Title Page
 Table of Contents
 Executive summary
 A crop modeling strategy for CIMMYT's...
 A physiologist's wish list for...
 Issues arising from the use of...
 Canopy development, radiation interception,...
 Module structure in CROPGRO...
 A methodology for linking spatially...
 Work group outputs
 Contact information for partic...

Group Title: Geographic information systems series - Natural Resources Group - 01-02
Title: Directions in modeling wheat and maize for developing countries
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00077515/00001
 Material Information
Title: Directions in modeling wheat and maize for developing countries proceedings of a workshop, CIMMYT, El Batán, Mexico, 4-6 May 1998
Series Title: Natural Resources Group geographic information systems series
Physical Description: vi, 42 p. : ill. (some col.), maps ; 28 cm.
Language: English
Creator: Grace, Peter Robert
White, Jeffrey W
International Maize and Wheat Improvement Center
Publisher: CIMMYT
Place of Publication: Mexico
Publication Date: c2001
Subject: Wheat -- Congresses -- Developing countries   ( lcsh )
Corn -- Congresses -- Developing countries   ( lcsh )
Genre: bibliography   ( marcgt )
conference publication   ( marcgt )
non-fiction   ( marcgt )
Bibliography: Includes bibliographical references.
Statement of Responsibility: Jeffrey W. White and Peter R. Grace, editors.
 Record Information
Bibliographic ID: UF00077515
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: oclc - 48460931
isbn - 9706480811


This item has the following downloads:

NRGGIS01-02 ( PDF )

Table of Contents
    Title Page
        Page i
        Page ii
    Table of Contents
        Page iii
        Page iv
    Executive summary
        Page v
        Page vi
    A crop modeling strategy for CIMMYT's research on sustainable wheat and maize production systems
        Page 1
        Page 2
        Page 3
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
    A physiologist's wish list for a robust wheat model
        Page 9
        Page 10
        Page 11
        Page 12
    Issues arising from the use of CERES for tropical maize
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
    Canopy development, radiation interception, and a simple model of maize productivity
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
    Module structure in CROPGRO v4.0
        Page 23
        Page 24
        Page 25
        Page 26
        Page 27
    A methodology for linking spatially interpolated climate surfaces with crop growth simulation models
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
    Work group outputs
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
    Contact information for participants
        Page 41
        Page 42
Full Text


Directions in Modeling Wheat and Maize for
Developing Countries

Proceedings of a Workshop, CIMMYT,
El Batin, Mexico, 4-6 May 1998

Jeffrey W. White and Peter R. Grace

Natural Resources Group

Geographic Information Systems
Series 01-02

CIMMYT (www.cimmyt.org) is an internationally funded, nonprofit, scientific research and training organization.
Headquartered in Mexico, CIMMYT works with agricultural research institutions worldwide to improve the
productivity, profitability, and sustainability of maize and wheat systems for poor farmers in developing countries. It
is one of 16 food and environmental organizations known as the Future Harvest Centers. Located around the world,
the Future Harvest Centers conduct research in partnership with farmers, scientists, and policymakers to help
alleviate poverty and increase food security while protecting natural resources. The centers are supported by the
Consultative Group on International Agricultural Research (CGIAR) (www.cqiar.orq), whose members include nearly
60 countries, private foundations, and regional and international organizations. Financial support for CIMMYT's
research agenda also comes from many other sources, including foundations, development banks, and public and
private agencies.

Future Harvest builds awareness and support for food and environmental research for a world with less poverty, a
healthier human family, well-nourished children, and a better environment. It supports research, promotes
partnerships, and sponsors projects that bring the results of research to rural communities, farmers, and families in
Africa, Asia, and Latin America (www.futureharvest.orq).

O International Maize and Wheat Improvement Center (CIMMYT) 2001. All rights reserved. The opinions expressed
in this publication are the sole responsibility of the authors. The designations employed in the presentation of
materials in this publication do not imply the expression of any opinion whatsoever on the part of CIMMYT or its
contributory organizations concerning the legal status of any country, territory, city, or area, or of its authorities, or
concerning the delimitation of its frontiers or boundaries. CIMMYT encourages fair use of this material. Proper
citation is requested.

Correct citation: White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing
Countries: Proceedings of a Workshop, CIMMYT El Batan, Mexico, 4-6 May 1998. NRG-GIS Series 01-02. Mexico,

Abstract: In a workshop sponsored by the Natural Resources Group (NRG) of the International Maize and Wheat
Improvement Center (CIMMYT), crop modelers, crop physiologists, and other scientists from around the world met
at CIMMYT headquarters in May 1998 to review the relevance of the CERES-Maize and CERES-Wheat models for
conditions in developing countries, to define an appropriate role for CIMMYT in promoting the use of crops models,
and to consider ways to develop a new generation of wheat and maize models, both for natural resource
management research and to link models to genetic research. In addition to invited papers, the publication
describes the conclusions of working groups.

ISSN: 1405-7484

ISBN: 970-648-081-1

AGROVOC descriptors: Maize; Wheats; Plant physiology; Food production; Natural resources; Resource
conservation; Sustainability; Simulation models; Research projects; Research institutions; Tropical zones;
Developing countries

AGRIS category codes: F01 Crop Husbandry

A50 Agricultural Research

Dewey decimal classification: 633.1

Printed in Mexico.


L. H arrington ............................................. ... ..... ... ..... ... ..... ... ............................ iv
Executive Sum m ary ........................................................................................................................ v

A Crop Modeling Strategy for CIMMYT's Research on Sustainable
Wheat and Maize Production Systems
J.W W h ite an d P.R G race ............................................................................................. ............ ............. 1
A Physiologists Wish List for a Robust Wheat Model
M .P R ey n o ld s .................................. ..... ... ..... ... ..... ... ..... .............. ................ ........ 9
Issues Arising from the Use of CERES for Tropical Maize
G E dm eades and J. B olaios ................................................................................. ................................ 13
Canopy Development, Radiation Interception, and a Simple Model of Maize Productivity
J. Bolafios ............................................................... 18
Module Structure in CROPGRO v4
C.H. Porter, J.W Jones, and P. W ilkens .............................. ........................ ........................... 23
A Methodology for Linking Spatially Interpolated Climate Surfaces
with Crop Growth Simulation Models
S.N C ollis and J.D C orbett ............................................................................................ ................... 28
W ork G group O utputs................................................................................................................................ 36
Contact Information for Participants .................................................................................... 41


Process-based models of crop growth and
development are among the most exciting tools for
agricultural research and development that are
emerging from the on-going information technology
revolution. Models can improve our understanding of
the complex processes underlying crop production,
particularly in water and nutrient management. Their
predictive power can help deal with issues that have
long proven difficult or intractable, such as system
sustainability or the impacts of global change driven
by forces such salinization, desertification, and an
increased concentration of greenhouse gasses.

Models, however, are only as good as the science and
data they are built from. While simulation models
have solid foundations in research on temperate
environments in developed countries, much less
model development and testing has occurred in
subtropical and tropical environments and for
production systems in developing countries.
Furthermore, production practices in developing
countries often differ dramatically from the high-
input monocultures of the North.

This workshop provides a timely opportunity to
examine potential applications of crop models in
relation to CIMMYT's mandate of promoting
sustainable improvements in the productivity of
maize and wheat systems for resource poor farmers.
Issues that we might hope to see addressed include:
* How well do available models deal with
conditions of low water, nitrogen, and phosphorus
availability that are frequent in smallholder
systems in developing countries?
* How can CIMMYT best assist in model testing
and development?
* What are the most promising or appropriate
applications for current models?

To help push model development forward in ways
that increase its relevance for research intended to
benefit developing countries, CIMMYT is pleased
to host its first formal, international workshop on
crop modeling.

We thank Jane Reeves and Mike Listman for the
style editing of this publication and Wenceslao
Almazan for the layout.

Larry Harrington
Natural Resources Group (NRG)

Executive Summary

The CIMMYT Wheat and Maize Modeling Workshop,
held from 4 to 6 May, 1998, at CIMMYT's
headquarters at El Batan, Mexico, had the broad goal
of examining crop modeling needs for research on
sustainable maize and wheat systems in developing
countries. Specific objectives included:
* Reviewing the current status of the CERES-Maize
and CERES-Wheat models, including working on
known problems with phenology and growth for
conditions prevalent in developing countries.
* Defining an appropriate role for CIMMYT in
promoting use of crops models.
* Considering strategies for a new generation of
wheat and maize models, both for natural resource
management research (e.g., SALUS) and to link
models to genetic research (e.g., a gene-based wheat
or maize model similar to GENEGRO).
The workshop began with presentations on various
aspects of wheat and maize modeling and management
of data for crop models. The following two days were
dedicated to informal working groups that examined
specific topics in detail. In most cases, this involved
running test data sets and modifying model codes.

The CIMMYT Wheat and Maize Modeling Workshop,
4 to 6 May, 1998, at CIMMYT headquarters in El
Batan, Mexico, examined crop modeling for research
on sustainable maize and wheat systems. Specific
objectives included reviewing the current status of the
CERES-Maize and CERES-Wheat models, examining
roles for CIMMYT in promoting use of crops models,
and considering strategies for a new generation of
wheat and maize models.

Potential applications of crop modeling at CIMMYT
include basic agronomy, systems research, and crop
improvement (paper by J.W White and P.R. Grace).
Most models now incorporate responses to available
soil moisture and nitrogen and can be used for a wide
range of research problems relating to these responses.
However, there is need to introduce or improve
routines for tillage, phosphorus uptake and utilization,

pH effects, and soil erosion. CIMMYT can play a
valuable role in facilitating access to quality sets of field
data in standard formats (e.g., the ICASA standards),
promoting more complete documentation of models, and
linking modelers with priority research applications.

From a "wish list" for wheat physiologists (M. Reynolds),
a key interest is to model canopy temperature differences
to help understand observed relations between canopy
temperature depression and grain yield under favorable
conditions. Other topics include lodging resistance and
use of genetic information to improve modeling of
vernalization and photoperiod effects on phenology.

In maize (G. Edmeades and J. Bolafios), a major
constraint is the failure of existing models to handle stress
effects on the anthesis-silking interval and on related
processes. CIMMYT's Maize Program has many datasets
that could serve for model development and evaluation.

The modular structure of CROPGRO (C.H. Porter, J.W.
Jones, and P. Wilkens) was presented as a promising
strategy to facilitate integration of information from
different disciplines, allow contributions from many
authors, provide flexibility in updating, and thus extend
the useful life of models. Key features of a modular
approach are that the modules should relate to real world
components or processes, that inputs and outputs should
represent measurable variables, and that communication
between modules should occur only through the inputs
and outputs.

There is great interest in linking models and GIS to
improve regional targeting of new technologies (S.N.
Collis and J. Corbett). One approach is to use interpolated
climate surfaces as a source of inputs for weather
generators. To reduce the potential number of simulations,
regional variation can be summarized as "effective
environments." These can be overlaid with soil layers and
simulations run on unique polygons. As an example,
performance of two maize cultivars were compared for
rainfed systems of eastern Africa. An ArcView extension
is available to automate much of this work.

In the hands-on modeling sessions, widespread concern
was noted about problems of version control for models
distributed in DSSAT. The next official version of
DSSAT was to be Version 3.5, which was released in
late 1998. Test versions of models for development
should be based on Version 3.5. However, they should
be identified as 3.6xx, where "xx" suffix identifies the
developer (e.g., "3.6PG" for Peter Grace). A facility is
needed in the code to allow for input and display of this

In CERES-Maize, the number of maize leaves is over-
estimated. Although the phyllochron interval (PHINT)
appeared in the cultivar file, the value was actually
"hard wired" at 75 degree days. In version 3.5, the
interval can be varied.

Files of cultivar-specific parameters in CERES (e.g.,
"MZCER980.CUL') provide no indication of reliability.
Ideally, the number of observations, types of data, and
method used should all be indicated. Tony Hunt strongly
urged that the list of cultivars be shortened to include
only the most reliable coefficients. Five generic cultivars
should also be provided.

