landscape evolution, Earth-Science Reviews, 150, 102–120,
graphical differences between static and mobile armour layers, Earth doi:10.1016/j.earscirev.2015.07.004.
Surface Processes, 36, 1321–1334, doi:10.1002/esp.2156. Minasny, B., and A. B. McBratney (1999), A rudimentary mechanistic
Markgraf, W., C. W. Watts, W. R. Whalley, T. Hrkac, and R. Horn (2012), model for soil production and landscape development, Geoderma, 90(1–
Influence of organic matter on rheological properties of soil, Applied Clay 2), 3–21, doi:10.1016/S0016-7061(98)00115-3.
Science, 64, 25–33, doi:10.1016/j.clay.2011.04.009. Minasny, B., and A. B. McBratney (2001), A rudimentary mechanistic
Marshall, J. S., and W. M. Palmer (1948), The distribution of raindrops model for soil formation and landscape development II. A two-
with size, Journal of Meteorology, 5, 165–166. dimensional model incorporating chemical weathering, Geoderma, 103
Martin, Y. E. (2007), Wildfire disturbance and shallow landsliding in (1–2), 161–179, doi:10.1016/S0016-7061(01)00075-1.
coastal British Columbia over millennial time scales: A numerical model- Minasny, B., and A. B. McBratney (2006), Mechanistic soil–landscape
ling study, Catena, 69(3), 206–219, doi:10.1016/j.catena.2006.05.006. modelling as an approach to developing pedogenetic classifications, Geo-
Martinez-Casasnovas, J. A. (2003), A spatial information technology derma, 133(1–2), 138–149, doi:10.1016/j.geoderma.2006.03.042.
approach for the mapping and quantification of gully erosion, Catena, 50 Minasny, B., A. B. McBratney, and S. Salvador-Blanes (2008), Quantita-
(2–4), 293–308. tive models for pedogenesis – A review, Geoderma, 144(1–2), 140–157,
Mataix-Solera, J., I. Gómez, J. Navarro-Pedreño, C. Guerrero, and R. doi:10.1016/j.geoderma.2007.12.013.
Moral (2002), Soil organic matter and aggregates affected by wildfire in Meriam, J. L., L. G. Kraige, and J. N. Bolton (2012), Engineering mech-
a Pinus halepensis forest in a Mediterranean environment, International anics: Statics, Wiley, Chichester, UK.
Journal of Wildland Fire, 11, 107–114, doi:10.1071/WF02020. Mitchell, A., and O. Hungr (2017), Theory and calibration of the Pierre
Matsuoka, N., and K. Moriwaki (1992), Frost heave and creep in the Sør 2 stochastic rock fall dynamics simulation program, Canadian Geotech-
Rondane Mountains, Antarctica, Arctic and Alpine Research, 24(4), nical Journal, 54(1), 18–30, doi:10.1139/cgj-2016-0039.
271–280. Mitchell, P. B. (1985), Some aspects of the role of bioturbation in soil
Mead, S. R., and P. W. Cleary (2015), Validation of DEM prediction for formation in south-eastern Australia, PhD thesis, Macquarie University.
granular avalanches on irregular terrain, Journal of Geophysical Research Mitchell, S. G., and E. E. Humphries (2015), Glacial cirques and the
(Earth Surface), 120, 1724–1742, doi:10.1002/2014JF003331. relationship between equilibrium line altitudes and mountain range height,
Mein, R. G., E. M. Laurenson, and T. A. McMahon (1976), Simple Geology, 43(1), 35–38, doi:10.1130/G36180.1.
nonlinear model for flood estimation, Journal of Hydraulic Division – Mitrovica, J. X., and A. M. Forte (2004), A new inference of mantle
ASCE 100(NHY11), 1507–1518. viscosity based upon joint inversion of convection and glacial isostatic
Meisina, C., and S. Scarabelli (2007), A comparative analysis of terrain adjustment data, Earth and Planetary Science Letters, 225(1–2), 177–189,
stability models for predicting shallow landslides in colluvial soils, Geo- doi:10.1016/j.epsl.2004.06.005.
morphology, 87, 207–223, doi:10.1016/j.geomorph.2006.03.039. Moglen, G. E., and R. L. Bras (1995), The importance of spatially
Melini, D., P. Gegout, G. Spada, and M. A. King (2015), REAR: heterogeneous erosivity and the Cumulative Area Distribution within a
A Regional Elastic Rebound calculator, GitHub. basin evolution model, Geomorphology, 12(3), 173–185.
References 307
Moglen, G. E., E. A. B. Eltahir, and R. L. Bras (1998), On the sensitivity National Academy of Science (NAS) (2010), Landscapes on the edge:
of drainage density to climate change, Water Resources Research, 34(4), New horizons for research of earth’s surface, National Academies Press,
855–862. Washington, DC.
Montelli, R., G. Noilet, F. A. Dahlen, G. Masters, E. R. Engdahl, and S. H. Navarre-Stichler, A. K., D. R. Cole, G. Rother, L. Jin, H. L. Buss, and S.
Hung (2004), Finite-frequency tomography reveals a variety of plumes in L. Brantley (2013), Porosity and surface area evolution during weathering
the mantle, Science, 303(5656), 338–343, doi:10.1126/science.1092485. of two igneous rocks, Geochimica et Cosmochimica Acta, 109, 400–413,
Montgomery, D. R., and W. E. Dietrich (1988), Where do channels doi:10.1016/j.gca.2013.02.012.
begin?, Nature, 336, 232–234. Nayyar, H., and D. Gupta (2006), Differential sensitivity of C3 and C4
Montgomery, D. R., and W. E. Dietrich (1989), Source areas, drainage plants to water deficit stress: Association with oxidative stress and antioxi-
density and channel initiation, Water Resources Research, 25(8), dants, Environmental and Experimental Botany, 58(1–3), 106–113,
1907–1918, doi:10.1029/WR025i008p01907. doi:10.1016/j.envexpbot.2005.06.021.
Montgomery, D. R., and W. E. Dietrich (1994), A physically based model Nearing, M. A., G. R. Foster, L. J. Lane, and S. C. Finkner (1989),
for the topographic control on shallow landsliding, Water Resources A process-based soil erosion model for USDA-Water Erosion Prediction
Research, 30(4), 1153–1171. Project Technology, American Society of Agricultural Engineers, 32(5),
1587–1593.
Montgomery, D. R., and E. Foufoula-Georgiou (1993), Channel network
source representation using digital elevation models, Water Resources Nelson, P. A., J. G. Vendetti, W. E. Dietrich, J. W. Kirchner, H. Ikeda, F.
Research, 29(12), 3925–3935, doi:10.1029/93WR02463. Iseya, and L. S. Sklar (2009), Response of bed surface patchiness to
reductions in sediment supply, Journal of Geophysical Research (Earth
Montgomery, D. R., K. M. Schmidt, W. E. Dietrich, and J. McKean
Surface), 114, F02005, doi:10.1029/2008JF001144.
(2009), Instrumental record of debris flow initiation during natural rainfall:
Nepf, H. (2012), Hydrodynamics of vegetated channels, Journal of
Implications for modeling slope stability, Journal of Geophysical
Hydraulic Research, 50(3), 262–279, doi:10.1080/00221686.2012.696559.
Research (Earth Surface), 114, F01031, doi:10.1029/2008JF001078.
Nesbitt, S. W., and A. M. Anders (2009), Very high resolution precipita-
Moody, J. A., R. A. Shakesby, P. R. Robichaud, S. H. Cannon, and D. A.
tion climatologies from the Tropical Rainfall Measuring Mission precipi-
Martin (2013), Current research issues related to post-wildfire runoff and
tation radar, Geophysical Research Letters, 36, L15815, doi:10.1029/
erosion processes, Earth-Science Reviews, 122, 10–37, doi:10.1016/j.
2009GL038026.
earscirev.2013.03.004.
Nicholas, A. P., and T. A. Quine (2010), Quantitative assessment of
Moore, J. G. (1987), Subsidence of the Hawaiian Ridge, in Volcanism in
landform equifinality and palaeoenvironmental reconstruction using geo-
Hawaii, edited by R. W. Decker, T. L. Wright, and P. H. Stauffer, USGS
morphic models, Geomorphology, 121, 167–183, doi:10.1016/j.
Professional Paper 1350.
geomorph.2010.04.004.
Moore, J. M., A. D. Howard, and A. M. Morgan (2014), The landscape of
Niklasson, M., and A. Granström (2000), Numbers and sizes of fires:
Titan as witness to its climate evolution, Journal of Geophysical Research
Long-term spatially explicit fire history in a Swedish boreal landscape,
(Planets), 119(9), 2060–2077, doi:10.1002/2014JE004608.
Ecology, 81(6), 1484–1499.
Moreno de Las Heras, M., P. M. Saco, and G. R. Willgoose (2012), Nimmo, J. R., and K. S. Perkins (2002), Aggregate stability and size
A comparison of SRTM V4 and ASTER GDEM for hydrological applica- distribution, in Methods of soil analysis, Part 4 – Physical methods, edited
tions in low relief terrain, Photogrammetric Engineering and Remote by J. H. Dane and G. C. Topp, pp. 317–328, Soil Science Society of
Sensing, 78(7), 757–766. America, Madison, WI.
Moreno de Las Heras, M., P. M. Saco, G. R. Willgoose, and D. J. Tong- Nolan, R. H., P. N. J. Lane, R. G. Benyon, R. A. Bradstock, and P. J.
way (2012), Variations in hydrological connectivity of Australian semiarid Mitchell (2014a), Changes in evapotranspiration following wildfire in
landscapes indicate abrupt changes in ecosystem functionality, Journal of resprouting eucalypt forests, Ecohydrology, 7, 1363–1377, doi:10.1002/
Geophysical Research (Biogeosciences), 117, G03009, doi:10.1029/ eco.1463.
2011JG001839.
Nolan, R. H., P. N. J. Lane, R. G. Benyon, R. A. Bradstock, and P. J.
Morin, R. H. (2005), Negative correlation between porosity and hydraulic Mitchell (2015), Trends in evapotranspiration and streamflow following
conductivity in sand-and-gravel aquifers at Cape Cod, Massachusetts, wildfire in resprouting eucalypt forests, Journal of Hydrology, 524,
USA, Journal of Hydrology, 316, 43–52, doi:10.1016/j. 614–624, doi:10.1016/j.jhydrol.2015.02.045.
jhydrol.2005.04.013.
Nolan, R. H., P. J. Mitchell, R. A. Bradstock, and P. N. J. Lane (2014b),
Mudd, S. M., and D. J. Furbish (2004), Influence of chemical denudation Structural adjustments in resprouting trees drive differences in post-fire
on hillslope morphology, Journal of Geophysical Research (Earth Sur- transpiration, Tree Physiology, 34(2), 123–136, doi:10.1093/treephys/
face), 109, F02001, doi:10.1029/2003JF000087. tpt125.
Munson, B. R., D. F. Young, and T. H. Okiishi (1998), Fundamentals of North Greenland Ice Core Project (NGRIP) (2004a), High-resolution
fluid mechanics, Wiley, Chichester, UK. record of Northern Hemisphere climate extending into the last interglacial
Murphy, B., J. Russell-Smith, and L. Prior (2010), Frequent fires reduce period, Nature, 431(7005), 147–151.
tree growth in northern Australian savannas: Implications for tree demog- North Greenland Ice Core Project (NGRIP) (2004b), North Greenland Ice
raphy and carbon sequestration, Global Change Biology, 16(1), 331–343, Core Project Oxygen Isotope Data, NOAA/NGDC Paleoclimatology Pro-
doi:10.1111/j.1365-2486.2009.01933.x. gram, Boulder CO, USA, IGBP PAGES/World Data Center for Paleocli-
Murray, A. B., and C. Paola (2003), Modelling the effect of vegetation on matology Data Contribution Series no. 2004-059.
channel pattern in bedload rivers, Earth Surface Processes and Landforms, Noy-Meir, I. (1973), Desert ecosystems: Environment and producers,
28(2), 131–143. Annual Review of Ecology and Systematics, 4, 25–51.
308 References
NRCCA (2016), Northeast Region Certified Crop Advisor (NRCCA) Parana Manage, N. (2017), Testing the hydrologic validity of downscaled
Study Resources, Cornell University, Ithaca, NY, http://nrcca.cals.cornell rainfall data for water security assessment, PhD thesis, University of
.edu/soil/CA2/CA0212.1–3.php. Newcastle, Callaghan, Australia.
Nyman, P., G. J. Sheridan, and P. N. J. Lane (2013), Hydro-geomorphic Pardini, G. (2003), Fractal scaling of surface roughness in artificially
response models for burned areas and their applications in land manage- weathered smectite-rich soil regoliths, Geoderma, 117(1–2), 157–167.
ment, Progress in Physical Geography, 37(6), 787–812, doi:10.1177/ Parizek, J. R., and G. H. Girty (2014), Assessing volumetric strains and
0309133313508802. mass balance relationships resulting from biotite-controlled weathering:
Nyman, P., G. J. Sheridan, J. A. Moody, H. G. Smith, P. J. Noske, and P. Implications for the isovolumetric weathering of the Boulder Creek Grano-
N. J. Lane (2013), Sediment availability on burned hillslopes, Journal of diorite, Boulder County, Colorado, USA, Catena, 120, 29–45, doi:10.1016/
Geophysical Research (Earth Surface), 118, 2451–2467, doi:10.1002/ j.catena.2014.03.019.
jgrf.20152. Parker, G. (1990), Surface-based bedload transport relation for gravel
Nyman, P., G. J. Sheridan, H. G. Smith, and P. N. J. Lane (2011), rivers, Journal of Hydraulic Research, 28(4), 417–436.
Evidence of debris flow occurrence after wildfire in upland catchments Parker, G., and P. C. Klingeman (1982), On why gravel beds streams are
of south-east Australia, Geomorphology, 125, 383–401, doi:10.1016/j. paved, Water Resources Research, 18(5), 1409–1423, doi:10.1029/
geomorph.2010.10.016. WR018i005p01409.
O’Brien, B. J., and J. D. Stout (1978), Movement and turnover of soil Parker, G., K. Sawai, and S. Ikeda (1982), Bend theory of river meanders:
organic matter as indicated by carbon isotope measurements, Soil Biology Part II, Nonlinear deformation of finite amplitude bends, Journal of Fluid
& Biochemistry, 10(4), 309–317, doi:10.1016/0038-0717(78)90028-7. Mechanics, 115, 303–314.
O’Callaghan, J. F., and D. M. Mark (1984), The extraction of drainage Parker, G., Y. Shimizu, G. V. Wilkerson, E. C. Eke, E. C. Abad, J. D.
networks from digital elevation data, Computer Vision, Graphics and Lauer, C. Paola, W. E. Dietrich, and V. R. Voller (2011), A new frame-
Image Processing, 28(3), 323–344, doi: 10.1016/S0734-189X(84)80011-0. work for modeling the migration of meandering rivers, Earth Surface
O’Donnell, A. J., M. M. Boer, W. L. McCaw, and P. F. Grierson (2011), Processes and Landforms, 36, 70–86, doi:10.1002/esp.2113.
Climatic anomalies drive wildfire occurrence and extent in semi-arid Parker, G., P. R. Wilcock, C. Paola, W. E. Dietrich, and J. Pitlick (2007),
shrublands and woodlands of southwest Australia, Ecosphere, 2(11), Physical basis for quasi-universal relations describing bankfull hydraulic
127, doi:10.1890/ES11-00189.1. geometry of single-thread gravel bed rivers, Journal of Geophysical
O’Donnell, A. J., M. M. Boer, W. L. McCaw, and P. F. Grierson (2014), Research (Earth Surface), 112, F04005, doi:10.1029/2006JF000549.
Scale-dependent thresholds in the dominant controls of wildfire size in Parsons, A. J., A. D. Abrahams, and S. H. Luk (1990), Hydraulics of
semi-arid southwest Australia, Ecosphere, 5(7), 93, doi:10.1890/ES14- interrill overland flow on a semi-arid hillslope, southern Arizona, Journal
00145.1. of Hydrology, 117, 255–273.
Ogawa, M. (2008), Mantle convection: A review, Fluid Dynamics Parsons, A. J., A. D. Abrahams, and J. R. Simanton (1992), Microtopo-
Research, 40, 379–398, doi:10.1016/j.fluiddyn.2007.09.001. graphy and soil-surface materials on semi-arid piedmont hillslopes, south-
Olive, J.-A., M. D. Behn, and L. C. Malatesta (2014), Modes of exten- ern Arizona, Journal of Arid Environments, 22, 107–115.
sional faulting controlled by surface processes, Geophysical Research Parsons, A. J., J. Wainwright, A. D. Abrahams, and J. R. Simanton (1997),
Letters, 41, 6725–6733, doi:10.1002/2014GL061507. Distributed dynamic modelling of interrill overland flow, Hydrological
Ollier, C. D., and C. F. Pain (1996), Regolith, soils and landforms, Wiley, Processes, 11(14), 1833–1859.
Chichester, UK. Parton, W. J., J. W. B. Stewart, and C. V. Cole (1988), Dynamics of C, N,
O’Reilly, S. Y., W. L. Griffin, Y. H. Poudjom Djomani, and P. Morgan P and S in grassland soils: A model, Biogeochemistry, 4(1), 109–131,
(2001), Are lithospheres forever? Tracking changes in subcontinental doi:10.1007/BF02180320.
lithospheric mantle through time, GSA Today, 11(4), 4–10. Paton, T. R., G. S. Humphreys, and P. B. Mitchell (1995), Soils: A new
Ouimet, R. (2008), Using compositional change within soil profiles for global view, CRC Press, Boca Raton, FL.
modelling base cation transport and chemical weathering, Geoderma, 145 Paulson, A., S. Zhong, and J. Wahr (2007), Inference of mantle viscosity
(3–4), 410–418, doi:10.1016/j.geoderma.2008.01.007. from GRACE and relative sea level data, Geophysical Journal Inter-
Oyama, T., and M. Chigara (1999), Weathering rate of mudstone and tuff national, 171, 497–508, doi:10.1111/j.1365-246X.2007.03556.x.
on old unlined tunnel walls, Engineering Geology, 55(1), 15–27, Peakall, J., and E. J. Sumner (2015), Submarine channel flow processes
doi:10.1016/S0013-7952(99)00103-9. and deposits: A process-product perspective, Geomorphology, 244,
Pack, R. T., D. G. Tarboton, and C. N. Goodwin (1998), The SINMAP 95–120, doi:10.1016/j.geomorph.2015.03.005.
approach to terrain stability mapping, in Eighth International Congress of
Peckham, S. D., E. W. H. Hutton, and B. Norris (2013), A component-based
the International Association for Engineering Geology and the Environ-
approach to integrated modeling in the geosciences: The design of CSDMS,
ment, edited by D. Moore and O. Hungr, pp. 1157–1165, Vancouver,
Computers & Geosciences, 53, 3–12, doi:10.1016/j.cageo.2012.04.002.
21–25 September.
Peltier, W. R. (1998), Postglacial variations in the level of the sea:
Panagos, P., P. Borrelli, K. Meusburger, C. Alewell, E. Lugato, and L.
Implications for climate dynamics and soil-earth geophysics, Reviews of
Montanarella (2015), Estimating the soil erosion cover-management factor
Geophysics, 36(4), 441–500.
at the European scale, Land Use Policy, 48, 38–50, doi:10.1016/j.
landusepol.2015.05.021. Peltier, W. R. (1999), Global sea level rise and glacial isostatic adjustment,
Global Planetary Change, 20, 93–123.
Pandey, S., and H. Rajaram (2016), Modeling the influence of preferential
flow on the spatial variability and time-dependence of mineral weathering Peltier, W. R. (2004), Global glacial isostasy and the surface of the
rates, Water Resources Research, 52, doi:10.1002/2016WR019026. ice-age Earth: The ICE-5G (VM2) model and GRACE, Annual Reviews
References 309
of Earth and Planetary Sciences, 32, 111–149, doi:10.1146/annurev. Placzkowska, E., M. Gornik, E. Mocior, B. Peek, P. Potoniec, B. Rzonca,
earth.32.082503.144359. and J. Siwek (2015), Spatial distribution of channel heads in the
Peng, J., L. Dan, and M. Huang (2014), Sensitivity of global and regional Polish Flysch Carpathians, Catena, 127, 240–249, doi:10.1016/j.
terrestrial carbon storage to the direct CO2 effect and climate change based catena.2014.12.033.
on the CMIP5 model intercomparison, PLoS ONE, 9(4), e95282, Plante, A. F., and W. J. Parton (2007), The dynamics of soil organic matter
doi:10.1371/journal.pone.0095282. and nutrient cycling, in Soil Microbiology, Ecology, and Biochemistry,
Penman, T. D., R. A. Bradstock, and O. Price (2013), Modelling the edited by E. A. Paul, pp. 433–470, Academic Press, Amsterdam.
determinants of ignition in the Sydney Basin, Australia: Implications for Poesen, J., J. Nachtergaele, G. Verstraeten, and C. Valentin (2003), Gully
future management, International Journal of Wildland Fire, 22(4), erosion and environmental change: Importance and research needs,
469–478, doi:10.1071/wf12027. Catena, 50(2–4), 91–133.
Perera, H. J., and G. R. Willgoose (1998), A physical explanation of the Pollen-Bankhead, N., and A. Simon (2010), Hydrologic and hydraulic
cumulative area diagram, Water Resources Research, 34(5), 1335–1345. effects of riparian root networks on streambank stability: Is mechanical
Perez, L., and S. Dragicevic (2012), Landscape-level simulation of forest root-reinforcement the whole story?, Geomorphology, 116, 353–362,
insect disturbance: Coupling swarm intelligent agents with GIS-based doi:10.1016/j.geomorph.2009.11.013.
cellular automata model, Ecological Modelling, 231, 53–64, Polyakov, V. O., and R. Lal (2004), Modeling soil organic matter
doi:10.1016/j.ecolmodel.2012.01.020. dynamics as affected by soil water erosion, Environment International,
Perron, J. T. (2011), Numerical methods for nonlinear hillslope transport 30, 547–556, doi:10.1016/j.envint.2003.10.011.
laws, Journal of Geophysical Research (Earth Surface), 116, F02021, Polychronaki, A., I. Z. Gitas, and A. Minchella (2014), Monitoring post-
doi:10.1029/2010JF001801. fire vegetation recovery in the Mediterranean using SPOT and ERS
Perron, J. T., and L. H. Royden (2013), An integral approach to bedrock imagery, International Journal of Wildland Fire, 23, 631–642,
river profile analysis, Earth Surface Processes and Landforms, 38, doi:10.1071/WF12058.
570–576, doi:10.1002/esp.3302. Popper, K. R. (1959), The Logic of Scientific Discovery, Hutchinson,
Peterman, W., and D. Bachelet (2012), Climate change and forest dynam- London.
ics: A soils perspective, in Soils and food security, edited by R. E. Hester Portenga, E. W., and P. R. Bierman (2011), Understanding Earth’s eroding
and R. M. Harrison, pp. 158–182, Royal Society of Chemistry, London. surface with 10Be, GSA Today, 21(8), 4–10, doi:10.1130/G111A.1.
Peterman, W., D. Bachelet, K. Ferschweiler, and T. Sheehan (2014), Soil Prentice, I. C., W. Cramer, S. P. Harrison, R. Leemans, R. A. Monserud,
depth affects simulated carbon and water in the MC2 dynamic global and A. M. Solomon (1992), A global biome model based on plant physi-
vegetation model, Ecological Modelling, 294, 84–93, doi:10.1016/j. ology and dominance, soil properties and climate, Journal of Biogeog-
ecolmodel.2014.09.025. raphy, 19(2), 117–134.
Petersen, J. F., D. Sack, and R. E. Gabler (2012), Physical geography, Price, F., and R. A. Bradstock (2011), Quantifying the influence of fuel
10th ed., Brooks/Cole, Belmont, CA. age and weather on the annual extent of unplanned fires in the Sydney
Petit, J. R., et al. (1999a), Climate and atmospheric history of the past region of Australia, International Journal of Wildland Fire, 20(1),
420,000 years from the Vostok ice core, Antarctica, Nature, 399(6735), 142–151, doi:10.1071/wf10016.
429–436, doi:10.1038/20859. Price, O. E., R. A. Bradstock, J. E. Keeley, and A. D. Syphard (2012), The
Petit, J. R. (1999b), Vostok Ice Core Data for 420,000 Years, NOAA/ impact of antecedent fire area on burned area in southern California coastal
NGDC Paleoclimatology Program, Boulder CO, IGBP PAGES/World ecosystems, Journal of Environmental Management, 113, 301–307,
Data Center for Paleoclimatology Data Contribution Series #2001-076. doi:10.1016/j.jenvman.2012.08.042.
Pfister, L., and J. W. Kirchner (2017), Debates – Hypothesis testing in Proffitt, G. T., and A. J. Sutherland (1983), Transport of non-uniform
