Professional Documents
Culture Documents
“DLO-Research Institute for Agrobiology and Soil Fertility, P.O. Box 14.
Wageningen 6700 AA, The Netherlands
ABSTRACT
In this paper, a pedigree of rhe crop growth .simulation modeis by. the
‘School of de Wit’ is presented. The origins and philosophy of this school
we trtlced,from de Wit’s classical publication on modelling photosynthesis
of leqf canopies in 1965. It is shown how changing research gouls und
priorities over the years have resulted in the evolution of u pedigree of‘
models that are similar in philosophy hut d@er in level of complexit!,, the
processes addressed and their,functionalityl. In the beginning, modelling HUT
motivated by the quest for scienttjic insight and the lvi.4 to quantify. and
integrate hiophysical processes to explain the observed vuriution in crop
gro\vth. Later, the emphasis of undfundingjor, agricultural research shifted
to\zlardsputting ucquired insights to practical and operutional use. Model
development became led by a demand,for tacticul und strategic decision sup-
port, ?,iell,fi,recusting, land zonation and explorative scenario studies. Model-
ling developments ,for dtyerent production situutions are illustrated using
the models the uuthors consider most important, i.e. BACROS, SD’C’ROS,
WOFOST, MACROS and LINTEL, but reference is also made to other
models. Finully, comments we made about the usefulness and ~pplicuhilitj’of’
these models qfter nearly 30 years of’tkveiopment. and some,future cour.ses
of action are suggested. Copyright t$ I996 Published h,, Elsrvier Science Lttl
INTRODUCTION
By the end of the 196Os, computers had evolved sufficiently to allow and
even to stimulate the first attempts to synthesize detailed knowledge on
171
172 B. A. M. Bouman, H. van Keulen, H. H. van Laar, R. Rabbinge
published by PUDOC between 1971 and 1993). Finally, the authors pre-
sent a personal review of the usefulness and applicability of these models
after more than 30 years of development.
I I
/I
VI
1.5 I’ 5 10 ton ha-’
Production level
Fig. 1. The relationship among potential, attainable and actual yield and defining, limiting
and growth reducing factors (Rabbinge, 1993).
coefficients that are typical for the processes that are described (de Wit,
1982b). The time coefficient, which has units of time, is the inverse of the
characteristic rate of a process. This inclusion of time differentiates them
from static models in which, for example, crop production is statistically
regressed on weather variables. Second, they divide the system under
study into hierarchical levels of organization, e.g. cells, organs, plants,
crop. These hierarchical levels exhibit characteristic behaviours that result
from the integration of lower-level processes (Loomis et al., 1979). For
instance, a leaf light-response characteristic is the result of processes at the
lower levels of cells and chloroplasts (Sinclair et al., 1977), and canopy
photosynthesis is the sum of photosynthesis of all the individual leaves (de
Wit, 1965). Different hierarchical levels can be combined in one model
provided that time-coefficients that are appropriate for each level of hier-
archy are used (de Wit, 1970). Mathematical modelling entails quantita-
tive integration of the mechanisms at the various hierarchical levels
to provide an explanation of system behaviour. Third, the system is
characterized by a set of state variables (e.g. weights) that are updated at
each iteration or time-step, by rate variables (e.g. carbon flux). The time-
step is typically one quarter of the time coefficient. Values of the rate
variables are calculated from information about the current state of the
system and from external, environmental (e.g. solar radiation) and
auxiliary variables (e.g. leaf area index) (Fig. 2). Fourth, the models are
explanatory because the calculations involving rate variables are based on
Crop growth simulation models 175
solar air
irradiation temperature
I I
structural
biomass
stor. organs
roots
I
I rate I
Fig. 2. Diagram of the relations in a typical “School of de Wit” crop growth model
(SUCROS) for potential production. Boxes indicate state variables, valves rate variables.
circles auxiliary variables, solid lines (arrows) the flow of matter and dotted lines the flow
of information.