To facilitate modeling of nitrogen mineralization in
CERES and CROPGRO, a new soil parameter file was
created, SOILN980.PAR, that externalizes many of the
coefficients needed for simulating the decomposition of
soil organic matter (one pool) and organic matter added
as residue or manure (three pools). If the file does not
exist in the data directory, it is created using default
values upon the first run of the model.

To handle tile drains, Bill Bachelor defined a soil layer
15 cm thick at the approximate depth of the drain.
Unfortunately, testing showed that introducing this layer
produced unexpected changes in model outputs.

Among changes suggested for subsequent releases of
CERES were:
* Improved thermal time calculations.
* Accounting for mass of dead leaves and their
subsequent incorporation into soil organic matter.
* Improved modeling of grain number in maize (based
on approach of Andre Du Toit).
* For wheat, Zadok stages should be output along with
standard phenology stages.
* Routines to simulate:
Conservation tillage.
Tile drainage.
Response to soil phosphorus.
* Generate solar radiation if reliable daily data are or
not available.
* Genetic coefficients for winter-kill, vernalization, and
* Improve the water balance routines both for root
uptake and estimates of potential evaporation.

Many researchers assume that Jones and Kiniry (1986)
is still an accurate description of the CERES models.
Much more effort is needed to update documentation.
Reports should be sent to the Hawaii for posting on the
list server. Links to modeling sites such as Michigan
State University, the University of Florida, and
elsewhere can easily be included.

The need to assemble quality data sets collected through
the IBSNAT project and other sources (e.g., GCTE)
arose several times. It is easy to criticize models, but
the models are only as good as the data sets they are
based upon.

A Crop Modeling Strategy for CIMMYT

A Crop Modeling Strategy for CIMMYT's

Research on Sustainable Wheat and Maize

Production Systems

J.W. White and P. R. Grace
Natural Resources Group, CIMMYT Mexico

This paper outlines crop improvement, strategic agronomy, and natural resource management
research concerns central to CIMMYT's mission and which models can help address. Authors
also discuss related issues, including model documentation and data management. Although
CIMMYT has little comparative advantage in model development per se, the center has much
to contribute to others' development and refinement efforts, to the availability of quality data,
and to the promotion of models and training in their use, and active participation in these
areas will ultimately result in better models for CIMMYT's own aims.


Process-based crop models can increase research
effectiveness by allowing researchers to formulate
hypotheses in a precise, quantifiable manner and by
converting complex hypotheses into quantifiable
estimates of crop response to management and
environment. These responses may have a temporal
scale ranging from a few minutes to multiple decades,
the latter scale being of particular relevance where
natural resource degradation is a concern.

Researchers at the International Maize and Wheat
Improvement Center (CIMMYT) recognize the
potential of crop models and have applied them to
studies of drought tolerance, yield potential, nitrogen
dynamics, farmer risk and impact of climate change,
among others. In particular, the Natural Resources
Group sees models as key tools for understanding
issues related to conserving the resource base of
maize- and wheat-based production systems.

Given the diversity of maize and wheat models available
and the wide range of potential applications, CIMMYT
needs a coherent strategy for developing and applying
crop models. This paper outlines the needs for models to
satisfy different research objectives and discusses related
issues, including model documentation and data

Model Needs, as Defined by Research
Objectives at CIMMYT

Possible applications of crop models at CIMMYT are
conveniently viewed in relation to research objectives
(Table 1). These objectives are divided into broad
categories of crop improvement, strategic agronomy, and
natural resource management, although there is much
potential for overlap, especially among the latter two.

Crop improvement
Applications of models to crop improvement efforts
include evaluating the impact of specific characteristics
on yield or other traits, defining crop ideotypes, and
examining genotype x environment interactions.

CIMMYT. Published in White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing Countries: Proceedings of a
Workshop, CIMMYT, El Batan, Mexico, 4-6 May 1998. NRG GIS Series 01-02. Mexico, D.F.: CIMMYT.

Directions in Modeling Wheat and Maize for Developing Countries

The importance of high-yielding wheat regions in
several developing countries means that simply
characterizing regional variation in yield potential is of
fundamental interest. However, in a recent attempt to
estimate potential yields in the Yaqui Valley, Sonora,
Mexico, the CERES-Wheat model gave maximum
yields of 9.5 t/ha, whereas the highest yields reported
from large plots in agronomy trials are around 11 t/ha
(K. Sayre, personal communication, 1998).

Improved modeling of maize and wheat phenology
should assist breeders in understanding how to match
phenology to different environments and lead to
identification of improved selection criteria, both for
high yield and stress conditions. Rapid progress in
molecular marker techniques for characterizing the
genetic makeup of cultivars suggests the possibility of
developing gene-based phenology models.

Research conducted at CIMMYT on the physiology of
maize has identified the control of grain set as a key
issue, both for yield potential and tolerance to water and

nitrogen deficits. The anthesis-silking interval provides
a useful field indicator of stress tolerance (Edmeades et
al. 1993; Bolafios and Edmeades 1996). Pollination is
not limiting. Partitioning to grain is reduced under
stress, causing grain abortion. Related to partitioning to
grain are prolificacy (formation of multiple ears) and
barrenness (where no ear is formed). Accurately
describing these responses may be essential for realistic
predictions of maize yields under the low yield
conditions typical of many smallholder systems in
Africa and Latin America.

CIMMYT has identified increasing nitrogen use
efficiency (NUE) as a breeding priority for maize.
Evidence to date suggests that variation in NUE is
related to efficiency of N mobilization from stems and
leaves to grains and not to variation in uptake (Bellows
1997; van Beem and Smith 1997). A detailed model of
interactions of N uptake and mobilization with growth
would allow researchers to examine possible tradeoffs in
selecting for high NUE.

Table 1. Needs for different crop models at CIMMYT, as related to research objectives.

Research objective Crop Type of model needed

Crop improvement

Define crop ideotypes. Maize Advanced process-based with improved handling of ear and
grain set (maize) and canopy temperature (wheat).

Understand cultivar differences Wheat Advanced process-based with improved handling of canopy
in canopy temperature, temperature (wheat).

Test hypotheses related to Maize, wheat Gene-based model with detailed treatment of processes.
cultivar differences specified
at the gene level.


Characterize climatic risk Maize, wheat Robust basic model with cultivar differences in phenology.
in rainfed systems.

Estimate potential yield loss due Wheat Robust basic model that simulates canopy geometry.
to growing wheat on beds.

Natural resource management

Estimate long term impact of Maize, wheat
zero tillage with different levels
of residue retention.

Estimate potential impact of Maize Model of legume species, preferably with ability to simulate
green manures relay-cropped interspecific competition.
with maize.

A Crop Modeling Strategy for CIMMYT

Breeding for increased yield potential is a priority of the
CIMMYT Wheat Program. Selection of erect semidwarf
cultivars gave a large initial yield increase when
combined with increased inputs. Recent efforts focus on
selecting for increased yield within semidwarf wheats.
Greater canopy temperature depression is associated
with increased grain yield (Amani et al. 1996), and there
is interest in using models to examine underlying
mechanisms and to assess whether the effectiveness of
this trait will vary with humidity in the different
production regions.

Lodging is also a major concern in high yield wheats.
Static models of the mechanics of lodging suggest ways
to increase lodging tolerance. These might be profitably
linked to process-based models to examine how
decreasing the plant's center of gravity and modifying
the mechanical properties of stems might affect overall
growth and partitioning to grain.

Natural resource management and
strategic agronomy
The Natural Resources Group (NRG) at CIMMYT was
established in 1996 to work on long-term issues of maize
and wheat crop production. Its activities complement
those of the maize and wheat agronomists in CIMMYT's
two commodity focused programs. The NRG's primary
focus is on issues related to conservation tillage and
residue management. To structure its activities, the NRG
follows a research framework developed from the
Center's previous experiences in farming systems
research (Table 2). Most steps of the framework involve
research where models can increase research efficiency.
The steps focusing on initial characterization and final
assessment of impact may involve many situations where
use of modeling is the only practical alternative.

For rainfed maize and wheat, a central issue is how to
improve or, better still, synchronize N availability,
particularly in soils with low N status and in systems
where external N inputs are low. Low inputs of organic

Table 2. Framework for research activities of the Natural Resources Group, CIMMYT, with suggestions
of relevance of crop modeling.

Research step

Examples of possible applications of models

Characterize incidence and pace
of resource degradation.

Understand processes underlying

Develop prototype solutions.

Conduct adaptive research with
farmer participation.

Assess adoption process.

Synthesize and extrapolate
to other regions.

Assess impact, including
consideration of the possible
consequences of inaction.

Estimate nutrient dynamics and implications of soil loss in maize-ground
nut systems of Malawi and Zimbabwe under current practices.

Characterize the dynamics of nitrogen and soil organic matter in rainfed
maize systems of Zimbabwe.

Evaluate potential impact of zero tillage with partial residue removal in
Jalisco, Mexico.

Identify regions in India where wheat production on raised beds is
feasible based on soil types.

Examine how suggested farmer modifications to soil fertility
management strategies respond to climatic risk in Malawi and Zimbabwe.

Provide feedback on how adoption of green manures in Central America
compares to expectations from research.

Estimate the potential impact of conservation tillage in regions of Mexico
that differ in rainfall and soil type.

Estimate the potential of different green manures to improve productivity
of maize systems in Malawi and Zimbabwe.

Simulate long term consequences of specific solutions vs. those of
present practice.

Directions in Modeling Wheat and Maize for Developing Countries

and inorganic fertilizers, combined with the use of grain
legumes or green manures, appear to be among the few
options for maintaining or improving system
productivity in many low-income, dryland
environments. If we can improve our ability to predict
changes in soil water storage and N availability in
response to various forms of tillage and residue
management (particularly surface application of
residues), we have the ideal tool for achieving this
source-sink synchronization. Unreleased modifications
to the surface residue management routines in CERES
have made significant improvements in this area. Also
available are a great deal of quality data that allow us to
include measurable entities such as microbial biomass in
our algorithms (Grace et al. 1994). The use of
biologically significant soil pools that are actually
measurable is becoming more prevalent in modeling
circles today and avoids the problem of model re-
calibration for each particular soil environment.

Models will be essential for examining the relative
benefits of different technologies, including accounting
for climatic risk and variation in soil types. While
single-season or even single-rainfall event simulations
may be of use in research on degradation processes, the
ability to simulate multiple years is a must, particularly
for investigating changes in soil physical properties. For
example, changing from conventional tillage to zero
tillage can have a significant impact on soil water and
nutrient cycling; a feature that many models fail to
capture at present.

In irrigated or high rainfall systems, loss of mineral N
from the soil system is a major problem. With respect to
gaseous losses, denitrification routines in current
models have been shown to be either too data intensive
or severely lacking in accuracy. With funding
opportunities increasingly available for predicting
greenhouse gas emissions, a significant overhaul of
these routines in CERES is being done at UC-Berkeley
(Riley, unpublished), incorporating water-filled pore
space routines using isotope data collected at CIMMYT-
Obregon. Ammonia volatilization is not dealt with at all
in nutrient cycling models but is a significant loss
mechanism, particularly where anhydrous ammonia is
applied to irrigation water. A significant amount of data
(and interest) on this mechanism are held at Lowry
Harper, USDA, for inclusion in the CERES models.

Management of phosphorus and soil pH are also
concerns, particularly for maize systems in Africa.
Whilst a P module is available in CERES and is
currently being modified (Gerakis et al. 1998; Daroub
et al. 1999), a critical issue is how to relate soil
measurements of available P to pool sizes in a
process-based model. More challenging still is how to
quantify effects of root architecture and mycorrhizal
associations. While the latter is perhaps beyond the
scope of CERES/CROPGRO models, the
implications of pH on soil processes are readily
known and could be incorporated with little problem.

Growing wheat on raised beds offers important
advantages in terms of fertilizer, irrigation, and weed
management. Effective modeling of this system
would require routines for canopy architecture and
two-dimensional movement of moisture and nutrients.
Relevant data are being collected at CIMMYT in
collaboration with Michigan State University, but
algorithms need to constructed and tested. The
erosion-runoff routines in CERES also appear weak,
and improvements are needed.