hydrology: Theory and practice, Water Resources Research, 53, sediments, Journal of Hydraulic Research, 21, 33–43.
doi:10.1002/2016WR020116. Prosser, I. P., and B. Abernethy (1996), Predicting the topographic limits
Philip, J. R. (1991), Hillslope infiltration, Water Resources Research, to a gully network using a digital terrain model and process thresholds,
27(1), 109–117. Water Resources Research, 32(7), 2289–2298.
Phillips, J. D. (1993), Stability implications of the state factor model of Prosser, I. P., W. E. Dietrich, and J. Stevenson (1995), Flow resistance and
soils as a nonlinear dynamical system, Geoderma, 58(1–2), 1–15. sediment transport by concentrated overland flow in a grassland, Geo-
morphology, 13, 71–86, doi:10.1016/0169-555X(95)00020-6.
Picard, K., B. Brooke, and M. F. Coffin (2017), Geological insights from
Malaysia Airlines flight MH370 search, EOS, 98, doi:10.1029/ Pumpanen, J., H. Ilvesniemi, and P. Hari (2003), A process-based model
2017EO069015. for predicting soil carbon dioxide efflux and concentration, Soil Science
Society of America, 67(2), 402–413, doi:10.2136/sssaj2003.4020.
Pierson, F. B., D. H. Carlson, and K. E. Spaeth (2002), Impacts of wildfire
on soil hydrological properties of steep sagebrush-steppe rangeland, Quillet, A., C. Peng, and M. Garneau (2010), Toward dynamic global
International Journal of Wildland Fire, 11, 145–151, doi:10.1071/ vegetation models for simulating vegetation–climate interactions and feed-
WF02037. backs: Recent developments, limitations, and future challenges, Environ-
mental Reviews, 18, 333–353, doi:10.1139/A10-016.
Pilgrim, D. H. (1976), Travel times and nonlinearity of flood runoff from
tracer measurements on a small watershed, Water Resources Research, Radoane, M., I. Ichim, and N. Radoane (1995), Gully distribution and
12(3), 487–496. development in Moldova, Romania, Catena, 24(2), 127–146, doi:10.1016/
0341-8162(95)00023-L.
Pilgrim, D. H. (1977), Isochrones of travel time and distribution of flood
storage from a tracer study on a small watershed, Water Resources Rajaram, H., and M. Arshadi (2016), A similarity solution for reaction
Research, 13(3), 587–595. front propagation in a fracture–matrix system, Philosophical Transactions
310 References
of the Royal Society of London A, 374, 20150424, doi:10.1098/ 1990 results, Internal Report 49, Supervising Scientist for the Alligator
rsta.2015.0424. Rivers Region.
Ramankutty, P., M. Ryan, R. Lawes, J. Speijers, and M. Renton (2013), Rinaldo, A., G. K. Vogel, R. Rigon, and I. Rodriguez-Iturbe (1995), Can
Statistical emulators of a plant growth simulation model, Climate one gauge the shape of a basin?, Water Resources Research, 31(4),
Research, 55, 253–265, doi:10.3354/cr01138. 1119–1127.
Rasmussen, E., R. A. Dahlgren, and R. J. Southard (2010), Basalt Ritchie, J. C., G. W. McCarty, E. R. Venteris, and T. C. Kaspar (2007),
weathering and pedogenesis across an environmental gradient in the Soil and soil organic carbon redistribution on the landscape, Geomorph-
southern Cascade Range California, USA, Geoderma, 154, 473–485, ology, 89, 163–171, doi:10.1016/j.geomorph.2006.07.021.
doi:10.1016/j.geoderma.2009.05.019. Robertson, A., E. Githumbi, and D. Colombaroli (2016), Paleofires and
Rawls, W. J., Y. A. Pachepsky, J. C. Ritchie, T. M. Sobecki, and H. models illuminate future fire scenarios, EOS, 97, 11, doi:10.1029/
Bloodworth (2003), Effect of soil organic carbon on soil water retention, 2016EO049933.
Geoderma, 116(1–2), 61–76. Rodrigues, M., P. Ibarra, M. Echeverría, F. Pérez-Cabello, and J. de la
Recking, A. (2010), A comparison between flume and field bed load Riva (2014), A method for regional-scale assessment of vegetation recov-
transport data and consequences for surface-based bed load transport ery time after high-severity wildfires: Case study of Spain, Progress in
prediction, Water Resources Research, 46, W03518, doi:10.1029/ Physical Geography, 38(5), 556–575, doi:10.1177/0309133314542956.
2009WR008007. Rodriguez-Iturbe, I., E. J. Ijjasz-Vasquez, R. L. Bras, and D. G. Tarboton
Refice, A., E. Giachetta, and D. Capolongo (2012), SIGNUM: A Matlab, (1992a), Power law distributions of discharge mass and energy in river
TIN-based landscape evolution model, Computers & Geosciences, 45, basins, Water Resources Research, 28(4), 1089–1093.
293–303, doi:10.1016/j.cageo.2011.11.013. Rodriguez-Iturbe, I., and J. M. Mejia (1974), On the transformation of
Reneau, S. L., D. Katzman, G. A. Kuyumjian, A. Lavine, and D. V. point rainfall to areal rainfall, Water Resources Research, 10(4), 729–735,
Malmon (2007), Sediment delivery after a wildfire, Geology, 35(2), doi:10.1029/WR010i004p00729.
151–154, doi:10.1130/G23288A.1. Rodriguez-Iturbe, I., and A. Porporato (2005), Ecohydrology of water-
Rengers, F. K., and G. E. Tucker (2014), Analysis and modeling of controlled ecosystems: Soil moisture and plant dynamics, Cambridge
gully headcut dynamics, North American high plains, Journal of Geo- University Press, Cambridge.
physical Research (Earth Surface), 119, 983–1003, doi:10.1002/ Rodriguez-Iturbe, I., and A. Rinaldo (2001), Fractal river basins: Chance
2013JF002962. and self-organization, Cambridge University Press, Cambridge.
Rethemeyer, J., C. Kramer, G. Gleixner, B. John, T. Yamashita, H. Flessa, N. Rodriguez-Iturbe, I., A. Rinaldo, R. Rigon, R. L. Bras, E. Ijjasz-Vasquez,
Andersen, M. J. Nadeau, and P. M. Grootes (2005), Transformation of and A. Marani (1992b), Fractal structures as least energy patterns – The
organic matter in agricultural soils: Radiocarbon concentration versus soil case of river networks, Geophysical Research Letters, 19(9), 889–892.
depth, Geoderma, 128(1–2), 94–105.
Rodriguez-Iturbe, I., and J. B. Valdes (1979), The geomorphologic struc-
Richter, D. D., and D. Markewitz (1995), How deep is soil?, Bioscience, ture of hydrologic response, Water Resources Research, 15(6),
45(9), 600–609, doi:10.2307/1312764. 1409–1420.
Richter, D. D., D. Markewitz, S. Trumbore, and C. G. Wells (1999), Rapid Roe, G. H. (2005), Orographic precipitation, Annual Review of Earth and
accumulation and turnover of soil carbon in a re-establishing forest, Planetary Sciences, 33, 645–671, doi:10.1146/annurev.earth.33.092203
Nature, 400(6739), 56–58, doi:10.1038/21867. .122541.
Riebe, C. S., J. W. Kirchner, and R. C. Finkel (2003), Long-term rates of Roe, G. H., and M. T. Brandon (2011), Critical form
chemical weathering and physical erosion from cosmogenic nuclides and and feedbacks in mountain-belt dynamics: Role of rheology as a
geochemical mass balance, Geochimica et Cosmochimica Acta, 67(22), tectonic governor, Journal of Geophysical Research (Solid Earth),
4411–4427, doi:10.1016/S0016-7037(03)00382-X. 116, B02101, doi:10.1029/2009JB006571.
Riebe, C. S., J. W. Kirchner, and R. C. Finkel (2004), Erosional and Roe, G. H., D. R. Montgomery, and B. Hallet (2002), Effects of oro-
climatic effects on long-term chemical weathering rates in granitic land- graphic precipitation variations on the concavity of steady-state river
scapes spanning diverse climate regimes, Earth and Planetary Science profiles, Geology, 30(2), 143–146.
Letters, 224(3–4), 547–562, doi:10.1016/j.epsl.2004.05.019.
Roe, G. H., D. R. Montgomery, and B. Hallet (2003), Orographic precipi-
Rieke-Zapp, D., J. Poesen, and M. A. Nearing (2007), Effects of rock tation and the relief of mountain ranges, Journal of Geophysical Research
fragments incorporated in the soil matrix on concentrated flow hydraulics (Solid Earth), 108(B6), ETG15, doi:10.1029/2001JB001521.
and erosion, Earth Surface Processes and Landforms, 32, 1063–1076,
Roe, G. H., K. X. Whipple, and J. K. Fletcher (2008), Feedbacks among
doi:10.1002/esp.1469.
climate, erosion, and tectonics in a critical wedge orogen, American
Rietkerk, M., M. C. Boerlijst, F. van Langevelde, R. HilleRisLambers, J. Journal of Science, 308, 815–842, doi:10.2475/07.2008.01.
van de Koppel, L. Kumar, H. H. T. Prins, and A. M. de Roos (2002), Self-
Roering, J. J. (2004), Soil creep and convex-upward velocity profiles:
organization of vegetation in arid ecosystems, American Naturalist, 160
Theoretical and experimental investigation of disturbance-driven sediment
(4), 524–530, doi:10.1086/342078.
transport on hillslopes, Earth Surface Processes and Landforms, 29,
Rigon, R., A. Rinaldo, I. Rodriguez-Iturbe, R. L. Bras, and E. Ijjasz- 1597–1612, doi:10.1002/esp.1112.
Vasquez (1993), Optimal Channel Networks – A framework for the study
Roering, J. J. (2008), How well can hillslope evolution models ‘explain’
of river basin morphology, Water Resources Research, 29(6), 1635–1646.
topography? Simulating soil transport and production with high-resolution
Riley, S. J., B. Gardiner, F. Hancock, and C. Uren (1991), Concentrated topographic data, Geological Society of America Bulletin, 120(9/10),
flow simulation experiments, waste rock dumps, Ranger Uranium Mine, 1248–1262, doi:10.1130/B26283.1.
References 311
Roering, J. J., J. W. Kirchner, and W. E. Dietrich (1999), Evidence Rumpel, C., J. Leifield, C. Santin, and S. Doerr (2015), Movement of
for nonlinear, diffusive sediment transport on hillslopes and implica- biochar in the environment, in Biochar for environmental management:
tions for landscape morphology, Water Resources Research, 35(3), Science, technology and implementation, edited by J. Lehmann and S.
853–870. Joseph, pp. 283–300, Routledge, Oxford.
Roering, J. J., J. W. Kirchner, L. S. Sklar, and W. E. Dietrich (2001), Russell-Smith, J., P. G. Ryan, and R. Durieu (1997), A LANDSAT-MSS
Hillslope evolution by nonlinear creep and landsliding: An experimental derived fire history of Kakadu National Park, monsoonal northern Austra-
study, Geology, 29(2), 143–146, doi:10.1130/0091-7613(2001)029<0143: lia, 1980–94: seasonal extent, frequency and patchiness, Journal of
HEBNCA>2.0.CO;2. Applied Ecology, 34, 748–766.
Roering, J. J., J. T. Perron, and J. W. Kirchner (2007), Functional relation- Ryan, S. A., K. A. Dwire, and M. K. Dixon (2011), Impacts of
ships between denudation and hillslope form and relief, Earth and Planet- wildfire on runoff and sediment loads at Little Granite Creek,
ary Science Letters, 264, 245–258, doi:10.1016/j.epsl.2007.09.035. western Wyoming, Geomorphology, 129, 113–130, doi:10.1016/j.
Roering, J. J., K. M. Schmidt, J. D. Stock, W. E. Dietrich, and D. R. geomorph.2011.01.017.
Montgomery (2003), Shallow landsliding, root reinforcement, and the Sachau, T., D. Koehn, and C. Passchier (2013), Mountain building under
spatial distribution of trees in the Oregon Coast Range, Canadian Geo- extension, American Journal of Science, 313, 326–344, doi:10.2475/
technical Journal, 40, 237–253, doi:10.1139/T02-113. 04.2013.03.
Rogers, N. (2007), An introduction to the structure and composition of the Saco, P. M., and M. Moreno de Las Heras (2013), Ecogeomorphic coevolu-
Earth, in An Introduction to Our Dynamic Planet, edited by N. Rogers, tion of semiarid hillslopes: Emergence of banded and striped vegetation
pp. 1–46, Cambridge University Press, Cambridge. patterns through interaction of biotic and abiotic processes, Water Resources
Román-Sánchez, A., T. Vanwalleghem, A. Peña, A. Laguna, and J. V. Research, 49(1), 115–126, doi:10.1029/2012WR012001.
Giráldez (2017), Controls on soil carbon storage from topography and Saco, P. M., G. R. Willgoose, and G. R. Hancock (2006), Spatial organiza-
vegetation in a rocky, semi-arid landscapes, Geoderma, doi:10.1016/j. tion of soil depths using a landform evolution model, Journal of Geophys-
geoderma.2016.10.013. ical Research (Earth Surface), 111, F02016, doi:02010.01029/
Roseberry, J. C., D. J. Furbish, and M. Schmeeckle (2012), A probabilistic 02005JF000351.
description of the bed load sediment flux: 2. Particle activity and motions, Saco, P. M., G. R. Willgoose, and G. R. Hancock (2007), Eco-geomorph-
Journal of Geophysical Research (Earth Surface), 117, F03032, ology and banded vegetation patterns in arid and semi-arid regions,
doi:10.1029/2012JF002353. Hydrology and Earth System Sciences, 11(6), 1717–1730, doi:10.5194/
Rothermel, R. C. (1972), A mathematical model for predicting fire spread in hess-11-1717-2007.
wildland fire, Rep. INT-115, Intermountain Forest and Range Experiment Saft, M., M. C. Peel, A. W. Western, J. M. Perraud, and L. Zhang (2016a),
Station, USDA Forest Service, Ogden, UT. Bias in streamflow projections due to climate-induced shifts in catchment
Rothermel, R. C. (1991), Predicting behavior and size of crown fires in the response, Geophysical Research Letters, 43(4), 1574–1581, doi:10.1002/
northern Rocky Mountains, Rep. INT-438, Intermountain Research Sta- 2015GL067326.
tion, USDA Forest Service, Ogden, UT. Saft, M., M. C. Peel, A. W. Western, and L. Zhang (2016b), Predicting
Roy, S. G., P. O. Koons, P. Upton, and G. E. Tucker (2015), The influence shifts in rainfall-runoff partitioning during multiyear drought: Roles of dry
of crustal strength fields on the patterns and rates of fluvial incision, period and catchment characteristics, Water Resources Research, 52,
Journal of Geophysical Research (Earth Surface), 120, 275–299, doi:10.1002/2016WR019525.
doi:10.1002/2014JF003281. Saft, M., A. W. Western, L. Zhang, M. C. Peel, and N. J. Potter (2015), The
Royden, L. H., and J. T. Perron (2013), Solutions of the stream power influence of multiyear drought on the annual rainfall-runoff relationship: An
equation and application to the evolution of river longitudinal profiles, Australian perspective, Water Resources Research, 51, 2444–2463,
Journal of Geophysical Research (Earth Surface), 118(2), 497–518, doi:10.1002/2014WR015348.
doi:10.1002/jgrf.20031. Saito, K., and T. Oguchi (2005), Slope of alluvial fans in humid regions of
Ruddiman, W. F. (2013), Earth’s climate: Past and future, 3rd ed., Japan, Taiwan and the Philippines, Geomorphology, 70, 147–162,
Freeman, New York. doi:10.1016/j.geomorph.2005.04.006.
Rüdiger, C., G. R. Hancock, H. M. Hemakumura, B. Jacobs, J. D. Kalma, Sala, O. E., W. J. Parton, L. Joyce, and W. K. Laurenrath (1988), Primary
C. Martinez, M. Thyer, J. P. Walker, T. Wells, and G. R. Willgoose production of the central grassland region of the United States, Ecology,
(2007), Goulburn River experimental catchment data set, Water Resources 69(1), 40–45.
Research, 43(10), W10403. Sallares, V., and A. Calahorrano (2007), Geophysical characterization of
Ruimy, A., L. Kergoat, A. Bondeau, and Postdam NPP Model Intercom- mantle melting anomalies: A crustal view, in Plates, plumes and planetary
parison Team (1999), Comparing global models of terrestrial net primary processes, Geological Society of America Special Paper 430, edited by G.
productivity (NPP): Analysis of differences in light absorption and light- R. Foulger and D. M. Jurdy, pp. 507–524, Geological Society of America,
use efficiency, Global Change Biology, 5(Suppl. 1), 56–64. Boulder, CO, doi:10.1130/2007.2430(25).
Rumpel, C., V. Chaplot, O. Planchon, J. Bernadou, C. Valentin, and A. Salvador-Blanes, S., B. Minasny, and A. B. McBratney (2007), Modelling
Mariotti (2006), Preferential erosion of black carbon on steep slopes long-term in situ soil profile evolution: Application to the genesis of soil
with slash and burn agriculture, Catena, 65, 30–40, doi:10.1016/j. profiles containing stone layers, European Journal of Soil Science, 58,
catena.2005.09.005. 1535–1548.
Rumpel, C., A. Ba, F. Darboux, V. Vhaplot, and O. Planchon (2009), Samonil, P., K. Kral, and L. Hort (2010), The role of tree uprooting in soil
Erosion budget and process selectivity of black carbon at meter scale, formation: A critical literature review, Geoderma, 157, 65–79, doi:10.1016/
Geoderma, 154, 131–137, doi:10.1016/j.geoderma.2009.10.006. j.geoderma.2010.03.018.
312 References
Sanchidrian, J. A., F. Ouchterlony, P. Segarra, and P. Moser (2014), Size Sharmeen, S., and G. R. Willgoose (2006), The interaction
distribution functions for rock fragments, International Journal of between armouring and particle weathering for eroding landscapes,
Rock Mechanics & Mining Sciences, 71, 381–394, doi:10.1016/j. Earth Surface Processes and Landforms, 31(10), 1195–1210,
ijrmms.2014.08.007. doi:10.1002/esp.1397.
Sankaran, M., et al. (2005), Determinants of woody cover in African Sharmeen, S., and G. R. Willgoose (2007), A one-dimensional model for
savannas, Nature, 438(7069), 846–849, doi:10.1038/NATURE04070. simulating armouring and erosion on hillslopes. 2. Long-term erosion and
Santi, P. M., V. G. deWolfe, J. D. Higgins, S. H. Cannon, and J. E. Gartner armouring predictions for two contrasting mine spoils, Earth Surface
(2008), Sources of debris flow material in burned areas, Geomorphology, Processes and Landforms, 32(10), 1437–1453, doi:10.1002/esp.1482.
96, 310–321, doi:10.1016/j.geomorph.2007.02.022. Sharples, J. J., G. J. Cary, P. Fox-Hughes, S. Mooney, J. P. Evans, M.-S.
Sauer, D., P. Finke, R. Sorensen, R. Sperstad, I. Schulli-Maurer, H. Hoeg, Fletcher, M. Fromm, P. F. Grierson, R. McRae, and P. Baker (2016),
and K. Stahr (2012), Testing a soil development model against southern Natural hazards in Australia: Extreme bushfire, Climate Change, 139,
Norway soil chronosequences, Quaternary International, 265, 18–31, 85–99, doi:10.1007/s10584-016-1811-1.
doi:10.1016/j.quaint.2011.12.018. Sheridan, G. J., P. N. J. Lane, and P. J. Noske (2007), Quantification of
Saxton, K. E., and W. J. Rawls (2006), Soil water characteristic estimates hillslope runoff and erosion processes before and after wildfire in a wet
by texture and organic matter for hydrologic solutions, Soil Science Soci- eucalyptus forest, Journal of Hydrology, 343, 12–28, doi:10.1016/j.
ety of America, 70, 1569–1578, doi:10.2136/sssaj2005.0117. jhydrol.2007.06.005.
Schaetzl, R. J. (2014), Professor Donald Johnson’s list of Landmark Shreve, R. L. (1967), Infinite topologically random channel networks,
Papers (1878–1998) in geomorphology and soil geomorphology – An Journal of Geology, 75, 178–186.
appreciation, Progress in Physical Geography, 28(1), 129–137, Shuin, Y., N. Hotta, M. Suzuki, and K. Ogawa (2012), Estimating the
doi:10.1177/0309133314522284. effects of heavy rainfall conditions on shallow landslides using a distrib-
Schaetzl, R. J., and M. L. Thompson (2015), Soils genesis and geomorph- uted landslide conceptual model, Physics and Chemistry of the Earth, 49,
ology, 2nd ed., Cambridge University Press, Cambridge. 44–51, doi:10.1016/j.pce.2011.06.002.
Schmidt, K. M., J. J. Roering, J. D. Stock, W. E. Dietrich, D. R. Montgom- Shull, D. H. (2001), Transition-matrix model of bioturbation and radio-
ery, and T. Schaub (2001), The variability of root cohesion as an influence nuclide diagenesis, Limnology and Oceanography, 46(4), 905–916.
on shallow landslide susceptibility in the Oregon Coast Range, Canadian Sidle, R. C. (1992), A theoretical model of the effects of timber harvesting
Geotechnical Journal, 38, 995–1024, doi:10.1139/cgj-38–5-995. on slope stability, Water Resources Research, 28(7), 1897–1910.
Schmidt, M. W. I., et al. (2011), Persistence of soil organic matter as an Sidman, G., D. P. Guertin, D. C. Goodrich, D. Thoma, D. Falk, and I. S.
ecosystem property, Nature, 478, 49–56, doi:10.1038/nature10386. Burns (2016), A coupled modelling approach to assess the effect of fuel
Schimel, D. S., B. H. Braswell, E. A. Holland, R. McKeown, D. S. Ojima, treatments on post-wildfire runoff and erosion, International Journal of
T. H. Painter, W. J. Parton, and A. R. Townsend (1994), Climatic, edaphic, Wildland Fire, 25, 351–362, doi:10.1071/WF14058.
and biotic controls over storage and turnover of carbon in soils, Global Silva, J. S., F. C. Rego, and S. Mazzoleni (2006), Soil water dynamics
Biogeochemical Cycles, 8(3), 279–293. after fire in a Portuguese shrubland, International Journal of Wildland
Schönbrodt, S., P. Saumer, T. Behrens, C. Seeber, and T. Scholten (2010), Fire, 15, 99–111, doi:10.1071/WF04057.
Assessing the USLE crop and management factor C for soil erosion Simonett, D. S. (1967), Landslide distribution and earthquakes in the
modeling in a large mountainous watershed in Central China, Journal of Bewani and Torricelli Mountains, New Guinea, in Landform studies from
Earth Science, 21(6), 835–845, doi:10.1007/s12583-010-0135-8. Australia and New Guinea, edited by J. N. Jennings and J. A. Mabbutt,
Schoorl, J. M., A. J. A. M. Temme, and T. Veldkamp (2014), Modelling pp. 64–84, Cambridge University Press, Cambridge.
centennial sediment waves in an eroding landscape – Catchment complex- Simunek, J., M. Senja, H. Saito, M. Sakai, and M. T. van Genuchten
ity, Earth Surface Processes and Landforms, 39(11), 1526–1537, (2008), The HYDRUS-1D software package for simulating the
doi:10.1002/esp.3605. one-dimensional movement of water, heat, and multiple solutes in
Schoorl, J. M., and A. Veldkamp (2001), Linking land use and landscape variably-saturated media, version 4.0, 315 pp, University of California.
process modelling: A case study for the Alora region (south Spain), Agri- Sitch, S., et al. (2003), Evaluation of ecosystem dynamics, plant geog-
culture, Ecosystems & Environment, 85(1–3), 281–292, doi:10.1016/s0167- raphy and terrestrial carbon cycling in the LPJ dynamic global vegetation
8809(01)00194-3. model, Global Change Biology, 9(2), 161–185.
Scott, K. M., and C. F. Pain (2009), Regolith science, Springer, Dordrecht, Six, J., H. Bossuyt, S. Degryze, and K. Denef (2004), Review: A history of
the Netherlands. research on the link between (micro)aggregates, soil biota, and soil organic
Sella, G. F., S. Stein, T. H. Dixon, M. Craymer, T. S. James, S. Mazzotti, matter dynamics, Soil Tillage Research, 79, 7–31, doi:10.1016/j.
and R. K. Dokka (2007), Observation of glacial isostatic adjustment in still.2004.03.008.
‘stable’ North America with GPS, Geophysical Research Letters, 34(2), Sklar, L. S., and W. E. Dietrich (1998), River longitudinal profiles
