C - Y W M: Fao-Wmo R S
C - Y W M: Fao-Wmo R S
C - Y W M: Fao-Wmo R S
ON
BY
RENÉ GOMMES
1. CHAPTER ONE : INTRODUCTION ...............................................................4
1
Strictly: variance.
2
Throughout this manual, the word management is used as a shortcut for on-farm management
decisions by farmers.
3
There is no trend either in some developed countries where yields are close to their potential
values and where any yield increase, though technically feasible, would be uneconomical.
Sakamoto and his colleague, in particular indices (Walker, 1989), are still being
used today, as they correspond as well to specific input data timeliness and user
demand requirements. Some new methods have been developed since the early
eighties, but it is difficult to say where the major methodological developments will
lead.
The author’s view is that we will witness an evolution along five axes:
• development of scale-specific models. Although almost all current models were
originally developed for modelling a crop at the scale of a field, some of them
are being used at the level of countries and even of the planet in global change
impact studies. The inputs used are almost meaningless at the national and
regional scales, even if - and sometimes precisely because - they have been
gridded. Model use for a particular purpose depends on whether its complexity
is appropriate to the question being asked and whether the model has been
tested in diverse environments (Boote et al., 1996). It is suggested that there is
an optimal model type with an adequate time step and number of input data at
all geographic scales. It could even be argued that the most detailed models will
be replaced at the farm level by direct sampling of assimilation rates and other
parameters (Fredrick and Lemeur, 1997), and that models could in the future
loose importance at that scale;
• development of non-parametric and rule based models, because of their
simplicity of calibration and use, and low cost . On the scale of models outlined
by Sakamoto and LeDuc, they come after deterministic models. A parallel
progress will be process-oriented models with outputs as probability density
functions. Both developments will be possible because of the availability of
faster computers;
• integration of real-time inputs in the models, provided either by direct (and
possibly automatic) measurements on crops and weather at ground stations, or
through aircraft or satellite remote sensing. We currently witness this evolution
for Leaf Area Index and evapotranspiration at the regional scale. This will
replace some of the model components which currently estimate the same
parameters;
• a greater inter-compatibility of models (Maracchi and Sivakumar, 1995). Users
should be in a position to assemble their models by using building blocks
(modules), for instance the root component of CERES, and the assimilation
block of WOFOST. The first step should be the harmonisation of data files in the
direction of a universal self-documented format;
• development of models which can take into account, at various scales, the
effect of weather factors which physically harm plants (frost, sandy wind, very
strong winds), and a better integration of crop, pathogens and competitors.
The current lecture notes are meant as a general introduction to the subject of
agrometeorological crop modelling. As indicated in the title, the focus is on the
effect of weather (as opposed to other factors) and on applications which fall
traditionally in the province of agrometeorology (as opposed to many other fields
like plant breeding).
After the present introduction (1), the volume is subdivided into 6 main chapters :
(2) Fundamentals; (3) an overview of model types; (4) quality checking of models;
(5) some method and tools for operational crop modelling; (6) agrometeorological
applications of models; (7) exercises, followed by the bibliographic references.
Chapters are arranged following a logical sequence, but the author is well aware
that several alternative options could have been chosen. For instance, chapter 3
(Overview of model types) could have been dealt with before chapter 2. Chapter 5,
dealing with special inputs such as those provided by remote sensing could have
been inserted in 3.5 (3.5.3. to 3.5.5), etc. This apparent redundancy, combined
with frequent cross referencing was maintained to allow a more flexible use of the
material presented. For instance, a less technical presentation of the subject could
skip section 2.2.
Fundamentals are subdivided into (i) the basic ecophysiological aspects, such as
water in plants, the conversion of radiant energy into biomass energy and the links
between CO2 assimilation and water use; (ii) the second part of the chapter deals
with model components, i.e. essentially with the practical implementation of the
basic ecophysiological aspects.
The overview of model types is subdivided into several somewhat arbitrary and
overlapping categories. Global biomass models, including the issue of potential
biomass determination, are followed by a short note on vegetation models and
statistical models. Next comes a note on the semi-empirical models used by FAO
for regional crop assessments. Simulation models proper, the process-oriented
models, receive the most detailed treatment : some of the best known models are
described, together with a short historical note and main features (CERES,
WOFOST, EPIC), starting with CropSyst, on which particular emphasis will be put
in the exercises.
Quality checking of models describes the procedures and caveats that a potential
model user must be aware of before a model can be implemented in a decision-
making process. This includes the general aspects of model evaluation,
verification, etc.
Under some methods and tools for operational crop yield modelling we insist on
more specific aspects like the growing number of uses of remote-sensing in
operational applications, area averaging and a short introduction to random
weather generators.
Agrometeorological applications of models elaborate on the potential uses of crop
models in agrometeorological practice, of which crop forecasting appears to be
one of the most important from a economic point of view. Two sections deal with
farm-level applications and institutional customers, like crop insurance.
The exercises provide hands-on experience with models. They include an annex
which provides a reminder about some of the physical units which are omnipresent
in crop modelling. The exercises focus on some specific aspects covered in the
lecture notes proper, and, conversely, some of the aspects presented in chapter 2
(fundamentals) were chosen because they easily lend themselves to a practical
implementation in exercises. The students will be provided with a diskette having
the relevant software and data.
Almost all exercises will be implemented in a spreadsheet. It is therefore essential
that the students be familiar with basic spreadsheet use, graphing, the use of
named cells, and regression.
The first exercises present some what-if experiments with basic equations taken
from the manual in order to show the influence of the parameters.
For the practical work with a model we use CropSyst, developed by C. Stöckle at
Washington State University (Stöckle et al., 1994 and other papers co-authored by
Stöckle; see references). CropSyst was selected for several reasons: CropSyst is
an up-to-standard model provided with a random weather generator; it is very
user-friendly and exists for DOS and Windows 954; it is multilingual; its is
completely free and can be downloaded from the WWW site of the Biological
Systems Engineering Department of Washington State University.
The current notes owe a lot to the participation of the author for several years in
SuGrAm, the Support Group for Agrometeorology of the Joint Research Center of
the European Union (JRC, Ispra, Italy), a think tank established to provide
expertise to the EU MARS programme (Monitoring Agriculture by Remote
Sensing); see Vossen and Rijks, 1995; Dallemand and Vossen, 1995; Rijks,
Terres and Vossen, 1998). MARS constitutes arguably the most advanced
regional crop forecasting system currently in operation, and the source of
significant methodological progress.
The author is indebted to several colleagues who have kindly accepted to provided
critical comments or specific inputs to this document : Michele Bernardi, FAO
Agrometeorology Group; Cédric Gommes, ANAST-HémoLiège, University of
Liège, Belgium; Graham Russell, Institute of Ecology & Resource Management
University of Edinburgh, United Kingdom; Jin Chul Shin, National Crop Experiment
Station, R.D.A, Suwon, Korea; Claudio Stöckle, Biological Systems Engineering
Department of Washington State University, USA; Iwan Supit, MARS project,
Institute of Space Applications, EU Joint Research Centre, Ispra, Italy; Bernard
Tychon, Fondation Universitaire Luxembourgeoise, Arlon, Belgium. Their
suggestions and contributions have allowed to improve the document and to
remove some inconsistencies and errors. Needless to say, the author accepts full
responsibility for remaining problems.
4
Contrary to other models, the DOS and Windows versions are functionally equivalent. It has been
the practice of many model developers to stop upgrading the DOS versions!
2. Chapter two : Fundamentals of crop modelling
2.1 Introduction
We start this introduction by borrowing the definition of a system from an old
volume of Schaum’s Outline Series: a system is an arrangement of physical
components connected in such a manner as to form and/or act as an entire unit
(Distefano et al., 1967)5. Crop models assume that plants are systems: they
behave as an entire unit. Roots feed the leaves mineral substances and water;
leaves assimilate CO2, and the stems provide the link and transport of substances
between the two, so that when leaves grow, roots can grow as well, and when
water is absorbed by the roots, it is made available to the whole plant and
transpired through the above-ground parts. The definition of a plant as a system
thus makes sense.
The structure of a system is the set of links and connections between its
components. Depending on the distance from which a system is looked at, the
level of details can vary tremendously. Models attempt to describe the structure
and the functioning of a system. Again, the word “description” can go from a
simple verbal description to a mathematical description (rarely going beyond
difference equations). When the mathematical description is coded into a
computer programme that will compute the evolution of the numerical values
attached to the model components, we talk of computer simulation : the machine
mimics the behaviour of reality at a given resolution and with a given time step.
The resolution encompasses not only a spatial scale, but also the correlative level
of detail of plant organs or cell components, biochemical and physiological
processes, etc. There is also a link between the time step and the spatial scale.
Before we proceed, we must observe that so-called crop-weather models are
more than just a description of the plant system. All of them include at least a soil
component where the behaviour of water and possibly nutrients and organic matter
is described. The system is thus larger than just a plant.
Models are characterised by state variables, rates, derived variables,
parameters, inputs and outputs.
State variables, a concept borrowed from physics, are variables which completely
describe the state of a system. This may be biomass (Wd, biomass of day d in g
m-2 ) , for instance. Rates express the speed at which the state variables change,
for instance ∆W is the rate of change of biomass in g m-2 d-1 .
Wd +1 = Wd + ∆W (1a)
Derived variables are computed from state variables, but they do not have the
same fundamental nature. For instance, if the density of leaves (expressed by the
5
In a broader sense, a system can also be a structured organisation of people, theory, methods
and equipment to carry out an assigned set of tasks, as in crop forecasting system when
considering all the logistics and institutional arrangements involved in crop forecasting.
Leaf Area Index, in m2 of leaves per m2 land area), is computed from biomass,
LAI is obviously less “central” in the model then W.
Assume that ∆W is a simple (and unrealistic) function of W and LAI:
∆W = K × LAId × Wd (1b)
where the proportionality factor K would have the unit m2(ground) m-2 (leaf) d-1 . K
would be a parameter. It would probably be crop specific and adjustable by the
user of the model. Next to crop parameters, there are also soil and other
parameters.
Finally, input variables are those which characterise external action on the system:
they include essentially the weather and management.
The order of the calculations 6 would thus be:
1. Start with Wd, the biomass at the beginning of day d
2. Compute LAId from biomass Wd, as well as all other variables that
depend on W
3. Compute ∆W= K LAId Wd
4. Compute the biomass at the beginning of the next day Wd+1 = Wd +
∆Wd
5. Start again from step 1 at the beginning of the next day.
Outputs are simply the variables which the user decides to look at. Some of them
will be used for model verification and validation, others, like yield, are of
immediate practical relevance.
6
This is in fact a difficult issue when models are complex and subdivided into modules (Supit et
al., 1994). The state variables are usually treated as global variables (always accessible
throughout the programme), while other variables tend to be “local”.
2.2 Ecophysiological aspects
Resistors in series r1 r2 r = r1 + r2
r1
Resistors in parallel 1/r = 1/r1 + 1/r2
R2
If we apply this to water flow in plants, we can write that the uptake of water through
the root surface is given by
7
This section is largely based on Hillel (1971) and Kramer (1983).
Water uptake =
Water potential in soil − Water potential at root surface (4)
Soil resistance
which we can rewrite, using the symbols defined in figure 1, as
PS − PR
Water uptake = (5)
rR
The water potential is a measure of how strongly water is bound to its medium. The
binding energy can be expressed in energy units per mass units (joule kg-1 ) or,
alternatively, as energy units per volume (joule m-3 ) which is equivalent to a
pressure (1 Pa is equivalent to 1 J m-3 and to 1 N m-2 ). Refer to the section on
units for details on the conversions (after the exercises of chapter 7). The unit of rR
depends on the units adopted for P S - P R. It usually has the dimension of s m-1 .
If the potential is very low (very negative), water is bound very strongly to the
constituents of its medium, and it will move to another medium only if its potential is
still lower (more negative).
Obviously, as there is no significant accumulation of water in a growing plant, but
rather a steady flow8, and the amount of water transpired equals the amount
absorbed through the root. We can thus write
PS − PR PR − PL PL − PA
Water flow = = = (6)
rR rP rA
In the older literature and in soil science texts, water potential is often expressed in
pressure units (bars and pascals, and their multiples and subdivisions, in particular
the MPa).
Soil water potentials are of the order of -0.1 to -15 bars (-10 to -1500 J Kg-1 ),
corresponding approximately to field capacity and to permanent wilting point. The
water potential values found in leaves are around -2000 J.Kg-1 (equivalent to -2
MPa). The water potential (also referred to as tension, suction, etc.) tends to be
high in the soil (not very negative and close to close to 0), and very low in the
atmosphere (very negative). Of the values of potential difference used in the
equation above (P S - PR, PR - PL and PL - PA ), the last is very significantly larger
than the others, sometimes by a factor of hundred, thus indicating the importance
of the atmospheric water demand). Typical values are 1000 J Kg-1 for PS - PR and
PR - P L but 50 KJ for P L - P A.
The water potential of air9 can be approximated by
RgTK e
PA = ln (7)
V e0
where Rg is the gas constant, TK the absolute temperature, V the partial molal
volume of water, e the actual water vapour pressure and e0 is the saturation water
vapour pressure at temperature TK. e/e 0 is thus the relative humidity.
Substituting the constants in the above equation (Rg = 8.31 J mol-1 K-1 , V=0.018
Kg mol-1 , TK = 273.15 + TC) leads to
e
PA = 461.66 (27315
. + TC )ln (8)
e0
This reduces to
e
PA = 135
. 105 ln (9)
e0
8
The amount of water held in plant tissues is normally very small compared with the amounts
transpired.
9
With SI units (Kg,m,s,K), this expression is equivalent to a pressure in Pa (Energy RT, divided
by a V). The Nernst equation states that one mole of water has an energy of Rg TK ln(e/e0),
equivalent to the sum of the energy of the bonds. To express this to energy per weight (J Kg-1), it
must be divided by the molar mass of water (0.018 Kg).
J Kg-1 at a temperature of 20°C. As an example, the water potential
corresponding to 80% relative humidity (ln(e/e 0)=ln(0.8)=-0.223) at 20°C is -30105
J Kg-1, while it drops to -217.8 KJ Kg-1 at 20 % moisture.
According to Kramer (1983) this equation greatly exaggerates the drop in potential
in the vapour phase between leaf and atmosphere. For instance, halving relative
humidity decreases vapour pressure by about 50%, but reduces the equivalent
water potential gradient nearly 25-fold.
There is no need, in this context, to insist on the components of the water potential.
For instance, the soil water potential is traditionally subdivided into osmotic,
pressure (or matric), capillary and gravitational.
In fact, the potential of free water (for instance the solution used for hydroponic
crops) and the potential of leaves when they are in the dark is nil.
In addition, plants can actively control their transpiration in response to their
environment, for instance by adjusting the opening of their stomata (figure 5, 2.2.4).
Plants do not only evaporate water, but also assimilate carbon, which is also
absorbed through the stomata. This is only one of the numerous interactions
between the various physiological functions of crops (2.2.3 for details). The
resistances as well vary a lot, even over short time intervals. For instance, soil
resistance is a function of soil water content and root development.
The leaf and air resistances are controlled by stomatal opening, water pressure
deficit and air movement, as in the equation below
where Cleaf and eleaf are the water vapour concentration and vapour pressure,
respectively, at the evaporating surfaces within the leaf and rleaf is the additional
diffusive resistance of the leaf. Transpiration is in Kg (water) m-2 s-1 , the
concentrations C are in Kg (water) m-3 and the resistances are in m s-1 . As the
vapour pressures are in hPa, the factor ku is symbolically added to account for the
differences in units between e and C.
2.2.2 CO2 assimilation10: conversion of radiation energy to biomass
energy
The radiant energy reaching the upper limit of the atmosphere on a horizontal
surface is known as Angot’s value (RA). It is also the amount of solar energy that
would reach the ground in the absence of the atmosphere.
Figure 2 shows that its average value is about 300 GJ Ha -1 day-1 between -60°
degrees southern latitude and 60° northern latitude, where most of the world
agriculture is concentrated. When RA is expressed in energy per year, the average
becomes 300 x 365 = 109500 GJ Ha -1 year-1 which we can round to 100 TJ Ha -1
year-1 . The detail of the calculations to obtain figure 2 is given in the exercises
under 7.2.
400
350
Equator
300
20 N
250
20 S
200
40 N
150
100 40 S 60 N
50 60 S
0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Only about half the radiant (solar) energy RA that reaches the upper limit of the
atmosphere actually reaches the ground (as RI) where it is available, among
others, for plant photosynthesis.
This is expressed by the well known formula by Ansgtröm:
n
RI = RA (a + b ) (11a)
N
10
Sections 2.2.2 and 2.2.3 are mainly after Russel (1977), de Wit et al. (1978) and van Keulen and
Wolf (1986).
where a and b are the empirical “Angström coefficients” which vary as a function of
type of climate, and n/N is the sunshine fraction, i.e. the actual hours of bright
sunshine divided by daylength. Values of a and b are close to 0.25 and 0.45 for dry
tropical areas (a is often taken as 0.29 cos ϕ, where ϕ is latitude). When n/N=0
(complete overcast), all light is diffuse. The maximum value for RI is about 70% of
RA.
The classical energy balance is usually written
RI (1-s)- B = H = P+ G+ A+ E (11b)
(all variables in units of W m-2 or J s-1 m-2 ) were s is the albedo (about 0.15 to 0.40:
15 to 40 % is reflected and radiated back in the atmosphere) and B is the
outgoing long-wave radiation leaving the surface11. As indicated, RI.s + B make
up about 50 % of RI. The difference H (the net radiation) can be used for other
processes such as photosynthesis (P, energy stored in biomass chemical bonds),
the evaporation of water from crop and soil (E), and for heating the crop and the
ground below it (G), as well as heating the air in contact with the soil and the crop.
The latter energy is known as “sensible heat” (A).
The maximum proportion of the net radiation H that can be used for photosynthesis
P is about 15 % over short periods of time, provided that the crop is well supplied
with water and perfectly managed ( P / H < 0.15). Over weeks or crop cycles, the
efficiencies are usually below 2 %, and 1% is often regarded as an average for the
fraction of the net radiation that is actually converted into biomass energy, to the
extent that the net radiation is often regarded as the sum of just evaporation by
crops and soil (i.e. evapotranspiration: latent heat) and sensible heat (equation
61).
Although photosynthesis does not play a significant role in the radiation balance, it
constitutes the source of all solar energy stored on earth, and the only source of all
food. Photosynthesis is directly driven by light. Net carbon dioxide absorption
during photosynthesis follows a characteristic light response curve (figure 3) given
by de Wit et al. (1978) as a function of absorbed photosynthetically active radiation
(PAR), RHC (see 3.1.1)
Elc. RHC
Fn = Fd + ( Fm - Fd )(1 - exp(- ))
Fm (12)
Where the symbols have the definition and units given in table 1.
Note that equation (12) gives short-term net assimilation in Kg CO2 Ha-1 (leaf)
hour-1 , thus assuming a huge horizontal leaf with an area of 1 Ha. The radiant flux
in the 400-700 nm range corresponds to the so-called Photosynthetically Active
Radiation (PAR), roughly equivalent to 50% of global radiation at ground level. The
Efficiency at the light compensation point is the slope of the Fn curve at the light
compensation point.
11
B stands for Brunt, the name of the scientist who computed B as a function of air temperature
(σT4), air moisture and n/N.
Table 1: Some characteristics of photosynthesis as driven by light, as a function of
the C3/C4 type12.
C3-plants C4-plants
Net assimilation Kg CO2 / Ha leaf / Fn
hour
Maximum rate of net Kg CO2 / Ha leaf / Fm 30 60
assimilation hour (15 to 50) (30 to 90)
Net assimilation in Kg CO2 / Ha leaf / Fd -3 -6
the dark hour
Absorbed radiant joule / m2 / s RHC
flux in the 400-700
nm range
Efficiency at light Kg C02 / Joule Elc 0.25 0.30
comp. point
Temperature- No Yes
dependent F m ?
The curve below illustrates the equation. Above the light compensation point,
assimilation exceeds respiration, and there is thus a net biomass accumulation .
C3 plants respond better at low intensities, but rapidly reach light saturation; C4
plants, which are more typical of tropical regions and include millet, sorghum,
sugar cane and maize, continue increasing the assimilation over a much large
range of radiation.