A systems perspective also requires the ability to
model other crops and weeds that may occur within
maize or wheat systems, whether in rotation or
intercropping situations. As mentioned above,
legumes represent the most important set of alternate
crops and include sources of grain, feed, and green
manure. The CROPGRO model simulates growth of
the relevant grain legumes, and intercropping
algorithms have been developed. Modeling of forage
and green manure species is largely constrained by a
lack of growth data from tropical environments. In
Asia, rice-wheat systems present a special challenge
due to the unusual soil conditions of paddy systems.
Other species that may be rotated with maize or
wheat include sugar cane, sunflower, cotton, jute,
cassava, potatoes, and tomatoes. Like most modeling
efforts where funds are at a premium, many of the
improvements listed above are not adequately
implemented for general use and are inevitably
forgotten or lost. This is a major area of concern if the
Decision Support System for Agrotechnology
Transfer (DSSAT) is going to continue to be a
product with credible outputs in agronomically

A Crop Modeling Strategy for CIMMYT

diverse agricultural systems. Other modeling efforts
will pass DSSAT by, as will funding agencies, unless a
coordinated approach to modification, testing, and, in
particular, a release strategy, perhaps Web-based, is

Prioritizing Model Development

The preceding review illustrates many potential
applications of models to issues relevant to CIMMYT
and its research partners. However, the review is
inadequate for priority setting because of overlap in
how model improvements benefit different research

Table 3 suggests priority areas for model development.
Criteria considered include importance, technical
feasibility, and availability of quality data. Highest
priority is given to improving the modeling of
phenology and basic soil processes, particularly
residue handling

Other Modeling Issues

Effective use of models does not depend solely on the
availability of suitable models. Factors contributing
greatly to the success of the IBSNAT project were that it
provided users with a software shell for managing field
data and simulation results, and that it established
standard data formats that could be used on a range of
crop models. Thus, CIMMYT should reflect on what its
role is and who its collaborators should be in developing
modeling tools and managing data for modeling

Complex versus simple models
Over the past decade, there has been an increase in
interest in simple models based on radiation use
efficiency (RUE) and harvest index conversions (e.g.,
Sinclair and Horie 1989). While we endorse the use of
simple approaches where feasible, our concern is that
apparent functionality can be an artifact of calibrating a
given model to a narrow range of conditions or
germplasm. For an institute with global research
concerns, preference should be given to models that
offer a framework for providing varying levels of
process detail.

Table 3. Priority areas for model development.

Type of model/subroutine needed Crop Possible sources Priority


Growth response to temperature Wheat CERES ***
Canopy temperature Wheat CROPGRO **
Canopy structure (bed system) Wheat CROPGRO *
Basic phenology and leaf development Maize CERES **
Gene-based treatment of phenology Maize, wheat CROPGRO, GENEGRO *
Contribution of senesced material to residue Maize, wheat CERES, APSIM *
Stress effects on grain set Maize **
Surface residue decomposition Maize, wheat SOCRATES, CENTURY ***
Improved runoff Maize, wheat APSIM **
Soil erosion Maize, wheat **
Phosphorus Maize EPIC, CENTURY *
pH, including response to liming Maize *
Tillage effects Maize, wheat **
Intercropping Maize APSIM **
Weed competition Maize APSIM *
Model of green manure legume species Mucuna, Canavalia, etc. CROPGRO, APSIM **

Directions in Modeling Wheat and Maize for Developing Countries

User interfaces
Although some modeling activities may primarily be of
use to CIMMYT scientists per se, most potential
applications need to be readily transferable to national
agricultural research systems (NARS) or non-
governmental organizations (NGOs). Indeed, our
partners will often be co-participants in model
development, evaluation, and application. Thus
CIMMYT should seek generic tools, which would
logically include a standardized model user interface.
DSSAT3 (Thornton et al. 1996) provides a well known
interface that many CIMMYT and NARS staff have
been exposed to. However, it is clear that in the near
future most users will expect a Windows-based,
graphical user interface. Several options for such
interfaces exist or are under development, including
APSFRONT (Veraart 1998) and GUICS (Acock et al.
1999). CIMMYT, as a potentially major user, should try
to participate in the development of improved user

A specialized group of user interfaces worthy of
separate consideration is that which allows models to be
combined with geographic information systems (GIS).

Documentation of models
Proper documentation of models is perhaps a higher
priority for CIMMYT than for model developers. The
Center must be prepared to transfer modeling
approaches to its own staff, NARS, and NGOs, all of
whom may insist upon explanations of the underlying
assumptions of a given model. Unfortunately, crop
models are notorious for poor documentation. We
attribute this to several factors:
Model developers are pressured to publish research
outputs, whereas model descriptions are mistakenly
seen as minor, "gray literature" publications.
Models are continually undergoing revision, rendering
documentation obsolete.
Current models typically have multiple authors, which
can make the coordination of publication difficult.

As a partial solution to the difficulties of model
documentation, we propose that a dynamic description
of models be maintained on a web site and that any
CIMMYT participation in model development include
requirements for contribution to this documentation.

Data management
CIMMYT is a logical source of crop data to develop
and evaluate maize and wheat models. The Center's
research data are dispersed over a wide range of
groups, and, in some cases, valuable data are only
available in summarized forms in research papers.
Initiatives are underway to assemble research data
into easily accessible databases. Like most
institutions, though, funding constraints limit the
speed of development.

The International Wheat Information System (IWIS)
provides ready access to results of CIMMYT
international wheat trials on CD-ROM (Fox et al.
1997). Besides grain yield, the data reported can
include days to heading and maturity, 1000-grain
weight, and basic summaries of management.
Unfortunately, no weather or soil profile descriptions
are given. A further limitation is that results of many
CIMMYT wheat experiments are not included
because they do not form part of the international
trial system.

Recognizing the deficiencies of IWIS, the CIMMYT
Wheat Program participates in the development of the
International Crop Information System (ICIS). This
system is still incomplete, but design specifications
include the ability to store a much wider range of data
and experimental designs than is possible with the
original IWIS design and to output data in model-
ready formats such the ICASA standards (Hunt et al.
2000). Another important innovation of ICIS is to
allow users to maintain local versions of the database
that can be uploaded to the central version at the users

Data from CIMMYT's Maize International Testing
Unit are also stored in electronic format; currently,
access is only possible through specific requests to
the Unit. The CIMMYT Maize Program is evaluating
options for developing a more flexible information
system, and it is likely that they will either join ICIS
or create a parallel system with similar functionality.
The Maize Physiology group is committed to
maintaining a set of quality data for model
development and evaluation.

A Crop Modeling Strategy for CIMMYT

The Natural Resource Group is developing the
Sustainable Farming Systems Database (SFSD) as a
specialized implementation of ICIS that can manage
long term and multiple-species trials. The SFSD should
eventually become a major repository of data for long
term trials and cropping systems research. ICIS thus
holds the promise of making a large amount of data
available in model-ready formats. However, parallel to
this effort, there is still need for CIMMYT researchers
to make more data available for modeling efforts.

Furthermore, although CIMMYT does not seem to have
a comparative advantage in maintaining soil and
meteorological data, the lack of ready sources of data is
a major impediment to model use. A dual strategy seems
appropriate here. One component is to collect quality
sets of soil and weather data. The other is to develop
interpolated surfaces of soil and climatic data that can
be used with crop models. CIMMYT has assisted in the
development of the Spatial Characterization Tool (SCT;
Corbett and O'Brien 1997). The SCT currently includes
surfaces of monthly precipitation and temperature from
Africa and Latin America, and surfaces for Asia are under
development (J. Corbett, personal communication, 2000).

Transfer of modeling skills to national
agricultural research systems
Nothwithstanding major projects such as IBSNAT
(Harrison et al. 1990) and SARP (Penning de Vries et al.
1991) and continued training efforts, models have seen
limited use by NARS except in small workshops, which
are usually organized by local agencies with international
trainers. Limited access to adequate hardware and
software and the turnover of NARS staff trained in
modeling are obvious constraints. Ready access to
adequate weather and soil data is also a concern.

Perhaps the greatest obstacle, however, is the lack of
understanding of the potential applications of models to
research concerns in developing countries. Underlying
this is an absence of well-documented case studies where
modeling has brought major benefits to NARS research
concerns. An appropriate strategy for CIMMYT may be
to promote workshops that focus on raising awareness of
the potential use of models, as opposed to transferring
modeling skills to a large number of researchers.


Reviewing the points above, we suggest that
CIMMYT should:

1. Participate in the continued development of
CERES-Maize, CERES-Wheat, and models such
as CROPGRO, which form part of our systems
research initiative as robust models for a diverse
set of research applications. Priority areas for
development include:
Externalization of model parameters.
Improved handling of phenology, leaf number,
and yield components under tropical
Development of more realistic handling of
crop residues and runoff.
Improved documentation of both models.
2. Support the development of models that provide
more detailed representations of processes related
to tillage and residue management.
3. Participate in the development of a more
mechanistic crop model, possibly based on
CROPGRO, that would include effects of canopy
architecture and would simulate cultivar
differences in canopy surface temperatures.
4. Participate in the development and implementation
of intercropping routines in ICASA-compatible
5. Promote adherence to ICASA standards for model
6. Maintain quality maize and wheat datasets for
model development and model evaluation.
7. Participate in efforts to make soil and
meteorological data widely available.
8. Promote proper documentation of crop models,
possibly managing a web site for documentation of
models used at CIMMYT.

Given current levels of resourcing, the actual writing
of model code at CIMMYT should focus only on key
problems where CIMMYT scientists have a clear
advantage. However, in most cases, a more
appropriate strategy is for CIMMYT to collaborate
in model development at other institutions.

Directions in Modeling Wheat and Maize for Developing Countries


Acock, B., Y.A. Pachepsky, E.V Mironenko, ED.
Whisler, and VR. Reddy. 1999. GUICS: A generic
user interface for on-farm crop simulation. Agronomy
Journal 91:657-665.
Amani, I., R.A. Fischer, and M.P. Reynolds. 1996.
Canopy temperature depression association with yield
of irrigated spring wheat cultivars in a hot climate.
Journal ofAgronomy and Crop Science 176:119-129.
Bellows, EE. 1997. Growth and productivity of maize
under nitrogen stress. In G.O. Edmeades, M.
Banziger, H.R. Mickelson, and C.B. Pefia-Valdivia
(eds.), Developing Drought and Low-N Tolerant
Maize. Proceedings of a Symposium, 235-240.
Mexico, D.E: CIMMYT.
Bolafios, J., and G.O. Edmeades. 1996. The importance
of the anthesis-silking interval in breeding for drought
tolerance in tropical maize. Field Crops Research
Edmeades, G.O., J. Bolafios, M. Hernandez, and
S. Bello. 1993. Causes for silk delay in lowland
tropical maize. Crop Science 33:1029-1035.
Corbett, J.D., and R.F. O'Brien. 1997. The Spatial
Characterization Tool. Texas A&M University
System, Blackland Research Center Report No. 97-
03. Texas, USA: Texas A&M University System. CD-
Daroub, S.H., A. Gerakis, J. T. Ritchie, and J. Ryan.
1999. Predicting phosphorus availability to wheat in
semi-arid areas using CERES-P AgronomyAbstracts,
p. 41.
Fox, PN., R.I. Magafia, C. Lopez, H. Sanchez, R.
Herrera, V Vicarte, J.W. White, B. Skovmand, and
M.C. Mackay. 1997. International Wheat information
System (IWIS), Version 2. Mexico, D.E: CIMMYT.