L02306, doi:10.1029/2006GL027081. and bedrock incision models: Stream power and the influence of
Shakesby, R. A. (2011), Post-wildfire soil erosion in the Mediterranean: sediment supply, in Rivers over rock: Fluvial processes in bedrock
Review and future research directions, Earth-Science Reviews, 105(3–4), channels, edited by K. J. Tinkler and E. E. Wohl, AGU, Washington,
71–100, doi:10.1016/j.earscirev.2011.01.001. DC.
Shakesby, R. A., and S. H. Doerr (2006), Wildfire as a hydrological and Sklar, L. S., and W. E. Dietrich (2001), Sediment and rock strength
geomorphological agent, Earth-Science Reviews, 74, 269–307, doi:10.1016/ controls on river incision into bedrock, Geology, 29(12), 1087–1090,
j.earscirev.2005.10.006. doi:10.1130/0091-7613(2001)029<1087:SARSCO>2.0.CO;2.
References 313
Sklar, L. S., and W. E. Dietrich (2004), A mechanistic model for river Stewart, V. I., W. A. Adams, and H. H. Abdulla (1970), Quantitative
incision into bedrock by saltating bedrock, Water Resources Research, pedological studies on soils derived from Silurian mudstones. II. The
40(6), W06301. relationship between stone content and apparent density of the fine earth,
Sklar, L. S., and W. E. Dietrich (2008), Implications of the saltation– Journal of Soil Science, 21, 248–255.
abrasion bedrock incision model for steady-state river longitudinal profile Stocker, T. F., D. Qin, G.-K. Plattner, M. M. B. Tignor, S. K. Allen, J.
relief and concavity, Earth Surface Processes and Landforms, 33(7), Boschung, A. Nauels, Y. Xia, V. Bex, and P. M. Midgley (Eds.) (2013),
1129–1151, doi:10.1002/esp.1689. Working Group I Contribution to the Fifth Assessment Report of the
Sklar, L. S., and W. E. Dietrich (2012), Correction to A mechanistic model Intergovernmental Panel on Climate Change, Cambridge University Press,
for river incision into bedrock by saltating bed load, Water Resources Cambridge.
Research, 48, W06902, doi:10.1029/2012WR012267. Stockmann, U., B. Minasny, and A. B. McBratney (2014), How fast does soil
Sklar, L. S., and J. A. Marshall (2016), The problem of predicting the size grow?, Geoderma, 216, 48–61, doi:10.1016/j.geoderma.2013.10.007.
distribution of sediment supplied by hillslopes to rivers, Geomorphology, Stockmann, U., et al. (2013), The knowns, known unknowns and
277, 31–49, doi:10.1016/j.geomorph.2016.05.005. unknowns of sequestration of soil organic carbon, Agriculture, Ecosystems
Sloan, S. W. (1987), A fast algorithm for constructing Delaunay & Environment, 164, 80–99, doi:10.1016/j.agee.2012.10.001.
triangulation in the plane, Advances in Engineering Software, 9(1), 34–55. Strahler, A. N. (1952), Dynamic basis of geomorphology, Bulletin of the
Smith, H. G., G. J. Sheridan, P. N. J. Lane, P. Nyman, and S. Haydon Geological Society of America, 63, 923–938.
(2011), Wildfire effects on water quality in forest catchments: A review Strahler, A. N. (1952), Hypsometric (area-altitude) analysis of erosional
with implications for water supply, Journal of Hydrology, 396, 170–192, topography, Bulletin of the Geological Society of America, 63, 1117–1142.
doi:10.1016/j.jhydrol.2010.10.043. Sullivan, A. L. (2009a), Wildland surface fire spread modelling,
Smith, P., et al. (1997), A comparison of the performance of nine soil 1990–2007. 1: Physical and quasi-physical models, International Journal
organic matter models using datasets from seven long-term experiments, of Wildland Fire, 18, 349–368, doi:10.1071/WF06143.
Geoderma, 81(1–2), 153–225. Sullivan, A. L. (2009b), Wildland surface fire spread modelling,
Smith, T. R., and F. P. Bretherton (1972), Stability and the conservation of mass 1990–2007. 2: Empirical and quasi-empirical models, International Jour-
in drainage basin evolution, Water Resources Research, 8(6), 1506–1529. nal of Wildland Fire, 18, 369–386, doi:10.1071/WF06142.
Solyom, P. B., and G. E. Tucker (2004), Effect of limited storm duration Sullivan, A. L. (2009c), Wildland surface fire spread modelling,
on landscape evolution, drainage basin geometry, and hydrograph shapes, 1990–2007. 3: Simulation and mathematical analogue models, Inter-
Journal of Geophysical Research (Earth Surface), 109, F03012. national Journal of Wildland Fire, 18, 387–403, doi:10.1071/WF06144.
Solyom, P. B., and G. E. Tucker (2007), The importance of the catchment Surkan, A. J. (1969), Synthetic hydrographs: Effects of network geometry,
area–length relationship in governing non-steady state hydrology, optimal Water Resources Research, 5(1), 112–128.
junction angles and drainage network pattern, Geomorphology, 88, Sweeney, K. E., J. J. Roering, and C. Ellis (2015), Experimental evidence
84–108, doi:10.1016/j.geomorph.2006.10.014. for hillslope control of landscape scale, Science, 349(6243), 51–53,
Sørensen, M. (2004), On the rate of aeolian sand transport, Geomorph- doi:10.1126/science.aab0017.
ology, 59(1–4), doi:10.1016/j.geomorph.2003.09.005. Syvitski, J. P. M., S. D. Peckham, R. Hilberman, and T. Mulder (2003),
Southard, J. B. (2006), 12.090 Introduction to fluid motions, sediment Predicting the terrestrial flux of sediment to the global ocean: A planetary
transport, and current-generated sedimentary structures, MIT OpenCour- perspective, Sedimentary Geology, 162, 5–24, doi:10.1016/S0037-0738(03)
seWare, Massachusetts Institute of Technology, Cambridge, MA. 00232-X.
Spada, G., G. Ruggieri, L. S. Sorenson, K. Nielson, D. Melini, and F. Tarboton, D. G. (1997), A new method for the determination of flow
Colleoni (2012), Greenland uplift and regional sea level changes from directions and upslope areas in grid digital elevation models, Water
ICESat observations and GIA modelling, Geophysical Journal Inter- Resources Research, 33(2), 309–319.
national, 189, 1457–1474, doi:10.1111/j.1365-246X.2012.05443.x. Tarboton, D. G., R. L. Bras, and I. Rodriguez-Iturbe (1989), Scaling and
Stallard, R. F. (1998), Terrestrial sedimentation and the carbon cycle: elevation in river networks, Water Resources Research, 25(9), 2037–2052.
Coupling weathering and erosion to carbon burial, Global Biogeochemical Tarolli, P. (2014), High-resolution topography for understanding Earth sur-
Cycles, 12(2), 231–257. face processes: Opportunities and challenges, Geomorphology, 216,
Stark, C. P., and F. Guzzetti (2009), Landslide rupture and the probability 295–312, doi:10.1016/j.geomorph.2014.03.008.
distribution of mobilized debris volumes, Journal of Geophysical Taylor, S. H., B. S. Ripley, F. I. Woodward, and C. P. Osborne (2011),
Research (Earth Surface), 114, F00A02, doi:10.1029/2008JF001008. Drought limitation of photosynthesis differs between C3 and C4 grass
Stark, C. P., and N. Hovius (2001), The characterization of landslide size species in a comparative experiment, Plant, Cell and Environment, 34,
distributions, Geophysical Research Letters, 28(6), 1091–1094, 65–75, doi:10.1111/j.1365-3040.2010.02226.x.
doi:10.1029/2000GL008527. Teles, V., G. de Marsily, and É. Perrier (1998), Sur une nouvelle approche
Steefel, C. I. (2009), CrunchFlow: Software for modelling multicomponent de modélisation de la mise en place des sédiments dans une plaine
reactive flow and transport, user manual, Earth Sciences Division, Law- alluviale pour en représenter l’hétérogénéité, Comptes Rendus de l’Acade-
rence Berkeley National Laboratory, Berkeley, CA, www.csteefel.com/ mie des Sciences Serie II Fascicule A – Sciences de la terre et des planètes,
CrunchFlowManual.pdf. 327(9), 597–606, doi:10.1016/S1251-8050(99)80113-X.
Steel, Z. L., H. D. Safford, and J. H. Viers (2015), The fire frequency- Temme, A. J. A. M., L. Claessens, A. Veldkamp, and J. M. Schoorl
severity relationship and the legacy of fire suppression in California (2011), Evaluating choices in multi-process landscape evolution models,
forests, Ecosphere, 6(1), 8, doi:10.1890/ES14-00224.1. Geomorphology, 125, doi:10.1016/j.geomorph.2010.10.007.
314 References
Temme, A. J. A. M., I. Peeters, E. Buis, A. Veldkamp, and G. Govers Torri, D., and J. Poesen (2014), A review of topographic threshold condi-
(2011), Comparing landscape evolution models with quantitative field data tions for gully head development in different environments, Earth-Science
at the millennial time scale in the Belgian loess belt, Earth Surface Reviews, 130, 73–85, doi:10.1016/j.earscirev.2013.12.006.
Processes and Landforms, 36(10), 1300–1312, doi:10.1002/esp.2152. Tranter, G., B. Minasny, A. B. McBratney, B. Murphy, N. J. McKenzie,
Temme, A. J. A. M., and T. Vanwalleghem (2016), LORICA – A new M. Grundy, and D. M. Brough (2007), Building and testing conceptual
model for linking landscape and soil profile evolution: Development and and empirical models for predicting soil bulk density, Soil Use and
sensitivity analysis, Computers & Geosciences 90, 131–143, doi:10.1016/ Management, 23(4), 437–443, doi:0.1111/j.1475-2743.2007.00092.x.
j.cageo.2015.08.004. Trumbore, S. (2000), Age of soil organic matter and soil respiration:
Temme, A. J. A. M., and A. Veldkamp (2009), Multi-process Late Qua- Radiocarbon constraints on belowground C dynamics, Ecological Applica-
ternary landscape evolution modelling reveals lags in climate response tions, 10(2), 399–411, doi:10.1890/1051-0761(2000)010[0399:AOSOMA]
over small spatial scales, Earth Surface Processes and Landforms, 34(4), 2.0.CO;2.
573–589, doi:10.1002/esp.1758. Trumbore, S. (2009), Radiocarbon and soil carbon dynamics, Annual
Tesemma, Z. K., Y. Wei, M. C. Peel, and A. W. Western (2015), Review of Earth and Planetary Sciences, 37, 47–66, doi:10.1146/
The effect of year-to-year variability of leaf area index on annurev.earth.36.031207.124300.
Variable Infiltration Capacity model performance and simulation of Tucker, G., S. Lancaster, N. Gasparini, and R. Bras (2001a), The channel-
runoff, Advances in Water Resources, 83, 310–322, doi:10.1016/j. hillslope integrated landscape development model (CHILD), in Landscape
advwatres.2015.07.002. erosion and evolution modelling, edited by R. S. Harmon and W. W. Doe,
Tessler, N., L. Wittenberg, and N. Greenbaum (2013), Soil water repel- pp. 349–388, Kluwer, New York.
lency persistence after recurrent forest fires on Mount Carmel, Israel, Tucker, G. E., and R. L. Bras (2000), A stochastic approach to modeling
International Journal of Wildland Fire, 22, 515–526, doi:10.1071/ the role of rainfall variability in drainage basin evolution, Water Resources
WF12063. Research, 36(7), 1953–1964.
Thomas, I. A., P. Jordan, O. Shine, O. Fenton, P. E. Mellander, P. Dunlop, Tucker, G. E., and G. R. Hancock (2010), Modelling landscape evolution,
and P. N. C. Murphy (2017), Defining optimal DEM resolutions and point Earth Surface Processes and Landforms, 35(1), 28–50, doi:10.1002/
densities for modelling hydrologically sensitive areas in agricultural catch- esp.1952.
ments dominated by microtopography, International Journal of Applied
Tucker, G. E., S. T. Lancaster, N. M. Gasparini, R. L. Bras, and S. M.
Earth Observation and Geoinformation, 54, 38–52, doi:10.1016/j.
Rybarczyk (2001b), An object-oriented framework for distributed hydro-
jag.2016.08.012.
logic and geomorphic modeling using triangulated irregular networks,
Thompson, S. E., C. J. Harman, P. Heine, and G. G. Katul (2010), Computers & Geosciences, 27(8), 959–973.
Vegetation infiltration relationships across climatic and soil type gradients,
Tucker, G. E., and R. L. Slingerland (1994), Erosional dynamics, flexural
Journal of Geophysical Research (Biogeosciences), 115, G02023,
isostasy, and long-lived escarpments – A numerical modeling study,
doi:10.1029/2009JG001134.
Journal of Geophysical Research (Solid Earth), 99(B6), 12229–12243,
Thonicke, K., A. Spessa, I. C. Prentice, S. P. Harrison, L. Dong, and C. doi:10.1029/94JB00320.
Carmona-Moreno (2010), The influence of vegetation, fire spread and fire
Tucker, G. E., and K. X. Whipple (2002), Topographic outcomes pre-
behaviour on biomass burning and trace gas emissions: Results from a
dicted by stream erosion models: Sensitivity analysis and intermodel
process-based model, Biogeosciences, 7, 1991–2011, doi:10.5194/bg-7-
comparison, Journal of Geophysical Research (Solid Earth), 107(B9),
1991-2010.
art. no.-2179.
Tilman, D. (1994), Competition and biodiversity in spatially structured
Turcotte, D. L., and G. Schubert (2002), Geodynamics, Cambridge Uni-
habitats, Ecology, 75(1), 2–16, doi:10.2307/1939377.
versity Press, Cambridge.
Ting, I. P., and W. E. Loomis (1965), Further studies concerning stomatal
Turowski, J. M., and D. Rickenmann (2009), Tools and cover effects in
diffusion, Plant Physiology, 40(2), 220–228, doi:10.1104/pp.40.2.220.
bedload transport observations in the Pitzbach, Austria, Earth Surface
Tisdall, J. M. (1996), Formation of soil aggregates and accumulation of Processes and Landforms, 34, 26–37, doi:10.1002/esp.1686.
soil organic matter, in Structure and organic matter storage in agricul-
Tushingham, A. M., and W. R. Peltier (1991), ICE-3G: A new global
tural soils, edited by M. R. Carter and B. A. Stewart, pp. 57–96, CRC
model of late Pleistocene deglaciation based upon geophysical predictions
Press, Boca Raton, FL.
of post-glacial relative sea level change, Journal of Geophysical Research
Tisdall, J. M., and J. M. Oades (1982), Organic matter and water stable (Solid Earth), 96(B3), 4497–4523.
aggregates in soils, European Journal of Soil Science, 33(2), 141–163.
Uijlenhoet, R. (2001), Raindrop size distributions and radar reflectivity–
Tomkin, J. H., M. T. Brandon, F. J. Pazzaglia, J. R. Barbour, and S. D. rain rate relationships for radar hydrology, Hydrology and Earth System
Willett (2003), Quantitative testing of bedrock incision Sciences, 5(4), 615–627.
models for the Clearwater River, NW Washington State, Journal of
Unger, P. W., and O. R. Jones (1998), Long-term tillage and cropping
Geophysical Research (Solid Earth), 108(B6), ETG 10, doi:10.1029/
systems affect bulk density and penetration resistance of soil cropped to
2001JB000862.
dryland wheat and grain sorghum, Soil Tillage Research, 45(1–2), 39–57,
Tongway, D. J., and J. A. Ludwig (1990), Vegetation and soil patterning in doi:10.1016/S0167-1987(97)00068-8.
semi-arid mulga lands of Eastern Australia, Australian Journal of Ecology,
USDA-ARS (1999), Soil quality test kit guide, USDA-ARS, Soil Quality
15, 23–34.
Institute.
Torn, M. S., S. E. Trumbore, O. A. Chadwick, P. M. Vitousek, and D. M.
USDA-ARS (2008), User’s reference guide: Revised universal soil loss
Hendricks (1997), Mineral control of soil organic carbon storage and
equation Version 2 (RUSLE2), USDA-ARS, Washington, DC.
turnover, Nature, 389(6647), 170–173, doi:10.1038/38260.
References 315
USGS (2015), PHREEQC (Version 3) – A computer program for speci- van Veen, J. A., and E. A. Paul (1981), Organic carbon dynamics in
ation, batch-reaction, one-dimensional transport, and inverse geochemical grassland soils. 1. Background information and computer simulation,
calculations, http://wwwbrr.cr.usgs.gov/projects/GWC_coupled/phreeqc/. Canadian Journal of Soil Science, 61(2), 185–201.
Valentin, C., J. Poesen, and Y. Li (2005), Gully erosion: Impacts, factors Vanwalleghem, T., J. Poesen, A. B. McBratney, and J. Deckers (2010),
and control, Catena, 63(2–3), 132–153, doi:10.1016/j.catena.2005.06.001. Spatial variability of soil horizon depth in natural loess-derived soils,
van Ash, T. J. W., and P. M. B. van Genuchten (1990), A comparison Geoderma, 157, 37–45, doi:10.1016/j.geoderma.2010.03.013.
between theoretical and measured creep profiles of landslides, Geomorph- Vanwalleghem, T., H. Saito, Y. Hayakawa, and T. Oguchi (2013), Inter-
ology, 3(1), 45–55, doi:10.1016/0169-555X(90)90031-K. action between soil formation and landslide occurrence, paper presented at
Vandaele, K., and J. Poesen (1995), Spatial and temporal patterns of EGU General Assembly 2013, Geophysical Research Abstracts,
soil-erosion rates in an agricultural catchment, Central Belgium, Catena, EGU2013-8665-1, European Geophysical Union, Vienna.
25(1–4), 213–226. Vanwalleghem, T., U. Stockmann, B. Minasny, and A. B. McBratney (2013),
Vandekerckhove, L., J. Poesen, and G. Govers (2003), Medium-term gully A quantitative model for integrating landscape evolution and soil formation,
headcut retreat rates in Southeast Spain determined from aerial photo- Journal of Geophysical Research (Earth Surface), 118, 1–17, doi:10.1029/
graphs and ground measurements, Catena, 50(2–4), 329–352, 2011JF002296.
doi:10.1016/S0341-8162(02)00132–7. Varadachari, C., A. K. Barman, and K. Ghosh (1994), Weathering of
Vandekerckhove, L., J. Poesen, D. O. Wijdenes, and G. Gyssels (2001), silicate minerals by organic-acids. 2. Nature of residual products, Geo-
Short-term bank gully retreat rates in Mediterranean environments, derma, 61(3–4), 251–268.
Catena, 44(2), 133–161, doi:10.1016/S0341-8162(00)00152-1. Veenstra, J. J., and C. L. Burras (2015), Soil profile transformation after
VandenBygaart, A. J., E. G. Gregorich, and B. L. Helgason (2015), Crop- 50 years of agricultural land use, Soil Science Society of America, 79,
land C erosion and burial: Is buried soil organic matter biodegradable?, 1154–1162, doi:10.2136/sssaj2015.01.0027.
Geoderma, 239–240, 240–249, doi:10.1016/j.geoderma.2014.10.011. Venevsky, S., K. Thonicke, S. Sitch, and W. Cramer (2002), Simulating
van der Beek, P., M. A. Summerfield, J. Braun, R. W. Brown, and A. Fleming fire regimes in human-dominated ecosystems: Iberian Peninsula case
(2002), Modeling postbreakup landscape development and denudational study, Global Change Biology, 8(10), 984–998.
history across the southeast African (Drakensberg Escarpment) margin, Jour- Ventsel, E., and T. Krauthammer (2001), Thin plates and shells: Theory:
nal of Geophysical Research (Solid Earth), 107(B12), art. no.-2351. analysis, and applications, Marcel Dekker, New York.
Van De Wiel, M. J., T. J. Coulthard, M. G. Macklin, and J. Lewin (2007), Verdon, D. C., A. S. Kiem, and S. W. Franks (2004), Multi-decadal
Embedding reach-scale fluvial dynamics within the CAESAR cellular variability of forest fire risk – Eastern Australia, International Journal of
automaton landscape evolution model, Geomorphology, 90, 283–301, Wildland Fire, 13(2), 165–171.
doi:10.1016/j.geomorph.2006.10.024. Vereecken, H., et al. (2016), Modelling soil processes: Review, key
van Dijk, A. I. J. M., L. A. Bruijnzeel, and C. J. Rosewell (2002a), Rainfall challenges and new perspectives, Vadose Zone Journal, 15(5),
intensity-kinetic energy relationships: A critical literature appraisal, Jour- doi:10.2136/vzj2015.09.0131.
nal of Hydrology, 261(1–4), 1–23, doi:10.1016/S0022-1694(02)00020-3. Vico, G., and A. Porporato (2008), Modelling C3 and C4 photosynthesis
van Dijk, A. I. J. M., A. G. C. A. Meesters, and L. A. Bruijnzeel (2002b), under water-stressed conditions, Plant and Soil, 313, 187–203,
Exponential distribution theory and the interpretation of splash detachment doi:10.1007/s11104-008-9691-4.
and transport experiments, Soil Science Society of America Journal, 66, Viles, H. A., L. A. Naylor, N. E. A. Carter, and D. Chaput (2008),
1466–1474, doi:10.2136/sssaj2002.1466. Biogeomorphological disturbance regimes: Progress in linking ecological
Van Eck, C. M., J. P. Nunes, D. C. S. Vieira, S. Keestra, and J. J. Keizer and geomorphological systems, Earth Surface Processes and Landforms,
(2016), Physically-based modelling of the post-fire runoff response of a 33, 1419–1435, doi:10.1002/esp.1717.
forest catchment in central Portugal: Using field versus remote sensing Vince, G. (2005), Tsunami seabed shows massive disruption, New Scientist,
based estimates of vegetation recovery, Land Degradation and Develop- www.newscientist.com/article/dn7465-tsunami-seabed-shows-massive-dis
ment, 27, 1535–1544, doi:10.1002/ldr.2507. ruption/.
van Grinsven, J. J. M., and W. H. van Riemsdijk (1992), Evaluation of Vincent, K. R., and O. A. Chadwick (1994), Synthesizing bulk-density for
batch and column techniques to measure weathering rates in soils, Geo- soils with abundant rock fragments, Soil Science Society of America
derma, 52(1–2), 41–57. Journal, 58, 455–464.
van Hooff, P. (1983), Earthworm activity as a cause of splash erosion in a Volkwein, A., K. Schllenberg, V. Labiouse, F. Agliardi, F. Berger,
Luxembourg forest, Geoderma, 31(3), 195–204. F. Bourrier, L. K. A. Dorren, W. Gerber, and M. Jaboyedoff (2011),
Vanmaercke, M., et al. (2016), How fast do gully headcuts retreat?, Earth- Rockfall characterisation and structural protection – A review, Natural
Science Reviews, 154, 336–355, doi:10.1016/j.earscirev.2016.01.009. Hazards and Earth System Sciences, 11, 2617–2651, doi:10.5194/nhess-
van Mantgem, P. J., et al. (2009), Widespread increase of tree mortality in 11-2617-2011.
the western United States, Science, 323(5913), 521–524, doi:10.1126/ Voroney, R. P. (2007), The soil habitat, in Soil microbiology, ecology, and
science.1165000. biochemistry, edited by E. A. Paul, pp. 25–52, Academic Press, Amsterdam.
Vanoni, V. A. (1975), Sedimentation engineering, ASCE, New York. Vrieling, A. (2006), Satellite remote sensing for water erosion assessment:
Van Oost, K., and M. M. Bakker (2012), Soil productivity and erosion, in A review, Catena, 65, 2–18, doi:10.1016/j.catena.2005.10.005.
Soil ecology and ecosystem services, edited by D. H. Wall, R. D. Bardgett, Wainwright, J. (2008), Can modelling enable us to understand the role of
V. Behan-Pelletier, J. E. Herrick, T. H. Jones, K. Ritz, J. Six, D. R. Strong, humans in landscape evolution?, Geoforum, 39(2), 659–674, doi:10.1016/
and W. H. van der Putten, pp. 301–314, Oxford University Press, Oxford. j.geoforum.2006.09.011.
316 References
Wainwright, J., and J. D. A. Millington (2010), Mind, the gap in Wechsler, S. P. (2007), Uncertainties associated with digital elevation
landscape-evolution modelling, Earth Surface Processes and Landforms, models for hydrologic applications: A review, Hydrology and Earth
35, 842–855, doi:10.1002/esp.2008. System Sciences 11, 1481–1500.
Wainwright, J., A. J. Parsons, J. R. Cooper, P. Gao, J. A. Gillies, L. Mao, Weissel, J. K., and M. A. Seidl (1997), Influence of rock strength proper-