12
The majority of plants fall into one of the C3 or C4 categories according to the number of carbon
atoms of one of the first acceptors of CO2 during photosynthesis. C3 plants respond well to
increased atmospheric CO2, while C4 plants, mainly tropical grasses (maize, sorghum,
sugarcane. millet) or halophytes, respond better to higher temperatures. Most agricultural plants
(cereals, legumes, vegetables) are adapted to lower temperatures and sunshine; they belong to
the C3 group. WUE tends to be higher in C4 plants.
Figure 3 : Net assimilation of carbon dioxide per hectare of leaf in C3 and C4
plants as a function of absorbed PAR (RHC). The values read from the graph for C3
plants at 400, 40 and 4 W m-2 are 28.82, 6.35 and -1.92 KgCO2 hectare-1 hour-1,
respectively. For 220 and 100 W m-2, they amount to 24.72 and 15.66 KgCO2
hectare-1 hour-1.
60
50
C3 C4
40
kgCO2/Ha/h
30
20
10
Radiation ( W/m2)
0
0 50 100 150 200 250 300 350 400 450 500
-10
2.2.3 From leaf to canopy
There are some major methodological difficulties linked with the translation of the
net photosynthesis at the level of a leaf (as above) to photosynthesis of an actual
crop. This is a typical scaling problem, of which there are many examples in the
general area of climate impact on agriculture !
We mentioned above that the equation, as written and with the units used,
describes photosynthesis of a huge leaf having an area of 1 Ha. This is to say that
the ratio between the leaf area and the ground area is exactly one, i.e. the Leaf
Area Index, or LAI, is 1. For actual crops and wild vegetation, values significantly
higher than 1 can be found.
In the next paragraphs, we follow the description of the scaling problem as given by
van Heemst (1986). We assume that the discussion below refers to
photosynthetically active radiation (PAR, the portion of the visible between 400
and 700 nm).
10 % of the PAR that hits the leave is reflected (mainly in the green part of the
spectrum: this is why leaves are green), 10 % just crosses the leaf, another 10% is
absorbed by various other pigments that play no part in photosynthesis, and the
remaining 70% is absorbed by the photosynthetic machinery in the leaf.
90
80
70
60
50
40
-0.7771LAI
T = 100e
30
20
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Leaf Area Index m2/m2
R = R0 e ( − k LAI )
(13)
where R is radiation at “depth” LAI in the canopy and R0 is the incident radiation at
the top of the canopy. k is the extinction coefficient. Note that the extinction
coefficient is larger for PAR than for visible light (compare with equation 40).
In theory, it is thus easy, knowing LAI and the extinction coefficient, to compute the
actual photosynthesis. Referring to figure 4, we note that 45 % of the light is still
available at the lower face of the first leaf (LAI = 1). This means that 55% has been
absorbed in the first layer. Of course, we know that part of this will be lost, but we
will ignore this for the time being. The second layer thus receives 45 % of the
incoming radiation, and transmits 20 (of the incoming radiation) to the lower layers,
thus retaining 25%.
It is now possible to compare the photosynthesis of the two layers with the one-
layer crop and the “superposed leaf” crop: still assuming that the available energy
was 400 W m-2 , the first layer has absorbed 55% of 400, thus 220 W m-2 , and the
13
About 10 % because some of the 10% that leaves the first leaf will be reflected etc. by the
second leaf. Strictly, only 70% of the 10% transmitted should be taken into account.
second 25% of 400, thus 100 W m-2 . The corresponding CO2 assimilation
amount to 24.72 and 15.66, thus a total of 40.38 KgCO2 hectare-1 hour-1 .
To summarise, we have examined three situations:
• one leaf, with a net assimilation of 28.82 KgCO2 hectare-1 (leaf) hour -1
• two superposed leaves, yielding 35.17 KgCO2 hectare-1 (leaf) hour -1
• “normal” leaves absorbing 40.38 KgCO2 hectare-1 (leaf) hour -1
It is clear that the real-world situation is the most efficient. Of course, in the words
of van Heemst (1986, p. 18), the reality is more complicated, as the influence of
direct and diffuse light, total leaf area, leaf angle distribution, leaf optical
properties and solar height on the light distribution within the canopy have to be
taken into account.. .
The light interception also varies according to the growth of the crop (Rosema et
al., 1998a). In the early stages, it can be assumed that the amount of light
intercepted is proportional to biomass. This corresponds to the exponential growth
phase, which persists until the surface is completely covered by vegetation.
In practice, there does not appear to be a simple analytical solution to the
computation of canopy photosynthesis based on leaf photosynthesis. An
interesting feature is that the multi-layered canopies are a mechanism to expose
more leaves to less intense radiation, i.e. non-saturating radiation. One of the
consequences is also that when figure 3 is drawn for actual crops (i.e. KgCO2
hectare-1 (field) hour -1 instead of KgCO2 hectare-1 (leaf) hour -1 as above), the linear
part of the curve extends into higher radiation intensities.
There are many techniques to estimate total canopy assimilation; they differ by the
vertical resolution in the canopy (number of layers, from 1 to many) and by the time
step adopted (typically from hours to 1 day), and by other details. WOFOST, for
instance (Supit et al. 1994), adopts three “horizons” in the canopy, corresponding
to 0.11 LAI, 0.5 LAI (mid-height) and 0.89 LAI, close to the ground. The total
canopy assimilation is obtained as the weighted average of the three “typical”
horizons above. WOFOST also computes light distribution and rates of
assimilation at different depth separately from direct and diffuse light.
Needless to say, the “other details” constitute the very core of the actual complexity
of the models and largely condition the input parameters required to operate
them. They also seriously complicate the model validation.
14
Also called stomata (plural) and stoma (singular).
opening, the stomatal aperture (Figure 5). The cuticle is a wax-like substance
covering the epidermis, i.e. the outer layer of cells literally paving the surface of
plant organs. The cuticle is almost impermeable, so that 80 to 90 % of the gases
are actually exchanged through the stomates. Their diameter is about 20 µm and
their density averages 300 per mm2 , although it may reach 1200 in some plants
(Kramer, 1983). The anatomy of the guard cells is such that they open when they
are turgid, and that they “automatically” close in the case of loss of cell water
pressure, for instance in the event of water stress.
The rates of diffusion of CO2 and of H2O is controlled by an Ohm-like equation (van
Keulen and van Laar, 1986; also refer to Box 1, 2.2.1):
(CO2)ext (CO2) int ÄCO2
VCO2 = = (14a)
rCO2 rCO2
( H 2O) int − ( H 2O)ext ∆H 2O
VH 2 O = = (14b)
rH 2O r H2 O
where round brackets denote concentrations (Kg m-3); V is the rate of diffusion (kg
m-2 s-1 ) and r is the resistance to CO2 and to H2O diffusion in s m-1 , respectively.
rCO2 is the sum of several resistances: ra the resistances of the boundary layer, the
layer of air immediately in contact with the leaf; rc, the stomatal resistance, and rm
the mesophyll resistance, the mesophyll being the “spongy” tissue 15 where
photosynthesis takes place. The mesophyll resistance is usually larger than the
sum of the two others; the standard value of rm is often taken as 440 s m-1 .
The stomatal cavity is normally saturated with water vapour, hence the diffusion is
normally toward the atmosphere when air moisture is less than 100%.
15
“Spongy” because there are lacunae between the cells to favour the movement of gases
between the cells and the stomates.
Due to the different molecular weights16 of CO2 and H2O we have
for the gas phases (ra and rc). Regarding the mesophyll resistance, things are a bit
more complicated (Kramer, 1983). The surface of the stomatal cavity is covered
by a film of water, which opposes no resistance to the penetration of water, but
when it comes to CO2, rm is significant17.
The relation between V C02 and V H20 can be expressed in several ways. One is the
transpiration coefficient which expresses the amount of biomass produced relative
to the amount of water evapotranspired. The orders of magnitude are about 200
Kg to 500 Kg of water for 1 Kg of biomass.
Another way to express the relation is the Water Use Efficiency WUE:
ÄCO2 ÄCO2
r r +r +r 0.64ÄCO2 r +r
WUE= CO2 = c a m = × c a
ÄH2O ÄH2O ÄH2O rc + ra + rm
rH2O 0.64(rc + ra )
(16)
This equation shows that, if weather is relatively windy so that ra ≈ 0 and ∆CO2 is
reasonably constant, water use efficiency depends mainly on air moisture and on
the relative value of rc and rm .
de Wit was among the first who recognised in the mid fifties that there is a direct
link between transpiration and plant productivity. Transpiration can be limited due
to short supply of water in the root zone, or by the amount of energy required to
vaporise the water. It can be said that plants growth (biomass accumulation) is
driven by the available energy, but that plants pay for the energy by evaporating
water. This one of the basic functions implemented in all crop models.
16
The speeds of diffusion are proportional to the square roots of the molecular weights of the
gases.
17
There are also biochemical limitations connected with the photosynthetic machinery.
.
1
0.9
50
0.8 100
440
R, relative assimilation
0.7
700
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Q, relative evapotranspiration
where
. (1 + γ )ra
18
a=
ra + rm (18);
18
LE, is thus the evaporative heat loss (J m-2 d-1), the product of E, the rate of water loss from a
surface (kg m-2 d-1) and L, the latent heat of vaporisation of water is 2.45 106 J kg-1 .
K BT T
∆WT = (19)
VPD
where ∆WT , the water limited growth, is in Kg (biomass) m-2 d-1 , KBT is the crop
specific biomass transpiration coefficient in kPa Kg (biomass) Kg-1 (water), T is
transpiration (Kg water m-2 . d-1 ) and VPD stands for daytime vapour pressure
deficit in kPa. The order of magnitude of KBT is 5 kPa kg biomass for 1 m3 of
water.
2.3 Model components
19
Refer to the section on units for the conversion from ppmv to weight per volume (appendix at the
end of the exercises).
20
Technically, this conversion is a chemical reduction.
21
The steps are (1) cyclical photo-phosphorilation during which light interacts with chlorophyll and
increases the energy level of a couple of electrons that will eventually return to chlorophyll over a
chain of redox-catalysers, while at the same time losing their energy which is stored in chemical
bonds in a chemical known as ATP; (2) photolysis of water: water molecules are split, making O2
available while hydrogen is stored and additional ATP is synthesised. Note that the oxygen
resulting from photosynthesis stems from water and not from CO2 ; (3) one CO2 molecule is
bound to a C5 compound (CO2 + C5 → C6 ) to yield a C6 molecule which will undergo a bewildering
chain of splits and additions, fuelled by the above-mentioned ATPs and the stored hydrogen,
eventually yielding C6H12O6 while the ATPs are being recharged in steps (1) and (2). The process
requires a permanent supply of CO2, water and energy.
Needless to say, photosynthesis can only proceed at the above-indicated rate if a
substantial renewal of air actually takes place (which is to say air is replace twice
per hour). Actual plants have Leaf Area Indices which exceed 1, and therefore
higher wind speeds are required.
If we consider a field of 1 Ha, the above rate (“one square m of leaf can produce 1
g of sugar in one hour”) amounts to 10 Kg of sugar and 15 Kg of CO2. If, in
addition, we assume that the LAI of the plant is 5 m2 (leaf) m-2 (ground), than the
assimilation rate could amount to the high but realistic value of about 75 Kg CO2
Ha -1 hour-1 (assuming linearity between 1 and 5 LAI).
The energy required to produce 1 mole of sugar, considering the various losses
that take place in the photosynthetic machinery, activation energies etc., the
process requires ≈4400 KJ mol-1 of light energy22 (the amount of energy refers to
the moles, not molecules), but when 1 mole of sugar is “burnt”, as in
C6H12O6 + 6 O2 + 6 H2O → 6 CO2 + 12 H20 + 2880 KJ mol-1 (23)
2880 KJ mol-1 are produced23. The efficiency of the process is about 65 % (Pilet,
1967). Note that oxygen is required, but, given its abundance in the atmosphere, it
is normally not limiting.
The sugars may be used in several different ways by the plants:
1. conversion to structural elements (cellulose, wood, fats) which are necessary for
plant growth and development; when plants die and decay, it is the structural
elements that last longest. Note that we refer to structural elements at the micro-
level (scale of the cell). Macro-structural aspects will be covered later under
partitioning 2.3.3;
2. polymerised into starch and more or less temporarily stored;
3. used as a source of energy to “operate”: synthesise other cell chemicals like
proteins and fats, to maintain its structure and functions, to absorb chemical
elements through the root system, etc.
The process through which energy is mobilised is respiration, often subdivided
into “maintenance” respiration and growth respiration.
Based on the treatment of the subject by van Heemst (1986), we can state that the
daily dry matter increase ∆W (Kg (DM) Ha -1 d-1 ) of a crop with a standing biomass
of W (Kg (DM) Ha -1 ) is given by
1
∆W = Effg ( FCO2 − W Respm )
15
. (24)
where F CO2 /1.5 is the gross sugar assimilation [Kg (sugar) Ha -1 d-1 ] derived from
the CO2 uptake; Effg is the conversion efficiency of sugar into DM [Kg (DM) Kg-1
22
According to Pilet (1967).
23
Note that, if the complete oxidation of sugar takes place in a living organism, the energy
produced is also less, among others because some energy (activation energy) has to be spent to
start the reaction.
(sugar)] , Respm is the relative maintenance respiration rate [Kg (sugar) Kg-1 (DM)
d-1 ]. The factor 1.5 comes from equation 22.
Orders of magnitude of Respm depend on the main storage chemicals in the plant
under consideration. For the starchy roots and tubers, where little conversion to
sugar is required, values are about 10 [g (sugar) per Kg of biomass]. Slightly
higher values (15) are given for cereals (which are richer in protein), and oil crops
are at 30. Effg varies from 0.75 [Kg (DM) Kg-1 (sugar)] to 0.70 and 0.50 for the
same crops as above, respectively (van Heemst, 1986). The preceding data
provide one of the reasons why tubers tend to out-yield cereals in DM terms.
Respm appears to be much more temperature dependent than, for instance, FCO2.
In the physiological range of temperatures, it doubles every 10°C (a first
approximation only; see section on phenology and sums of Degree-Days 2.3.2).
Here again, there is a direct link with the common observation that high night-time
temperatures are not conducive to high yields.
2.3.2 Phenology
Phenology qualitatively describes the successive stages in the development of
plants, from seed germination to flowering to maturity. Since the stages (also
called phenophases) are very plant dependent, it is difficult to define a universal
system to describe phases, as only the main stages (germination, differentiation of
flowers, flowering, seed formation, seed filling, maturity) are common to almost all
flowering plants. Additional stages like branching, stem elongation, rosette
formation, etc. all come between germination and flowering, but have no degree of
generality. BASF, Bayer, Ciba-Geigy and Hoechst (BBCH) have developed a
unified coding system that attempts to cover all crops from cereals 24 to beans,
potatoes, vegetables and grape vines (Strauss et al., 1994).
Phenology can be modelled based on vernalization, photoperiod, thermal
response and intrinsic earliness (Cao and Moss,1997), most of which are plant
specific. The Intrinsic earliness is conditioned by the genetic features of the plant,
and it has constituted a main target for breeders, as earliness is one of the
mechanisms to avoid (evade) several difficulties linked with adverse factors like
drought or early fall frost. Photoperiod and vernalization are qualitative responses
of seeds or young plants that require the exposure to a cold period of a certain
length and intensity before they can develop properly.
The accurate determination of phenology has a important economic role per se,
for instance in all impact assessments and planning of farm operations (for
instance labour requirement at harvesting time, or the determination of frost risk at
critical crop stages; D'Antuono and Rossini, 1995). All models include a phenology
component. Although they differ usually by few technical details, some phenology
models are know to perform better than others. For instance, Piper et al. (1996)
compared SOYGRO V5.42 and CROPGRO V3.0 for their ability to predict
phenology of soybean accurately. The role of phenology goes often beyond the
mere qualitative assessment phases (e.g. determination of the start of flowering or
the end of root growth). Many models use a direct link between phenology and LAI,
so that the whole process of assimilation becomes in fact dependent on the
accuracy of the phenology (Chapman et al., 1993, and the QSUN sunflower
model).
Crop phenology plays an important part whenever several species interact, either
“naturally”, as in the case of weeds, insects or pathogens, or as a result of
management. The phenology of pests and diseases is essential when assessing
risks or evaluating losses (Rickman and Klepper, 1991), or in the optimisation of
multiple cropping systems by selecting cropping sequences and their
management (Tsai et al., 1987)
2.3.2.1 Photoperiod
Plants can be categorised as long-day plants, short-day plants and day-neutral
plants. Flower differentiation is initiated in long-day plants by a threshold of day
length below which the plants will not flower. Above the threshold, there is an
24
Even within the groups, there is little homogeneity. For instance, some crops tiller (sorghum,
wheat) while others do not (maize)
optimum daylength. Similarly, short-day plants will not flower is the day exceeds a
threshold. To some extent, the photoperiodic response is independent of growth: if
plants are grown outside the optimum time of the year, they may flower in very early
stages (millet) or never flower at all if the proper daylength is not available.
2.3.2.2 Vernalization
Vernalization can be seen as the need for seeds or plants to be exposed to a cold
threshold between T1 and T2 (T1 < T2). It also constitutes a mechanism to avoid
frost damage. Temperatures below T1 will kill the plant, while if temperature stays
above T2, plants will not develop. This may be combined with the duration of
exposure: a shorter exposure is sometimes sufficient close to T1, while the
vernalization duration is much longer close to T2.
(25)
T is average daily temperature, Tb is the base temperature below which no
development takes place, and Tu is an upper threshold temperature above which it
is assumed that temperatures ceases to have an effect on development. For
instance, the sum of temperatures from sowing to emergence could be 100°,
meaning that, with a base temperature of 10°, the plant would emerge after 10
days at 20°.
Among the criticisms to this approach, one can list the following:
• it assumes that growth varies linearly with temperature. While this may hold as a
first approximation within a limited range of temperatures, it is well known that
actual rates go through a maximum. The linearity assumption is closely related
to the concept of Q10, the factor by which the speed of a chemical reaction
25
Or from some conventional date before planting is the planting date is to be determined in
temperate and cold climates.
increases if temperature is raised by 10°C, as in the equation
rate o f reaction a t ( T + 1 0 ) °
Q 10 = (26);
rate o f reaction a t T °
• it assumes that the temperatures Tb and Tu are constants and do not change
over the cycle;
• the timing of warm spells is indifferent, whereas it is well known that the timing of
abnormal temperatures does matter;
• the effects of other quantitative and qualitative factors is not taken into account
(radiation).
On the positive side, we note that
• there is huge corpus of data showing that the concept provides useful
indications as long as temperatures remain within the “normal range”;
• the approach gives a very simple solution to the problem of some stresses. It is
well known that plants like sorghum, when water stressed, may considerably
delay their development (Ben Mechlia and Carroll, 1989, illustrate the fact for
oranges). The introduction of a high threshold Tu can somehow account for this,
as water stress periods are also characterised by high temperatures.
Over the years, many equations have been proposed to substitute the SDD
method, for instance the well know biometeorological time scale of Robertson
(1968).
In practice, the proposed equations often have too many parameters, or, like the
SDD, do not perform well around the extremes. Interestingly (but not surprisingly),
there is a marked scale effect: when equations describing chemical processes, or
micro-physiological ones are transposed to the field, they usually cease to perform
properly. The equation below, recently published by Xinyou Yin et al. (1995) seems
to give reasonable results under field conditions (which does not mean that it
would yield useful results at a regional scale).
The graph below (figure 7) illustrates the behaviour of a more sophisticated model
proposed by Xinyou Yin et al. (1995), based on the beta distribution. The
development rate DR is given by
Development rate
0.014
0.012
0.01 Tu
0.008
0.006
Tb
0.004
To
0.002
0
0 5 10 15 20 25 30 35 40 45
Temperature (C)
2.3.3 Biomass partitioning
Like respiration, biomass partitioning, i.e. the distribution of the formed biomass
among organs (leaves, roots...) is done at an energetic cost, which is one of the
reasons why assimilation and partitioning are often discussed jointly.
Partitioning is a crop characteristic. Many models include thus a “partitioning table”
which contains the necessary information. At early stages (just sown seed), the
biomass stored in the cotyledons is used for root development and cotyledon
elongation until they reach the surface of soil , when they start photosynthesising
and assimilating. At this early stage, plants usually develop roots and leaves
simultaneously at about the same rate. Roots then normally stop growing and most
assimilates go to the formation of leaves. Eventually, grain formation and filling
takes over the main role a sink of assimilates.