Gerakis, A., S. Daroub, J.T. Ritchie, D.K. Friesen, and
S.H. Chien. 1998. Phosphorus simulation in the
CERES models. AgronomyAbstracts, p. 14.
Grace, P.R., D.C. Godwin, and J.T. Ritchie. 1994.
Improving soil carbon and nitrogen dynamics in
CERES models. AgronomyAbstracts, p. 18.
Harrison, S.R., P.K. Thornton, and J.B. Dent. 1990. The
IBSNAT project and agricultural experimentation in
developing countries. ExperimentalAgriculture
Hunt, L.A., G. Hoogenboom, J.W. Jones, and J.W.
White. 2000. ICASA files for experimental and
modeling work. Online at http://icasanet.org/
standards/index.html (verified 25 August 2000).
Penning de Vries, EWT., H.H. van Laar, and M.J.
Kropff (eds.). 1991. Simulation and Systems Analysis
and Rice (SARP). Wageningen: Pudoc.
Sinclair, T.R., and T. Horie. 1989. Leaf nitrogen,
photosynthesis and crop radiation-use efficiency: a
review. Crop Science 29:90-8.
Thornton, P.K., H.W.G. Booltink, and J.J. Stoorvogel.
1996. DSSAT Version 3.1: Spatial analysis. In G.Y.
Tsuji and G. Uehara (eds.), DSSAT Version 3, 1-78.
Honolulu, Hawaii: University of Hawaii.
van Beem, J., and M.E. Smith. 1997. Variation in
nitrogen use efficiency and root system size in
temperate maize genotypes. In G.O. Edmeades, M.
Banziger, H.R. Mickelson, and C.B. Pefia-Valdivia
(eds.), Developing Drought and Low-N Tolerant
Maize. Proceedings of a Symposium, 241-244.
Mexico, D.F.: CIMMYT.
Veraart, V 1998. APSFront User Manual. Online at

A Physiologist's Wish List

A Physiologist's Wish List for a Robust

Wheat Model

M.P. Reynolds
Wheat Program, CIMMYT, Mexico

The major objective of crop modeling is to explain yield variation by combining physiological
processes with environmental data in a quantifiable framework; however, models frequently fall
short of this objective. It is suggested that modeling specific physiological processes might be a
more achievable goal, in addition to providing building blocks for more comprehensive models.
Some reasonably well studied processes that could be modeled include the relationship between
stomatal aperture and yield, or the evaluation of alternative grain filling strategies such as stay-
green versus remobilization of stem reserves. Lodging, which can have a major impact on wheat
yield, would lend itself quite readily to being modeled, since it is a largely mechanical process.
Where genetic bases of physiological responses are partially understood, as in the case of Ppd and
Vrn on phenological development, a gene-based modeling approach might permit extrapolation
whereby phenological patterns that optimize source-sink balance could be determined.


The ideal crop breeding model would identify specific
trait or gene combinations that optimize yield in a given
environment. In reality, our understanding of the
interaction between genotype and environment is
incomplete at all levels of integration, whether we are
talking agronomic traits, physiological process, or genes.
Nonetheless, traditional plant breeding has very
successfully exploited yield testing to improve adaptation
of wheat to most environments. Some geneticists suggest
that, in a short time, using new techniques such as
functional genomics, a sufficiently comprehensive
understanding of the genetic basis of yield will enable
cultivars to be designed at the allelic level. Somewhere
between the two extremes, physiologists attempt to
identify traits (usually representing more than one gene)
that, in the right genetic background, will enhance yield.
The basic principal of crop modeling is to put these traits
into a quantifiable framework. While existing models are
not sophisticated enough to accurately simulate
differences in genetic yield potential, a step in this
direction would be to improve our understanding of

genetic differences in the components of crop growth.
Working models of specific crop functions that relate to
yield could be used as the building blocks for a robust
wheat model for crop performance.

Modeling Stomatal Aperture
Related Traits

A number of stomatal aperture related traits (SATs),
such as stomatal conductance (COND), canopy
temperature depression (CTD), and carbon-isotope
discrimination, have been shown to be strongly
associated with performance in wheat (Fischer et al.
1998; Reynolds et al. 1994). There is considerable
interest in the possibility of using SATs as indirect
selection criteria in conventional breeding to identify
physiologically superior lines in early generations. A
modeling procedure may have application in: 1)
quantifying potential improvement in breeding
efficiency associated with applying selection pressure
for SATs in different environments, and 2) predicting
the physiological mechanisms that underlie the
association of yield potential with expression of SATs.

CIMMYT. Published in White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing Countries: Proceedings ofa
Workshop, CIMMYT El Batan, Mexico, 4-6 May 1998. NRG GIS Series 01-02. Mexico, D.F.: CIMMYT.

Directions in Modeling Wheat and Maize for Developing Countries

In the first case, important factors that would influence
the impact of selection for SATs on breeding efficiency
include: the typical range of SATs values associated
with genetic variability in a given environment,
accuracy of instrumentation, measurement errors due to
environment (e.g., wind, soil heterogeneity), heritability
of SAT, genetic correlation of SAT with yield, and
genetic gains associated with visual selection (Table 1).
All of these factors can be estimated empirically in
different environments. The model could be used, for
example, to compare the relative efficiency of selection
for different SATs, or to estimate the intensity of SATs
measurements required to maximize genetic gains in a
given environment (Table 1).

The second example predicting the physiological
mechanisms that underlie the association of yield with
SATs is more challenging, since SATs tend to be
complex traits affected by many plant characteristics
(Table 2). Capacity for vascular transport may influence
evapotranspiration rate and stomatal aperture, and
metabolic capacity (e.g., rate ofC fixation) has a
feedback effect on stomatal conductance. Partitioning to
yield (i.e., harvest index) may also influence stomatal
conductance during grainfilling, since grain number and
filling rate influence the demand for photo-assimilates.
Environment can also influence SATs, for example, via
the effect of temperature on metabolism (Keeling et al.
1994). Another example would be low soil water
potential, which can reduce stomatal conductance via
root signals even before water stress is detectable in the
leaf (Davies and Zhang 1991). If these processes and
their interactions could be accurately simulated and
extrapolated, the results might provide a basis for
evaluating the extent to which SATs can be used to
select for increased yield potential in different

Modeling Alternative Grain Filling
Strategies: Stay-Green versus
Remobilizing Stem Reserves

If you ask a breeder whether it is more desirable to
select for the stay-green trait or remobilization of
soluble stem carbohydrates to the grain, the answer is
likely to be yes to both. However, Blum (1998) argues
that the two traits may be mutually exclusive. Since it is
difficult to experimentally quantify the relative effect of
these traits, simulation modeling might be a useful

approach to establish, in theory, the relative trade-off
between the two strategies. The model could be built
using empirical observations of assimilation rate during
terminal grain filling, calculations of potential
photosynthesis based on the range of chlorophyll
content observed, and estimates of soluble stem
carbohydrates available for translocation taking into
consideration respiratory costs (Table 3).

Table 1. Examples of inputs and outputs for a
model simulating the association of stomatal
aperture related traits (SATs) with yield.

Range of SATs associated with genetic variability in
a given environment.
Accuracy of measuring instrument.
Measurement errors due to environmental variation.
Heritability of SAT.
Genetic correlation of SAT with yield.
Genetic gains associated with visual selection.

Compare the relative efficiency of different SATs.
Estimate how many measurements of SATs are
required to maximize genetic gains.

Table 2. Inputs and outputs for a model
predicting physiological mechanisms
contributing to expression of SATs.

Vascular capacity: may not always be sufficient to
meet evaporative demand or sink demand.
Partitioning: sink demand effects regulation of
photosynthesis and stomatal conductance.
Metabolic processes: e.g., soluble starch synthase is
inhibited by heat.
Soil water potential: may reduce stomatal
conductance via root signals.
Environmental parameters: temperature, radiation,
relative humidity.

Optimal stomatal conductance to maximize
assimilation rate.
Indication of the physiological mechanisms limiting
assimilation rate.

A Physiologist's Wish List

Traits Determining Lodging Tendency in

Recent research has suggested that lodging is a function
of the dissipation of kinetic energy by stems (Farquhar
et al. 1997). Observations of the oscillatory motion of
wheat stems in response to artificial wind gusts suggest
that some wheat lines dissipate kinetic energy through
vibration, while others transmit the moment to the base
of the plant, inducing root lodging. Traits involved in
this process include the center of gravity of the plant,
the elasticity of the stem that is modified by its
anatomical structure, as well as its biochemical
composition (e.g., cellulose and silica contents). If a
simulation model was designed that used the measurable
physical and anatomical information, and accurately
predicted transmission of force to the base of the plant
over a range of variables (including wind intensity as
well as plant traits), it may be extrapolated to provide
information on plant traits that might reduce lodging
tendency at current yield levels as well as for the higher-
yielding wheat lines (Table 4).

Modeling Phenology to Optimize Source-
Sink Balance

The phenological development of a crop plays an
important role in determining fertility and grain number.
In theory, it should be possible to manipulate the relative
duration of phenological stages, such that yield is
optimized in a given environment. For example, in high
yielding environments, it has been suggested that an
increased duration of the rapid spike-growth phase may
increase partitioning of assimilates to the spike,
improving fertility and optimizing source-sink balance
(Slafer et al. 1996).

The genes involved in determining response to
photoperiod (Ppd) and vernalization (Vrn) have a major
effect on determining phenological response to the
environment, and hence a gene-based modeling
approach may be feasible as our knowledge of gene
action increases. In addition, since developmental rate is
strongly affected by temperature, phenological models
can be further improved by using actual plant
temperatures. Assuming that temperature is sensed by
leaves, it is possible to estimate leaf temperature from
air temperature if vapor pressure deficit and water
availability are known (Table 5).

Table 3. Inputs and outputs for a model simulating
stay-green vs. stem reserve mobilization.

Net assimilation rate during terminal grain filling.
Leaf chlorophyll content in late grain filling.
Soluble stem carbohydrates available.
Translocation efficiency.
Sucrose to starch conversion efficiency.

Best strategy genetically for maximizing grain filling.

Table 4. Possible inputs and outputs for a model
simulating lodging tendency.

Center of gravity of wheat stem.
Stem elasticity.
Anatomical structures of the wheat canopy: spikes,
awns, stems, leaves, roots.
Biochemical composition of stem, e.g., cellulose and
silica contents.

Predict transmission of force to the base of plant over a
range of plant traits.
Evaluate plant traits to reduce lodging tendency at
current yield levels.
Extrapolate for the higher yielding wheat lines, new
plant types.

Table 5. Modeling the impact of phenological
patterns on source-sink balance.

Environmental: photoperiod, radiation, temperature, soil
water, RH, calculated plant temperatures.
Specific genes and alleles: Ppd, Vrn, Eps.
Physiological: cardinal phenological stages, potential
number of grains, grain weight potential.

Information on how genes and environment interact to
determine duration of phenological stages.
Information on strategic deployment of genes to
optimize source-sink balance.

Directions in Modeling Wheat and Maize for Developing Countries

Integrating Small Models

The examples given above represent reasonably well
studied areas of physiology. By definition we can only
model processes which we at least partially
understand. Nonetheless, connections between these
areas become apparent. One of the outputs for the
stomatal aperture model is leaf temperature, which is
an input for the model on phenological development.
The output for the model on alternative grain filling
strategies could be used as input for the model on
stomatal aperture, etc. Eventually, models of discrete
physiological processes may be integrated to build a
more robust model for yield.


Blum, A. 1998. Improving wheat grain filling under
stress by stem reserve mobilisation. Euphytica
Davies, W.J., and J. Zhang. 1991 Root signals and the
regulation of growth and development in drying soil.
Annual Review of Plant Physiology and Plant
Molecular Biology 42:55-76.
Farquhar, T, J. van Beem, H. Meyer, and M.P. Reynolds.
1997. Lodging resistance of wheat assessed by video
image analysis, pp. 143. AgronomyAbstracts.
Fischer, R.A., D. Rees, K.D. Sayre, Z. Lu, A.G. Condon,
A. Larque Saavedra, and E. Zeiger. 1998. Wheat yield
progress is associated with higher stomatal
conductance, higher photosynthetic rate and cooler
canopies. Crop Science 38:1467-1475.
Keeling, P.L., R. Banisadr, L. Barone, B.P. Wasserman,
and A. Singletary. 1994. Effect of temperature on
enzymes in the pathway of starch biosynthesis in
developing maize and wheat grain. Australian Journal
ofPlant Physiology 21:807-827.
Reynolds, M.P., M. Balota, M.I.B. Delgado, I. Amani,
and R.A. Fischer. 1994. Physiological and
morphological traits associated with spring wheat
yield under hot, irrigated conditions. Australian
Journal ofPlant Physiology 21:717-30.
Slafer, G.A., D.F. Calderini, and D.J. Miralles. 1996.
Yield components and compensation in wheat:
opportunities for further increasing yield potential. In
M.P. Reynolds, S. Rajaram, and A. McNab (eds.),
Increasing Yield Potential in Wheat: Breaking the
Barriers, 101-133. Mexico, D.F.: CIMMYT.