J. D. Orford, and P. G. Knight (2015), The concept of transport capacity in ties on escarpment retreat across passive continental margins, Geology, 25
geomorphology, Reviews of Geophysics, 53, 115–1202, doi:10.1002/ (7), 631–635, doi:10.1130/0091-7613(1997) 025<0631:IORSPO> 2.3.
2014RG000474. CO;2.
Wainwright, J., A. J. Parsons, E. N. Müller, R. E. Brazier, and D. M. Welivitiya, W. D. D. P. (2017), A next generation spatially distributed
Powell (2009), Response to Hairsine’s and Sander’s ‘Comment on “A model for soil profile dynamics and pedogenesis, PhD thesis, University of
transport-distance based approach to scaling erosion rates”’: Parts 1, 2 and Newcastle, Callaghan, Australia.
3 by Wainwright et al., Earth Surface Processes and Landforms, 34(6), Welivitiya, W. D. D. P., G. R. Willgoose, G. R. Hancock, and S. Cohen
886–890. (2016), Exploring the sensitivity on a soil area-slope-grading relationship
Wainwright, J., A. J. Parsons, E. N. Müller, R. E. Brazier, D. M. Powell, and to changes in process parameters using a pedogenesis model, Earth System
B. Fenti (2008a), A transport-distance approach to scaling erosion rates: 1. Dynamics, 4, 607–625, doi:10.5194/esurf-4-607-2016.
Background and model development, Earth Surface Processes and Land- Wells, T., P. Binning, and G. R. Willgoose (2005), The role of moisture
forms, 33, 813–826, doi:10.1002/esp.1624. cycling in the weathering of a quartz chlorite schist in a tropical environment:
Wainwright, J., A. J. Parsons, E. N. Müller, R. E. Brazier, D. M. Powell, Findings of a laboratory simulation, Earth Surface Processes and Landforms,
and B. Fenti (2008b), A transport-distance approach to scaling erosion 30(4), 413–428, doi:10.1002/esp.1149.
rates: 2. Sensitivity and evaluation of MAHLERAN, Earth Surface Pro- Wells, T., P. Binning, G. R. Willgoose, and G. R. Hancock (2006),
cesses and Landforms, 33, 962–984, doi:10.1002/esp.1623. Laboratory simulation of the salt weathering of schist: I. Weathering of
Wainwright, J., A. J. Parsons, E. N. Müller, R. E. Brazier, D. M. Powell, schist blocks in a seasonally wet tropical environment, Earth Surface
and B. Fenti (2008c), A transport-distance approach to scaling erosion Processes and Landforms, 31(3), 339–354, doi:10.1002/esp.1248.
rates: 3. Evaluating scaling characteristics of MAHLERAN, Earth Surface Wells, T., G. R. Hancock, C. Dever, and C. Martinez (2013), Application
Processes and Landforms, 33, 1113–1128, doi:10.1002/esp.1622. of RothPC-1 to soil carbon profiles in cracking soils under minimal till
Walker, J. P., and G. R. Willgoose (1999), On the effect of digital cultivation, Geoderma, 207–208, 144–153, doi:10.1016/j.
elevation model accuracy on hydrology and geomorphology, Water geoderma.2013.05.018.
Resources Research, 35(7), 2259–2268. Wells, T., G. R. Hancock, C. Dever, and D. Murphy (2012), Prediction of
Walker, J. P., G. R. Willgoose, and J. D. Kalma (2001), One-dimensional vertical soil organic carbon profiles using soil properties and environmen-
soil moisture profile retrieval by assimilation of near-surface measure- tal tracer data at an untilled site, Geoderma, 170, 337–346, doi:10.1016/j.
ments: A simplified soil moisture model and field application, Journal of geoderma.2011.11.006.
Hydrometeorology, 2(4), 356–373. Wells, T., G. R. Willgoose, and P. Binning (2007), Laboratory simulation
Wang, E., M. J. Robertson, G. L. Hammer, P. S. Carberry, D. Holzworth, of the salt weathering of schist: II. Fragmentation of fine schist particles,
H. Meinke, S. C. Chapman, J. N. G. Hargreaves, N. I. Huth, and G. Earth Surface Processes and Landforms, 32(5), 687–697, doi:10.1002/
McLean (2002), Development of a generic crop model template in the esp.1450.
cropping system model APSIM, European Journal of Agronomy, 18, Wells, T., G. R. Willgoose, and G. R. Hancock (2008), Modelling weathering
121–140. pathways and processes for salt induced fragmentation of quartz-chlorite
Wang, Y., and F. Tonon (2011), Dynamic validation of a discrete element schist, Journal of Geophysical Research (Earth Surface), 113, F01014,
code in modeling rock fragmentation, International Journal of Rock Mech- doi:10.1029/2006JF000714.
anics & Mining Sciences, 48, 535–545, doi:10.1016/j.ijrmms.2011.02.003. West, A. J. (2012), Thickness of the chemical weathering zone and
Warburton, J., D. G. Milledge, and R. Johnson (2008), Assessment of implications for erosional and climatic drivers of weathering and for
shallow landslide activity following the January 2005 storm, northern carbon-cycle feedbacks, Geology, 40(9), 811–814, doi:10.1130/G33041.1.
Cumbria, Proceedings of the Cumberland Geological Society, 7, 263–283. West, N., E. Kirby, P. Bierman, and B. A. Clarke (2014), Aspect-depend-
Wardle, D. A. (1992), A comparative-assessment of factors which influ- ent variations in regolith creep revealed by meteoric 10Be, Geology, 42(6),
ence microbial biomass carbon and nitrogen levels in soil, Biological 507–510, doi:10.1130/G35357.1.
Reviews of the Cambridge Philosophical Society, 67(3), 321–358, Western, A. W., R. B. Grayson, G. Blöschl, G. R. Willgoose, and T. A.
doi:10.1111/j.1469-185X.1992.tb00728.x. McMahon (1999), Observed spatial organization of soil moisture and its
Wasson, R. J., R. K. Mazari, B. Starr, and G. Clifton (1998), The recent relation to terrain indices, Water Resources Research, 35(3), 797–810.
history of erosion and sedimentation on the Southern Tablelands of south- Whipple, K. X. (1997), Open-channel flow of Bingham fluids: Applica-
eastern Australia: Sediment flux dominated by channel incision, Geo- tions in debris-flow research, Journal of Geology, 105(2), 243–262,
morphology, 24(4), 263–372, doi:10.1016/S0169-555X(98)00019-1. doi:10.1086/515916.
Watson, F. G. R., R. A. Vertessy, T. A. McMahon, B. G. Rhodes, and I. S. Whipple, K. X. (2009), The influence of climate on the tectonic evolution
Watson (1999), The hydrologic impacts of forestry on the Maroondah of mountain belts, Nature Geoscience, 2, 97–104, doi:10.1038/ngeo413.
catchments Rep. 99/1, Cooperative Research Centre for Catchment
Whipple, K. X., G. S. Hancock, and R. S. Anderson (2000), River
Hydrology, Melbourne.
incision into bedrock: Mechanics and relative efficacy of plucking, abra-
Watts, A. B. (2001), Isostasy and flexure of the lithosphere, Cambridge sion, and cavitation, Geological Society of America Bulletin, 112(3),
University Press, Cambridge. 490–503.
References 317
Whipple, K. X., and B. J. Meade (2004), Controls on the strength of Willett, S. (1999), Orogeny and orography: The effects of erosion on the
coupling among climate, erosion, and deformation in two sided, frictional structure of mountain belts, Journal of Geophysical Research (Solid
orogenic wedges at steady state, Journal of Geophysical Research (Earth Earth), 104(B12), 28957–28981.
Surface), 109(F1), F01011, doi:10.1029/2003JF000019. Willett, S. D., R. Slingerland, and N. Hovius (2001), Uplift, shortening,
Whipple, K. X., and G. E. Tucker (1999), Dynamics of the stream-power and steady-state topography in active mountain belts, American Journal of
river incision model: Implications for height limits of mountain ranges, Science, 301(4–5), 455–485.
landscape response timescales, and research needs, Journal of Geophys- Willgoose, G. R. (1989), A physically based channel network and catch-
ical Research (Solid Earth), 104(B8), 17661–17674. ment evolution model, PhD thesis, Department of Civil Engineering,
Massachusetts Institute of Technology, Cambridge, MA. Also published
Whipple, K. X., and G. E. Tucker (2002), Implications of sediment-flux-
as Willgoose, G. R., R. L. Bras, and I. Rodriguez-Iturbe (1989),
dependent river incision models for landscape evolution, Journal of Geo-
A physically based channel network and catchment evolution model,
physical Research (Solid Earth), 107(B2), art. no.-2039.
Tech. Report 322, Ralph M. Parsons Laboratory, Dept. of Civil Engineer-
White, A. F., and S. L. Brantley (2003), The effect of time on the weathering ing, MIT, Cambridge, MA.
of silicate minerals: Why do weathering rates differ in the laboratory
Willgoose, G. R. (1994a), A physical explanation for an observed area-
and field?, Chemical Geology, 202(3–4), 479–506, doi:10.1016/j.
slope-elevation relationship for declining catchments, Water Resources
chemgeo.2003.03.001.
Research, 30(2), 151–159.
White, A. F., A. E. Blum, M. S. Schulz, T. D. Bullen, J. W. Harden, and
Willgoose, G. R. (1994b), A statistic for testing the elevation characteris-
M. L. Peterson (1996), Chemical weathering rates of a soil chronose-
tics of landscape simulation models, Journal of Geophysical Research
quence on granitic alluvium: I. Quantification of mineralogical and surface
(Solid Earth), 99(B7), 13987–13996.
area changes and calculation of primary silicate reaction rates, Geochimica
et Cosmochimica Acta, 60(14), 2533–2550. Willgoose, G. R. (1995), A preliminary assessment of the effect of vege-
tation on the long-term stability of the proposed above-grade rehabilitation
White, A. F., M. S. Schulz, D. A. Stonestrom, D. V. Vivit, J. Fitzpatrick, strategy at Ranger Uranium Mine, Open File Report 119, Environmental
T. D. Bullen, K. Maher, and A. E. Blum (2009), Chemical weathering of a Research Institute of the Supervising Scientist, Jabiru, Australia.
marine terrace chronosequence, Santa Cruz, California. Part II: Solute
Willgoose, G. R. (1997), A hydrodynamic particle tracking algorithm for
profiles, gradients and the comparisons of contemporary and long-term
simulating settling of sediment, Mathematics and Computers in Simula-
weathering rates, Geochimica et Cosmochimica Acta, 73, 2769–2803,
tion, 43, 343–349.
doi:10.1016/j.gca.2009.01.029.
Willgoose, G. R. (2001), Erosion processes, catchment elevations and
White, A. F., M. S. Schulz, D. V. Vivit, A. E. Blum, D. A. Stonestrom, and landform evolution modelling, in Gravel Bed Rivers 2000, edited by P.
S. P. Anderson (2008), Chemical weathering of a marine terrace chron- Mosley, pp. 507–530, Hydrology Society, Christchurch, New Zealand.
osequence, Santa Cruz, California I: Interpreting rates and controls based
Willgoose, G. R. (2005a), Mathematical modeling of whole-landscape
on soil concentration–depth profiles, Geochimica et Cosmochimica Acta,
evolution, Annual Review of Earth and Planetary Sciences, 33, 443–459.
72(1), 36–68, doi:10.1016/j.gca.2007.08.029.
Willgoose, G. R. (2005b), User manual for SIBERIA (Version 8.30),
White, A. F., M. S. Schulz, D. V. Vivit, A. E. Blum, D. A. Stonestrom, and
www.telluricresearch.com/siberia_8.30_manual.pdf.
J. W. Harden (2005), Chemical weathering rates of a soil chronosequence
Willgoose, G. R. (2009), TELLUSIM: A Python plug-in based computa-
on granitic alluvium: III. Hydrochemical evolution and contemporary
tional framework for spatially distributed environmental and earth sciences
solute fluxes and rates, Geochimica et Cosmochimica Acta, 69(8),
modelling, paper presented at 18th World IMACS/MODSIM Congress,
1975–1996, doi:10.1016/j.gca.2004.10.003.
Cairns, Australia, 13–17 July 2009.
Wickens, G. E., and F. W. Collier (1971), Some vegetation patterns in the
Willgoose, G. R. (2010), Assessment of the erosional stability of encapsu-
Republic of Sudan, Geoderma, 6, 43–59, doi:10.1016/0016-7061(71)
lation caps and covers at the millennial timescale: Current capabilities,
90050-4.
research issues and operational needs, paper presented at Proceedings of
Wickert, A. D. (2015), Open-source modular solutions for flexural isos- the Workshop on Engineered Barrier Performance Related to Low-level
tasy: gFlex v1.0, Geoscientific Model Development Discussions, 8, Radioactive Waste, Decommissioning, and Uranium Mill Tailings facil-
4245–4292, doi:10.5194/gmdd-8-4245-2015. ities, NUREG/CP-0195, Office of Nuclear Regulatory Research, Rockville,
Wilcock, P. R., and J. C. Crowe (2003), Surface-based transport model for Maryland, 3–5 August 2010.
mixed-size sediment, Journal of Hydraulic Engineering, 129(2), 120–128, Willgoose, G. R. (2011), Modelling bushfire impact on hydrology: The
doi:10.1061/ASCE 0733-9429(2003)129:2(120). implications of the fire modelling approach on the climate change impact,
Wilkinson, M. T., G. S. Humphreys, J. Chappell, K. Fifield, and B. Smith in MSSANZ MODSIM 2011, pp. 3664–3670, Modelling and Simulation
(2003), Estimates of soil production in the Blue Mountains, Australia, using Society of Australia and New Zealand, Perth.
cosmogenic 10Be, Advances in Regolith: Proceedings of the CRC LEME Willgoose, G. R. (2015), The interactions between evolving soils and
Regional Regolith Symposia, 2003, Cooperative Research Centre for Land- landforms, and the importance of relative response times, in EGU General
scape Environments and Mineral Exploration (CRC LEME), Canberra, Meeting, pp. EGU2015–8220, EGU, Vienna.
http://crcleme.org.au/Pubs/Advancesinregolith/Wilkinson_et_al.pdf. Willgoose, G. R., R. L. Bras, and I. Rodriguez-Iturbe (1990), A model of
Wilkinson, M. T., P. J. Richards, and G. S. Humphreys (2009), river basin evolution, EOS, 71(47), 1806–1807.
Breaking ground: Pedological, geological, and ecological implications of Willgoose, G. R., R. L. Bras, and I. Rodriguez-Iturbe (1991a), A coupled
soil bioturbation, Earth-Science Reviews, 97, 257–272, doi:10.1016/j. channel network growth and hillslope evolution model. 1. Theory, Water
earscirev.2009.09.005. Resources Research, 27(7), 1671–1684.
318 References
Willgoose, G. R., R. L. Bras, and I. Rodriguez-Iturbe (1991b), A coupled 537, US Government Printing Office, Washington, DC. http://naldc.nal
channel network growth and hillslope evolution model. 2. Nondimensio- .usda.gov/download/CAT79706928/PDF.
nalization and applications, Water Resources Research, 27(7), 1685–1696. Witten, T. A., and L. M. Sander (1981), Diffusion-limited aggregation,
Willgoose, G. R., R. L. Bras, and I. Rodriguez-Iturbe (1991c), A physical a kinetic critical phenomenon, Physical Review Letters, 47(19),
explanation of an observed link area-slope relationship, Water Resources 1400–1403.
Research, 27(7), 1697–1702. Wohl, E. E., R. O. Hall, K. B. Lininger, N. A. Sutfin, and D. M. Walters
Willgoose, G. R., and Y. Gyasi-Agyei (1995), New technology in hydrol- (2017), Carbon dynamics of river corridors and the effects of human
ogy and erosion modeling for mine rehabilitation, paper presented at alterations, Ecological Monographs, doi:10.1002/ecm.1261.
APCOM XXV Application of Computers and Operations Research in Woods, S. W., and V. N. Balfour (2008), The effect of ash on runoff and
the Mineral Industries, Australian Institute of Mining and Metallurgy, erosion after a severe forest wildfire, Montana, USA, International Jour-
Brisbane. nal of Wildland Fire, 17, 535–548, doi:10.1071/WF07040.
Willgoose, G. R., and G. R. Hancock (1998), Revisiting the hypsometric Wu, S., R. L. Bras, and A. P. Barros (2006), Sensitivity of channel profiles
curve as an indicator of form and process in transport-limited catchments, to precipitation properties in mountain ranges, Journal of Geophysical
Earth Surface Processes and Landforms, 23(7), 611–623. Research (Earth Surface), 111, F01024, doi:10.1029/2004JF000164.
Willgoose, G. R., and G. R. Hancock (2010), Applications of long-term Wu, W., and R. C. Sidle (1995), A distributed slope stability model for
erosion and landscape evolution models, in Handbook of erosion model- steep forested basins, Water Resources Research, 31(8), 2097–2110.
ling, edited by R. P. C. Morgan and M. A. Nearing, pp. 339–359, Wiley-
Wyllie, D. C. (2014), Calibration of rockfall modeling parameters, Inter-
Blackwell, Oxford.
national Journal of Rock Mechanics & Mining Sciences, 67, 170–180,
Willgoose, G. R., G. R. Hancock, and G. A. Kuczera (2003), A framework doi:10.1016/j.ijrmms.2013.10.002.
for the quantitative testing of landform evolution models, in Predictions in
Yang, C. T. (1973), Incipient motion and sediment transport, Journal of
geomorphology, edited by P. R. Wilcock and R. M. Iverson, pp. 195–216,
Hydraulic Division – ASCE, 99(HY10), 1679–1704.
American Geophysical Union, Washington, DC.
Yeteman, O., E. Istanbulluoglu, and A. R. Duvall (2015a), Solar radiation
Willgoose, G., and G. Kuczera (1995), Estimation of subgrid scale kin-
as a global driver of hillslope asymmetry: Insights from an ecogeomorphic
ematic wave parameters for hillslopes, Hydrological Processes, 9(3–4),
landscape evolution model, Water Resources Research, 51, 9843–9861,
469–482.
doi:10.1002/2015WR017103.
Willgoose, G. R., G. A. Kuczera, and B. J. Williams (1995), DISTFW-
Yetemen, O., E. Istanbulluoglu, J. Flores-Cervantes, E. R. Vivoni, and R. L.
NLFIT: Rainfall-runoff and erosion model calibration and model uncer-
Bras (2015b), Ecohydrologic role of solar radiation on landscape evolution,
tainty assessment suite, Research Report 108.03.1995, Department of
Water Resources Research, 51, 1127–1157, doi:10.1002/2014WR016169.
Civil Engineering and Surveying, University of Newcastle, Australia,
Callaghan. Yoo, K., R. Amundson, A. M. Heimsath, and W. E. Dietrich (2005),
Erosion of upland hillslope soil organic carbon: Coupling field measure-
Willgoose, G. R., and H. J. Perera (2001), A simple model for saturation
ments with a sediment transport model, Global Biogeochemical Cycles,
excess runoff generation based on geomorphology, steady state soil mois-
19, GB3003, doi:10.1029/2004GB002271.
ture, Water Resources Research, 37(1), 147–156.
Yoo, K., R. Amundson, A. M. Heimsath, and W. E. Dietrich (2006),
Willgoose, G. R., and S. J. Riley (1998a), Application of a catchment
Spatial patterns of soil organic carbon on hillslopes: Integrating geo-
evolution model to the prediction of long term erosion on the spoil heap at
morphic processes and the biological C cycle, Geoderma, 130(1–2),
Ranger Uranium Mines: Initial analysis, Supervising Scientist Report 132,
47–65.
Australian Government Publishing Service, Canberra. Originally pub-
lished as Willgoose, G. R., and S. J. Riley (1993), Application of a Yoo, K., and S. M. Mudd (2008), Toward process-based modeling of geo-
catchment evolution model to the prediction of long-term erosion on the chemical soil formation across diverse landforms: A new mathematical frame-
spoil heap at Ranger Uranium Mine, Open File Report 107, Office of the work, Geoderma, 146, 248–260, doi:10.1016/j.geoderma.2008.05.029.
Supervising Scientist, Jabiru. Yu, C.-K., and L.-W. Cheng (2013), Distribution and mechanisms of
Willgoose, G. R., and S. J. Riley (1998b), An assessment of the long-term orographic precipitation associated with Typhoon Morakot, Journal
erosional stability of a proposed mine rehabilitation, Earth Surface Pro- of Atmospheric Sciences, 70(9), 2894–2915, doi:10.1175/JAS-D-12-
cesses and Landforms, 23, 237–259. 0340.1.
Willgoose, G. R., and S. Sharmeen (2006), A one-dimensional model for Yu, N., B. Boudevillian, G. Delrieu, and R. Uijlenhoet (2012), Estimation
simulating armouring and erosion on hillslopes. 1. Model development of rain kinetic energy from radar reflectivity and/or rain rate based on a
and event-scale dynamics, Earth Surface Processes and Landforms, 31(8), scaling formulation of the raindrop size distribution, Water Resources
970–991, doi:10.1002/esp.1398. Research, 48, W04505, doi:10.1029/2011WR011437.
Williams, B. J. (2006), Hydrobiological modelling: Processes, numerical Yu, Y. Y., P. A. Finke, H. B. Wu, and Z. T. Guo (2013), Sensitivity
methods and applications, Lulu Press, Newcastle, Australia. analysis and calibration of a soil carbon model (SoilGen2) in two contrast-
ing loess forest soils, Geoscientific Model Development, 6, 29–44,
Wilson, C. J., K. J. Crowell, and L. J. Lane (2006), Surface erosion doi:10.5194/gmd-6-29-2013.
modelling for the repository waste cover at Los Alamos National Labora-
Zaitlin, B., and M. Hayashi (2012), Interactions between soil biota and the
tory Technical Area 54, Material Disposal Area G, Rep. LA-UR-05–7771,
effects on geomorphological features, Geomorphology, 157–158,
Los Alamos.
142–152, doi:10.1016/j.geomorph.2011.07.029.
Wischmeier, W. H., and D. D. Smith (1978), Predicting rainfall erosion
Zhang, J. J., and L. R. Bentley (2005), Factors determining Poisson’s ratio,
losses – A guide to conservation planning, USDA Agriculture Handbook
Research Report 62, University of Calgary, CREWES, Calgary.
References 319
Zhang, K., J. S. Kimball, and S. W. Running (2016), A review of remote Zhong, S., A. Paulson, and J. Wahr (2003), Three-dimensional finite-element
sensing based actual evapotranspiration estimation, Wiley Interdisciplin- modelling of Earth’s viscoelastic deformation: Effects of lateral variations in
ary Reviews – Water, 3(6), 834–853, doi:10.1002/wat2.1168. lithospheric thickness, Geophysical Journal International, 155(2), 679–695,
Zhang, W., and D. R. Montgomery (1994), Digital elevation model grid doi:10.1046/j.1365-246X.2003.02084.x.
size, landscape representation and hydrologic simulations, Water Zhou, X., E. Istanbulluoglu, and E. R. Vivoni (2013), Modeling the
Resources Research, 30(4), 1019–1028. ecohydrological role of aspect-controlled radiation on tree-grass-shrub
Zhang, W., J. Niu, V. L. Morales, X. Chen, A. G. Hay, J. Lehmann, and T. coexistence in a semiarid climate, Water Resources Research, 49,
S. Steenhuis (2010), Transport and retention of biochar particles in porous 2872–2895, doi:10.1002/wrcr.20259.
media: Effect of pH, ionic strength, and particle size, Ecohydrology, 3, Zolezzi, G., R. Luchi, and M. Tubino (2012), Modelling morphodynamic
497–508, doi:10.1002/eco.160. processes in meandering rivers with spatial width variations, Reviews of
Zhang, Z. F., A. L. Ward, and J. M. Keller (2009), Determining the Geophysics, 50, RG4005, doi:10.1029/2012RG000392.
porosity and saturated hydraulic conductivity of binary mixtures, Pacific
Northwest National Laboratory, Oak Ridge.
Index
Acid mine drainage, 135 Interdecadal Pacific Oscillation, 26, 248