26
In fact, for cereals the ratio sometimes refers to the quantity of product upon total above-ground
biomass.
2.3.4 Root growth
The treatment of roots, if compared with the level of detail most models adopt of
for light interception and the processes taking part above the ground, is best
termed elementary in most cases27.
One common approach is to assume some relation between root biomass
obtained from partitioning of the photosynthetates, root weight (or root length ) and
the depth currently reached. The weakest points usually regard the distribution of
roots over the profile, which is often assumed to be simply linear or negative
exponential between the surface and the maximum rooting depth. It is amazing that
validated models can predict yields and water balance rather well, while at he
same time simulating rather unrealistic root distribution and biomass (for an
example with CERES, see Savin and Satorre, 1994). This clearly points at other
unrealistic assumptions, including in the above-ground component.
More sophisticated models do exist, which are able to simulate the relative root
accumulation at certain depth due to nutrient or water distribution (Adiku et al.,
1996; Asseng et al., 1997), or were calibrated against actually sampled root
distributions in the soil (Pages and Pellerin, 1996).
Adiku et al. consider that the vertical extension of roots (root front extension) is
more or less constant, while root proliferation, i.e. their lateral extension, is more
directly conditioned by soil conditions and follows a logistic curve (which assumes
a “saturation” value).
27
For instance, the behaviour of oxygen in soils receives little attention, although it is certainly
comparable in its complexity to CO2 absorption by leaf canopies.
2.3.5 Water balance
2.3.5.1 General
The soil water balance is the universal tool to estimate soil moisture storage and
its availability to plants.
Let us assume that the discussion below refers to 1 m2. In this case, 1 mm of
rainfall corresponds to 1 kg (or litre) of water.
The soil is subdivided into several layers (sometimes only one) characterised by a
water holding capacity28 (WHC); see figure 9. The order of magnitude of WHC
depends on soil granulometry, composition, compaction, etc. 1 mm (water) cm-1
(soil) constitutes an average value. Refer to Table 2 for sample values and refer to
2.2.1 for the corresponding water potentials.
Soil water content can be expressed as volumetric water content [m3 (water) m-3
(soil) or, as above, mm (water) cm-1 (soil)] or as gravimetric soil moisture [two
options, either Kg (water) Kg-1 (dry soil ) or Kg (water) Kg-1 (wet soil). Other
parameters frequently requested as model inputs are the bulk density which is the
apparent density of dry soil [g (soil) cm-3 (soil)]
When water enters the soil, it fills the first layer to capacity, after which it spills over
into the second layer, etc. The bottom of the last layer coincides with the maximum
rooting depth. Water that spills over beyond rooting depth is usually termed deep
percolation. Additional water movements are sometimes considered, for instance
the rise of water from deep layers due to capillarity, and (rarely) lateral movements.
An empirical factor often used is soil bypass coefficient (0 to 1) which is useful
when soils crack and fraction of the water just flows through the soil. This is one of
those empirical factors which can be conveniently resorted to force a model to
behave according to observation, sometimes regardless of their actual physical
significance. More examples will be given under chapter 4 (4.1).
Plant roots start growing in the first layer, and gradually extend throughout the
profile to occupy the whole volume up to a maximum, given either as a crop or as a
soil characteristic (chemical - pH, sulphates- or physical barriers, including the
presence of rock or the water table).
28
Usually defined as the difference between field capacity and permanent wilting point.
Plants will gradually extract water from all the layers (transpiration). Soil
evaporation occurs only from the top of the first layer, although some models make
provision for air channels which are responsible for some evaporation from deeper
layers.
2.3.5.2 Infiltration
One of the main difficulties in practice is the determination of the amount of water
which actually enters the soil (effective rainfall Reff). Part of the rainfall can be
intercepted by vegetation (for instance between the leaves and the stems of
cereals) or it can just be lost laterally by surface runoff (particularly on sloping
terrain). In addition, infiltration rates depend on the “history” of the soil and the
successive wetting cycles do not behave identically, nor are soil properties
identical in the drying and wetting phases. Numerous methods have been
designed to estimate Reff, going from the very “theoretically correct” to the very
empirical. The latter will be illustrated below.
Two popular methods are those of USDA Soil Conservation Service (USDA,
1967), and the runoff curve number approach of the USDA-SCS (USDA, 1972).
According to the first method:
(29)
For years the USDA Soil Conservation Service (SCS; refer to Maidment, 1992)
has been using an empirical approach to determine runoff, known as the runoff
curve number approach. The method refers to one storm, hence both rainfall Prec
and runoff Q refer to the accumulated values from the beginning of the storm. The
derivation of the equations is quite simple; they are based, among others, on the
assumption that no runoff occurs until rainfall exceeds an initial threshold (all rainfall
infiltrates, is stored in the soil, percolates below it, or otherwise held in the basin).
( Prec − 0.2 S ) 2
Q= (30)
Prec + 0.8S
1000 − CN
S = 254
. (31)
CN
where S is the potential retention of water in the catchment area (in mm; mainly soil
storage in agricultural areas) and CN is the curve number obtained from published
tables as a function of general hydrological and land-use characteristics of the
area.
Figure 9 : Schematic representation of water flow in a three layer soil model with
no lateral movements of water in the soil.
The soil is first characterised by assigning a land use, treatment or practice, and
general hydrologic condition (poor, fair, good). Examples are “fallow, straight row,
poor”, or “pastures, nil, fair”, or “row crop, contoured, good”. Once the categories
above have been identified, the user has to assign the area being modelled to a
hydrologic soil group, defined according to their intake of water when they are
thoroughly wet and receive precipitation from long-duration storms (Table 3 and
figure 10).
The four hydrologic soil groups are:
Group A - Soils having a high infiltration rate (low runoff potential) when thoroughly
wet. These consist mainly of deep, well drained to excessively drained
sands or gravelly sands. These soils have a high rate of water transmission;
Group B - Soils having a moderate infiltration rate when thoroughly wet. These
consist chiefly of moderately deep or deep, moderately well drained or well
drained soils that have moderately fine texture to moderately coarse texture.
These soils have a moderate rate of water transmission;
Group C - Soils having a slow infiltration rate when thoroughly wet. These consist
chiefly of soils having a layer that impedes the downward movement of
water or soils of moderately fine texture or fine texture. These soils have a
slow rate of water transmission;
Group D - Soils having a very slow infiltration rate (high runoff potential) when
thoroughly wet. These consist chiefly of clays that have a high shrink-swell
potential, soils that have a permanent high water table, soils that have a clay
pan or clay layer at or near the surface, and soils that are shallow over
nearly impervious material. These soils have a very slow rate of water
transmission.
Figure 10 : USDA-SCS relation between storm runoff and rainfall for curve numbers
between 100 and 20 (heavy lines). Thin lines indicate the intermediate CN values
(95, 85 etc.).
CN = 100
CN = 90
CN = 80
CN = 70
H − H pwp
Hr =
H fc − H pwp (34)
Different types of curves have been used; some of them are schematically in
Figure 11 below.
Figure 11: Relative evapotranspiration as a function of Hr, the relative soil moisture
The less empirical approach would be to use soil water potential curves (soil
suction) or soil water retention curves, but this is usually done only in the more
specialised models (McCoy, 1991), probably because the potentials are difficult
to sample and to derive from other soil features with some degree of accuracy
(Whisler and Landivar, 1988).
Finally, ET0 is often multiplied by an empirical coefficient to account for the fact that
actual transpiration may exceed the potential values by a factor up to 20 or 30 %.
The factor is called the Crop Coefficient (K c). Kc is one of those empirical factors
which should, in theory, no longer be necessary with the Penman-Monteith
evapotranspiration formula. Yet, the coefficient is still very present in the literature,
although, as Choudhury et al. (1994) observe, variations in soil evaporation can
introduce considerable scatter in the relationship between the crop coefficient and
leaf area index.
TD = DD + GD (35)
With Nmax being the maximum crop nitrogen concentration and NW being the actual
concentration,
DD = W( Nmax − NW ) (36)
and
GD = N max × ∆W (37)
with the following units Kg (DM) Ha -1 for W and ∆W, and Kg (N) Kg-1 (DM) for Nmax
and NW.
Needless to say, N is taken from all soil layers and depends on the availability of
nitrogen, its chemical form (NO3-, NH4+), the availability of water (as nitrogen is
dissolved in water), the root length, etc. A nitrogen budget is computed, and
nitrogen is allowed to move between soil layers, and sometimes to be leached out.
Figure 12: Response of crop yield (Tonnes Ha-1) to nitrogen uptake [Kg (N) Ha-1]
There are several techniques to quantify the effect of nitrogen uptake on yields.
The response curve of yield to nitrogen is of the saturation type (figure 12). The
curve can be made to condition maximum CO2 absorption, resulting in the
expected response.
An alternative approach is to partition the absorbed N into the various plant
organs, usually according to their “normal” protein concentrations. As plants and
their organs are characterised by a structural ratio between protein and the
constituents (for instance carbohydrate), the growth can be constrained by the
failure of the available N-supply to allow for the normal development of the plant
organs.
CropSyst adopts the following expression for nitrogen limited growth (all biomass
parameters on a daily basis):
N pcrit - N p
∆WN = ∆W (1 - )
N pcrit - N p min (38)
where ∆W is the water and radiation limited growth, Npcrit is a critical nitrogen
concentration, Np is current plant nitrogen concentration and Npmin is the minimum
plant concentration.∆
3. Chapter three : Overview of models
This chapter provides an overview of the types of tools available for the
assessment of climate and weather impact on agriculture and plant production.
They vary in their complexity from general relations to crop specific models, their
scope (scale and field of application), and institutional context.
Different models have also vastly different input data and data processing
requirements, and therefore their practical application implies the mastery of tools
and technologies from the very simple (pencil and paper) to the very complex
(computer network with automated real-time data collection).
The following 6 categories are schematically identified:
• Global biomass models (3.1): descriptive methods relating biomass to
environmental conditions at a very generalised scale. They are suitable for
global climatological and agroclimatological studies. Their practical
applications are relatively limited but for the determination of potential yield,
although their relevance for fundamental and research aspects is beyond doubt;
• Vegetation models (3.2) attempt to describe vegetation behaviour, over
various scales, from global to local. This is a field which groups rather different
models, and an area where modelling is probably more difficult than in other
fields, among others because of the larger number of species involved;
• Statistical models (3.3) are not covered in detail; they are just mentioned here
as they still constitute a major tools for many crop-weather impact assessments.
So-called statistical models are simply formulae linking some agronomic
parameter, for instance yield with other factors, environment-, climate-,
economics- or management-related. If the variables are chosen properly,
statistical “models” can perform efficiently at little cost;
• Empirical models for regional applications (3.4): models requiring relatively
few inputs which can be used for monitoring and forecasting crops over large
areas. This includes very “applied” models;
• Simulation models (3.5): mostly process-oriented models which actually
attempt to simulate the actual interactions between plants and their environment
based on chemical, physical, physiological and anatomical data and principles
covered in chapter 2. Due to their detailed nature, they constitute essential
research tools for understanding crop response, and their most obvious
applications at the very fine spatial resolution, i.e. the field.
• Other models (3.6): this category covers methods of weather impact
assessments that do not fall into any of the previous categories
where H is the global net radiation (50 TJ/year per Ha, or 50 106 MJ/year per Ha;
see 2.2.2) and EffH, Effa and Effc are conversion efficiencies:
• EffH is the climatic efficiency, i.e. the fraction of photosynthetically active
radiation (PAR) relative to H, usually in the range from PAR/H=0.33 to 0.5
according to location and season, although 0.5 seems to be adopted by most
authors for practical applications (Lauciani and Ponticiello, 1993) ;
• Effc is the conversion efficiency of absorbed PAR (PARa) to biomass
produced. Effc is very plant dependent and varies over the life cycle. Typical
values are in the range of 3 g of DM per MJ for C4 plants, and about 2 g per MJ
for C3 plants;
• Effa, the absorption efficiency depends on leaf area index (LAI: m2 (leaf) m-2
(ground surface)) and canopy geometry. This parameter is also called
“interception efficiency” It is thus, again, dependent on plant type.
In general,
3000
NPP T = (43)
1 + e (1.315-0.119TC )
N P P P =3000 (1 - e -0.000664 P r e c ) (44)
where TC is average annual temperature in °C and Prec is annual precipitation in
mm. According to whether T or P is limiting, the lowest value of NPPT and NPPP is
eventually retained.
Table 5 shows some typical values of NPP as a function of TC and P.
Temperature Precipitation
TC Nmm NPPP
P
P
T
0 60 0
3
5
5 9 100 193
8
2
10 1 250 459
4
0
6
15 1 500 848
8
4
6
20 2 1000 1456
2
3
1
25 2 1500 1892
5
2
1
30 2 2000 2205
7
1
5
In 1972, Lieth and Box (Lieth 1972, 1973, 1975) have also proposed the equation
below which links productivity and annual potential evapotranspiration in mm:
NPPX = a (1 − e− bX ) (47)
where X stands for Prec or ETa. The coefficients are given below:
a b
Rainfall (mm) 2909 0.000688
Actual ET (mm) 3511 0.000778
29
Biomes are very broad environments, usually of world-wide importance, under the same climate.
Examples for the main biomes are the tundra, temperate forests, equatorial rainforests, coastal
areas...
disturbance by fire excludes grasses as a dominant plant type if soil conditions are
too wet. Otherwise, the plant type with the highest NPP is selected as the dominant
plant type.
At a very different scale, ALMANAC (Agricultural Land Management Alternative
with Numerical Assessment Criteria) is a predictive and process oriented model ,
but it is also able to simulate competition between 2-10 plant species. It includes
detailed functions for water balance, nutrient cycling, plant growth, light
competition, population density effects, and vapour pressure deficit effects. t has
enough detail to be general across locations and species but is not so complex
that independent users cannot apply it to their situations (Kiniry et al., 1992). It is
thus not surprising that many models restrict themselves to two or three plants: a
crop in monoculture, with one or two competing weeds (Kropff et al., 1992).
The potential geographic distribution of plants, pests and diseases can be
assessed using specific models such as CLIMEX. Based on gridded climatic
data, such models define how conducive the conditions in a given grid-cell are for
the development of a specific organism with known ecophysiological
requirements. The information can then be turned into potential or risk maps using
a GIS (Baker, 1996). At the field level, models incorporating weed control
strategies, in combination with weed population dynamics, provide means to
simulate the most cost effective control measures (Holzmann and Niemann,
1988).
3.3 Statistical “models”30
The most common application of statistical “models” is the use of multiple
regression techniques to estimate crop yield: a regression equation (usually linear)
is derived between crop yield and one or more agrometeorological variables, for
instance
Yield = 5 + 0.03 Rain March − 0.10 TC , June (51)
with yield in tons Ha -1 , March rainfall in mm and June temperature in °C. Beyond
their simplicity, their main advantage is the fact that calculations can be done
manually, and in the fact that data requirements are limited. The main
disadvantage is their poor performance outside the range of values for which they
have been calibrated. They often also lead to unrealistic forecasts when care is not
taken to give greater priority to the agronomic significance than to statistical
significance. The equation above, for instance, suggests that low March rainfall (a
negative factor) could be corrected by below zero temperatures in June (frost),
which obviously does not make sense. Another disadvantage is the need to
derive a series of equations to be used in sequence as the cropping season
develops. For an overview of regression methods, including their validation, refer
to Palm and Dagnelie (1993) and to Palm (1997a).
Many of the disadvantages of the regression methods can be avoided when value-
added variables are used instead of the raw agrometeorological variables, as is
done in the FAO method (see section below). Such a value-added variable would
be, for instance, actual crop evapotranspiration, a variable known to be linked
directly with the amount of solar radiation absorbed by the plant under satisfactory
water supply conditions or light water stress (2.2.4).
30
Taken from Gommes, 1998b.
3.4 Semi-empirical models for regional applications: the FAO method
There appears to be a gap between the global scale described above (3.1) and
the field scale. It is suggested that the approach used by FAO and a number of
developing countries for crop forecasting at the national level occupies an
intermediate niche, both in terms of input requirements and ease of validation31.
The section thus describes the FAO crop modelling and forecasting philosophy,
based largely on Gommes et al., 1998.
The word “philosophy” is preferred to “methodology” because the position of FAO
has been to propose a general framework of which the totality, or only some
elements, can be adopted by the countries for their national crop forecasting
methodology for food security. It is also felt that “philosophy” stresses the fact that,
when operating in a field with many partners (economists, marketing experts,
nutritionists, statisticians, demographers, etc.), the most serious problems are not
technical but organisational and institutional: co-ordination of the participants and
integration of different sectoral approaches.
31
Validation, in this context, covers basically the statistical calibration of the model, as the
underlying processes can hardly be verified at the considered scale.
32
Normalised difference vegetation index. A satellite index which is roughly and not too far from
linearly correlated with standing living biomass. Under normal circumstances, the condition of
natural vegetation and crop condition are related. Refer to 5.1.2.1.
33
The yield function is usually an equation, linear in most variables, which was obtained by
multiple regression of a combination of time series and cross-sectional data.
It is stressed that the derivation of the indices above constitute a major difference
with the process-oriented models: they constitute some of the variables that will be
used in an empirical multiple-regression type “model”, to estimate yields, as
described under 3.3. However, because they derive from an agrometeorological
analysis, they constitute highly value-added variables.
Figure 14 : The flow of data in FAO-promoted crop forecasting systems for food
security.
Station yields are then area-averaged using, for instance, NDVI as a background
variable (see 5.3.3), possibly adjusted with other yield estimated provided by
national statistical services, multiplied by planted area to yield a district production
estimate 34.
As indicated, according to countries, variants to this general scheme can be
introduced at almost every step. The technical options were adopted mainly to
reduce computing overhead and bypass, for the time being, some problems which
are still difficult to handle in the context of developing countries. More details will be
given below, but simple, even elementary solutions are sometimes preferable to
complex solutions for which the necessary inputs are not available and must
virtually be guessed. It is also suggested that a codified system and reproducible
approach, even if very far from perfect, is preferable to no system.
To illustrate the previous point: many countries estimate crop production35 by
calling a meeting of knowledgeable people (grain board, statistics,
34
In practice, the situation is slightly more complicated as “station yields” have themselves been
calibrated against agricultural statistics which are given by administrative units.
35
In most developing countries there are not many alternatives to agrometeorological crop
forecasting, with or without remote sensing inputs. Some countries in the Sahel conduct rapid
agrometeorological services) and, through bargaining, eventually reach an
agreement on the current crop production estimate. No specific methodology is
followed, and strong political bias - conscious or otherwise - is often a basic
ingredient in the forecast.
Under such circumstances, any “system” which will avoid political bias and ensure
at least a reasonable degree of consistency from year to year and from place to
place is to be preferred.
estimates based on interviews with farmers. Other countries have developed biometric systems
based on measured crop indices (plant density, maize cob size). In some countries agricultural
statistics are so uncertain that the agrometeorological forecasts are taken as final yield and
production figures. The agrometeorological approach usually gives best results in semi-arid areas
where the water deficit is the main limiting factor. It performs poorly in some mountainous areas
where (i) farming does not follow a homogeneous pattern, (ii) coverage by the weather stations is
insufficient and (iii) water surplus, or pests and diseases, tend to be the main limiting factor(s).
Simple statistical (trend) models perform very poorly in semi-arid countries, where the inter-annual
variability of yields reaches very high values. This being said, after an initial spell of enthusiasm,
the hope to use direct correlations between satellite indices and yields as a forecasting tool, was
gradually abandoned. The methods worked only in few countries, if given the help of additional
data collected at ground level.
36
Gridding of actual data, for instance weather data for short time intervals, is the typical example
where we feel that the available techniques have not reached the a level of reliability which would
justify our transferring the methodology to national services in developing countries.
yield functions which, in addition, avoid to touch on the most difficult issue of
geographic scale effects;
• tools are modular, i.e. the crop forecasting system uses a number of software
tools that carry the analysis from the data to the final production estimate (see
Gommes, 1995, for a more detailed account of the software). Depending on the
local conditions, national services can choose between different tools (for instance
for area averaging). Any specific tools can be changed without touching the whole
structure of the system: the system remains light and easily upgradable and
maintainable. This is facilitated by standardisation37 through common file names
and structures and early reduction38 of RS images (Snijders, 1995). What this
means is that the users, who are responsible for carrying out the analyses and the
forecasts, need not worry about the technical (remote-sensing technical) aspects
of satellite inputs.