Use of CERES for Tropical Maize

Issues Arising from the Use of CERES for

Tropical Maize

G. Edmeades' and J. Bolahos2
'CIMMYT Maize Program, Mexico
2CIMMYT Maize Program, Guatemala

The authors describe drawbacks of the CERES model that limit its usefulness in representing the
performance of tropical maize. These include the difficulty of measuring thermal time from
emergence to the end of the juvenile phase in the field and the model's inability to predict the
performance of tropical highland maize or to accurately account for the effects of different seeding
densities, key stresses, or variations in daylength. They mention information collected by
CIMMYT scientists that bears on the model's genetic coefficients for maize, and list potentially
useful datasets available from the CIMMYT Maize Program. Finally, they caution potential users
about taking model outputs (sophisticated graphs, extremely precise figures) too seriously or
considering numerical outputs as actual "data."

Our Interest in the
CERES-Maize Model

The CERES-Maize model can be a valuable tool for
examining potential yields and the probability and
extent of yield loss due to drought, low soil fertility, and
high temperatures in given environments. The
International Maize and Wheat Improvement Center
(CIMMYT) can assist in the development of a version
of CERES-Maize for tropical environments
characterized by genotypes and production systems that
range in elevation from sea level to over 3,000 masl and
from latitudes of 0 to 30 N or S. Such a model would
allow us to examine an array of management options
that aim to increase productivity while preserving the
natural resource base of soil and water in complex
maize-based cropping systems. In tropical regions, open
pollinated varieties (OPVs) are still used extensively, but
are gradually being replaced by hybrids that exhibit
higher yield potentials under most conditions. Both
types of germplasm carry specific resistances to tropical
diseases and are generally highly photoperiod sensitive.
To further these goals, the modeling efforts of
CIMMYT's Maize Program have focused on:

1. Quantifying genetic coefficients for tropical
cultivars (OPVs and hybrids).
2. Developing and publishing data sets of
representative tropical cultivars.
3. Determining whether the model satisfactorily
explains genetic differences in stress tolerance (in
particular, drought, low nitrogen, plant density,
and acid soil tolerance), especially during
flowering, when barrenness occurs and ears per
plant fall below a value of 1 (i.e., when a
significant proportion of plants within the crop
have no ears at all).
4. Determining how well the nitrogen submodels
predict N uptake, distribution and redistribution
among plant parts under N and water stress.
5. Developing a more effective selection strategy in
breeding for environments where abiotic stresses
are common, using data from 3) and 4).
6. Improving the definition of major production
environments (mega-environments) for maize in
the tropics through the development of better
tropical maize models. Such a definition is needed
to prioritize research in CIMMYT's Maize

CIMMYT. Published in White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing Countries: Proceedings of a
Workshop, CIMMYT El Batan, Mexico, 4-6 May 1998. NRG GIS Series 01-02. Mexico, D.F.: CIMMYT.

Directions in Modeling Wheat and Maize for Developing Countries

Problems with CERES-Maize version 3.1

In using CERES-Maize v3.1, we encountered various
areas where the performance of the model was
unrealistic. Examples include:

* Leaf number predictions were often around 28-35 on
late maturing cultivars and 20-25 on early maturing
cultivars, rather than the observed values of 22-25 and
15-19, respectively. The use of a phyllochron value
(PHINT) of around 45-50Cd places a cap on leaf
number in the planned release of CERES-Maize v3.5,
which will enable us to more realistically predict
flowering dates. We see some genetic variation in
PHINT in the field, but it is usually not larger than
10-15% of the mean.

* Genetic coefficient P1 (thermal time from emergence
to the end of the juvenile phase, i.e., the start of the
photoperiod sensitive phase) has a strong effect on
final leaf numbers but is very difficult to measure in
the field. We have measured P1 experimentally using
plants grown under artificially lengthened days. Our
measured values must be corrected for thermal time
to germination (around 110Cd) and for time from the
end of the juvenile phase to TI (TI taken to be the
point when the tassel reaches 0.5 mm in length; about
5 days or 850Cd). This suggests that our observed

values should be reduced by around 1950Cd. Similarly,
photoperiod sensitivity measured in the field averages
around 2.7 d/h from sowing to TI in lowland tropical
genotypes but can vary up to 6 to 7 d/h in highly
sensitive varieties. Note that our values are computed
by regressing changes in time to TI on photoperiod
between the daylengths of 13 to 15 h, rather than from
12.5 h, as assumed in CERES-Maize. If the critical
photoperiod was indeed 13 h, and we wish to predict
the effects of photoperiod in a site like Tlaltizapan
under 13.3 h summer photoperiods, our values have to
be reduced by multiplying them by a factor of 0.3/0.8
or 0.38. Furthermore the determination of P1 is
difficult. In environments where photoperiod does not
vary greatly, an alternative would be to use final leaf
number and/or days to anthesis, which are more easily

Indications from growth chamber data on photoperiod
sensitive genotypes are that P, (the critical photoperiod
above which flowering is delayed by lengthening days)
is a function of temperature such that P, is greater in
cool conditions than in warm conditions. This helps to
explain the apparent reduction of about 35% in
photoperiod sensitivity in sensitive germplasm in
winter when nights are cooler, as observed in the field
in Tlaltizapan, Mexico (Table 1). Note, however, that a
similar growth-chamber-based analysis confirmed that
the value of P, for a Corn Belt hybrid B73 x Mo 17 was
virtually unaffected by temperature and remained
around 12.5 h, as assumed by CERES-Maize.

Table 1. Comparison of sensitivity to daylengths >13 h, as measured for 40 diverse maize cultivars over
summer and winter at Tlaltizapan, Mexico. Data are thermal times from time of sowing to tassel
initiation (TI), anthesis (AD), and final leaf number (FLN). T.ma in summer and winter are 310 and 300C,
respectively; corresponding figures for Tmin are 190 and 120C, respectively.

TT to TI d(TT to TI) TT to AD d(TT to AD) FLN d(FLN)
(Cd) (Cd/h) (Cd) (Cd/h) (If) (If/h)
All genotypes (N=40)
Summer 360 57 993 103 18.7 1.82
Winter 430 40 1,049 70 20.1 1.05
Lowland tropical (N=15)
Summer 359 72 978 133 19.2 2.37
Winter 452 55 1,072 83 20.9 1.41
Temperate (N=7)
Summer 365 16 1,000 36 18.0 0.69
Winter 434 18 1,054 37 19.8 0.33
Source: Edmeades et al. (1994).

Use of CERES for Tropical Maize

* There may be problems with how heat units are
calculated for highland-derived genotypes. Growth
room comparisons suggest that the Top for
development rate from sowing to tassel initiation is
around 20 to 21 C for true highland genotypes,
compared to around 300C for lowland tropical,
temperate, and subtropical genotypes (Ellis et al.
1992). During flowering, Top of highland genotypes
seems to be around 260C. Therefore, classical systems
of thermal time calculation, assuming a Top of 30C,
will not work well for highland-adapted germplasm,
nor will predictions of phenology in diverse

* Stress at the time of flowering extends the anthesis-
silking interval (ASI), and a close relationship (r=-
0.60) is observed between ASI and grain yield
(Bolafios and Edmeades 1993, 1996). This response is
caused by delayed silk emergence under drought, high
plant density, and low soil fertility (Lafitte and
Edmeades 1995), and thus is a widespread problem
where phenology and production of maize are being
predicted for environments where stress at flowering
is common (as in many tropical environments). We
believe that variation in ASI reflects the rate of growth
of the ear (Edmeades et al. 1993). When we select for
reduced ASI under stress, plants with more rapid ear
growth rates and fewer spikelets are obtained;
therefore, ASI reflects partitioning to the ear. If
relationships can be determined between kernels per
plant (or per m2) and an integral of crop growth rate
during the flowering period (estimated as 120Cd
before anthesis to 2500Cd after anthesis ), the slope of
the regression of kernel number on crop growth
during that period may indicate differences in
tolerance to stress during flowering.

* At low plant densities, parameters for kernel number
may also be used to describe a genetic tendency to
prolificacy (more than one fertile ear per main stem).
At present, CERES-Maize does not accurately predict
barrenness, nor does it then grow out barren plants as
a separate population to the fertile plants within the
crop community. As a consequence, the number of
ears per plant is poorly predicted, and density
experiments conducted using the model provide
unrealistic results. Furthermore, plant population is
one of the most important factors in yield
determination in maize (Bolafios 1992; Duvick 1997).

For the CERES model to have applicability in maize
cropping system research, it has to offer improved
simulation of maize response to both high and low

* If CERES is used to reproduce Denmead and Shaw's
(1962) classic experiments on the sensitivity of maize to
water stress at different development stages, grain yield
only begins to decline 8 to 9 days after pollination,
affecting grain size but not grain number. CERES-Maize
currently shows no effect of water stress on tassel or ear/
ovule development (which control ASI, barrenness, etc.).
Ear development, which sets the number of ovules to be
pollinated, should be sensitive to water stress from
around a third to halfway through development; i.e.,
from initiation to silking. Grain number should decrease
in response to water stress.

* Since the model was developed using silking date rather
than anthesis date as a key stage in phenology, anything
that increases ASI will result in an error in predicting
physiological maturity of the crop. It is important that
anthesis date be used as the definitive milestone of
development for these sorts of environments.

* The RUE value of 4.5 (or 5) used by CERES-
Maize appears to be too high for tropical maize.
Respiration is high, and peak intense radiation is
conducive to photo-oxidation.

Information That We Have Collected:
Genotype Coefficients

* Information on P1 and P2 has been collected from
experiments comprising approximately 100 common
tropical lowland, subtropical, and highland varieties,
with some common Corn Belt checks, under 4
daylengths that varied in 1.5-h steps from the ambient
(11.5-13.3 h) to 17.5 h. Unfortunately, we could not
estimate thermal time from silking to physiological
maturity in these same genotypes, since long days
greatly disturb kernel set. Thus, much less data are
available on estimates of P5. From the breeder's
viewpoint, collecting data on time from sowing to
50% anthesis or silking, final leaf number, and time
to brown husk or 50% leaf senescence would be a lot
easier than obtaining P1, P2, and P5. So why not express
sensitivity in terms of rate of change in time to anthesis
or in final leaf number, per hour increase
in photoperiod?

Directions in Modeling Wheat and Maize for Developing Countries

* A small amount of data on G2 (maximum kernel
number per ear) is available, although kernel number
per ear under normal competitive conditions is often
measured in many of our trials and typically is around
400 kernels under densities of 5 plants/m2. G3 has
been measured, but we usually cannot assume that it is
a maximum rate. Typically it averages around 8 mg
kernel/d at 5 plants/m2 for many of our cultivars and
seems little affected by artificially reducing kernels
per plant. While G2 and G3 are treated as maximum
values, our incentive to measure them or even attempt
to estimate them accurately remains small.

* Experimental data on PHINT (rate of appearance of
visible leaves versus thermal time) has been reported
in a large number of trials (see above).

Datasets Available from the CIMMYT
Maize Program

Various datasets suitable for evaluating maize models
have been produced at CIMMYT and eventually should
be available in DSSAT format. Grouped by the scientist
who conducted the trials, they are as follows:

A. Scott Chapman
* Twelve genotypes (four lowland tropical, three
subtropical, two temperate, and three highland),
representing several maturities and including eight
OPVs and four hybrids, were evaluated in four
environments. Biomass and leaf area were measured
sequentially, together with hourly readings of
temperature and daily totals of total radiation, but no
soil parameters were recorded. Photoperiod responses
were measured under four daylengths, but data from
this trial were not processed.
* A subset of four genotypes was measured for growth
and green leaf area under drought, high plant density,
and low N. Water extraction under drought was also
* Data was almost completely processed from raw form
in Excel spreadsheet format rather than DSSAT files.