TABLE 1
Steps in the Conceptual Phases of Model Development: Conceptual or Preliminary,
Comprehensive and Summary
1 Formulation of objectives
2 Definition of system limits
3 Conceptualization of the system (states, rates, auxiliary variables, forcing vari-
ables etc.)
Comprehensive phase
they were quickly recognized as powerful tools for exploring situations and
possibilities of crop production that were almost impossible to investigate
using the conventional methods and techniques of experimentation.
Because comprehensive models are typically complex and hardly accessible
to potential users, this stimulated the development of summary models. A
summary model can be regarded as a model of a (comprehensive) model, in
which essential elements are simplified and aspects that are only marginally
important are ignored. Summary models have typically been developed
in response to application-oriented research questions (e.g. tactical and
strategic). Their use, for example, may be in decision support systems for
pest and disease management and for plant nutrient management.
19’10
1975
ARID CROP
PAPRAN
1985
1990
1995
Fig. 3. Pedigree of crop growth simulation models of the “School of de Wit”, 1965-1995.
Models in bold boxes have been “lead” models for the development of other crop models.
Model names are explained in the text.
played a role or where water was the limiting factor, knowledge of the
relative importance of the constituent processes allowed the derivation of
effective summary models. For other production situations and levels
however, e.g. where shortage of nitrogen was the limiting factor, there was
still insufficient basic knowledge and the first preliminary and comprehen-
sive models had first to be developed.
The summary models initiated in the 1980s are increasingly being used
operationally as a result of the demand by policy makers and land managers
for data that can only be produced by models. Typical applications include
agro-ecological zonation, regional yield forecasting and scenario studies for
exploring the effect of environmental or socioeconomic changes on agri-
culture. Moreover, the increasing pressure from agricultural funding agen-
cies to ‘prove’ the operational applicability of modelling has led to research
being driven by the development of new technology. This change of
emphasis has introduced new requirements for models and has highlighted
the importance of software quality, an issue that had been recognized in the
early years of modelling, e.g. by Arnold & de Wit (1976).
LZNTUL For many studies at scales ranging from the regional to the
global, existing summary models needed further simplification because the
availability and quality of model input data were often found to be more
constraining than knowledge of the basic processes incorporated in them.
Spitters (1990) argued that SUCROS could be further simplified by
incorporating only those processes that affect the major determinants of
growth, and laid the foundations for a modelling approach that would
later be baptised LINTUL (Light INTerception and UtiLization; Spitters
& Schapendonk, 1990; Kooman, 1995). LINTUL was the first deviation
from the photosynthesis-based models of the De Wit school. In the
LINTUL models, total dry matter production is calculated using the
Monteith approach (Monteith, 1969, 1990) in which crop growth rate is
calculated as the product of interception of radiation by the canopy and a
light-use efficiency (LUE), which should more correctly be called a dry
matter: radiation quotient (Russell et al., 1989). The LUE can often be
considered constant over the growing season and a property of the crop of
interest. For regional studies, LINTUL-type models have the advantage
that data input requirements are drastically reduced and model para-
meterization is facilitated. The LINTUL approach was used, at the
request of the International Potato Center (CIP), for the agro-ecological
characterization of global potato production to help target research at
production problems in those regions where potato cultivation is most
Crop growth simulation models 185
promising (van Keulen & Stol, 1995). Recently, Penning de Vries et al.
(1995) used the LINTUL approach in a world food study in which
potential and water-limited food production was estimated for the year
2040 for 15 major regions of the world.