37
This issue was addressed by a recent meeting organised by FAO (FAO, 1995).
38
Image reduction here refers to the corrections (geometric, collocation, radiometric, etc.) which
must be made on the images before they can be used for applications.
3.5 Simulation models
This being said, the following general areas of application can be identified:
• research, to understand the actual behaviour of plant-environment and, in the
more complex cases, plant-plant-environment interactions. Models offer also
one of the methods to determine the order of magnitude of some variables and
crop constants which are not accessible to experimental determination.
• training at all levels, to illustrate the behaviour of crops when exposed to varying
environmental conditions and management options. This is probably the only
area where fancy user interfaces are required, as in some farm simulation
games based on sound physical, eco-physiological and economic principles;
• management at farm-scale, and other operational applications: to assist
farmers to improve the planning and timing of their operations. With the advent
of precision farming, a completely new class of applications is being
considered.
• regional applications, mostly in the area of planning. This sector of activities,
often at the margin of agriculture proper (e.g. river basin management) is under-
developed.
The two last, regional and operational applications, are also characterised by their
data requirements, i.e. real-time, or near real-time data are required, which adds a
significant constraint.
39
http://www.std.com/vensim/VBROCH.HTM, http://www.powersim.com/,
http://www.greenhat.com/,
http://www.ranchvision.com/.
3.5.3 Input data40
The list of input data used by crop models grows with model development and
with the introduction of new data types and sources. The data belong to the
categories of weather and climate, crop ecophysiology and phenology, agronomy
, pedology and terrain, as well as economic data. The data are characterised by
specific sources, both technical and institutional, and by sampling frequencies. It is
also stressed that, according to the scope of the model (see above: research,
training, management and planning), the input data may vary (Hough et al., 1998).
Model sophistication is paralleled by the sophistication of input requirements and
the level of spatial detail , but no necessarily by the accuracy of the outputs. The
empirical rule seems to be that monthly data are adequate with statistical “models”
operating at a synoptic scale (roughly 1°x1° grids) (Sakamoto et al., 1977) while
the scale of the field requires daily data.
Sometimes there are several versions of the same model, with different input
requirements, or the same model can use the data which are actually available.
For instance, CropSyst accepts three levels of weather inputs:
• option 1 : daily rainfall , maximum and minimum temperature ;
• option 2: daily rainfall , maximum and minimum temperature, radiation;
• option 3: daily rainfall , maximum and minimum temperature, radiation,
maximum relative humidity, minimum relative humidity and windspeed;
according to data availability, the model will use a simple temperature based
model (option 1), Priestley-Taylor (1972; Option 2), or Penman-Monteith (Penman,
1948; Monteith, 1965; option 3). It is well understood that option 3 is the preferred
one, but the model can still operate with limited data, which is an advantage.
WOFOST 6.0, the version implemented in the EC Crop Growth Monitoring System
(CGMS) of the MARS project normally uses daily weather data on a 50 km x 50
km grid. It also includes an option to use average (monthly) weather data. Daily
data are then derived through interpolation, except rainfall, which is generated
using a built-in mathematical rainfall generator.
The pre-processing of data is a crucial issue in operational crop modelling, and
includes data aggregation and disaggregation, the estimation of missing data and
the related problem of the spatial interpolation of agroclimatic data (see 5.3).
As described in chapter 5, there is now a tendency to rely more and more on
remotely sensed data (RS data). There are several reasons to that, among others
the following:
• RS data provide a global spatial coverage in their native format, while most
ground data must go through a error-prone spatial interpolation procedure;
• several types of weather data and, to some extent, crop data, can be
determined based on RS sources, for instance radiation, surface temperatures,
vegetation indices, phenology and even, according to some authors, even
disease impact on crop condition (Nilsson, 1997);
40
This is covered with more detail in chapter 5 (5.1 and 5.2).
• RS data are usually reliable and easy to obtain, as the number of institutional
partners is limited and data collection and dissemination is largely automated;
• RS data already allow to skip several steps in some model components. For
instance, if RS could reliably estimate actual crop evapotranspiration and
phenology, the models could do without the calculation of those variables, or the
variables could be adjusted based on the RS observations. Needless to say,
RS inputs are more useful at the regional scale, although there is a potential to
derive local values from pixel values, a procedure usually simpler than
computing area-wide values from point data.
Among other “new” sources of data is weather radar (Hough, 1998; see 5.2), the
spatial resolution of which (kilometric pixels) is usually adequate for local studies if
used in combination with ground data.
3.5.4 Sub-models
In this context, the wording “sub-model” does not refer to actual model
components, but rather to very specialised models of physiological processes
which are too complex to be included in operational models; they contribute to
improving the coarser approaches by giving better insight into the processes and
by the development of more realistic simplifying assumptions. An example would
be a model like 2DLEAF which simulates the processes of CO2, O2, and water
vapour diffusion in an intercellular space and boundary layer, evaporation from
cells' surface, assimilation of CO2 , and stomatal movements (Pachepsky and
Acock, 1994; Acock, 1994).
Another example is the Root Zone Water Quality Model (RZWQM; Nokes et al.,
1996): a computer model developed to simulate water, chemical, and biological
processes in the root zone of agricultural management systems, or the programme
given by Jones et al. (1991), although the level of detail is significantly less than in
2DLEAF. For a detailed biophysical root model, refer to McCoy (1991).
41
RWG outputs are obviously based on the statistical properties of real world weather data from
the same location.
Of the associated models and tools, Geographic Information Systems have now
become ubiquitous. GIS techniques, normally in conjunction with geostatistical
software, are used to prepare the spatial input data for the regional applications;
they are used after model runs to format and present the output and analyses.
Finally, particularly for the regional applications, it is necessary to estimate planting
dates. This an be done using a variety of techniques, from actual observations to
the very empirical approaches (“farmers plant when 60 mm of rain have fallen in 4
days, and the 4-day period is not followed by more than 8 consecutive dry days in
the two following weeks”) to sophisticated models involving soil trafficability and
short range weather forecasts.
42
This is almost the structure adopted under 2.3 for model components.
∑ (CC − MCI )
i
2
6
MBI = 1 − (52)
MCI
where the index i indicates the 6 components.
The extreme values that are possible for MBI are -1.23... (very unbalanced and low
complexity with values 0 0 0 0 0 3) to 1 (very balanced model: all CCi identical,
except when MCI = 043). The average value expected for MCI is 1.5, and 0.27 for
MBI44. Note that the balance of (3 3 3 3 3 0) is 0.55, a value much higher that for (0
0 0 0 0 3) as the MBI tends to be higher in complex models.
While recognising that the procedure just described is a subjective one, it is
interesting to note that the most complex model studied by Jones and his
colleagues is WHEAT (MCI = 2.8) followed by PNUTGRO (MCI = 2.7), and by
CERES-maize and CERES-wheat (both at MCI = 2.3). The lowest complexity was
found in an early version of RICEMOD (MCI = 0.8) , while version 3.00 of the same
model reaches 1.8.
Regarding balance, there tends to be a qualitative correlation with MCI. The
lowest MBI was TOBACCO (-0.28), followed by SIMREW (MBI = -0.10) and the
above-mentioned early version of RICEMOD.
According to the authors, EPIC, ALMANAC, NTRM and WOFOST compare with
CERES.
It would be interesting to critically evaluate the concept of MBI with a view to
reducing its subjectivity, possibly by taking into account the number of inputs and
parameters. Most models studied by Jones et al. are, as indicated, relatively early
versions, and the number of inputs is amazingly low when judging by the more
recent models (see 4.1 in this document) where inputs and parameters are often
counted by the hundreds (Jones list an average of about 5 weather inputs, 10 crop
inputs and 2 to 10 soil inputs per soil layer). A revision of their paper should cover
more models (in particular the WOFOST family) and take into account the number
of inputs and parameters, as well as the number of internal variables. We suggest
that due attention should also be given to the ad hoc parameters, modular
construction and the ability to evolve.
3.5.8.1 CropSyst45
CropSyst (Cropping Systems Simulation Model) is a multi-year, multi-crop, daily
time step crop growth simulation model, developed with emphasis on a friendly
user interface, and with a link to GIS software and a weather generator (Stöckle,
1996). Link to economic and risk analysis models is under development.
43
In which case it will be difficult to talk about a model at all!
44
The average MBI value is only indicative. It was obtained by Monte-Carlo simulation.
45
The section is quoted from the WWW site maintained by C. Stöckle at the university of
Washington,
http://www.bsyse.wsu.edu/cropsyst/
The model’s objective is to serve as an analytical tool to study the effect of
cropping systems management on crop productivity and the environment. For this
purpose, CropSyst simulates the soil water budget, soil-plant nitrogen budget,
crop phenology, crop canopy and root growth, biomass production, crop yield,
residue production and decomposition, soil erosion by water, and pesticide fate.
These are affected by weather, soil characteristics, crop characteristics, and
cropping system management options including crop rotation, cultivar selection,
irrigation, nitrogen fertilisation, pesticide applications, soil and irrigation water
salinity, tillage operations, and residue management.
An advanced user-friendly interface allows users to easily manipulate input files,
verify input parameters for range errors and cross compatibility, create
simulations, execute single and batch run simulations, customise outputs, produce
text and graphical reports, link to spreadsheet programs, and even select a
preferred language for the interface text.
Simulations can be customised to invoke only those modules of interest for a
particular application (e.g., erosion and nitrogen simulation can be disabled if not
desired), producing more efficient runs and simplifying model parameterization.
The model is fully documented (Stöckle and Nelson, 1994, and successive
updates) , and the manual is also available as a help utility from the CropSyst
interface. CropSyst executable program, manual, and tutorials can be retrieved
directly over the Internet (http://www.bsyse.wsu.edu/cropsyst)
Additional references to the model include Pala et al., 1996; Stöckle et al., 1994;
Stöckle et al., 1996; Badini et al., 1997; Stöckle and Debaeke, 1997, and
Pannkuk et al., 1998.
3.5.8.3 EPIC
EPIC, the Environmental Policy Integrated Climate (formerly Erosion Productivity
Impact Calculator), and SWAT , the Soil and Water Assessment Tools can both be
found and downloaded from of the Blackland Research Center (Temple, Texas)
which belongs to TAMUS (Texas A&M University System). EPIC is well
documented and has been used in a number of publications ( Williams and Berndt,
1977; Williams et al., 1989; Sharpley and Williams,1990; Williams et al., 1990).
The models home page is at the TAMU:
http:/ / brcsun0. tamu.edu / epic /
http:/ / brcsun0. tamu.edu / swat / swat
The description below is drawn from the EPIC version 5300 fact sheet (Mitchell et
al., 1995).
The objective of the model is to
• assess the effects of soil erosion on productivity;
46
Most of the present information stems from the ECOBAS site.
• predict the effects of management decisions on soil, water, nutrient, and
pesticide movement and their combined impact on soil loss, water quality and
crop yields for areas with homogeneous soils and management47.
The model components include:
• A detailed soil water balance, including lateral subsurface flow and snow melt
(rarely found in other models), water and wind erosion ;
• Detailed N and P budget (loss in runoff, leaching, including for the organic
phases, mineralisation and uptake);
• Pesticide fate and transport;
• Crop growth and yield for over 20 crops;
• Crop management (drainage, irrigation, fertiliser application;
• Economic accounting and
• Waste management.
EPIC operates at 1-daily steps and is driven by soil, weather, tillage and crop
parameters data supplied with the model. Soil profile can be subdivided into up to
ten layers.
The model’s first steps go back to 1980; it is being continuously updated and
adapted to a variety of uses and situations, in particular climate change studies,
farm-level planning and soil-loss assessment. There is an Australian sugarcane
model (AUSCANE) derived from EPIC. Like the other popular models, EPIC now
comes with a weather generator.
The main application of EPIC remains in erosion studies, and thanks to the
weather generator, the model can be run for thousands of years to capture the
effect of this very slow process.
47
Note that this limitation is a very real one and applies to almost all process-oriented models.
Few, however - including the developers - confine the use to their model to homogeneous areas.
48
Information about all the DSSAT/CERES products is available directly from the listed WWW site
at the University of Hawaii (http://everex.ibsnat.hawaii.edu/). The site provides links to the related
products.
chickpea, tomato and pasture; SOYGRO, PNUTGRO, BEANGRO, SUBSTOR-
potato) are accessible in DSSAT. The crop models were developed to assess the
influence of weather and management practices on crop growth and development.
One interesting feature of DSSAT is the development of standards for data
collection and formats for data acquisition and exchange. This allows any crop
model of the family to share and access common soils and weather data. Each of
the models are processed-based and simulate the daily growth and development
of crops as influenced by daily weather. Multiple season simulation provides
cumulative probability analysis for risk management.
Among the listed models of the family, several have been integrated into a higher
level modelling environment known as CropSys (Crop Systems), for instance
those of the CERES-type. The models currently in CropSys include: CERES-
Barley, CERES-Maize, CERES-Millet, CERES-Rice, CERES-Sorghum, and
CERES-Wheat.
49
A nice acronym: the Latin word ceres designates harvests, wheat and bread; from Ceres, the
Roman goddess who taught men agriculture.
3.6 Other process-oriented models
There are many crop specific models, as well as models usable for a larger range
of crops, for instance legumes. For a short list of models by crops, refer to Jones
et al., 1997. Many models, as indicated, are now part of a more comprehensive
package including several auxiliary decision making tolls. An example is
GOSSYM, a cotton growth model. Its main originality is its association with
COMAX, an expert system (Whisler and Landivar, 1988). Refer to the sources
indicated above (3.5.1) for details, i.e. mainly CAMASE and ECOBAS.
3.7 Rule-based systems
Rule-based systems include the whole spectrum from simple descriptive
thresholds to expert systems. We suggest that they are particularly useful in
assessing qualitative and indirect effects of weather on crops. The first part of this
section on descriptive models is taken from Gommes (1998b).
The simplest descriptive methods are those that involve one or two thresholds. A
hypothetical example is given in the table below (Table 6).
50
This is not unrelated with the typical “weather types” described by meteorologists.
→ Global atmospheric circulation → Local weather →
Local crop yield) where each step introduces new uncertainties. As mentioned
above, this chain of interactions can also be seen as a “complex” starting with the
El Niño - Southern Oscillation (ENSO) index. In southern Africa, for instance, warm
El Niño events are associated with an premature start of the rainy season,
followed by a drought at the time of flowering of maize, the main crop grown in the
area. This pattern usually results in good vegetative growth, followed by drought
induced crop losses. Cane et al., (1994) have found good relations between El
Niño parameters (i.e. the very beginning of the causal chain) and maize yields in
Zimbabwe, which constitutes a good illustration of the concepts described in this
section on “descriptive” methods.
Descriptive methods have also been used successfully to estimate the quality of
agricultural products such as wine. Given that the concept of “quality” is difficult to
describe in quantitative terms51, the non-parametric approach of the is probably
the most suitable.
Expert systems are more complex (Russell et al., 1997). They use the techniques
of artificial intelligence to infer the impact of environmental conditions on crop
yield. To do so, they require a base of data, a knowledge-base and an “inference
engine” which is the software which constitutes the interface between the data and
the users. A knowledge base includes all the normal database functions, but has
additional functionality in terms of the way questions can be asked. For instance, a
knowledge base “knows” synonyms, it knows orders of magnitude (“low yield”),
understands contexts (general information, e.g. properties of a group of plants, for
instance grasses) and is normally able to perceive implicit information. Implicit
information is the information normally associated with a category, like humic
gleysol (pH, drainage properties, depth, texture, etc.).
The inference engine controls the reasoning used to answer queries. Knowledge
bases can use the outcome of one rule as an input for another. Below we quote an
example adapted from Russell (Russell and Muetzelfeldt, 1998), the author of a
very detailed wheat knowledge base for Europe, which at the same time illustrates
the concept and shows the usefulness of knowledge bases in crop-weather
modelling:
what are the consequences of high temperatures in March on wheat yield in
Spain ?
The expert system must first “understand” what is meant by high temperatures,
next it must “know” at what phenological stage wheat will be in Spain at the said
time. Finally, the programme must “understand” the concept of Mediterranean
region If no specific data are available for Spain, the system will “know” that Italy,
Greece and Southern France are part of the same region and that some data can
be borrowed from there.
The European wheat knowledge base puts special emphasis on the identification
of alarm situations, based on research and expert knowledge. As such, a
knowledge base constitutes a unique monitoring tool as it is unlikely that any of the
51
Quality of wine is described by a combination of pH, sugar contents and types, acid types,
concentration of tannins, colour, etc.
other types of models will be able to perceive the more complex environmental
interactions and sequences, such as the example quoted under sensitivity analysis
(4.6): a succession of very warm days at the beginning of flowering of orchard
crops, followed by a week of heavy rain, which will have several indirect effects,
like poor pollination.
Expert systems can be combined with the traditional process-oriented models.
Kamel et al. (1995) have developed a tool to support the regional management of
irrigated wheat in Egypt which captures local expertise through the integration of
expert system technology and a crop simulation model (CERES). The system can
improve the selection of sowing date and variety, pest monitoring, identification
and remediation and harvest management, and may allow better utilisation of
resources, especially water.
For an easily accessible rule-based model (PLANTGRO) the reader can consult
the following WWW site : http://www.ozemail.com.au/~chackett/.
4. Chapter four : Checking the quality of models
The current versions of models like EPIC, CERES and WOFOST use about 50
crop characteristics, around 25 parameters to describe soils, plus 40 or so
management and miscellaneous parameters. In comparison, the daily weather
variables, which actually drive the models, are usually just 5 or 6 (rainfall, minimum
and maximum temperatures, windspeed, radiation and air moisture). The internal
variables used by WOFOST amount to about 260, of which half are crop variables,
30% are soil variables and 20% are weather variables (including all the
astronomic variables like daylength, Angot’s value etc.).
Output variables can, in principle, be any of the internal model variables. The EPIC
manual, for instance, list 180 between input parameters and output variables. In
comparison, CropSyst uses “only” 50 input parameters.
All process-oriented models more or less openly use ad hoc variables to force the
models to behave like the experimental data. It is not always easy to decide which
variables are ad hoc without digging deeply into the operation of the models, which
is possible only with the models for which detailed documentation and often the
source code is available. The ad hoc variables are sometimes grouped under a
category of “miscellaneous” variables, or they have names like “reduction factor”,
“adjusted rate”, “correction factor” or “coefficient of crop yield sensitivity to water
Mitchell et al., 1995) has a “factor to adjust crop
canopy resistance in the Penman equation” and a “nitrogen leaching factor”. One
of the best examples remains the Crop Coefficient (2.3.5).
Savin et al. (1994) compared some root variables of CERES-wheat with actual
conditions. While the model accurately predicted crop development and yield, it
over-predicted root depth by 90 cm at terminal spikelet and by 50 cm at booting,
anthesis and mid grain-filling periods. This indicates that an accurate prediction
can actually be made by less than perfect models. The reasons for this can be
tentatively found in the ad hoc coefficients that adjust for model imperfections, for
instance the use of laboratory determined constants on field crops, and the fact
that models are mostly calibrated against proxies, not the actual crop variables. A
very common example is soil moisture: a model can correctly simulate soil
moisture even if root behaviour and water relations are very far from reality, like
with one-layer soil models.
Errors often mutually cancel out, and so do errors in the parameters of a model. In
the example above too large a root length can be compensated for by too small an
absorption rate.
4.2 Model evaluation, validation and cross-validation.
Before models can be put to work in assisting with decisions in the real-world, the
user must be reasonably confident that the model describes actual crop responses
to weather and management with a degree of precision that is sufficient for the
intended application. The standard wording usually resorted to in this context
includes validation, calibration, verification, etc. (Penning de Vries et al., 1995), but
there is no consensus in their actual acceptation. Some terms (like accuracy)
appear to be used only in particular contexts such as crop forecasting.
In fact, maybe with the exception of evaluation, the meaning of some of the words
varies according to the type of model, and the concepts themselves have a
subjective element.