Aeolian transport, 96 interglacial, 6, 26, 307
Agent modelling, 16–17, 26, 299, 312 North Atlantic Oscillation, 26
Amazon, 68, 196, 297 CLORPT, 19
Angle of repose, 160, 209, 214, 219–220, 278 Coevolution, 243, 270, 287–288, 304, 311
ARMA models, 37–38 Colloid, 163, 294, 296
Armouring. See also Mixed size sediment transport models Coulomb, 199, 201, 216–217, 302
framework-supported, 64 Courant number, 255
hiding function, 63–64 Critical Zone, ix, 289, 291, 294
matrix-supported, 64 Crust
relative mobility, 63, 65 buoyancy, 193–195, 197, 201, 210
ASTER, 259, 307 compensation depth, 191
continental, 190–192, 197–198, 202–203, 303
Banded vegetation, 243, 311 flexural isostasy, 282, 314, 317
Biosequestration of C, 90 flexural parameter, 195
Bioturbation, 8–9, 90, 101, 109, 112–114, 116–117, 135–136, flexural rigidity, 193, 282
152, 154, 160–161, 167–171, 178–179, 182–183, 185–186, forebulge, 195–197
189, 253, 271, 299, 305–306, 312, 317 lithospheric thickness, 195, 198, 294, 319
ants, 109, 111–114, 116, 160, 305 oceanic, 190–192, 203
earthworms, 109–110, 305, 315 plate tectonics, 190, 293, 296
gophers, 109, 114, 299, 302 thin plate theory, 201, 302
pigs, 254, 300 Cumulative area diagram (CAD), 31, 44, 259, 267, 269, 286, 309
prairie dogs, 114
termites, 109, 113–114, 160, 305 Debris flows, 212–213, 217–219, 223, 227, 295, 307–308, 312
tree throw, 90, 92, 109–111, 114, 117–118, 178–180, collapse, 90, 111, 117, 119, 126, 174, 186, 214, 216, 218, 226
187–188, 228, 253–254, 281, 299 dilation, 90, 118–119, 188, 216, 218
wombats, 109, 114 runout, 215, 217–218, 222–223, 303
Delaunay triangulation, 203, 313
Catchment divide migration, 13, 15, 18, 29, 43–44, 85–86, Deposition-limitation, 56, 61
224–225, 282, 284 Detachment-limited erosion, 40–41, 58–59, 68, 70, 263
Cellular automata, 16, 39, 215, 245, 309 bedrock incision, 58
Char. See Charcoal detachment rate, 58, 61, 226
Charcoal, 9, 163, 165–166, 171, 244, 251, 301 Diffusion-limited aggregation (DLA), 39
Chi plot, 42 Discrete element model, 221, 223, 316
Clay, 28–29, 65, 88, 116–117, 152, 157–158, 163–166, 172, DVM. See DGVM
174–175, 183, 185–186, 217, 242, 271, 293, 298, 306 Dynamic equilibrium, xi, 40, 43, 45, 70, 72, 75, 144, 200, 259,
Cliff retreat, 223–226, 283 265
Climate variability, 4, 26, 36, 55, 72, 87, 152, 183, 236, 244, 249, Dynamic Global Vegetation Models (DGVMs), 231, 238, 254
253, 299
El Niño, 26, 248 Effective parameterisation, 18, 139
320 Indian Ocean Dipole, 26 effective parameter, 25, 71–72
Index 321
Einstein-Brown, 54, 97 Ice sheet, 196