B. Anne Elings
* Five genotypes (all lowland tropical genotypes
varying in maturity, two common to the Chapman set)
were evaluated specifically under stress (three drought

levels and three N levels) in highland, subtropical,
and lowland environments. Measures similar to the
Chapman set were taken, with full phenology,
limited soil data (0-30 cm, 30-60 cm, 60-90 cm),
and extensive data on N content of plant parts
available at most sites.
* Maximum kernel growth rate experiments were
conducted on two entries, one OPV and one hybrid.
* Data on the five genotypes under three N levels,
and from an unstressed site at another location, was
in DSSAT format, although the model will not run
properly with the files as they are at present.

C. Greg Edmeades: Photoperiod responses
* Approximately 100 genotypic responses (time,
thermal time to TI, AD, and delta leaf number) were
cataloged using lights at ambient, ambient + 1.5 h,
ambient + 3 h, and ambient + 4.5 h as the four
photoperiods, and using soil temperature (5 cm)
from sowing to TI, and air temperature measured in
plot for the rest (Tbase 7 C, Tmax 30 C, broken
stick response for thermal time, computed hourly).
* Two hundred inbred lines were measured under two
photoperiods (ambient and 17.5 h) for a crude
assessment of responsiveness.
* Recombinant inbred lines of a cross CML9
(sensitive) x A632 Ht (almost insensitive) were used
to map photoperiod sensitivity with molecular
markers. Data are currently being analyzed by a
PhD student (Ms. Rkia Moutiq) at Iowa State
University, USA.

D. Jorge Bolaihos: Genetic parameters for a set
of lowland and highland cultivars
* A set of 16 lowland and 14 highland maize cultivars
(including OPVs and hybrids) was evaluated in at
least 6 environments, each ranging in altitude and
average temperature (lowland locations from 60 to
1,000 masl; highland locations from 1,000 to 2,300
masl) to determine phenological genetic
parameters. Both datasets include: weekly
determination (4 to 5 times) of the number of
initiated leaves; final number of leaves; days to
50% anthesis, 50% silking, and 50% physiological
maturity; rate of kernel growth; development of
kernel milk line; final yield and components;
senescence patterns with leaf chlorophyll meter;
and daily recordings of maximum and minimum
temperatures. Datasets are being analyzed and will
be available in 2001.

Use of CERES for Tropical Maize

Possible Problems with the Datasets

* Regarding the meteorological data, there is some
indication that radiation sensors have drifted with time
and are giving questionable values. This should be
discussed before CIMMYT data are altered, given the
unusually hazy conditions that prevail in winter at
* Soil profile data are usually lacking from our studies,
with the exception of Anne Elings' data and a subset of
Scott Chapman's data.
* Systematic errors in the way photoperiod responses
were measured may have slightly increased estimates
of sensitivity.
* Most of our data relate to OPVs, but increasingly these
data are being used to predict hybrid responses.
Hybrids will generally out-yield OPVs by 15-25%, and
show increased kernel numbers, longer filling periods,
and delayed leaf senescence.
* An unspecified amount of irrigation water has been
applied to most of our trials by furrow irrigation,
usually to maintain the trials in a nonstressed state. In
most cases, it is better to assume full irrigation, unless
otherwise specified.

General Questions Regarding Models

A model is a conceptual representation that can be
precisely formulated for a given event or series of related
events. The law of gravity, for example, is a simple
model: objects are attracted to Earth at 9 m/s2. Every
model is as good as the formulations it contains. Over the
last decade, there has been an explosion of crop
simulation models in terms of scope and sophistication.
There is a danger in that many users do not have the
required systems perspective and take the numerology
and graphicology generated by the models at face value.
This creates a large potential for misuse of crop
simulation models. As a result, there is a need to reflect
on how models can best serve potential end users, in this
case, maize agronomists, since many question the utility
of modeling.

Using data from more than 2,700 experiments from
CIMMYT's international trials (breeding nurseries and
agronomic plots) where maize yield was measured,
Crossa et al. (1993) determined an average standard error
of the difference (s ) of 500 kg/m2. This implies that for a
given treatment A to be statistically different to a given
treatment B, the two mean yields must differ by at least 1

t/ha (LSD = t s ). The value 500 kg/m2 is the best objective
measure of the hidden uncertainty in the field determination
of maize yield. If this is the best we can measure in reality
(using our best instruments of detection), should we be
concerned about the extreme detail that is presented in
computer outputs? Is the apparent precision of model output


Bolafios, J. 1992. Desarrollo del follaje, interception de
radiacion y un modelo simplificado de la productividad
potential del maiz. In Sintesis de Resultados
Experimentales 1991, Volumen 3, 215-224. Guatemala,
Guatemala: Programa Regional de Maiz.
Bolafios, J., and G.O. Edmeades. 1993. Eight cycles of
selection for drought tolerance in lowland tropical maize.
II. Responses in reproductive behavior. Field Crops
Research 31:253-268.
Bolafios, J., and G.O. Edmeades. 1996. The importance of the
anthesis-silking interval in breeding for drought tolerance
in tropical maize. Field Crops Research 48:65-80.
Crossa, J., S. Chapman, and H. Barreto. 1993. Pattern
analysis and parameters of experimental precision of
historical maize data. Internal document. Mexico, D.E:
Denmead, O.T, and R.H. Shaw. 1962. Availability of soil
water to plants as affected by soil moisture content and
meteorological conditions. Agronomy Journal 54:385-389.
Duvick, D.N. 1997. What is yield? In G.O. Edmeades, M.
Banziger, H.R. Mickelson, and C.B. Pefia-Valdivia (eds.),
Developing Drought- and Low N-Tolerant Maize.
Proceedings of a Symposium, March 25-29, 1996,
CIMMYT El Batan, Mexico, 332-335. Mexico, D.E:
Edmeades, G.O., J. Bolafios, M. Hernandez, and S. Bello.
1993. Causes for silk delay in lowland tropical maize. Crop
Science 33:1029-1035.
Edmeades, G.O., S.C. Chapman, and H.R. Lafitte. 1994.
Photoperiod sensitivity of tropical maize cultivars is
reduced by cool night temperatures. AgronomyAbstracts
Ellis, R.H., R.J. Summerfield, G.O. Edmeades, and E.H.
Roberts. 1992. Photoperiod, temperature, and the interval
from sowing to tassel initiation in diverse cultivars of
maize. Crop Science 32:1225-1232.
Lafitte, H.R, and G.O. Edmeades. 1995. Stress tolerance in
tropical maize is linked to constitutive changes in ear
growth characteristics. Crop Science 35:820-826.

Directions in Modeling Wheat and Maize for Developing Countries

Canopy Development, Radiation Interception,

and a Simple Model of Maize Productivity

Jorge Bolanos
Maize Program, CIMMYT, Guatemala

A conceptual model for maize crop productivity that uses canopy radiation interception and its
conversion to biomass is described. Crop duration, seasonal radiation, the fraction of radiation
intercepted by foliage, the efficiency of conversion into biomass, and the portion partitioned to the
harvestable part of the crop are all needed for estimating maize grain yield. Model assumptions are
illustrated using schematic examples derived from tropical maize (Zea mays L.) under varying
environmental conditions. The model also provides an adequate and easy conceptual framework for
ex-ante examination of agronomic and breeding strategies for increasing crop productivity.


This paper presents a simple model of the productivity
of annual crops, where crop yield is the result of net
carbon dioxide assimilation and the partitioning of
assimilates to the harvestable part of the crop, integrated
over the duration (emergence to harvest) of the crop
cycle. Stated mathematically:

Y = HIA dt

where Y is harvestable yield, HI is the harvest index or
the proportion of Y in the total biomass, A is the net
assimilation rate per unit land area, and t is time
(Monteith 1990; Loomis and Connor 1992).

Overall assimilation is the product of source size
integrated over the time span in question, multiplied by
source intensity. In unstressed crops, the major
determinant of biomass production is the amount of
radiation intercepted over the crop cycle (Loomis and
Connor 1992). Biomass accumulation is the product of
incident solar radiation, the fraction intercepted by
green leaf area, the efficiency with which the radiation
is used, and partitioning to the harvestable part of the
crop. Therefore, grain yield (Y) can be expressed as:

Y tx Rs x %RI x ExHI

where t is crop duration in days, Rs is average daily
solar radiation (MJ/m2/d), %RI is the percentage of the
total seasonal radiation intercepted by the canopy, E is
the average radiation use efficiency (g/MJ), and HI is
harvest index.

Given no change in crop duration, there are two options
to increase biomass: 1) increase radiation interception
through faster early-leaf-area development and/or slower
senescence, or 2) increase average radiation use
efficiency for all or part of the season.

This simple model can predict grain yield with relative
accuracy ( 1-2 t/ha) if provided with sound estimates
of %RI, E, and HI, taking into account the underlying
assumptions of model parameters. It also provides an
adequate conceptual framework for the ex-ante
examination of the potential impact of different
breeding or agronomic strategies on crop productivity.

The Radiation Environment

In the tropics (0 to 30 latitude), incident daily solar
radiation (Rs) can vary from about 10 to 25 MJ/m2d,
due mostly to changes in latitude, day of year, and
degree of cloudiness (Gates 1980). Photosynthetically
active radiation (PAR) is normally around 400 to 700
nm or 48 to 52% of total R (Loomis and Connor 1992).

CIMMYT. Published in White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing Countries: Proceedings ofa
Workshop, CIMMYT, El Batan, Mexico, 4-6 May 1998. NRG GIS Series 01-02. Mexico, D.F.: CIMMYT.

A Simple Model of Maize Productivity

The solar constant (incident radiation just outside the
earth's atmosphere) is around 30 MJ/m2/d, but
atmospheric attenuation typically decreases Rs to around
20 to 25 MJ/m2/d on clear, sunny days and to around 5
tol0 MJ/m2/d on cloudy, rainy days, under typical
tropical conditions. Solar radiation is normally
measured with pyranometers, although it can be
predicted from latitude, day of year, and ratio of bright
to cloudy hours (Gates 1980). For example, using a
daily average of 20 MJ/m2/d and a total crop duration of
120 d, total available radiation would be 2,400 MJ/m2
for the season.

Canopy Development and
Radiation Interception

For a given crop, the pattern of canopy development
depends on planting density; spatial arrangement; leaf
architecture; rate of leaf initiation, expansion, and
senescence; and many other genetic and environmental
factors. Maize is normally planted at densities ranging
from 4 to 7 plants/m2 (Fischer and Palmer 1984). In
annual crops, canopy radiation interception starts at the
time of emergence and increases as foliage expands.
During the early stages of crop growth, much of the
incident radiation is not intercepted by the canopy, thus
biomass production is limited by source size (Hsiao
1982). Given the architecture of most maize cultivars,
complete radiation interception occurs with leaf area
indices (LAI; m2 of foliage per m2 of land area) ranging
from 3 to 4 for cultivars with large, lax leaves to 5 to 6

S Flowering

0 20 40 60

for cultivars with small, narrow, erect leaves (Fischer and
Palmer 1984; Loomis and Connor 1992). Percent radiation
interception decreases during crop maturation and grain
filling due to leaf senescence and loss of green leaf area.

Along with crop water use, the cumulative amount of
radiation intercepted throughout the season is the most
important factor determining total biomass and grain
production, more than the intensity at any given moment
(Monteith 1990; Loomis and Connor 1992). Total crop
duration also helps govern the seasonal amount of
radiation intercepted by most crops. If even mild levels
of stress occur early in the season, the effects normally
compound over time and drastically reduce the seasonal
amount of intercepted radiation (Hsiao 1982).

These principles are illustrated using three schematic cases
(A, B, C) of maize canopy development, reflecting different
management and/or environmental conditions (Fig. 1).
Percentages shown refer to the proportion of total incident
radiation intercepted by each canopy from emergence to
harvest; i.e., the area under each curve (%RI).