Evaluation seems to be the most neutral term. It means nothing more than
assessing the value of a model: how realistic are the model components? What is
their level of detail? How many fudge factors (ad hoc correction factors, adjusted
rates, etc.) had to be introduced by the model developers to make it behave
adequately. What do model developers actually call an adequate behaviour? How
balanced is the model (see 3.5.7)? Evaluation is the qualitative mental exercise,
based on subjective perception as well as, but not necessarily, on quantitative
tests (validation, calibration), that will eventually lead a potential user to declare the
model fit for his intended purpose.
Validation is more difficult. There is no such thing as a definitive or final validation
of a model. A model must be validated at the same spatial scale and with the
same type of data as those that will be available in operational work. In other
words, validation is the sequence of tests and checks that convince the user that
the model is valid for the intended purpose. If a model has been validated for many
different circumstances, the potential user can decide that it may perform properly
even under his own conditions, but he should not take it for granted.
Cross-validation is the comparison of one model against another taken as
standard.
4.3 Verification
According to Penning de Vries et al. (1995) the term designates the inspection of
the internal consistency of the model and its software implementation. In practice,
verification is very difficult to carry out, as it implies the access to the computer
source code and the full model documentation. Most successful models go through
a number of versions (sometimes starting twenty years ago) and are, in practice,
permanently “under construction”.
Some of the recent models have reached such a level of complexity that it is
almost impossible to verify them but for their authors. Fortunately, for most
reputable models, the source code is actually available freely over the internet or,
at request, from the authors. This will at least allow the potential users to examine
the algorithms.
The best practice, therefore, is to develop models in successive “stable” versions,
which are fully documented and published, while the authors develop and debug
the forthcoming version. As many scientists usually work on different parts of the
model, the most efficient approach is to develop the models components in
separate and largely independent modules. This will also facilitate verification and
allow users to select the level of complexity and the specific algorithms of their
simulations by choosing the appropriate modules.
We are still far away from this approach, among others because of the diverging
traditions of modelling schools (Wageningen, CERES), and because fully modular
models pose some very challenging technical problems.
4.4 Calibration
Calibration and fine-tuning are the same concept. Assuming that the author or the
user of a model is satisfied with the algorithms, the next step is to submit the
model to the ordeal by real-world data: the model is run repeatedly with actual
inputs to see if it mimics reality sufficiently well. The actual data are usual referred
to as calibration data or training data. The greater the variety of training data, the
greater the chances that the model will be well behaved under new conditions.
Calibration is done differently for different categories of models. For models
developed for educational purposes, it is sufficient if the models behaviour is
qualitatively consistent.
Research oriented models are more difficult to calibrate, as all the variables of the
crop-environment system should take values in agreement with reality. This is
impossible to verify in practice. It is also recalled that many crop constants were
obtained under laboratory conditions on single organs, and that rates observed on
whole plants under actual conditions may sometimes differ by orders of magnitude
from those observed under controlled environments. In addition, rates and
concentrations may vary significantly between the beginning and the end of an
experiment. A good example is provided by some of the current work on “double”
CO2 effects on plants: initial response is high, but the plants adapt (the
phenomenon is known as down-regulation of photosynthesis; see Wolfe and
Erikson, 1993; Allen et al., 1996). This is nothing but another manifestation of a
temporal and spatial scale problem.
In addition to the scaling problem, we have already underlined that very often
models are not calibrated against actual plant variables but against proxies, i.e. it
is the result of the crop-environment interaction that is modelled rather than the
processes themselves. Next to the example with soil water given above (4.1), we
could mention that crop yield models are calibrated against final yields, never
against daily biomass accumulation. Also note that the time step almost universally
adopted for models is the day, when plant processes actually take place at much
shorter scales.
In practice, models are all characterised by some parameters and thresholds the
value of which is not really known nor very precisely fixed, nor constant over the life
of crops (for instance, the base temperature for sums of temperatures, of leaf
water potentials, etc.). It is clear that such values can be adjusted without
contravening to the rules of plant physiology. Such adjustments may, however,
have an effect on the final yields or other model behaviour, so that they provide one
of the ways to calibrate a model.
Next come the ad hoc factors (4.1), often introduced because they are the only way
to make a model perform realistically; they constitute the second “button” that can
be used to fine-tune a model.
The calibration is thus done by trial and error, among others by plotting the errors
affecting some key crop or soil variables against the model parameters. Given the
complexity of models and the number of variables involved, it is virtually impossible
to be certain that the best set of parameters was eventually selected. The “best” is
to be understood in the double statistical and crop/physiological sense.
Needless to say, if next to statistical and physiological criteria we also adopt
agronomic criteria and constraints (like using the model at a different geographic
scale than the one for which it was developed), the “best” parameters may still be
different.
In multiple regression models, calibration takes the meaning of finding the values
of the coefficients that provide the best statistical fit to a set of experimental data
(usually yields). Within the limits of the models, rigorous methods exist to optimise
the values of the coefficients (see Palm and Dagnelie, 1993, and Palm, 1997a and
1997b). There is a large variety between the options to compute the regressions
proper and to verify the suitability for the intended purpose, which is usually crop
forecasting. One of them is known as jack-knifing, another is split-series.
Jack-knifing computes the regression with all the data but one set, and verifies that
the model can forecast the yield (or other parameters) of the missing set. The
same procedure is then repeated for all the sets in succession, lending the
procedure some statistical weight. With split series, the sets are subdivided into
two groups (often the first and the second half of a time series, but he concept is
more general) and the coefficients are determined for each set. If the model is of
general applicability, the coefficients should not be very different in the two groups.
The procedure can be repeated with other groups of sets.
In all statistical calibration, one of the main problems is the proper balance
between statistical and agronomic significance, essentially the orders of
magnitude and the signs of the coefficients.
The two extremes of model calibration thus seem to be process-oriented models
with trial and error calibration and empirical statistical “models” with statistically
optimal calibration.
52
This is the definition of an extreme event.
5. Chapter five : Some methods and tools for operational
crop modelling
5.1.2.1 Definitions
All vegetation indices are based on the fact that plants are green: they reflect a
much larger proportion of white sunlight in the green part of the spectrum than in
the blue and in the red. In fact, plants contain varying proportions of Chlorophyll-a
(absorbing mainly the blue between 0.38-0.45 µm and the red around 0.675 µm)
and Chlorophyll-b (0.41-0.47 µm and 0.61 µm). Plants reflect a much larger
proportion of light in the near infra-red, so that they appear normally very bright
when seen through a near infra-red filter or sensor. Figure 17 compares the
reflectance54 of different types of surfaces in the red and near infra-red.
53
Radiance is the amount of energy received by a sensor (radiometer) expressed in power units
(W or J s-1). The term is sometimes used to express flux density, expressed in W m-2 sr-1. In
practice, the sensors onboard an aircraft or a satellite measure electrical currents which are
converted into radiances using calibration tables.
54
The reflectance is the percentage of the incident radiation (including light) that is reflected. The
word is also sometimes used to indicate the intensity of reflected radiance.
Figure 17 : Reflectance spectrum55 of stressed vegetation , vegetation in good
health (veg. Norm), soil (S) and water H2O. The arrow indicates the second (red)
peak of absorption of chlorophyll.
It is obvious that the spectra of the different surfaces illustrated in Figure 17 are
characteristic for the surface and constitute their “spectral signature”. In principle, it
should thus be easy to identify the surfaces simply by comparing their reflectance.
Things are not so easy, for several reasons. To start with, the spectral signatures
are obtained experimentally on the ground under known conditions of irradiance56.
The radiance measured by satellites have undergone qualitative changes
(wavelength) and changes in intensity in the atmosphere. Further, the irradiance
varies as well as a function of atmospheric conditions and angle of incidence
(which is to say: the time of the day).
Many vegetation indices are thus “normalised” to correct for at least the most
obvious atmospheric and soil effects, as in the popular normalised difference
vegetation index (NDVI): it is defined as
NIR − R
NDVI =
NIR + R (53a)
where NIR stands for the reflectance in the near infra-red region (roughly from 0.8
to 1.1 µm) and R is the reflectance in the red part of the spectrum. Examples of the
calculations are given in Figure 18.
55
The remote sensing jargon also calls e reflectance spectrum spectral reflectance.
56
Irradiance is the incident radiance, for instance and normally as sunlight.
Figure 18 : Computation of the normalised difference vegetation index based on the
red (R) and near infra-red (IR) reflectance for soil (a) and vegetation (b).
In theory NDVI varies from -1 to 1; in practice, values for water are negative, bare
soil covers the range from 0 to 0.12, where sparse vegetation starts.
NDVI can be computed whenever red and near infra-red reflectance (i.e. reflected
radiance) are available, from such satellites as the LANDSAT-TM, SPOT and
NOAA-AVHRR.
The most popular source for satellite NDVI is the NOAA series polar-orbiting sun-
synchronous 57 satellites, starting with TIROS-N (1978). The most recent satellite of
the series (NOAA-14 at 833 Km) was launched in 1994, but NOAA-11 (launched
1988) and NOAA-12 (launched 1991) are still operating.
Their AVHRR (Advanced Very High Resolution Radiometer) measures radiance in
5 wavelength bands, also known as “channels”:
• Channel 1 (visible): 0.58 - 0.68 µm, mainly used for daytime cloud/surface and
vegetation mapping;
• Channel 2 (NIR): 0.725 - 1.10 µm; surface water, ice, snow melt, and vegetation
mapping;
• Channel 3 (SWIR; Short Wave IR): 3.55 - 3.93 µm; surface temperature, night-
time cloud mapping;
57
Polar-orbiting sun-synchronous: the satellite is on an orbit passing over the pole, and it is
synchronised with the sun, i.e. for a given location it passes every day at the same times. The
period of a circular orbit is easy to calculate: for an altitude h in thousand Km (i.e. 580 km = 0.58
Mm), the period T in hours is given by T=0.0875 √(6.37 + h)3.
• Channel 4 (TIR, thermal infra-red): 10.50 - 11.50 µm; surface temperature, day
and night cloud mapping;
• Channel 5 (TIR): 11.4 - 12.4 µm; surface temperature, day and night cloud
mapping.
NDVI is computed by calculating the ratio of the VI (vegetation index, i.e., the
difference between Channel 2 and 1) and the sum of Channels 2 and 1. Thus NDVI
= (channel 2 - channel 1) / (channel 2 + channel 1), using a series of corrections
(geometric, radiometric and declouding).
The products are available in three different formats: HRPT, LAC and GAC:
• HRPT, High Resolution Picture Transmission. HRPT data are full resolution
image data transmitted to a ground station as they are collected. The resolution
is 1.1 km at the satellite nadir;
• LAC, Local Area Coverage. LAC are full resolution data that are temporarily
stored onboard for subsequent transmission to the ground. The resolution is the
same as HRPT, but LAC data have been stored prior to their transmission;
• GAC, Global Area Coverage. GAC data are derived from a sample averaging
of the full resolution AVHRR data (1.1 km by 4 km).
NDVI can be obtained at the LAC and GAC resolutions.
5.1.2.2 Vegetation indices for monitoring
There are numerous actual and potential applications of NDVI in operational crop
modelling. Below (Figure 19) is an example of one of the most straightforward
applications.
0.4
0.35
0.3
NDVI
0.25
0.2
0.15
0.1
J A S O N D J F M A M J J A S O N D J F M A M J J A
07-1982 / A- 1984
The figure illustrates the use of NDVI to derive phenological information in the
absence of ground data. Note that due to the presence of clouds, many readings
have to be discarded. The operational procedure is, therefore, to use only the
maximum values for each pixel during a certain period (for instance ten days). This
removes only part of the variability from NDVI series, as can be seen in figure 19. It
is therefore also a common practice to smoothen the NDVI curves, often by
drawing an envelope around the cloud of observations.
Assuming that crops and natural vegetation are synchronised, planting can be
assumed to take place at dekad D when NDVI starts increasing again after the dry
season, or a fixed number of dekads after D (for instance D+4), or when NDVI
crosses an locally determined threshold...
NDVI and other vegetation indices can, of course, be computed at any scale.
Companies in the USA have started providing very detailed NDVI maps of farms
on a subscription basis.
One of the companies (Emerge58) provides 1-m resolution geo-referenced digital
imagery covering customers' fields. See figure 20 for an example.
Figure 20 : A farm in the Midwest (USA) in July 1997. The top left figure image is a
photograph (visible light), while top right photographs indicates a vegetation index
coded on a thermal scale. As the grey shades do not properly reflect the original
colours, the main “warm” and “cold” areas has been indicated on the left. The
zoomed area at the bottom shows the metric resolution of the original product,
with a road crossing the image.
This information can be printed or directly read into mapping software. The
acquisition and processing techniques allow “temperature” and visible maps to be
available to customers on their internet sites within 48 hours after the data have
been sampled using a small aircraft. The company stresses that the maps allow
farmers to visualise the variability of their field and areas of water or pest stress or
58
Emerge, trademark of TASC, Inc., a subsidiary of Litton Industries
(http://www.emerge.wsicorp.com/emerge/info.html)
otherwise sub-optimal management. Obviously the products could be use also for
early production estimates in combination with ground weather stations.
We conclude this section by mentioning the derived defined by Kogan (Kogan,
1995, 1997; Unganai et Kogan, 1998) based on NDVI further normalised by
comparing them with their recorded extremes:
NDVIi − NDVI min
VCI = 100 (53b)
NDVI max − NDVI min
and
BTi − BTmin
TCI = 100 (53c)
BTmax − BTmin
VCI is the Vegetation Condition Index and TCI is the Thermal Condition Index.
They vary in opposite directions. The BT are brightness temperatures computed
from AVHRR channel 4 (10.3 à 11.3 µm). The channel is more sensitive to drought
because it is less affected by atmospheric moisture than channel 5. The analyses
carried out by Kogan have shown that during drought years, BT is significantly
higher than in normal years, which entails a close relationship between TCI and
crop yield anomalies.
Note that Kogan (Unganai et Kogan, 1998) uses a linear combination between
VCI and TCI directly estimate maize yields.
where EffH is the climatic efficiency (0.5: PAR total light ), Effc is the conversion
efficiency of absorbed PAR (PARa) to biomass produced (also called the
radiation use efficiency or light to biomass conversion efficiency, etc.), and Effa is
the absorption or interception efficiency. Both Effa (see below) and Effc (5.1.3.1)
can benefit from remote sensing inputs.
We have mentioned that Effa is related to LAI. Various authors quoted by Laguette
have developed empirical relations linking Effa directly to NDVI, for instance
for millet:
for wheat
NIR
Eff a = a ( + b)
R (55)
and for rice
2
0.9 − NDVI
Eff a = 1 −
0.9 − NDVI soil (56)
59
For remote sensing, the infrared wavelengths are often subdivided into near infrared (0.7-1.3
microns), middle infrared (1.3-3.0 microns) and far infrared (7.0-15.0 microns). Far infrared is
sometimes referred to as thermal or emissive infrared.
using satellite data (Kustas and Norman, 1996; Bastiaansen et al., 1996). The
surface temperature Ts is the temperature of the layer of air immediately in contact
with the leaf, and Ta is the conventional air temperature measured in a screen.
The interest in the difference between surface and air temperatures goes back to
the work on Jackson’s CWSI (Crop Water Stress Index ; Jackson et al., 1981)
which is defined as
LE
CWSI = 1 − (59)
LE p
where LE is the actual potential evapotranspiration and LEp is the potential
evapotranspiration, usually referred to a short time interval.
The concept of CWSI is identical to the FAO Water Satisfaction Index (WSI)
introduced by Frère and Popov (1979) as a monitoring tool derived from a water
balance calculation over the whole cycle 60.
Jackson and his colleagues found that an alternative formulation for CWSI is
Ta − Tsmin
CWSI =
Ts m a x − Tsmin (60)
with Ta = air temperature and the various Ts parameters represent the surface
temperature minimum and maximum values. Ts represents the effective
temperature at which the processes in the leaves take place. It can be determined
using the AVHRR channels 4 and 5 with the proper geometric, radiometric,
atmospheric and declouding corrections using a linear combination technique
(so-called split-window technique), with an accuracy of 2 to 3 degrees. The
geostationary meteorological satellites like METEOSAT and GMS also have a TIR
channel that can be used to estimate Ts and derive evapotranspiration (Rosema
et al., 1998b).
It can be shown (Laguette, 1995, 1997) that, for a given net radiation, there is a
direct link between actual evapotranspiration and Ts-Ta.
In the net radiation balance equation already discussed under 2.2.2
H = P + G+ A + E (11)
the terms G and P play a minor part compared to A, the sensible heat and E, the
evapotranspiration; the equation thus reduces to
H ≅ A+ E (61)
It appears that A is directly related to Ts -Ta, and that the partitioning of H between
A and E is of the form
60
Needless say that the are marked scaling effect when passing from CWSI to WSI.
where b and c are coefficients depending on the Stefan-Boltzman constant, the
turbulent exchange coefficient and the volumetric heat capacity of air.
Ts -Ta, as a measure of the sensible heat flux, indicates how much energy is
available for evapotranspiration. A large difference indicates reduced
evapotranspiration and stomatal closure, and therefore a reduced conversion
efficiency Effc. A small difference indicates that water supply is adequate and that
plants actually evapotranspire, that most energy goes into E , under which
conditions the ratio of A to E is about 0.1. In general low temperatures thus
indicate healthy and photosynthesising crops.
The conversion efficiency Effc can be shown to be linked to the CWSI and thus to
Ts and Ta.
Although the relation between the solar radiation reaching the ground and clouds is
far from direct (Li et al., 1995), the development of more or less empirical methods
is progressing, usually with much better results than with rainfall (Lourens et al.,
1995; Wald, 1996; Supit and van Kappel, 1998), among others because the role
of clouds in radiation interception is far more direct than in rainfall production. In
addition, the methods, once they have been properly calibrated, apply in tropical
and temperate countries alike.
The original approach was to try and estimate rainfall based on CCD only,
assuming a constant intensity Ri (mm hour-1 ). Because of the large spatial and
temporal variability of Ri , the method is now being replaced by more or less real-
time calibration against ground data, using CCD as an auxiliary variable in the
spatial interpolation of raingauge measurements.
The major methodological issues regarding rainfall estimation and CCD can be
listed as:
F rains are known exactly only for a given duration only for a very limited area
around raingauge. According to the period covered (hours, days, dekads,
months), the radius within which a raingauge provides a representative sample
varies from a couple of hundred meters to 200 km;
F CCD indices cover a METEOSAT or GMS pixel (about 50 km2 at the equator),
and correspond to a discontinuous sample in time (one observation every 30
minutes);
F a plot of rainfall as a function of CCD thus compares two rather different
variables, both of which are used as proxies for a third unknown variable, the
average pixel rainfall. One of the consequences is a rather poor correlation61
between CCD and rainfall, and usually not usable for rainfall estimation. Newer
and significantly more efficient techniques are now available (see 5.3.3);
F for short time intervals (one day), an additional difficulty is the difference
between the time covered by rainfall measurements (09 GMT to 09 GMT the
next day) and the satellite images.
F It remains however that CCDs are associated with rainfall and that they provide
a useful monitoring tool, as shown in figure 21. This figure covers one of several
“homogeneous” rainfall units of the SADC region which are regularly published
by the regional monitoring system.
61
Il existe une limite théorique au coefficient de corrélation de l’ordre de 0.7 (50% de la variabilité
62
Brightness temperature is the temperature of a blackbody radiating the same amount of energy
per unit area at the wavelengths under consideration as the observed body.
5.2 Weather radar
Weather radar, like microwave satellites, operates in the centimetric range. The
technique basically measures rainfall intensity within a radius of about 100 Km
around the station, sometimes less. Its main advantage is that the spatial
distribution of rain over short time intervals can be determined with a significantly
better accuracy that with any other technique. As with satellite rainfall estimates,
the best results are achieved over relatively long time intervals (days and beyond)
after calibration against ground data. A reference quoted by Keane (1998)
indicates that in shower conditions, a radar calibrated against two raingauges over
1000 Km2 achieved the same accuracy as 50 raingauges.
Figure 22 illustrates a typical rainfall radar image.
63
A Digital Terrain Model is a grid of elevations.
In the first all input parameters are gridded and then input to the model at each
gridpoint. This is the approach adopted by the EU MARS project (Dallemand and
Vossen, 1995; Rijks et al., 1998). The alternative method (implemented by FAO:
Gommes et al., 1998) is to run the models only with actual data, but to
subsequently interpolate the yields using external variables (like NDVI) to guide the
interpolation. Both methods have advantages and disadvantages in terms of
reliability of input data and ease of use.