Enrichment factor, erosion, 65, 167, 173, 235, 285 Infiltration, 8–9, 28–32, 34, 49, 73, 89, 92, 121, 125, 127–128,
Eulerian modelling, 13, 15–17, 61–62, 213, 215 132, 136–139, 141–143, 145, 148, 151, 163–164, 169–171,
Evaporation, 9, 26, 28–31, 34, 36, 38, 47, 240–242, 247, 281, 173–175, 177, 187, 218, 229, 231, 236, 238, 240–243, 252,
303, 305 254, 281, 285, 314
Evapotranspiration, 28–29, 143, 232, 238, 240, 248, 250, 305, Infiltration-excess, 29–30, 32, 73, 142
307, 319 Infinite slope stability model, 207, 209, 211
Interrill erosion, 49, 52
Factor of safety, 208–213 Inverse soil texture hypothesis, 242
Faulting Isostasy, 191–192, 195, 197
normal fault, 190, 304
offset fault, 201 Kinematic wave runoff routing, 32–33, 35, 271, 318
rift valley, 190
strike-slip fault, 190, 203 Lagrangian modelling, 13, 15–16, 61–62, 213, 215, 222
Fickian diffusion, 20, 50, 110, 135, 148–149, 152, 154, 159, 161, Landslide, 13, 15–16, 60, 179, 205–207, 209–212, 214–215, 249,
169, 185, 218, 298 259, 283, 289, 293–295, 297, 299, 301, 306, 312–313, 315–
Fire 316
Fire Danger Coefficient, 247 Leaf Area Index (LAI), 231, 233, 242
Fire Danger Index, 248 Leaf fall, 116, 235, 251
frequency-area relationship, 249 Leaf litter, 8–9, 32, 51, 55, 165–166, 177, 228, 231, 235, 251,
human initiation, 249 254, 285
Kuczera Curve, 250 LIDAR, 1, 41, 259
lightning initiation, 247–248, 292, 304 Locality of models, 16, 194
Nestorov Index, 247 nonlocal, 16, 61
Normalised Difference Vegetation Index (NDVI), 250
post-fire dry ravel, 252 Macropores, 28, 109, 117, 138, 159–160, 175, 210, 252, 254
post-fire erosion, 251 Manning equation, 32–33, 54, 72, 77
post-fire hydrology, 251 Mantle
post-fire soil carbon, 252 mantle convection, 190–191, 244, 293, 306, 308
post-fire soil structure, 252 mantle plume, 190, 199, 307
post-fire water quality, 251 viscosity, 192–193, 197–198
recovery, 249, 251–252 Markov chain rainfall models, 37–38
savannah, 252 Mars, 10, 219, 288–289, 304
Fission track dating, 283 Meandering, 6–7, 20, 75–81, 176, 288–289, 301–302, 304, 308,
Floodplains, 6, 20, 75–76, 78–81, 176–177, 301, 304 319
SOC sequestration, 175 Minerals
Fragmentation, 60, 67, 89, 96–97, 102, 106–108, 117, 121, 133, aluminium hydroxide, 141
184, 186, 216, 220–222, 227, 254, 299, 316 biotite, 115, 120, 141, 186, 298, 301, 308
feldspar, 115, 120–121, 142
Gilgai soils, 160 gibbsite, 141
Glacial Isostatic Adjustment (GIA) models, 193, 197, 199, 204, 313 iron silicate, 186
Glacial unloading, 5, 198, 225, 282–283 kaolinite, 141–142, 186
GLONASS, 1 pyrite, 135
GPS, 1, 199, 312 quartz, 115, 186, 294, 316
GRACE, 1, 193, 300, 308 secondary minerals, 8, 53, 89, 99, 119, 186, 306
Green-Ampt infiltration model, 30–31, 34, 137, 241 titanium oxides, 119
Groundcover, 8–9, 229–231, 235–236, 244, 251, 263, 282 vermiculite, 186
Gully erosion, 74, 86, 167, 254, 290, 293, 306 zircon, 119
Mixed size sediment transport models, 62
Hadley cell, 190 fractional transport models, 62
Hawaii, 195, 303, 307 Model calibration
Headward growth network model, 39 erosion plots, 263, 265
Herbivores, 231, 234–236 overparameterisation, 263
Hypsometric curve, 21, 45, 266–267, 318 paleo-reconstruction, 265
Hypsometric integral, 21, 45–46, 267, 295 rainfall simulator, 69, 96, 263
322 Index
Model equifinality, x, 3, 263, 307 SHALSTAB, 210, 297