Increasing insight
Crop growth modelling started 30 years ago with the aim of increasing our
insight into crop growth processes by a synthesis of knowledge expressed
186 B. A. M. Bouman, H. van Keulen, H. H. van Laar, R. Rabbinge
Operational applications
Yield prediction
Validated models can be used in application-oriented research by scientists
and in operational applications where their users are managers or other
non-scientists. Most of the application-oriented research using the crop
growth models of the ‘School of de Wit’ has been related to the problem
of yield prediction (Seligman, 1990) including world food production
studies (Buringh et al., 1979; Penning de Vries et al., 1995), agro-ecologi-
cal zonation (Aggarwal, 1993; van Keulen & Stol, 1995) and explorations
of the effects of climate change on crop production (Wolf, 1993; Matthews
et al., 1995). Operational applications include the use of the crop growth
monitoring system (CGMS) by the Joint Research Centre for producing
monthly yield predictions for the regions of the EU (Vossen, 199.5), and
the use of WOFOST by Dutch consultancy agencies in land use planning
projects (personal communication). Two reasons may be postulated for
the relative success of crop models in yield prediction. First, models are
the only means of systematically exploring the production potential of
Crop growth simulation model.7 187
Plant breeding
Crop growth models have been used in plant breeding to simulate the
effects of changes in the morphological and physiological characteristics of
crops and thus to aid in the identification of ideotypes (Donald, 1968) for
different environments (Dingkuhn et al., 1993; Hunt, 1993; Kropff er al.,
1995). Hunt (1993) and Palanisamy et al. (1993) suggested that crop
growth models that have been parameterized for new cultivars in field
experiments can be used to simulate the long-term yield stability of these
cultivars at a location under the expected range of climatic conditions.
This technique holds out the promise of reducing the cost of breeding
programmes by limiting the number and years of expensive, multi-location
trials that are currently required to ensure statistical reliability. A recent
review of literature on the use of modelling in potato breeding led
Ellis&he & Hoogendoorn (1995) to the conclusion that ‘simulation
modelling can contribute to the efficiency of potato breeding programmes,
because modelling analyses complex characteristics, indicates the most
promising components for selection, can forecast plant growth under var-
ious conditions, including biotic and abiotic stress, and helps, therefore.
188 B. A. M. Bouman, H. van Keulen, H. H. van Laar, R. Rabbinge
Crop management
Crop growth simulation models have been used in numerous studies to
help farmers in day-to-day, i.e. tactical, decision making. They have been
used to investigate the effects of management options such as sowing time,
plant population density, irrigation timing and frequency and fertiliser
applications in different environmental conditions on long-term mean
yield and yield probability (e.g. Ungar, 1990; Carberry et al., 1992, 1993;
Keating et al., 1993; Aggarwal et al., 1994; Aggarwal & Kalra, 1994;
Rotter & Dreiser, 1994). In some cases, these studies have stimulated field
experimentation to test the outcomes predicted in the simulations (ten
Berge et al., 1994). However, the operational application of crop growth
models to support tactical decision making has generally not yet been
successful (Seligman, 1990) with the notable exception of the areas of
irrigation scheduling and water management (van Keulen & Penning de
Vries, 1993) and pest and disease management (e.g. the Epidemics Predic-
tion and Prevention System (EPIPRE); Rabbinge & Rijsdijk, 1983).
Treatment of the latter type of model is outside the scope of this paper and
readers are referred to relevant publications elsewhere, (e.g. Rabbinge et
al., 1990; Kropff & Lotz, 1992; Teng & Savary, 1992).
Recently, researchers have started to apply the results of crop models to
tactical decision making using knowledge based systems such as expert
systems and decision support systems. These software systems have been
promoted since the mid-1980s as a major breakthrough, opening new
horizons in decision support (Schiefer & da Silva, 1995). However, despite
continuing reports of successful developments and prototyping, Hilhorst
& Manders (1995) found that the overall acceptance of the technology in
agriculture in The Netherlands is still limited. They suggested that a main
reason for this lack of acceptance was the knowledge-intensive nature of
such systems. Such systems are of strong interest to research organizations
and this has resulted in: (i) an undue emphasis on problems of a scientific
rather than a practical interest; (ii) poor functionality for non-specialist
users; and (iii) an evolutionary development path which does not accord
with modern software engineering standards. However, these deficiencies,
which might equally well apply to the failure of crop growth models in
Crop growth simulation mod& 189
operational tactical decision support (see above), are now being addressed.