∑ X i d ia
X P = 1
n (63)
∑ d ia
1
64
This section and the next (SEDI) are based on Gommes and Hoefsloot, 1998.
The concepts of this interpolation method and software implementing the
technique have been described by Hoefsloot, 1996. The SEDI functions were
recently incorporated into the WINDISP_3 software (Pfirman and Hogue, 1998)
SEDI is a simple and straightforward method for 'assisted' interpolation. The
method can be applied to any parameter of which the values are available for a
number of geographical locations, as long as a 'background' field is available that
has a negative or positive relation to the parameter that needs to be interpolated.
Three requirements are a prerequisite for the application of the SEDI method:
1. The availability of the parameter to interpolate as point data at different
geographical locations (e.g. rainfall, potential evapotranspiration, crop yields). In
the present case of statistical variables, they were assigned a co-ordinate
corresponding to the centre of gravity of the administrative unit;
2. The availability of a background parameter in the form of a regularly spaced
grid (or field) for the same geographical area (e.g. the above-mentioned NDVI
variables, altitude).
3. A monotonous relation, at least locally, between the two parameters (negative
or positive; Yield/NDVI is positive, temperature/altitude would be negative). A
Spearman rank correlation test can reveal whether a relation exists, and how
strong this relation is.
The SEDI method yields the parameter mentioned under point 1 as a field (i.e. an
image covering the whole area under consideration).
Let us illustrate the method below using rainfall and CCD: it is implemented in
three steps (i) extracting CCD values from the satellite image and calculating the
ratio of point and image values; (ii) gridding the ratios to form a regularly spaced
grid, using any method, for instance inverse distance weighting or kriging; (iii)
multiplying the grid of interpolated ratios with the grid of CCD (image) pixel by
pixel to obtain an estimated rainfall grid (image).
There are several variants of the interpolation method thus described, for instance
the one described by Herman et al., 1997. The authors include an estimate of
orographic rain based on clouds with a relatively warm top. Note that the
described methods apply only to tropical conditions.
65
The method applies mostly to high value and mostly wind pollinated crops such as grapes.
Airborne pollen is sampled and calibrated against production in the surrounding area. The method
is currently under-developed regarding the physico-physiological emission and capture of pollen by
plants as a function of environmental conditions, transportation of pollens by air, the trapping
efficiency including trap behaviour and effect of atmospheric agents, sp. rain.
• insufficient knowledge of agro-pedo-meteorological growth conditions and yield
for the various regions of Europe;
• poor timeliness of some of the inputs.
It is suggested that an additional point could be mentioned, maybe the most
serious one: the very long “distance” between the raw weather data and the final
yield estimate at the regional scale. The “distance” would be measured in terms of
pre-processing (indirect estimation of radiation, area averaging for many
variables, etc.) and the processing by the internal machinery of the models. The
reporter suggests that many process-oriented models are too complex for regional
applications. Sensitivity analysis normally refers to model parameters, not to the
input data, in particular the weather data which are “given”. It would be most
interesting to artificially contaminate the input data with a random factor or
increasing magnitude to see what fraction of estimated detrended yield can
actually be assigned to weather. The suggested answer is that the yield values
would not be very dependent on some of the weather inputs.
Figure 23 : Yield forecast F at week 8, and harvest H at week 18. The black dot
represents the estimated value, together with its confidence interval.
The main difference between crop modelling per se and yield forecasting is the
fact that a forecast needs future data, i.e. an estimation of weather data between
now, the time of the forecast, and the harvest. This is illustrated in figure 23.
Several techniques can be used: either one uses “normal” weather, or the
historical data that have occurred between F and the time of harvest, or one uses a
random weather simulator. The two last approaches are preferable in that they
provide not only a yield estimate, but also a confidence interval.
In particularly favourable conditions, when more reliable long range forecasts are
available, a run can be done with no actual observations long in advance. Meinke
and Hammer (1997) use an ENSO forecast to extend the range of their models.
This information is available shortly after sowing a crop and at least 3-5 months
before harvest.
An actual example of the evolution of a Irish potato forecast is shown in Figure 24,
based on a figure given by MacKerron (1992).
Figure 24 : Evolution of a white potato forecast using actual 1985 data and the
long-term average ( after MacKerron, 1992).
72
71
70
69
68
67
Potato yield (T/Ha)
66
65
64
63
62
61 1985 60.2
60 Average
59 Actual
58
57
5 7 9 11 13 15 17 19 21 23
Weeks after planting
66
The coefficient of determination is the square of the coefficient of correlation. It expresses which
percentage of the variance is accounted for by the trend.
The middle curve shows the detrended yield (using the quadratic trend). This is the
yield that will be used to calibrate a regional crop forecasting model. The lower
curve shows the ratio between the yield of the current year and the average of the
yields of the 4 preceding years, assuming that the trend is not significant over such
short period. The advantage of this approach is that no trend has to be
determined, and no hypothesis has to be made about the shape of the trend.
Figure 25 : Yield of total paddy in the Korean Republic between 1960 and 1994
(based on FAO statistics). The top curve (a) indicates the actual yields with their
linear and quadratic trends; the middle curve (b) is the detrended yield, i.e. the
difference (residual) between actual yield and the quadratic trend; the lower curve
(c) shows the ratio between the yield of year N upon the average of the 4 years
from N-1 to N-4.
When there is a marked trend in the weather variables, it is probably better not to
detrend the time series, but to add time as one of the variables in the calibration
process (Swanson and Nyankori, 1979; Vossen, 1990), which we can write as
where Yy is the yield of year y computed from a function f1 of time and a function f2
of simulation model outputs (Supit, 1997b). An additional reason for adopting this
approach is that management is both time and weather dependent. For a more
detailed discussion of the concepts, refer to the quoted paper by Supit (1997b).
Equation 64 corresponds exactly to what we referred to under 3.4.1 and 3.4.2 as a
yield function.
6.2 Farm-level applications
67
A simple example of this could be, for instance, a threshold of air moisture or sunshine duration
to decide on pest risk, or a threshold of salt content of water to decide on irrigation-salinity risk.
Normally, other parameters (economic) also play an important part.
profitable cropping strategy. The authors stress the importance of accurate soil
moisture information.
68
Other methods include historical and geological analogues.
estimated separately, and constitutes one of the inputs in Moore and Osgood's
approach.
69
Largely taken from Gommes, 1997a. Although pests and diseases are not the object of this
section, it is worth noting that many models developed in the general field of plant pathology can
often be associated to the crop-weather models in impact assessments and warning systems.
For an overview of such models, refer to Seghi et al., 1996. Most of them are typical developed
country applications, where both data availability and good communications permit their
implementation in a commercial farming context.
• forecasts can be issued early and at regular intervals from the time of planting
until harvest. As such, they constitute a more meaningful monitoring tool
than the monitoring of environmental variables (e.g. rainfall monitoring);
• forecasts can often achieve a high spatial resolution, thus leading to an
accurate estimation of areas and number of people affected.
Due to the large number of institutional and technical partners involved in EWS,
interfacing between disciplines has been a crucial issue. For instance, crop prices
are usually provided as farmgate or marketplace prices, food production and
population statistics cover administrative units, weather data correspond to points
(stations) not always representative for the agricultural areas, satellite information
comes in pixels of varying sizes, etc. GIS techniques, including gridding, have
contributed towards improving links in the “jungle” of methods and data (Gommes,
1996).
Figure 26 : Recent variations of wheat, rice and maize prices between 1990 and
1996 (fixed 1996 CIF 71 US$ prices). The tics on the X-axis represent the beginning
of the respective years.
70
The main factors affecting prices are world production forecasts, speculation, weather, stocks
and the time of the year.
71
Cost Insurance and Freight.
A similar, but more dramatic situation occurred in 1977 with coffee prices when
they reached their all-time high due to low stocks and frost in some of the main
producing areas in Brazil (Brazil produces about 28 % of the world output of which
more than half comes from São Paulo and Minas Gerais).
Commercial forecasts are now available for a subscription. CROPCAST
(CROPCAST, 1994), for instance, provides estimates not only for yields, but also
for production, areas, stocks, crop condition and futures prices.
At a more local scale, many food processing plants depend on the production in
their area, and such production is linked to the seasonality of production for most
crops (canning of fruit and vegetables, sugar from sugar beet, cotton fibre
processing, oil from sunflowers and oil palm72, etc.). It is important that not only the
volume to be processed be known in advance, but also the timing.
Next to forecasting yields and production, there is thus a second category of
forecasts regarding phenology, especially maturity date: many fruits are still
harvested by hand, and the logistics of hiring the labour, storing and transporting
the produce, and marketing it is best planned as long as possible in advance. The
application has the largest potential for high-value fragile crops like grapes (Due
et al., 1993; Riou, 1994), vegetables73 (Bazlen et al., 1996) and flowers (when they
are grown in the open).
72
Oilpalm and other palms pose a series of very specific forecasting problems due to the very long
lag between flower initiation and harvest. This period usually covers 3 years and more. In addition,
probably more than in other plants, qualitative factors are very critical, for instance the effect of
temperature on sex differentiation (only female flowers produce seeds, thus oil). See Blaak (1997)
for details.
73
The paper by Bazlen also includes an example of a “biometric” forecast combined with a more
classical agrometeorological approach. In biometric forecasts, some characteristic size is
A new category of forecast has been gaining importance over the recent years:
forecasting quality of products. This regards not only the very impressionistic74
wine market (Desclée, 1991; Ashenfelter et al., 1995; Jayet and Mathurin, 1997),
but also some cereals entering an industrial process where, for instance,
starch/protein ratios should ideally remain within a relatively narrow range.
measured on a plant at a typical time (eg. cob length in maize) and used as a forecasting variable,
alone or in combination with other factors.
74
“Impressionistic” because next to quality proper (defined by pH, tannin content, sugar, colour,
etc.) the manipulation of demand plays a prominent role, particularly during average and mediocre
years (see Ashenfelfter et al., 1995).
based essentially on the drying conditions. One of the reasons why models have to
be used is the absence of a structure that measures, stores and markets forage
crops that is comparable to grain crops. In addition, field surveys are significantly
more expensive to carry out than forecasts.
7. Chapter seven : Exercises
RA = Sc ∫ cos(z) dz (65a)
z at sunrize
J = round (304
. M − 31, 0) + Date ; (66)
for the fifth day of a dekad (Dek = 1, 2, 3) in a month, use
J + 10
δ = − 2 3.4 5 cos( 2 π ) (69)
365
δ is in degrees (but the arguments of the trigonometric functions are in radians).
75
Unless otherwise specified, the formulae in this section are from Supit et al., 1994.
Before proceeding, it is necessary to compute the seasonal offset of the sine of
the solar height (SS) and the amplitude of the sine of solar height (CC) . Both are
derived directly from declination δ and latitude λ as follows:
24 CC + SS
D = Dr + 2 × × arctan (71)
π CC − SS
where D r is a daylight refraction correction factor. As the sun becomes visible due
to refraction of sunlight by the atmosphere before it actually crosses the horizon,
the actual day length is slightly longer than the astronomical day length by Dr (about
10 to 15 minutes, or 0.16666... to 0.25 decimal hours).
7.3 Beta distribution model for development rate
Using equation of Xinyou Yin et al. (1995),
Next compare NPP as a function of rainfall as given by Lieth’s Miami model and
the curves listed by White et al. (1992) :
VERIFICATION
Sum of above 2324.8 2379 2779 2512 2642 2684.8
Sum 1 Jan - harvest 2431.3 2397 2847 2549 2662 2686.5
AVERAGE TEMPERATURE
Transpl. to heading 22.729 22.57 25.94 23.95 25 24.718
1/D 0.0118 0.011 0.012 0.012 0.012 0.012
Figure 32 : Empirical determination of base temperature for Ilpoon paddy rice
grown at Suwon (Korean Republic) between 1992 and 1997: coefficient of variation
of the length of 3 phases as a function of temperature. The curves correspond to
planting to transplanting (P → T), transplanting to heading (T → F) and heading to
harvest (F → M).
Figure 33 : Rate of development (d-1) of Ilpoon paddy rice grown at Suwon (Korean
Republic) between 1992 and 1997 between planting and transplanting. The lowest
point (7,0) is not an observed point but corresponds to the base temperature.
IF(T< Base,0,
IF(T> Cutoff,Cutoff -Base,T-Base)) (72)
• Compute the average sum of temperatures, the standard deviation and the
coefficient of determination over the 6 years available, from 1992 to 1997. By
modifying Tb it will be observed that the Coefficient of variation goes through a
minimum at 7 °C. Note that we call this the “statistical base temperature”
because there may also be a physiological base temperature. The plot of the
coefficient of variation against the base temperature is shown in Figure 33;
• compute average temperatures over the time from planting to transplanting ,
and plot the rate of development (DR) against temperature. The rate of
development is simply the reciprocal of the length of the phase in days. If the
phase is long, DR is low, and vice-versa. Do not forget to include the value of Tb
for which DR is nil, by definition. The plot is shown in Figure 33. What are the
rates of development and the time from planting to transplanting at 10 °C, 15
°C, 20 °C, 25 °C and 30 °C ? Compute the three corresponding Q10, and
discuss the concept;
• based on the plot of DR as a function of temperature, make some guesses
about the optimum and cut-off temperatures (To and Tu). It is suggested that the
optimum is between 25 °C and 30 °C, which puts the cut-off temperature at
least to 35 °C;
• using the values just determined for the base and cut-off temperatures (7°C and
35 °C), determine the sums of temperatures for the given phenophases. The
value below are found:
0.5 2.5
0.45
2
0.4
0.35
Biomass (Kg DM / square m)
1.5
0.3
LAI
0.25 1
0.2 Root
Leaf
0.5
0.15 Tuber
Total
LAI
0.1
0
0.05
0 -0.5
1 21 41 61 81 101 121 141 161
Days
The values for the other days of the cycle (assumed to be 150 days) have to be
linearly interpolated between the values above using the Fill and Series options of
the spreadsheet. Please pay attention to the fact that only values for two of organs
must be interpolated (e.g. roots and leaves); the third has to be computed as 1 -
root - leaf (why?).
The model is an extremely simple one : LAI is obtained from the leaf biomass
applying a specific leaf area (SLA) factor provided in the table below (table 10).
The specific daily assimilation Ws [Kg (DM) m-2 (leaf)] is assumed to be constant,
hence the daily assimilation rate [Kg (DM) m-2 (ground)] is simply the product:
∆W = Wr × LAI (73)
In the next step we partition ∆W into roots, leaves and tuber according to the
partitioning table, resulting in ∆Wr, ∆Wl and ∆Wt.
On the second and on all subsequent days, the new root and leaf biomass are
computed as
100 − Rd
Wrd = (Wrd −1 + ∆W r ) (74)
100
and
100 − R l
Wl d = (Wl d − 1 + ∆Wl ) (75)
100
Wt d = Wt d −1 + ∆W t (76)
where Wrd indicates root biomass on day d, etc. The factors (100-X)/100, where X
(Rd and Rl ) is a decay factor in percent can be regarded as equivalent to
maintenance respiration or conversion efficiency. The factor was introduced to
“force” root death. Another technique which is often implemented in models
assigns leaves a life expectancy expressed in Degree-Days.
Table 10 : Parameters and variable used for a simple root crop simulation with a
spreadsheet
Table 11 : Spreadsheet layout for simple root crop simulation with a spreadsheet.
The results of the simulation are given in figure 34. Seminar participants will vary
the values of the model parameters given in table 10 to study their effect on the
simulation.
7.7 Simple soil-crop water balance
The purpose of this exercise is to compute a soil water balance for an irrigated
maize crop with a cycle of 120 days (emergence at day d=1, harvest at day
d=120), the growth of which is given by
LAI m a x
LAI =
1 + 220 × G S F × e − 0 .1d (77)
in m2 (leaf) m-2 (ground) where LAImax = 3.3 and GSF (Growth Shape Factor) = 1.
The student is invited to experiment with this curve and to determine the
significance of the GSF, including negative values!
Whenever sufficient moisture is available, crop transpiration T0 and evaporation E0
from soil are proceeding at the potential rate ET0 which is assumed to be constant
at ET0 = 6 mm d -1 .
ET0 is partitioned into E 0 and T0, which are given by
LAI
T = ET 0
0 (78)
LAI max
LAI
E 0 = ET 0 (1− ) (79)
LAI max
The soil is a 1 layer soil with a maximum water holding capacity (WHC) of 150
mm and a depth of 1 m from where evaporation takes place and which is also the
depth eventually reached by the roots.
No ET takes place during rain, which is to say that water infiltrates first, and only
subsequently is it being depleted by evaporation and transpiration. Moisture from
rainfall and irrigation (total: Prec mm) distributes homogeneously over the whole
profile. No horizontal nor ascending vertical movements of water take place from
below 1 m. Initial soil moisture is 75 mm. All water infiltrates instantaneously. The
soil water behaviour is described by
H d = H d −1 + Prec if H d < W H C
Hd = W H C if H d > W H C (80)
Table 12 : Rainfall amounts (mm) for simple soil water balance simulation with a
spreadsheet
Ta = Hr × T0 and Ea = Hr × E0
2
(81)
As soil mositure is depleted, both actual T (Ta) and Ea drop to zero, with the
consequence that
Figure 36 : Some simulated model variables (soil moisture, water supply as rainfall
and irrigation, potential transpiration and potential evaporation).
160 8
H
140 R+Irr 7
Tmax
120 6
Emax
100 5
80 4
60 3
40 2
20 1
0 0
0 10 20 30 40 50 60 70 80 90 100 110 120
The following inputs, parameters and variables will be used (tables 13a and 13b).
The parameters will be assigned names in the spreadsheet so that they can be
used to experiment with their numerical values.
Table 13a : Inputs and parameters and variables used for a simple soil water
balance simulation using a spreadsheet.
Variable Unit
RI = Precip + Irrigation mm
LAI m2(leaf)m-2 (ground)
T0 Maximum transpiration mm
E0 Maximum evaporation mm
Ta Actual transpiration mm
Ea Actual evaporation mm
Hr Relative soil moisture -
Hd Actual soil moisture mm
∆ Wd Biomass growth Kg (DM) day-1
Wd Accumulated biomass Kg (DM) m-2
Yield tons (grain) Ha -1
Although soil moisture Hd is, strictly speaking, only one variable, it will be
computed in several steps, which we will identify as Hd(1), Hd(2) and Hd(3).
The final value on a given day is Hd(3) and quite obviously Hd-1 (3) is the final value
of the previous day.
The steps are listed below.
1. New soil moisture is soil moisture at the end of the previous day (Hd previous)
plus water supply RI, i.e. the sum of rainfall and irrigation. If the sum exceeds
capacity, it is set to capacity:
Hd(1) = IF(Hd previous+RI > WHC, WHC, Hd previous+RI)
Note that on day 1 Hd previous is H0, the initial soil moisture - and one of the
parameters -, and Hd-1 (3) on the second and all subsequent days;
2. Potential and actual values of evaporation and transpiration (E 0, T0, Ea and
Ta) are computed next, and will subsequently be subtracted from Hd(1),
leading to Hd(2):
Hd(2) = Hd(1) -Ta - E a
3. Since the resulting value of Hd(3) might drop below 0 (although this is
unlikely, given the way they tend to 0 when relative soil moisture drops close
to 0), we apply the final EXCEL function
Hd(3) = IF(Hd(2) < 0, 0, Hd(2))
The seminar participants will prepare the layout of the spreadsheet in such way
that the parameters are at the top of the spreadsheet proper, leaving enough
space right of the parameters to display a graph showing soil moisture (as in
figure 38).
The spreadsheet proper will be arranged in the logical order presented in table 14.
Note that the table is transposed in comparision with the previous exercise (table
11).
Table 15 : Layout of spreadsheet for simple soil water balance simulation. RI, the
sum of rainfall and irrigation is the total water supply. Hd-1(3) stands for soil
moisture at the end of the previous day. The initial biomass (seed) is assumed to
be negligible.
The final part of the exercise will be a series of graphs showing the sensitivity of
yield to the different parameters (in particular the GSF; do not forget negative
values; pay particular attention to the behaviour between -0.05 and 0.05) and a
validation of the model using soil moisture considering that the values in table 15
were actually measured.