Maize crop A exemplifies a crop that received very good
agronomic management (high fertility, weed control, etc.),
so the canopy rapidly covers the ground, reaching almost
full (>95%) cover around flowering. Even under such ideal
environmental conditions, the canopy intercepts only 54%
of all available radiation (Fig. 1). The dotted line in the
figure represents a cultivar with stay-green, which captures
only a further 4% radiation (58% of the total).

80 100 120

Days after planting

Figure 1. Canopy interception (%) of incident radiation as a function of days for three schematic cases
(A, B, C) of maize canopy development reflecting different management and/or environmental
conditions. (Refer to text for explanation of each case.) Numbers shown refer to the proportion of total
radiation intercepted by each curve (i.e., the area under each curve).

Directions in Modeling Wheat and Maize for Developing Countries

Maize crop B is typical of bad agronomic management
and/or environmental stresses occurring during crop
establishment (e.g., poor weed control, low fertility, pests),
so the canopy never fully covers the ground. Crop B
intercepts only 32% of the total radiation available; 22%
less than crop A (Fig. 1). In addition, incident radiation
intercepted by weeds (rather than the maize canopy) results
in an equivalent usage of soil water and nitrogen by the
weeds, meaning less are available to the maize. If
intercepted by dry bare soil, incident radiation will increase
crop evapotranspiration (ET) by inter-row advection.

Maize crop C exemplifies a severely stunted canopy due to
bad agronomic management and/or environmental stresses
(drought, low fertility, weed competition, etc.) beginning
during early crop establishment. The negative effects
accumulate over time, so crop C intercepts only 19% of the
radiation available; 35% less than crop A.

The three cases show that, even with good agronomic
management and no environmental stresses, monocropped
maize canopies will intercept a maximum of only 55-60%
of seasonally available radiation. This is because canopy
cover early in the season is incomplete and canopies
senesce during grain filling and crop maturation. In
addition, only relatively small changes (5-10%) can be
achieved with improved agronomy and/or breeding.
Environmental stresses and/or bad agronomic management
can reduce cumulative %RI significantly (crop B
intercepted only 32% of all available radiation and crop C a
mere 19%). Due to the exponential nature of canopy
growth during early development, the effects of even very
mild stresses early in the season can accumulate over time,
leading to substantial reductions in total %RI (Hsiao 1982).

Estimates of %RI for curves A, B, and C (Fig. 1) agree
reasonably well with reports: 55-60% under well-watered
and fertilized conditions, and 30-40% under drought or low
N conditions (Muchow and Davis 1988; Muchow 1989a,b).
In a study of 100 tropical and temperate S1 maize lines, the
average RI% was 39% and varied between 30% and 46%
among entries (Chapman and Edmeades 1996).

Radiation Use Efficiency and Productivity

By intercepting radiation, canopies produce biomass and
use water through photosynthesis, hence the close
relationship between radiation interception, biomass

production, and water use found in many crops under
many different environmental and management
conditions (Monteith 1990). Theoretically, the potential
productivity of a crop surface is 1.7 g/MJ (3.4 g/MJ
PAR), though the highest reported rates are only 50-60%
of this potential (Loomis and Connor 1992). In general,
radiation use efficiency ("E", or g biomass produced per
MJ intercepted) averages 0.5-1.0 g/MJ for C3 crop species
and 1.0-1.5 g/MJ for C4 species, depending on the
photosynthetic capacity of the canopy and the
biochemical composition of the biomass produced
(Loomis and Connor 1992). In general, E will decrease
with reductions in leaf N and will normally be higher
during the vegetative phase than during grain filling.
Therefore, E for well managed, high-N maize will be
around 1.2 g/MJ, but for stressed, N-deficient maize
(typical of low input, marginal agriculture), this value can
fall below 0.5 g/MJ.

For temperate maize, E values of 1.2-1.5 g/MJ are
common for well-managed crops with high N levels
(Muchow et al. 1990; Loomis and Connor 1992).
However, temperate environments are characterized by
more radiation per unit thermal time than tropical
environments, which are typified by warm temperatures,
rapid phenological development, short crop duration, and
higher respiration rates (Fischer and Palmer 1984).

Tropical maize hybrids have E values of around 1.0-1.2 g/
MJ under well-watered and well-fertilized conditions and
about 0.4-0.6 g/MJ under drought or low N conditions
(Muchow and Davis 1988; Muchow 1989a,b). The same
authors reported a common linear relationship between E
and specific leaf N for maize and sorghum under varying
levels of N, after correcting for differences in specific
leaf area between the crops (Muchow and Davis 1988).

Bolafios and Edmeades (1993) reported E values of 0.8
and 0.4 g/MJ for the tropical maize population Tuxpefio
Sequia under well-watered and droughted conditions,
respectively. However, trials were conducted on a site
with known iron deficiencies and during the off, winter
season. In an unpublished study comprising 10 CIMMYT
tropical maize populations, synthetics, and varieties under
well-managed conditions, E averaged 1.0 g/MJ with low
variability among entries (H.R. Lafitte, personal
communication, 1995). In another study with 100 tropical
and temperate S, lines, E averaged 1.1 g/MJ and ranged
from 0.8 to 1.5 g/MJ (Chapman and Edmeades 1996).

A Simple Model of Maize Productivity

Grain Production and Harvest Index

Grain production depends on the partitioning of
assimilates to the harvestable portion. Under unstressed
conditions, HI values from 50 to 55%, 40 to 45%, and 30
to 35% are reasonable for temperate hybrids, improved
tropical, and unimproved maize cultivars, respectively
(Fischer and Palmer 1984). Stresses during flowering and/
or grain-filling can reduce HI.

Maize is unique among cereals in that male and female
inflorescences are on separate parts of the plant and begin
and end development at different times, with a relatively
limited overlap. Consequently, in the two-week period
either side of flowering, maize is especially sensitive to
stresses such as drought, increased plant density, reduced
leaf area caused by N stress, and long periods of shady
weather. Under such conditions, HI can fall below 10%
(Fischer and Palmer 1984). During grain filling, yield
reductions and reduced grain weight occur largely because
of reduced photosynthetic rates and accelerated foliar
senescence caused by drought or low N.

Simple model of maize productivity
In summary, grain yield (Y) can be estimated as:
Y tx Rs x %RI x ExHI

t =

crop duration in days, which varies with genotype,
photoperiod, and temperature. Typical values are 80
d for extra-early, 100 d for intermediate, and 120 d

for late-maturing tropical germplasm under
lowland conditions.
Rs = average daily solar radiation (MJ/m2/d), which can
vary from 10 to 20 MJ/m2/d due mainly to changes
in cloudiness and day of year for any given latitude.
%RI = the percentage of seasonal radiation intercepted by
the canopy (area under the curve of percent
radiation interception from emergence to harvest),
ranging from 60% (upper limit, no stress) to 20%
(lower limit, stress).
E = the average seasonal radiation use efficiency (g/
MJ), which can range from 1.2 g/MJ (high N status,
healthy foliage) to 0.5 g/MJ (low N, stressed).
HI = Harvest index, ranging from 50-55% for temperate
hybrids, 40-45% for improved tropical cultivars,
30-35% for unimproved landraces, to 0-5% due to
stress and barrenness.

Using moderate values for these parameters based on
the information above, grain yield can be estimated for
cases A, B, and C (Fig. 1). Crop A intercepted 54% of
available radiation (120 d x 20 MJ/m2d x 0.54 = 1,296
MJ/m2). With an E value of 1.0 g/MJ (high N), the
model predicts a total biomass of approximately 13.0 t/
ha or a grain yield of 5.8 t/ha, assuming an HI of 0.45.
This agrees with the yield potential of many tropical
cultivars achieved under good agronomic management.
For a cultivar with improved stay-green (i.e., 4% more
interception), an extra 1.0 t/ha of biomass or 0.4 t/ha of
grain would be produced.

Crop B intercepted 32% of all incoming radiation
(2,400 x 0.32 = 768 MJ/m2). With an E value of 0.8 g/
MJ (slightly lower than crop A because of some stress),
the model predicts 6.1 t/ha of biomass and 2.5 t/ha of
grain, assuming an HI of 0.40 (slightly lower than crop
A due to stress).

Crop C intercepted only 19% of total radiation (2,400 x
0.19 = 456 MJ/m2). With an E value of 0.6 g/MJ (even
lower than crop B), the model predicts a total biomass
of 2.7 t/ha and grain yield of 1.0 t/ha, assuming an HI
of 0.35 (slightly lower than crop B).

A crop surface with 100% radiation interception, an Rs
of 20 MJ/m2 d, and an E of 1.0 g/MJ would produce
200 kg biomass/ha/d. Simple estimates such as those
above are surprisingly effective in predicting actual
grain yield for many maize production systems.

Evapotranspiration: The Water Cost
of Productivity

Canopies use water when they intercept radiation. It
takes the equivalent energy of 2.4 MJ/m2 of radiation to
evaporate 1 mm of water from the surface (heat of
vaporization of 1 g of water at 200C is 683 calories)
(Gates 1980). As long as the surface is wet or acts as a
wet surface (e.g., green foliage, vegetation, grasslands),
over 90% of incident Rs will be dissipated as
evapotranspiration (ET) and very little through sensible
heat. The opposite occurs if the surface is dry, whereby
incident radiation will dissipate as sensible heat
through increases in temperature (e.g., compare the
temperature of dry and wet beach sand at midday).

Directions in Modeling Wheat and Maize for Developing Countries

Therefore, the proportion of the surface acting as a wet
surface can be used to estimate the fraction of incoming
Rs that will be dissipated as ET. This is the underlying
basis for crop coefficients (Kc) used to calculate crop
ET from potential ET (Dorenboos and Pruitt 1984). The
proportion of the surface not acting as a wet surface will
not dissipate radiation as ET but as sensible heat. In the
cases of advection, when the evapotranspiring surface is
surrounded by extensive dry areas, ET can exceed Rs by
20-40% (Gates 1980).

Using these guidelines, together with reasonable
assumptions, one can quite easily estimate crop water
requirements for different environments. In the tropics,
potential ET normally ranges from 3-4 mm/d under
cloudy conditions and 5-6 mm/d under summer, tropical
conditions, to 7-8 mm/d under hot, arid conditions
(Gates 1980). Average ET ranges from 100 to 160 mm
per month, in most tropical environments.

The Nutrient Cost of Productivity

Maize needs to absorb nutrients from the soil to support
productivity, as described above. Typically, N is the most
limiting element in agroecosystems (Loomis and
Connor 1992). Young, recently-expanded maize foliage
has around 3% N concentration on a dry matter basis.
With a specific leaf weight of 6 mg/cm, each m2 of
foliage requires 18 g ofN. Therefore, the foliage needed
for complete radiation interception (LAIs of 4-5) per
hectare has a cost of 70-90 kg N/ha.

Nitrogen concentration decreases with crop age. Maize
seedlings can have 5% N, whereas at around flowering,
a healthy, well-fertilized maize canopy can have
approximately 2.0-2.5% N (Loomis and Connor 1992).
Maize grain has around 1.5% N (10-11% protein).
Faced with N limitations, the maize plant responds by
making less total grain, rather than by varying the N
concentration (Lemcoff and Loomis 1986). In other
words, N content in maize grain has only a small range
of variation, roughly from 1.2 to 1.6 % N (protein = N x

Maize stover can have around 0.8-1.2% N, depending on
crop history and conditions. Therefore, a grain yield of
6.0 t/ha (1.5% N for 90 kg N/ha) and 7.0 t/ha of stover
(HI=46%; 1.0% N for 70 kg N/ha) requires 160 kg N/ha.
If this amount of N is not available, then the productivity
mentioned above will not be sustained.