Model testing SIBERIA, ix, xi, 40, 65, 80–81, 83, 85–86, 226, 257, 260,
confidence limits, 4–5, 44, 267–269 262–263, 265, 267, 270, 272, 286, 300, 317
t-test, 269 SIGNUM, 257, 310
validation, 2, 10, 122, 257–258, 260, 266–268, 291, SINMAP, 210, 308
300, 316 SoilGen2, 184, 298, 318
verification, 257–258 SOMKO, 167, 299
Modelling tools SSSPAM, 55, 81, 98, 102, 184, 188, 272, 290
BATEA, 269 SWAT, 52, 68
BMI, 256–257 Tellusim, 256–257
C, 1, 256–257 USLE, 35, 42, 51, 229–230, 235, 263, 312
Cython, 256–257 WEPP, 4, 51–52, 58, 64, 68–69, 86, 116, 230, 263, 304
Fortran, 1, 256–257 Monte Carlo simulation, 4–5, 25, 206, 269, 290
GLUE, 265–266, 268–269, 283 Mountain building
GRASS, 204 critical wedge theory, 199, 310
Kalman filter, 269 frontal accretion, 199–200
Matlab, 1, 38, 256–257, 290, 310 taper theory, 199
Python, 1, 38, 204, 256–257, 290, 317 underplating, 200
Models
APSIM, 237, 303, 316 Net Primary Production (NPP), 233
ARMOUR, 55, 81, 86, 96, 183, 268, 290 Numerical diffusion, 162, 224, 226
BIOME, 238, 254
CAESAR, 81, 262, 300, 315 Optimal channel network model (OCN), 40
CATGraSS, 234 Optimum bioclimatic state, 241–242
CENTURY, 163, 165–166, 168, 295, 301, 306 Orography and rainfall, 27, 317
CHILD, 81–82, 161, 226, 257, 314
CREAMS, 4, 42, 51, 116, 303 Particle tracking, 16–17, 317
CrunchFlow, 121, 123, 139, 313 Pedogenesis, 10, 88, 97–98, 110, 183, 185, 187, 189, 209, 292,
DELIM, 81 294–295, 301, 303, 306, 310, 316
dSLAM, 210 Pedon, 2, 270–271
FARSITE, 245–248, 298 Pedoturbation, 109–110
gFlex, 204, 317 Penman-Monteith equation, 28, 240
GOLEM, 42 Perturbation analysis, 22, 70
HYDRUS, 31, 137, 174 Philip infiltration model, 31, 241
IBIS, 295 Plants
ICE-3G, 193, 314 C3 photosynthesis, 242
Landlab, 257, 301 C4 photosynthesis, 242
LAPSUS, 231, 256, 271 CAM photosynthesis, 242
LEACHC, 184 carbohydrates, 240
LORICA, 271–272, 314 functional types (PFT), 238
LPJ, 177, 234, 238, 248, 254, 312 nutrient limitation, 187
mARM, 81, 97–99, 102–103, 118, 162, 183–185, 188, 277, recruitment, 235, 237, 244
290, 295 respiration, 9, 30, 88, 120, 122, 133–134, 179, 231, 233, 238,
MD-STAB, 211–212 296, 314
MILESD, 118, 188, 271 root zone, 28, 30, 34, 110, 136, 138, 241
MUSLE, 51 stomata, 240, 242
PESERA, 52, 247–248, 298 Plateau, 223, 225
PHREEQC, 140, 315 Pluto, 10, 288, 305
QUEL, 42, 45, 267, 288 Poisson ratio, 194
REAR, 204, 306 Pore pressure, 210–211, 213, 218–219, 223
Reg-FIRM, 247–248 Porosity, 11–12, 30, 88–90, 109–110, 112, 116–118, 137, 139,
RockyFor3D, 220, 222 142–143, 174–175, 178–179, 188, 205, 209, 241, 292, 294,
RUSLE, 4, 51, 229, 263, 297 298, 307, 319
RUSLE2, 229–230, 314 Precipiton, 15, 213, 215, 296
SGS, 237, 303 Pumice, 220, 295
Index 323
Radioisotopes Soils
10
Be, 202, 317 aggregates, 49, 65, 116, 172, 174, 235, 252, 314
137
Cs, 170 amino acids in SOM, 116, 167
14
C, 168, 172 bulk density, 11, 28, 88–90, 92, 102, 109, 113, 116–117, 155,
210
Pb, 170 167, 169–170, 175–176, 178–179, 183, 185–187, 208–209,
cosmogenic nuclides, 17, 91–92, 154, 160, 178, 310, 317 213, 216, 245–246, 292, 314–315
Rainflow, 49, 52, 230 carbohydrates in SOM, 116, 167
Rainsplash, 26, 49–52, 58, 86, 111, 160, 163–164, 175, 229–230, coevolving soils, 19
235–236 decomposition rate aging, 166–167
Ranger Mine, ix–x, 96, 102, 260, 288, 293, 298, 310, 317–318 fast soils, 19
Rayleigh-Benard cell, 190 humus, 116, 163
Regolith, 58, 88, 90, 179, 294, 296, 298–299, 303, 316 lignin in SOM, 116, 165–167
Restitution coefficient, 220–222, 293–295, 297 organic carbon (SOC), 65, 163–175, 177, 187, 237, 252,
Rheology, 88, 155, 157–158, 201, 204, 288, 300, 310 284–286, 296, 301, 303, 310, 313–314, 316, 318
Bingham fluid, 155, 157–158, 217, 316 organic matter (SOM), 2, 8–9, 65, 88, 120, 122, 163, 165,
granular flow, 217 167, 171–177, 179, 182, 187, 228, 235, 238, 241,
Herschel-Bulkley fluid, 155 252–254, 271, 281, 285–286, 293, 299, 301–302, 309,
Maxwell fluid, 197 312–315
Newtonian fluid, 155–158, 197, 292–293 slow soils, 19
viscoplastic, 155, 158 SOC dating, 172
viscous flow, 13, 152, 155–156, 158, 162, 179, 190–193, 197, structure, 8–9, 28, 92, 117, 163, 173–175, 243, 251–253, 285,
202, 204, 216–218, 283 292, 306
Richards equation for infiltration, 31, 137, 184 water-holding capacity, 88–89, 173, 181, 187, 228, 237, 241,
Rill erosion, 52, 155, 237 281
Rockfall, 219, 315 Solifluction, 96, 152, 158
Spreading centre, 190
Saprolite, 8, 11–13, 60, 80, 88–90, 92, 117–119, 123–124, 126, State-space modelling, 14, 99, 115, 167, 183
128, 134, 136, 143–144, 146–148, 150, 152–155, 158, 178– Strahler network classification, 39
179, 181–182, 184–185, 188, 205–206, 208–211, 216, 254, Subduction, 190–191, 195
270, 272, 304
Saturation-excess, 29–31, 73, 96, 143, 210, 259, 279, 318 Terrestrial crust. See Continental Crust
SCORPAN, 19 Time of concentration, 29
Scree slope, 223, 226 Tin Camp Creek, 19, 269, 291, 300
Sea level, 6, 18, 118, 188, 190, 193, 197, 308, 313–314 Titan, 288, 294, 307
Sediment transport Tokunaga network classification, 39, 44
bedload, 53, 55, 60, 62, 265, 307–308, 314 Tools-Cover model, 59–61, 66–67, 314
dissolved load, 6, 8, 53, 119, 143 Topologically distinct random channel networks (TDRN), 39
suspended load, 53, 265 Transpiration, 9, 28–29, 34, 136, 138, 233, 238, 240–241, 254,
Sheet erosion, 73–74, 155 307
Shuttle Radar Topography Mission (SRTM), 1, 82, 259, 300, 307 Transport-limited erosion, 40, 53, 58–59, 69, 76, 83, 271–272,
Slope-area relationship, 31, 33, 40–43, 45, 52, 60, 72–73, 75, 282
144, 184, 201, 259, 263, 265, 267, 269, 278–279, 283, 286, Triangulated irregular network (TIN), 11, 82–83, 203, 226,
288 310
Soil atmosphere, 30, 88–89, 134, 141, 165, 182, 233
Soil moisture, 9, 26, 28, 30–31, 34, 36, 87, 90–92, 99, 138, 157, Vegetation canopy, 8–9, 51, 229–235, 238, 244, 246, 250, 252,
159, 163–166, 173, 177, 181, 184, 211, 231, 248, 251, 254, 263
270, 280, 282, 297, 316, 318 Vegetation cover factor, 51, 229, 231
field capacity, 28, 34, 137–138, 175
wilting point, 28, 30, 34, 138, 175 Water-limitation, 187, 236, 238, 240, 242, 280–281
Soil production function (SPF), 91–95, 99, 144, 146, 153, 158, Weathering
178–179, 182, 185–186, 270–271, 299, 301 carbonation, 116
exponential SPF, 92, 144, 182, 185, 270–271 congruent weathering, 122
humped SPF, 91–95, 125–132, 136, 139, 146, 148–149, 151, hydration, 89, 96, 116
179, 185, 272–273, 299, 301 incongruent weathering, 122
reversed exponential SPF, 134, 272, 276 isovolumetric weathering, 126, 308
324 Index
Weathering (cont.) specific surface area (SSA), 42, 88, 120–121,