Table 16 : “Observed” soil moistures to validate a simple soil moisture balance
using a spreadsheet
Day 9 21 35 51 58 78 112
Soil moisture 119 138 121 81 71 102 93
Critically assess the model and list strong and weak points, as well as areas for
improvement.
7.8 Introduction to CropSyst
Option 1 : simple ETP, requires daily rainfall , maximum and minimum temperature
in ASCII formatted as
Day_number_(1-365) rainfall(mm) Tx(C) Tn(C) space separated, like in
1 0 32.1 23.4
2 10.1 34.0 26.2
Option 2: ETP according to Priestley-Taylor. Input data as above, plus radiation in
MJ m-2 day-1 .
1 0 32.1 23.4 3.2
2 10.1 34.0 26.2 2.6
Option 3: ETP according to the Penman-Monteith approach. Inputs as above, plus
maximum relative humidity (%), minimum RelHum (%) and windspeed (m/s)
1 0 32.1 23.4 3.2 92 82 1.5
2 10.1 34.0 26.2 2.6 95 80 2.1
Nitrogen uptake adjustment is an empirical factor to adjust for nitrogen uptake per
unit root length. Nitrogen availability adjustment should be left at 1. Amount of
residual nitrogen per soil layer is the amount that will not be extracted from the soil
layers by the model. If set to 0, all N can be used.
7.8.3.3 HARVEST-INDEX
Sensitivity to water stress during flowering and during grain filling: use 0 if stress at
those stages is insignificant or can be ignored. Translocation factor is the fraction
of above-ground biomass at flowering that can be traslocated to grain.
Note that LAI and SLA must be consistent with observed yield data and harvest
index.
Fraction of maximum LAI at physiological maturity includes senescent and green
leaves. SLA applies only to above-ground biomass. Extinction coefficient is
defined under 2.2.3 (also see last paragraph of this section). Leaf duration
sensitivity to water stress: during water stress, leaf temperature increases, and the
life of leaves shortens; the parameter indicates ho sensitive leaves are in this
respect.
Stem/leaf partition is slightly more difficult. Claudio Stöckle explains it as follows
(personal communication): the following equation represents the LAI production as
a function of above ground biomass:
where LAI = Leaf area index (m2 leaf / m2 ground), SLA = specific leaf area (m2
leaf / kg leaf biomass), B = Aboveground biomass (Kg biomass / m2 ground) and
p = stem-leaf partitioning coefficient (m2 ground / kg biomass).
p has the inverse units of B to that (1 / denominator) is a unitless fraction. When B
is very small (near zero), the fraction is essentially one and near all biomass is
"leaf biomass". As B increases, the fraction become less and less than 1,
implying that less of the total aboveground biomass is leaf biomass. CropSyst
uses the derivative of this equation is used to calculate the daily amount of LAI
produced as a function of "today’s biomass" and currently accumulated biomass.
25
20
15
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4
Biomass (Kg / sqm)
The extinction coefficient is used directly to partition ET into transpiration and soil
evaporation. Values vary from 0.35 (vertical leaves, or erectophile) to 0.65 for
heliotropic leaves). “We have found that, from theoretical considerations, PAR
extinction coefficient is around 40% larger than the coefficient for solar radiation.
This concept is used to determine PAR interception” (C. Stöckle).
DegC-days are accumulated from planting. Thermal time does not accumulate
below the base temperature or above the cut-off temperature. Phenologic
sensitivity to water stress indicates how cycle length reacts to water stress (i.e.
increased SDD; use 0 is cycle lengthens).
newton newton × m
pascal = 2
=
m m3
joule 0001
. joule
= 3 = (84)
m kg
The newton (N) is the force required to accelerate 1 kg by one meter per second
per second (1 N = 1 Kg m s-2 ). Much work on plant water potential still uses the
bar which is equivalent to 105 N m-2 so that 1 bar is 100 J Kg-1 ; 1 millibar is 102
pascal, or 1 hectopascal (hPa).
On the other hand, a pressure is an energy per volume, as shown above: 1 J kg-1
is 1000 Pa or 1 kPa or 100 bar.
Work, energy, power, radiation and evaporation
The joule is the practical unit of work; it is equal to a force of one newton (N) acting
over a distance of 1 meter, thus 1 J = 1 N m. Since energy is “stored” work, the
units used for energy and work are the same.
The joule has superseded the still ubiquitous calorie, a cgs unit, representing the
quantity of heat required to raise one gram of water through one degree Celsius.
One calorie is 4.18 Joule and the specific heat of water at 15°C is 4185.5 joule
-1
..
The watt (W) is the practical SI unit of power. It is the power dissipated when 1
joule is expended in 1 second (W = J/s). 1 J = 2.78 10-7 kWh and 1 W = 0.2388
cal s-1 .
Radiation (solar, incoming short-wave, outgoing long-wave, net radiation, etc.) is
now mostly expressed in J m-2 d-1 . Alternative expressions are W m-2 .
For instance, Stefan’s law, which states that the energy emitted by a full radiator is
proportional to the fourth power of its absolute temperature:
B = σT 4
(85)
where B is in W m-2 , the Stefan-Boltzman constant σ = 5.57 10-8 W m-2 K-4 and
the absolute temperature is T (°K).
The solar “constant”, i.e. the amount of solar energy which reaches the upper
atmosphere is now known to be variable (thanks to the measurements carried out
by satellites - ERB (Earth Radiation Budget, 1978), ERBS (Solar Maximum
Mission satellite, 1980), NOOA9 in 1984, and NOOA10 in 1986 - and close to
1366 W m-2 (which is approximately 2 cal cm-2 min-1 ).
The solar constant amounts to about 400 TJ hectare-1 year-1 at the equator. Due to
a number of factors (rotation of the earth and alternating light and darkness
periods, albedo of the earth, absorption in the atmosphere, clouds etc.) only about
50 TJ hectare-1 year-1 is actually available at plant level.
In crop modelling, energy absorbed or emitted by a surface is often expressed in
mm of water, i.e. the amount of energy required to evaporate (condense) 1 mm of
water spread over 1 square meter (1 litre of water per square m). L, the latent heat
of vaporisation of water76 is 2.45 106 J kg-1 . The product of L and E, the rate of
water loss from a surface (kg m-2 d-1 ), LE, is thus the evaporative heat loss (J m-2 d-
1
).
Physiologists in discussing metabolism use the large calorie or Kilocalorie which
is in fact 1000 calories. For example, a dish listing its energy content as 128 Kcal
or 537 KJ, will release 537000 J when burnt in a bomb calorimeter.
Biomass energy content is usually set at 15 GJ per ton at 20% moisture.
Others
Atmospheric concentrations of CO2 are often expressed in ppmv, equivalent to
parts per million in volume, or millilitres per cubic meter. Since one mole (M g)
occupies 22.4 litres under standard conditions
M mg
1 ppmv = (86)
22.4 m3
the current CO2 concentration of 360 ppmv amounts to 360 x 44 / 22.4 ≈ 700 mg
per m3.
76
The latent heat varies with temperature. It is 2.46 MJ Kg-1 at 10°C and 2.42 MJ Kg-1 at 30°C
Multiples
The common multiple-prefixes are well known (milli, kilo, etc.). The list below
indicates the larger multiples with their symbols
77
An updated version (December 1996) with the newer software was included in the proceedings of
a training workshop organised by the Fondation Universitaire Luxembourgeoise in Arlon (Belgium)
in September 1996.
Gommes, R., and P. Hoefsloot, 1998. Gaps in maps, estimation of missing data in agricultural
statistics maps. Proc. of the the COST (EU Committee on Science and Technology)
meeting in Volterra (Italy, Sep. 1997) on the Spatial Interpolation of Climatic and
Meteorological Data. In press.
Gommes, R., F.L. Snijders and J.Q. Rijks, 1998. The FAO crop forecasting philosophy in national
food security warning systems pp. 123-130, in: D. Rijks, J.M. terres and P. Vossen
(Eds), 1998.
Guillot, B., 1995. Satellite et précipitations. Constraintes techniques et physiques, analyse de
quelques méthodes, problèmes de recherche et de validation. Satellite and
precipitations. Technical and physical restrictions, analysis of some methods, research
and validation problems. Veille Climatique Satellitaire, ORSTOM, 55:27-58.
Hashmi,M.A., L.A. Garcia and D.G. Fontane, 1995. Spatial estimation of regional crop
evapotranspiration. Transaction ASAE, 38(5):1345-1351.
Haskett, J.D., Y.A. Pachepsky and B. Acock,. 1997. Increase of CO2 and climate change effects
on Iowa soybean yield, simulated using GLYCIM. Agronomy Journal, 89(2):167-176.
Haxeltine, A., and I.C. Prentice, 1996a. BIOME3: An equilibrium terrestrial biosphere model
based on ecophysiological constraints, resource availability and competition among
plant functional types. Global Biogeochemical Cycles. 10: 693-709.
Haxeltine, A., I. C. Prentice and I. D. Cresswell, 1996b. A coupled carbon and water flux model to
predict vegetation structure. Journal of Vegetation Science, 7: 651-666.
Herman, A., V.B. Kumar. P.A. Arkin and J.V. Kousky, 1997. Objectively determined 10-day
African rainfal estimates for famine early warning systems. Int. J. Remote Sensing,
18(10):2147-2159.
Hess, T., 1996. A microcomputer scheduling program for supplementary irrigation. Computers and
Electronics in Agriculture, 15(3)233-243.
Hijmans R.J., I.M. Guiking-Lens, C.A. van Diepen, 1994. WOFOST 6.0: User's guide for the
WOFOST 6.0 crop growth simulation model. Wageningen, DLO Winand Staring Centre,
Technical Document 12. 1994-146 pp.-48 figs.- 78 ref. ISSN 0928-0944
Hillel, D., 1971. Soil and water, physical principles. Academic Press, New-York, 288 pages.
Hoefsloot, P. 1996. IGT manual, Ver. 1.10. Working paper series N. 5. SADC/FAO,
GCPS/RAF/296/NET, Harare, 53 pp. Programme and manual are retrievable from
FTP://FTP.FAO.ORG/SDRN/IGT.
Holzmann, A., and P. Niemann, 1988. Prediction of Viola arvensis infestations based on
parameters of population dynamics.Zeitschrift fur Pflanzenkrankheiten und
Pflanzenschutz, Sonderheft 11, pp. 91-96.
Hoogenboom, G., J.W. Jones, L.A. Hunt, P.K. Thornton and G.Y. Tsuji, 1994 . An integrated
decision support system for crop model applications. American Society of Agricultural
Engineers paper No. 943025, 23 pp.
Hough, M., 1998. The incorporation of radar with raingauge observations. Pp 247-254 in: D. Rijks,
J.M. terres and P. Vossen (Eds), 1998.
Hough, M.N., R. Gommes, T. Keane and D. Rijks, 1998. Input weather data, pp. 31-55, in: D.
Rijks, J.M. terres and P. Vossen (Eds), 1998.
Hudson, G., and H. Wackernagel, 1994. Mapping temperature using kriging with external drift:
theory and an example from Scotland. Internat. J. Climatology, 14:77-91.
Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment,
25:295-309.
Hutchinson, M.F., 1991. The application of thin plate smoothing splines to continent-wide data
assimilation. In : J.D. Jasper (ed), Data assimilation systems, BMRC research report N.
27, Bureau of meteorology, Melbourne, pp. 104-113
Hutchinson, M.F., 1994. Stochastic space-time weather models from ground-based data. Agric.
Forest Meteorol., 73:237-264.
Hutchinson, M.F., and J.D. Corbett, 1995. Spatial interpolation of climate data using thin plate
smoothing splines. Pp 211-224 in : FAO, 1995.
Jackson, R.D., S.B. Idso, R.J. Reginato and P.J. Pinter, 1981. Canopy temperature as a crop
stress indicator. Water Resources Res., 17:1133-1138.
Jayet, P.A., and J. Mathurin, 1997. Wine market regulation and bargaining within the European
Union. Bulletin l'OIV (France), 70:(795-796):419-444..
Jerrard, H.G., and D.B. McNeill, 1986. A dictionary of scientifc units, including dimensionless
numbers and scales. Chapman and Hall, London and New York. 222 pp.
Jones, C.A., and J.R. Kiniry. 1986. CERES-Maize: A simulation model of maize growth and
development.Texas A&M University Press, College Station, Texas.
Jones, C.A., W.L. Bland, J.T. Ritchie and J.R. Williams, 1991 . Simulation of root growth.
Agronomy, 31:91- 123.
Jones, P.G., P.K. Thornton and P. Hill, 1997. Agro-meteorological models: crop growth and stress
indices. Pp. 53-65 in: EU/FAO, 1997.
Jordan, J.A. (Jr.) and L.J. Shieh, 1981. Information science technology and agricultural
management. Pp 166-178 in A. Weiss (ed) , 1981.
Kamel, A., K. Schroeder, J. Sticklen, A. Rafea, A. Salah, U. Schulthess, R. Ward and J. Ritchie,
1995. Integrated wheat crop management based on generic task knowledge-based
systems and CERES numerical simulation. AI Applications, 9(1):17-28.
Karim, Z., S.G. Hussain, M. Ahmed M and L. Erda, 1996. Assessing impacts of climatic
variations on foodgrain production in Bangladesh. Water, Air, and Soil Pollution, 92(1-2):
53-62.
Keane, T., 1998. Analysis of Dublin Airport radar rainfall amounts, pp. 255-258 in : D. Rijks, J.M.
Terres and P. Vossen (Eds), 1998.
Kiniry, J.R., J.R. Williams. P.W. Gassman and P. Debaeke, 1992. A general, process-oriented
model for two competing plant species. Transactions of the ASAE., 35(3):801-810.
Klein, G.H., 1997. Tools for detection of hidden dynamics in seasonal development during
phenologically opaque periods. Internat. J. Biometeorol., 40(3):123-127.
Kleinhenz, B., and V.Gutsche, 1996. Validation and utilization of the forecasting models
SIMCERC and SIMPHYT in plant protection advice. Statens Planteavlsforsog report No.
15, pp. 49-58, Denmark.
Kogan, F., 1997. Global drought watch from space. Bul. Am. Met. Soc., 78(4):621-636.
Kogan, F.N., 1995. Droughts in the late 1980s in the United States as derived from NOAA polar-
orbiting satellite data. Bul. Am. Met. Soc., 76(5):655-668.
Körner, Ch., 1995. Towards a better experimental basis for upscaling plant responses to elevated
CO2 and climate warming. Plant, Cell and Envir., 18:1001-1010.
Kramer, P.J., 1983. Water relations of plants. Academic Press, New-York, 489 pages.
Kropff, M.J., C.J.T. Spitters, B.J. Schneiders, W. Joenje and W. de Groot W, 1992. An eco-
physiological model for interspecific competition, applied to the influence of
Chenopodium album L. on sugarbeet. Weed Research Oxford, 32(6):451-463.
Kumar, L. and J.L Monteith. 1981. Remote sensing of crop growth, In: M. Smith, Ed., Plants and
daylight spectrum. Academic press, London. pp. 133-144.
Kustas, W.P., and J.M. Norman, 1996. Use of remote sensing for evapotranspiration monitoring
over land surfaces. Hydrological sciences, 41(4): 495-516.
Laguette, S., 1995. Combined use of NOAA-AVHRR indices and agrometeorological models for
yield forecasting, pp. 197-208 in: Dallemand and Vossen, 1995.
Laguette, S., 1997. Utilisation des données NOAA-AVHRR pour le suivi du blé à l’échelle de
l’Europe. Thèse présentée pour l’obtention du grade de docteur de l’ENGREF le 28
1997, Ecole Nationale du Génie Rural, des Eaux et des Forêts, Montpellier. 167 pp.
Laguette, S., A. Vidal and P. Vossen,1998. Using NOAA_AVHRR data to forecast yields at the
European scale, pp. 131-146 in: D. Rijks, J.M. terres and P. Vossen (Eds), 1998.
Landau,S., and and V. Barnett, 1996. A comparison of methods for climate data interpolation, in
the context of yield predictions from winter wheat simulation models. Aspects of Applied
Biology, 46:13-22.
Lauciani, E., and A. Ponticiello, 1993. Photosynthetically active radiation (PAR): empirical
determination and literature survey. FAO, Rome. Agrometeorology Series Working
paper N.7, 85 pp.
Laughlin, G.P., M.F. Hutchinson and B.G. Mackey, 1993. An intuitive approach to analysing small
point-source spatial data sets. Internat. J. geographical information systems, 7(1):21-38.
Le Toan, T., H. Laur, E. Mougin and A Lopes,1989. Multitemporal and dual polarisation
observatons of agricultural vegetation covers by X-band SAR images. IEEE transactions
on geoscience and remote sensing, November 1989, 27(6):709-718.
Li, Zhanqing, H.W. Barker and L. Moreau, 1995. The variable effect of clouds on atmospheric
absorption of solar radiation. Nature, 376: 486-490.
Lieth, H. 1972. Modelling the primary productivity of the world. Nature and Resources, UNESCO,
VIII, 2:5-10.
Lieth, H. 1973. Primary production: terrestrial ecosystems, Human Ecology, 1(4)303-332.
Lieth, H. 1975. Modelling the primary productivity of the world, Primary productivity of the
biosphere, Springer-Verlag, Berlin. pp. 237-263.
Lourens,U.W., C.M. van Sandwyk, J.M. de Jager and J. Van den Berg, 1995. Accuracy of an
empirical model for estimating daily irradiance in South Africa from METEOSAT
imagery. Agric. Forest Meteorol., 74: 75-86.
Lyamchai, C.J., T.J. Gillespie and D.M. Brown, 1997. Estimating maize yield in northern Tanzania
by adapting SIMCOY, a temperate-zone simulation model. Agric. Forest Meteorol.,
85(1-2):75-86.: 1 2, 75 86, 26 ref.
MacKerron, D.K.L., 1992. Agrometeorological aspects of forecasting yields of potato within the
E.C.. EUR 13909 EN, Official Publ. EC, Luxembourg, pp. 247.
Mahmood, R., 1997. Impacts of air temperature variations on the boro rice phenology in
Bangladesh: implications for irrigation requirements. Agric. For. Meteorol., 84(3-4):233-
247.
Maidment, D.R., 1992. Handbook of hydrology. McGraw Hill. Pages numbered by chapters.
Maracchi, G., and M.V.K. Sivakumar, 1995. Coordeination and harmonization of databases and
software for agroclimatic applications : discussion document. Pp. 11-38 in : FAO, 1995.
Marcus, K.G. and L.J. Heitkemper, 1997. Commodity market planning worldwide. Pp. 45-49 in:
EU/FAO, 1997.
Matthews, R.B., M.J. Kropff, T. Horie and D. Bachelet, 1997. Simulating the impact of climate
change on rice production in Asia and evaluating options for adaptation. Agric.
Systems, 54(3):399-425.
McCoy, E.L., 1991. A bioenergetic model of plant root growth in soil. Agric. Systems, 37(1):1-23.
Meinke, H. and G.L. Hammer. 1997. Forecasting regional crop production using SOI phases: an
example for the Australian peanut industry. Australian J. Agric. Research, 48(6):789-
793.
Mitchell, G., R.H. Griggs, V. Benson, J. Williams, B. Vanicek and D. Dumesnil, 1995. EPIC
User’s guide, Texas Agricultural research center (Blackand Research Center), USDA-
ARS (Grassland, Soil and water Research Laboratory), USDA (Natural Rsource
Conservation Service), Temple, Texas, USA. About 200 pp (depending on font).
Monteith, J.L., 1965. Evaporation and environment, in: G.E. Fogg (Ed.) The state and movement of
water in living organisms. Cambridge University Press, London.
Moore, P.M., and R.V. Osgood. 1987. Evaluation of a model for predicting sucrose yields following
hurricane damage to sugar-cane in Hawai. Agric. For. Meteorol., 40(2):191-197.
Myers, D.E., 1994. Statistical methods for interpolating spatial data. J. Appl. Science &
Computations, 1(2):283-318
Nemani,R., L. Pierce, S. Running and S. Goward, 1993. Developing satellite-derived estimates of
surface moisture status. J. Appl. Meteorol. 32: 548-557.
Nieuwenhuis, G.J.A, T. Van der Wal, C.A. Mücher and A., 1998. Integrated use of high and low
resolution satellite data and crop growth models, pp. 147-159 in : D. Rijks, J.M. Terres
and P. Vossen (Eds), 1998.