Three other reasons can explain this particular failure of models. First,
because one of the major sources of yield variation in agriculture is the
variability in weather conditions, success of decision support systems
depends largely on their ability to predict future weather. Even today,
weather predictions are at best reasonably accurate for only a few days
ahead and decision support systems have to rely on probability analyses
using long-term historical or generated weather data (van Keulen &
Penning de Vries, 1993). Second, the lack of accurate input data for soil
and crop characteristics, particularly in respect of their spatial variability,
is often a constraint on successful model applications (van Noordwijk &
Wadman, 1992). Third, models are often used for field conditions whereas
they were developed for rather strictly defined hypothetical production
situations (potential production, water-limited production, etc.) in
uniform fields. In farmers’ fields, several limiting and yield-reducing fat-
tors may occur simultaneously, so that the conditions fall outside the
boundary conditions or domain of validity of the models. This raises the
modeller’s dilemma that for ease of application in a particular practical
situation, models should be as simple as possible and require only a small
number of input data, but that on the other hand, they should be complex
and flexible enough to be able to represent the complex effects of the wide
range of potentially interacting yield-limiting and yield-reducing factors
that might be important for the crop of interest. For situations of poten-
tial production, the summary models developed from the comprehensive
crop growth models of the ‘School of de Wit’ satisfactorily predict crop
behaviour. However, although the processes of photosynthesis and growth
respiration are satisfactorily modelled mechanistically, aspects of main-
tenance respiration and morphogenesis (e.g. organ growth, assimilate
partitioning and leaf area development) are still not well understood and
little progress has been made since the release of BACROS and ARID
CROP. In water-limited conditions, the main effect of water shortage on
the reduction in photosynthesis is well understood and has been satisfac-
torily incorporated in summary models. However, recent experiences in
modelling rice growth in the SARP project have indicated the need for
further study and the inclusion of crop-specific adaptation mechanisms
(Wopereis, 1993).
Major gaps still exist in our knowledge of the effects of nutrient-limita-
tion and it is not yet possible to use mechanistic models directly for farm
level applications (van Keulen & Stol, 199 1; van Keulen & Penning de
Vries, 1993). Therefore the operational use of deterministic models that can
handle the even more complex situations that typify actual farming condi-
tions is still a long way off, and poses new challenges for the years ahead.
190 B. A. M. Bouman, H. van Keulen, H. H. van Laar, R. Rabbinge
POSTSCRIPT
REFERENCES
Alberda, Th. (1968). Dry matter production and light interception of crop
surfaces. IV. Maximum herbage production as compared with predicted
values. Netherlands Journal of Agricultural Science, 16, 142-53.
Alberda, Th. & Sibma, L. (1968). Dry matter production and light interception of
crop surfaces. III. Actual herbage production in different years as compared
with potential values. Journal of British Grassland Society, 23,206-l 5.
Alberda, Th., Seligman, N. G., van Keulen, H. & de Wit, C. T. (eds) (1992). Food
from dry lands: an integrated approach to planning of agricultural develop-
ment. Kluwer Academic Publishers, Dordrecht, The Netherlands. 211 pp.
Aggarwal, P. K. (1993). Agro-ecological zoning using crop growth simulation
models: characterization of wheat environments in India. In Systems
approaches for sustainable agricultural development, eds F. W. T. Penning de
Vries, P. Teng 8z K. Metselaar. Kluwer Academic Publishers, Dordrecht, The
Netherlands. pp. 97-109.
Aggarwal, P. K. & Kalra, N. (1994). Analyzing the limitations set by climatic
factors, genotype, and water and nitrogen availability on productivity of
wheat II. Climatically potential yields and management strategies. Field
Crops Research, 38, 93-103.
Aggarwal, P. K., Kalra, N., Singh, A. K. & Singha, S. K. (1994). Analyzing the
limitations set by climatic factors, genotype, and water and nitrogen avail-
ability on productivity of wheat I. The model description, parameterization
and validation. Field Crops Research, 38, 73-91.
Crop growth simulation models 191