leachate, 8–9, 123, 125–126, 128–131, 134–135, 137–138, 124, 132–133, 141, 185, 271,
140–141, 143, 145–147, 181, 184 282, 301
oxidation, 116, 135, 186, 294 supply-limited, 126
particle roughness, 120 transport-limited, 125
rate-controlled, 125 Weathering-limited erosion, 67
saturation index, 129 Width function, 22, 33, 46

Principles of Soilscape and Landscape Evolution ( PDFDrive )

Uploaded by

Copyright:

Available Formats

Principles of Soilscape and Landscape Evolution ( PDFDrive )

Uploaded by

Document Information

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

Principles of Soilscape and Landscape Evolution ( PDFDrive )

Uploaded by

Copyright:

Available Formats

Principles of Soilscape and Landscape Evolution

Computational models are invaluable in understanding the

G A R R Y W I L L G O O S E is a professor in the Faculty of Engineering

If it moves, model it! There is no better synthesis of all the

Cambridge University Press is part of the University of Cambridge.

2 Constructing a Landscape Evolution Model – Basic Concepts 11

3 A Brief Hydrology and Geomorphology Primer 26

4 Erosion and Other Water-Driven Processes 49

5 Soils: Constructing a Soilscape Evolution Model – Basic Concepts 87

6 Soils: Soil Depth 90

7 Soils: Physical Weathering and Soil Particle Fragmentation 96

8 Soils: Chemical Weathering 119

9 Soils: Slow Soil Flow and Creep 152

10 Soils: Colloids and Soil Organic Carbon 163

11 Soils: Constructing a Soilscape Evolution Model – Details and Examples 178

12 Tectonics and Geology 190

13 High-Slope Gravity Processes 205

14 Vegetation and Wildﬁre 228

15 Constructing a Landscape Evolution Model – Details 255

16 Examples of Landscape Evolution Modelling and Final Thoughts 270

FIGURE 1.2: A schematic of the

(b) soil organic matter and carbon transport, (c) water

2.1 Overview for erosion, the sign convention deﬁned to be consistent

FIGURE 2.1: Schematic of the

Stþ1 ¼ St þ ðU PÞjt¼t Δt (2.11)

zero erosion anywhere since ∂q ∂x ¼ 0 with x positive in the

crushed rock that can break down rapidly and faster

h between a and ^ a is simply a function of the nonlinearity of

2.4.2.4 Example 4: Variance of a Single

An alternative derivation can be performed for catch-

E ¼ K 2 Qm2 Sn2 (3.20)

When we substitute Equation (3.17) for discharge and

(3.20). All that is required is that the erosion processes

and following the derivation above for dynamic At ¼ ðn Þ =n

that is concave down everywhere. For hillslopes where

soils cemented by secondary minerals) and entrainment 4.2.1 Transport-Limited Erosion

longer time scales. It is thus reasonable to assert, given ∂z

FIGURE 4.6: Contours of

experiments (Bagnold, 1936). The main difference is that

4.3 Temporal Averaging of Erosion

The transport-limited and detachment-limited erosion

qp qp 3 qp T p It should be noted that while Willgoose et al. (1989,

The solution can be generalised to the case where m 6¼ 2

sensitivity to Manning n. If one of these equations deﬁnes 4.4.3 Temporal Dynamics

and f are typically about 0.5 and 0.4, respectively (Millar,

stream in the channel (v  K v Q0:1 ; Equation 3.4). Substi-

(a) (b) n i+1

Once the grid discretisation is determined, the next

1. Determine the D8 directions ∂z ∂q ∂ ∂  m mδ n

FIGURE 4.14: The schematic of

diffusion (Equation 9.20). Perron used a Taylor series

5.1 Overview dioxide concentrations in the atmosphere ranged from a

6.1 Introduction this chapter typically assumes that a distinct interface

changes to bioturbation and related processes, which will

FIGURE 6.2: As soil thickness

0.4 FIGURE 6.3: Model results

0.3 the Oregon Coast Range. The

Weathered 7.3.1 Dynamics of a Single Soil Layer

• Adjusting the layers in response to erosion at the sur-

shown. While all the examples in Table 7.1 have no mass

h between a and ^ a is simply a function of the nonlinearity of

stream in the channel (v K v Q0:1 ; Equation 3.4). Substi-

1. Determine the D8 directions ∂z ∂q ∂ ∂ m mδ n

where the right-hand side is the rate of change with