Nieuwenhuis,G.J.A., A.J.W. de Wit, T. van der Wal, C.A. Mücher, and H.J.C. van Leeuwen, 1996.
Integrated use of high and low resolution satellite data and crop growth models. DLO N.
127, 130 pp.
Nilsson, H.E., 1997. Remote sensing for assessment of disease impact. Pp. 123-142 in FAO/EU,
1997.
Nokes, S.E., F.M. Landa and J.D. Hanson, 1996. Evaluation of the crop growth component of the
root zone water quality model for corn in Ohio. Transactions of the ASAE., 39(3):1177-
1184.
Pachepsky, L.B., and B. Acock. 1994. Model of photosynthesis accounting for leaf anatomy.
Proceedings of 24th Annual Workshop on Crop Simulation. March 15-18, 1994. Raleigh,
NC. 18 pp.
Pages, L., and S. Pellerin, 1996. Study of differences between vertical root maps observed in a
maize crop and simulated maps obtained using a model for the three-dimensional
architecture of the root system. Plant and Soil, 182(2):329-337.
Pala, M., C.O. Stöckle and H.C. Harris.1996 Simulation of durum wheat (Triticum turgidum ssp.
Durum) growth under different water and nitrogen regimes in a Mediterranean
environment using CropSyst. Agricultural Systems, 51:147-163.
Palm, R., 1997a. Les modèles de prévision statistique: cas du modèle Eurostat-Agromet. Pp. 85-
108 in: Tychon, B., and V. Tonnard, 1997. Estimation de la production agricole à une
échelle régionale. Official Publications of the EU, EUR 17663, Luxembourg. 202 pp.
Palm, R., 1997b. Agrometeorological models: regression and trend analysis. Pp. 67-76 in:
EU/FAO, 1997.
Palm, R., and P. Dagnelie, 1993. Tendance générale et effets du climat dans la prevision des
rendements agricoles des différents pays des C.E. Official Publications of the EU, EUR
15106, Luxembourg. 132 pp.
Pannkuk, C.D., C.O. Stöckle and R.I. Papendick, 1998. Evaluating CropSyst simulations of wheat
mangement in a wheat-fallow region of the US Pacific northwest. Agricultural systems,
57(1). In press.
Pawer, M., 1993. Statistical validation of ecological models. Ecological modelling, 68:33-50.
Penman, H.L., 1948. Natural Evaporation from Open, Bare Soil and Grass. Proc. Royal Soc.
(London) A193:120-145.
Penning de Vries, F., M. Jansen and K. Metselaar, 1995. CAMASE guidelines for modelling.
CAMASE__NEWS (Newsletter of Agro-ecosystems Modelling) extra edition, November
1995. Downloaded from the CAMASE home page at Wageningen Agricultural University
(http://www.bib.wau.nl:80/camase/).
Petty,G.W., and W.F. Krajewski, 1996. Satellite estimation of precipitation over land. Hydrological
sciences, 41(4): 433-451.
Pfirman, E., and J. Hogue, 1998. WINDISP_3. Programme and manual are retrievable from
FTP://FTP.FAO.ORG/SDRN/WINDISP3.
Phillips,D.L., J. Dolph and D. Marks, 1992. A comparison of geostatistical procedures for spatial
analysis of precipitation in mountainous terrain. Agric. forest meteorol., 58(1-2): 119-
142.
Pilet, P.A., 1967. L’énergie végétale. Que Sais-je?, PUF, Paris. 126 pp.
Piper, E.L., K.J. Boote, J.W. Jones, S.S. Grimm, 1996. Comparison of two phenology models for
predicting flowering and maturity date of soybean. Crop Science, 36(6):1606-1614.
Plentinger, M.C. and F.W.T Penning de Vries, 1995. CAMASE register of agro-ecosystem
models, version 1. AB-DLO, Wageningen, Netherlands.179 pp.
Priestley, C.H.B. and R.J. Taylor. 1972. On the Assessment of Surface Heat Flux and
Evaporation Using Large Scale Parameters. Monthly Weather Review. 100:81-92.
Rickman, R.W., and B. Klepper, 1991. Environmentally driven cereal crop growth models. Ann.
Rev. Phytopathol., 29:361-380.
Rijks, D., 1997. Feedback between yield forecast and farm operations - a matter of weight? Pp.
41-44 in: EU/FAO, 1997.
Rijks, D., J.M. terres and P. Vossen (Eds), 1998. Agrometeorological applications for regional
crop monitoring and production assessment, Official Publications of the EU, EUR
17735, Luxembourg. 516 pp.
Riou, C., 1994. The effect of climate on grape ripening: application to the zoning of sugar content
in the European Community. Le déterminisme climatique de la maturation du raisin:
application au zonage de la teneur en sucre dans la communaute europeenne. Official
Publications of the EU, EUR 15863 FR/EN, Luxembourg. 322 pp. + 4 maps
Robertson, G.W., 1968. A biometeorological timescale for a cereal crop involving day and night
temperature and photoperiod. Int. J. Biometeorol., 12:191-223.
Rosema, A., R.A. Roebeling and D.A. Kashasha, 1998b. Using METEOSAT for water budget
monitoring and crop early warning, pp. 161-175, in: D. Rijks, J.M. Terres and P. Vossen
(Eds), 1998.
Rosema, A., R.A. Roebeling, A. Van Dijk, G.J.A. Nieuwenhuis, J. Huyghen, D.A. Kashasha,
1998a. ACMP agromet and crop monitoring project in the SADC region. Netherlands
Remote Sensing Board (BCRS) Report 96-13, Delft, 108 pp.
Russel, E.W., 1977. Soil conditions and plant growth. English Language Book Society and
Longman, London, 849 pp.
Russel, G., and R.I. Muetzelfeldt, 1998. Development of a crop knowledge base for Europe. Pp.
207-224, in: D. Rijks, J.M. terres and P. Vossen (Eds), 1998.
Russell, G., R.I. Muetzelfeldt and K. Taylor, 1998. Crop Knowledge base system. Official
Publications of the EU, EUR 17697, Luxembourg. Pages numbered by chapter.
Russell, J.S., and M.K. Wegener, 1990. Risk analysis of irrigated sugarcane using responses
generated by the AUSCANE model. Proc. Int. symposium on climatic risk in crop
production: models and management in the semi-arid tropics and subtropics, Brisbane
,Australia , 2-6 july 1990. CSRIO, Brisbane, Australia. Pp. 88-89.
Rustagi, N.K., 1988. Crop insurance in India: an analysis. Publishing Corp., India. xvi + 143 pp.
Sakamoto, C., 1995. Manual of agrometeorological tools in SADC Early Warning Systems.
SADC/FAO Regional Early Warning Systems for Food Security, GCPS/RAF/270/DEN.
Pages numbered by chapter.
Sakamoto, C.M, N.D. Strommen and S.K. LeDuc, 1977 Synoptic vs. Cooperative climate data as
inputs for wheat estimates. In: Proc. 13th Conf. Agric. For. Meteorology., Am. Meteorol.
Soc. Purdue University. Indiana. Quoted by Sakamoto and LeDuc, 1981.
Sakamoto, C., and S. LeDuc, 1981. Sense and non-sense: statistical crop growth and yield
models. Pp. 185-200 in A. Weiss (Ed), 1981.
Savin, R., A.J. Hall and E.H. Satorre, 1994. Testing the root growth subroutine of the CERES-
Wheat model for two cultivars of different cycle length. Field Crops Res., 38(3):125-133.
Seghi, L., S. Orlandin and B. Gozzini, 1996. Survey of simulation models of plant diseases. FMA,
CeSIA, IATA-CNR, Florence, Italy, 127 pp.
Seguin, B., 1992. Utilisation combinée de données satellitaires et de modèles
agrométéorologiques. Pp 199-206 in: Proc. Of the Conf. on the application of remote
sensing to agricultural statistics. 26-27 Nov. 1991, Villa Carlotta, Belgirate, Italy. Office
for Official publications of the EC, EUR 14262, Luxembourg. 380 pp.
Seino, H., and Z. Uchijima, 1992. Gobal distribution of Net Primary Productivity of terrestrial
vegetation. J. Agr. Met., 48(1): 39-48.
Selirio, I.S., and D.M. Brown, 1997. An overview of modelling forage yield and quality for crop
insurance purposes in Canada. Pp. 247-254 in: EU/FAO, 1997.
Sharpley, A. N. and Williams, J. R. 1990. EPIC - Erosion Productivity Impact Calculator: Vol. I.
Model Documentation. USDA Technical Bulletin No. 1768. 235 pp.
Shisanya, C.A. and H. Thuneman, 1993. The use of dynamic crop production simulation model in
quantifying agricultural resources: an example from Machakos district, Kenya. J.
Eastern African Research and Development, 23:176-191.
Smith, M., 1992. Cropwat, a computer programme for irrigation planning and management. FAO,
Rome, Irrigation and Drainage Paper N. 46. 126 pp.
Snijders, F.L., 1991. Rainfall monitoring based on Meteosat data - a comparison of techniques
applied to the western Sahel. Int. J. Remotes Sensing, 12(6):1331-1348.
Snijders, F.L., 1995. ARTEMIS and the outside world. File formats and naming conventions. In:
FAO, 1995: 139-146.
Stein,A., G. Kocks, J.C. Zadoks and H.D. Frinking, 1994. A geostatistical analysis of the spatio-
temporal development of Downy Mildew Epidemics in cabbage. American
Phytopathological Society, 84(10):1227-1238.
Stewart, J.I., 1988. Response farming in rainfed agriculture. Wharf foundation press,
Davis,California,USA. x + 103 pp.
Stöckle, C.O. 1996. GIS and simulation technologies for assessing cropping systems
management in dery environments. Am. J. Alternative Agric., 11(2-3):115-120.
Stöckle, C. O. , S. Martin and G. S. Campbell. 1994. CropSyst, a cropping systems model:
water/nitrogen budgets and crop yield. Agricultural Systems ,46:335-359.
Stöckle, C. O. and R. L. Nelson. 1994. Cropsyst User’s manual (Version 1.0). Biological Systems
Engineering Dept., Washington State University, Pullman, USA.
Stöckle, C. O. , M. Cabelguenne, and P. Debaeke. 1996. Validation of CropSyst for water
management in southwestern France using submodels of different levels of complexity.
European J. Agron., 7:89-98.
Stöckle, C. O. and P. Debaeke. 1997. Modeling crop N requirement: A critical analysis. European
J. Agron., 7:161-169.
Stott, Z., 1996. Predicting sugar beet yields using satellite data. SPOT Magazine, 26:14-16.
Strauss, R., H. Bleiholder, T. van den Boom, L. Buhr, H. Hack, M. Hess, R. Klose, U. Meyer and
E. Weber, 1994. Einheitliche Codierung der phänologischen Entwicklungstadien mono-
un dikotyler Pflanzen. BBA-BSA-IVA, Basel, Switzerland. 26 pp with several plates.
Supit, I., A.A. Hooijer, C.A. van Diepen (Eds), 1994. System description of the wofost 6.0 crop
simulation model implemented in CGMS. Vol. 1, Theory and algorthms. Office for official
publications of the EC, Agriculture Series, EUR15956EN, Luxembourg, 144 pp.
Supit, I., 1997a. Predicting national yields using acrop simulation and trend model. Agric. For.
Meteorol., 88:199-214.
Supit, I., 1997b. National wheat production volume forecasting with models including influences of
market prices and inputs such as plant protection and fertilisers. Application to the
European Community. Pp. 109-144 in: B. Tychon and V. Tonnard, (Eds.), Estimation de
la production agricole à une échelle régionale. Office for official publications of the EU,
Luxembourg, EUR 17663, 199 pp.
Supit, I., and R. Van Kappel, 1998. A simple method to estimate global radiation . Solar Energy,
in press.
Swanson, E.R., and J.C. Nyankori, 1979. Influence of weather and technology on corn and
soybean trends. Agric. Meteorol., 20:327-342.
Tabor,J.A., A.W. Warrick, D.A. Pennington and D.E Myers, 1984. Spatial variability of nitrate in
irrigated cotton: I.petioles. J. Soil Science Soc. of America, 48(3):602-607.
Tabor,J.A., A.W. Warrick, D.E. Myers and D.A. Pennington, 1985. Spatial variability of nitrate in
irrigated cotton: II.soil nitrate and correlated variables. J. Soil Science Society of
America, 49(2):390-394.
Tada, M., and Y. Morooka, 1995. Supply and demand analysis for rice in Peninsular Malaysia -
with special reference to direct seeding in the Muda area. JIRCAS Journal, 2(2):1-11.
Takezawa, K., and Y. Tamura, 1991. Use of smoothing splines to estimate rates of development.
Agric. forest meteorol., 57(1-3):129-145.
Takle, E., 1995. Use of physically based meteorological models for enhancing agroclimatic
databases. Pp. 291-294 in : FAO, 1995.
Thomson, S.J., and B.B. Ross, 1996. Model-based irrigation management using a dynamic
parameter adjustment method. Computers and Electronics in Agriculture, 14(4):269-
290.
Thornton, P.K., H.W.G. Booltink and J.J. Stoorvogel, 1997. A computer program for geostatistical
and spatial analysis of crop model outputs. Agronomy J., 89(4):620-627.
Toure, A., D.J. Major, and C.W. Lindwall, 1995 . Sensitivity of four wheat simulation models to
climate change. Canadian J. Plant Sci., 75(1):69-74.
Tsai, Y.J., J.W. Jones, and J.W. Mishoe, 1987. Optimizing multiple cropping systems: a systems
approach.Transactions of the ASAE, 30(6):1554-1561.
Uchijima, Z. and Seino, H., 1985. Agroclimatic evaluation of net primary productivity. J. Agr. Met.,
40(4): 343-352.
Unganai, L., and F. Kogan, 1998. Drought monitoring and corn yield estimation in southern Africa
from AVHRR data. Remote Sens. Envir., 63:219-232.
USDA, 1967. Irrigation Water Requirements, Technical Release N. 21, Soil Conservation Service
Engineering Division, 83 pp.
USDA, 1972. National Engineering Handbook, Hydrology section, chapters 4-10. USDA-SRC,
Washington.
van Diepen, C.A., J. Wol, H. van Keulen, C. Rappoldt, 1989. WOFOST: a simulation model of crop
production. Soil Use and Management, 5(1):16-24.
Van Heemst, H.D.J., 1986. Physiology principles, in: van Keulen and Wolf (1986), 13-26.
van Keulen, H., and H.H. van Laar, 1986. The relation between water use and crop production, in:
van Keulen and Wolf (1986), 117-129.
van Keulen, H., and J. Wolf (eds), 1986. Modelling of agricultural production: weather, soils and
crops. Simulation monographs, Pudoc, Wageningen, 478 pp.
van Kraalingen D.W.G., 1991. The FSE system for crop simulation. Simulation reports CABO-TT
23, CABO-DLO, WAU-TPE Wageningen, 77 pp.
van Leeuwen, H.J.C. and J.P.G.W Clevers, 1994. Synergy between optical and microwave remote
sensing for crop growth monitoring. Proc. sixth international symposium physical
measurements and signatures in remote sensing, Val d'Isère, France, pp 1175-182.
Vanclooster, M., P. Viaene, J. Diels and J. Feyen. 1995. A deterministic evaluation analysis
applied to an integrated soil-crop model. Ecological Modelling, 81(1-3), 183-195.
Vossen, P., 1990. Comparative statistical validation of two ten-day water use models and three
reduction hypotheses for yield assesment in Botswana. Agric. For. Meteorol. 51:177-
195.
Vossen, P., and D. Rijks (Eds.), 1995. Early crop yield assessment of the EU countries: the
system implemented by the Joint Research Centre. Office for Official Publications of the
EU, Luxembourg, EUR 16318EN, 182 pp.
Wagner, W., J. Noll, M. Borgeaud, H. Rott, 1999. Monitoring soil moisture over the Canadian
Prairies with the ERS Scatterometer, IEEE Trans. Geosci. Remote Sensing, in press
(January 1999 issue).
Wagner, W., G. Lemoine, M. Borgeaud, H. Rott, 1999. A study of vegetation cover effects on ERS
Scatterometer Data, IEEE Trans. Geosci. Remote Sensing, in press.
Wald,L., 1996. Some comments on the use of satellite data for assessing solar radiation
parameters, annex to Potentials of images from geostationary satellite for the
assessment of solar energy parameters. In Proceedings of the Workshop 'satellites for
solar resource information', held in Washington, D.C., 10-11 April 1996. Edited by R.
Perez and E. Maxwell, National Renewable Energy Laboratory, Boulder, CO., USA, 23
p.24-38.
Walker, G.K., 1989. Model for operational forecasting of western Canada wheat yield. Agric.
Forest Meteorol., 44(3-4), 339-351.
Weisensel, W.P., G.C. van Kooten, R.A. Schonry, 1991. Relative riskiness of fixed vs. flexible
crop rotations in the dryland cropping region of Western Canada. Agribusiness ( New
York), 7(6):551-562.
Weiss, A. (Ed), 1981. Computer techniques and meteorological data apllied to problems of
agriculture and forestry: A workshop. Proc. Of workshop held at Anaheim, California, 30-
31 March 1981.NSF ATM-8020422 grant. AMS, ASAE and ASA. 384 pp.
Whisler, F.D., and J.A. Landivar, 1988. A sensitivity test of COMAX/GOSSYM to some soil
physical properties.Proceedings, Beltwide Cotton Production Research Conferences, 3-
8 Jan. 1988, New Orleans, Louisiana, USA. Memphis, Tennessee, USA, National
Cotton Council of America. 85 pp.
White, I.D., D.N. Mottershead and S.J. Harrison, 1992. Environmental Systems, an introduction
text. Chapman & Hall, London, 616 pp.
White, I.D., D.N. Mottershead and SJ Harrison, 1992. Environmental Systems, an introduction
text. Chapman & Hall, London, 616 pp
Williams, J. R. and H.D.Berndt. 1977. Determining the Universal Soil Loss Equation's Length -
Slope Factor for Watersheds. Soil Erosion: Prediction and Control, Soil Conservation
Society Journal, p. 217-225.
Williams, J. R., P.T. Dyke, W.W. Fuchs, V.W. Benson, O.W. Rice, and E.D. Taylor, 1990. EPIC
- Erosion Productivity Impact Calculator. Vol. II. User Manual. A. N. Sharpley and J. R.
Williams, Editors. USDA Technical Bulletin No 1768. 127 pp.
Williams, J.R., C.A. Jones, J.R. Kiniry, and D.A. Spanel, 1989. The EPIC Crop Growth Model.
Trans. ASAE 32(2):497-511.
WMO/CTA, 1992. La radio rurale et la diffusion des informations agrométéorlogiques. Compte-
rendu d’un atelier tenu sur invitation du Gouvernement du Mali, Bamako, 18-22 mai
Wolf, J. 1993. Effects of climate change on wheat production potential in the European
Community. Eur. J. Agronomy. 2(4):281-292.
Wolf, J. And C.A. van Diepen, 1994. Effects of climate change on silage production potential in the
European Community. Agric. For. Meteorol. 71(1-2):33-60.
Wolf, J., and C.A. van Diepen, 1995. Effects of climate change on grain maize yield potential in
the European Community. Climatic Change, 29(3): 299-331.
Wolfe, D.W., and J.D. Erikson, 1993. Carbon dioxide effects on plants: uncertainties and
implications for modelling crop response to climate change. Pp 153-178 in: Kaiser and
Drennen, 1993. Agricultural Dimensions of Global Climate change. St. Lucie Press,
Delray Beach, Florida. 311 pp.
Wu, Y., C.M. Sakamoto, and D.M. Botner. 1989. On the application of CERES-Maize model to
the North-China Plain. Agric. Forest Meteorol., 49:9-22.
Xiniou Yin, M.J. Kropff, G. McLaren and R.M. Visperas, 1995. A nonlinear model for crop
development as a function of temperature. Agric. For. Meteorol., 77:1-16.
Yoshida, S., 1981. Fundamentals of rice crop science. IRRI, Los Baños, Laguna, Philippines. 269
pp
Zentner, R.P., F.B. Dyck, K.R. Handford, C.A. Campbell and F. Selles, 1993. Economics of flex-
cropping in southwestern Saskatchewan. Can. J. Plant Sci., 73:749-767.