Bolafios, J., and G.O. Edmeades. 1993. Eight cycles of
selection for drought tolerance in lowland tropical
maize. I. Responses in grain yield, biomass and
radiation utilization. Field Crops Research 31:233-252.
Chapman, S.C., and G.O. Edmeades. 1996. Differences in
radiation use efficiency among lines in a tropical maize
population. Paper presented at 8th Australian Agronomy
Conference, University of South Queensland
Toowoomba, Australia, 30 Jan-2 Feb, 1996.
Doorenbos, J., and W.O. Pruitt. 1984. Crop water
requirements. FAO Irrigation and Drainage Paper No.
24. Rome, Italy: FAO.
Gates, D. 1980. Biophysical Ecology. New York, New
York: Verlag.
Fischer, K.S., and A.FE. Palmer. 1984. Tropical maize. In
PR. Goldsworthy and N.M. Fischer (eds.), The
Physiology of Tropical Field Crops, 213-248. New York:
John Wiley & Sons.
Hsiao, T.C. 1982. The soil-plant-atmosphere continuum in
relation to drought and crop production. In Drought
Resistance in Crops with Emphasis on Rice, 39-52. Los
Bafios, Philippines: IRRI.
Lemcoff, J.H., and R.S. Loomis. 1986. Nitrogen
influences on yield determination on maize. Crop
Science 26:1017-1022.
Loomis, R.S., and D.J. Connor. 1992. Crop Ecology:
Productivity and Management in Agricultural Systems.
Cambridge, U.K.: Cambridge University Press.
Monteith, J. 1990. Steps in crop climatology. Conference
Paper No. 477. Patancheru, India: ICRISAT.
Muchow, R.C. 1989a. Comparative productivity of maize,
sorghum and pearl millet in a semi-arid tropical
environment. I. Yield potential. Field Crops Research
Muchow, R.C. 1989b. Comparative productivity of maize,
sorghum and pearl millet in a semi-arid tropical
environment. II. Effects of water deficits. Field Crops
Research 20:207-219.
Muchow, R.C., and R. Davis. 1988. Effect of nitrogen
supply on the comparative productivity of maize and
sorghum in a semi-arid tropical environment. II.
Radiation interception and biomass accumulation. Field
Crops Research 18:17-30.
Muchow, R.C., TR. Sinclair, and J.M. Bennett. 1990.
Temperature and solar radiation effects on potential
maize yield across locations. Agronomy Journal 82:338-

Module Structure in CROPGRO v4

Module Structure in CROPGRO v4.0

C.H. Porter', J. W. Jones', and P Wilkens2
'Agricultural and Biological Engineering Department, University of Florida, USA
2International Fertilizer Development Center, Mussel Shoals, Alabama, USA

As crop simulation models become more complex, a modular modeling approach facilitates
collaboration among researchers and improves capability of models. A modular approach
has been adopted for the CROPGRO v4.0 model. Modules, which are separated along
disciplinary lines, can be independently developed, tested, and "plugged into" the model
with minimal or no modification of other modules.


Module Definition and Structure

As new components are added to crop growth models to
expand their capabilities, the models become
increasingly complex. This generates the need for a
modular structure for the crop models so that new
components can be added, modified, and maintained
with minimal effort. A modular approach facilitates the
ability to integrate knowledge from different disciplines,
thereby improving the prediction capability of the

A modular approach, based on methods used in the
Fortran Simulation Environment (FSE) software, has
been implemented for the CROPGRO model
(Kraalingen 1995). The model structure includes
modules or groups of linked subroutines that represent
separate disciplinary functions within the model. This
modularization allows greater flexibility in future
updates to the model; modules can be added, modified,
or replaced with little impact to the main program or
other modules.

The CROPGRO model has recently undergone
restructuring to the modular format described herein.
Modules have been developed for phenology, soil water
balance, pest damage, plant growth and partitioning,
photosynthesis, and soil nitrogen functions. A general
description of this modular approach is available at the
website for the International Consortium for
Agricultural Systems Applications (ICASA; http://

Acock and Reynolds (1989) proposed criteria for a
generic modular structure for crop models. Three of their
criteria are:
1. Modules should separate easily along disciplinary lines.
2. Modules should have a minimum number of input and
output variables.
3. Modifying one module should not necessitate changing

The following guidelines, based on the approach of
Kraalingen (1995) and adapted by Kenig and Jones
(1997), are proposed for the construction of modules.
Each module should:
1. Read its own parameters
2. Initialize its own variables
3. Accept variables passed to it from other modules and
the environment
4. Pass variables that are computed within the module
5. Own its set of state variables
6. Compute rates of change for its state variables
7. Integrate its state variables
8. Write its own variables as output
9. Operate when linked to a dummy test program

Thus, all data input, initialization of variables, rate
calculations, integration calculations, and output of data
related to a specific function are handled within a single
module. Modules should run as stand-alone models when
linked to an appropriate driver program.

CIMMYT. Published in White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing Countries: Proceedings ofa
Workshop, CIMMYT, El Batan, Mexico, 4-6 May 1998. NRG GIS Series 01-02. Mexico, D.F.: CIMMYT.

Directions in Modeling Wheat and Maize for Developing Countries

Figure 1 illustrates the modular format used in the
CROPGRO model, in which each module has the
following six components:

1. Run initialization
2. Seasonal initialization
3. Rate calculations
4. Integration
5. Output
6. Final

The main program (CROPGRO.FOR) contains six calls to
each module to accomplish each component of
processing. Control of processing within the program is
regulated with the DYNAMIC variable. Each module is
called once at the beginning of simulation with
DYNAMIC set equal to RUNINIT, resulting in execution
of the run initialization portion of the module. The
seasonal initialization (DYNAMIC=SEASINIT) is used
for initialization of variables at the beginning of each
season of a multi-season simulation. During the daily time
loop, each module is called three times: once each for rate
calculation (DYNAMIC=RATE), integration calculations
(DYNAMIC=INTEGR), and daily output
(DYNAMIC=OUTPUT). A final call to each module is
made to close input and output files (DYNAMIC=FINAL)
after all seasonal simulations are complete. Submodules
may be called, as needed, from modules to perform similar
processing components.

Module #1
Run Initialization Input Files
Seasonal Initialization
Rate Calculations
Output Output Files

Module #2
Run Initialization Input Files
U Seasonal Initialization
Rate Calculations
Output Output Files

Figure 1. Structure of main program and modules.




0 .

Module Structure in CROPGRO v4

The FORTRAN code used for directing calls to a
module from the main program is presented in Figure 2.
Figure 3 lists typical codes used to control processing
within a module.

Run initialization (RUNINIT)
At the beginning of each simulation, modules are called
to input data from files and to initialize variables prior
to daily simulation. During this phase of processing,
each module reads input data from the CROPGRO input
data files (e.g., IBSNAT35.INP, SBGRO980.SPE,
SBGRO980.ECO, SBGRO980.SBT, etc.). Some of the

variables that are read as input from the modules (such
as simulation switches and soil characteristics) could
have been passed to the module from the main program
as arguments but, instead, are read directly from the
input files and treated as local variables. This eliminates
the need for COMMON blocks, while reducing the
number of arguments passed from the calling routine.
This section also performs initialization or computation
of variables that need to be set only once per simulation.
Submodules are called to perform initialization and
input calculations as required.


CALL MODULE1(argl, arg2, .. ., RUNINIT)
CALL MODULE2(argl, arg2, .. ., RUNINIT)

Begin Seasonal Loop
CALL MODULE1(argl, arg2,...., SEASINIT)
CALL MODULE2(argl, arg2,...., SEASINIT)

Begin Daily Loop
Rate Calculation Section
CALL MODULE1(argl, arg2,.... RATE)
CALL MODULE2(argl, arg2,...., RATE)

Integration Section
CALL MODULE1(argl, arg2,.... INTEGR)
CALL MODULE2(argl, arg2,...., INTEGR)

Output Section
CALL MODULE1(argl, arg2,.... OUTPUT)
CALL MODULE2(argl, arg2,...., OUTPUT)

End Daily Loop
End Seasonal Simulation Loop
CALL MODULE1(argl, arg2,.... ,FINAL)
CALL MODULE2(argl, arg2,...., FINAL)

End of Program

Figure 2. Fortran code showing module processing within main program.

Directions in Modeling Wheat and Maize for Developing Countries

Seasonal initialization (SEASINIT)
When multi-seasonal simulations are performed, each
season of simulation must be initialized independently.
This is done in the seasonal simulation section.

Rate calculations (RATE)
Rate calculations are updated at the beginning of the
daily time loop. This ensures that rates of change of
state variables for a given day of simulation are all based
on values of these state variables for a common point in
time (e.g., the end of the previous day). Submodules are
called to compute rate calculations as needed.

Integration calculations (INTEGR)
The integration portion of the model updates state
variables throughout the model for each day of
simulation using the rates calculated previously.

Daily output (OUTPUT)
Daily output data are written to files in this section.

Final section (FINAL)
The FINAL section of processing is used to close all
input and output files and to write simulation

SUBROUTINE MODULEl(argl, arg2, arg3, ..., DYNAMIC)

Run Initialization Section


Seasonal Initialization Section

Rate Calculation Section

Integration Section
< Update state variables>

Output Section
< Write daily output>

Final Section

End of Module


Figure 3. Fortran code showing module structure.

Module Structure in CROPGRO v4

Programming Guidelines for Modules

A general list of guidelines used in programming the
modules was developed for the creation of modules for
the CROPGRO model:
* Eliminate GO TO statements, which make it difficult
to follow code sequence. GO TO statements can
usually be replaced with IF-THEN-ELSE or DO-loop
* Eliminate COMMON blocks. These can be replaced
with argument lists, which explicitly call out the flow
of data to and from modules and subroutines. In
future versions of FORTRAN, the COMMON blocks
will be eliminated and are considered to be an
obsolescent feature of the language.
* Label input and output lists for each module and
subroutine. This labeling has been done in the
CROPGRO modules by grouping variables in the
argument lists by input, input/output, or output
function for the module or subroutine.
* Describe each module or subroutine as it is called.
* Include a list of variable definitions in each module or
* Read input variables from files rather than pass to
modules in the argument list. This reduces the number
of arguments passed to and from modules.

Modifications to Main Program

As modules are added to the CROPGRO model, the
main program is modified, as necessary, to perform the
appropriate calls to each module. With the addition of
modules, the main program is used less for reading
input data and initializing variables, thus many of the
subroutines that were previously called from the main
program are eliminated. For example, rather than
initializing and reading phenology variables with calls to
subroutines such as IPECO, INPHEN, IPIBS, and
IPCROP, the main program calls the PLANT subroutine,
with variable DYNAMIC=RUNINIT. In turn, the
PLANT subroutine calls subroutine PHENOL, which
performs the initialization and data input functions for
the phenology module. The PLANT subroutine is called
from the main program for each of the six phases of
processing corresponding to the values of the
DYNAMIC variable.

Simulation Run Times

Preliminary bench tests comparing CROPGRO v4.0,
with three modules, to the previous CROPGRO v3.5
indicate that run times are not increased and may
actually be slightly decreased by the use of the
modular format. Kenig (1998, unpublished) reports
that simulation run-times using the modular
TOMGRO v3.0 model are significantly reduced
when compared with run times produced by the
nonmodular format code.


Acock, B., and J.F Reynolds. 1989. The rationale for
adopting a modular generic structure for crop
simulators. Acta Horticulturae 248:391-396.
Kenig, A., and J.W. Jones. 1997. Model structure for
dynamic crop-greenhouse simulations. In I.
Seginer, J.W. Jones, P. Gutman, and C.E. Vallejos
(eds.), Optimal Environmental Control for
Indeterminate Greenhouse Crops. Final Report,
BARD Research Project IS-1995-91RC, Chap. 11-4.
Technion, Haifa, Israel: Agricultural Engineering
Kraalingen, D.W.G. van. 1995. The FSE system for
crop simulation, version 2.1. Quantitative
Approaches in Systems Analysis Report No. 1.
Wageningen: AB/DLO, PE.

Directions in Modeling Wheat and Maize for Developing Countries

A Methodology for Linking Spatially

Interpolated Climate Surfaces with Crop

Growth Simulation Models

S.N. Collis andJ.D. Corbett
Blacklands Research Center, Texas Agricultural Experiment Station, Temple, Texas, USA.

When linked to spatial data, a crop simulation model can characterize the adaptation of
germplasm or management practices, providing detailed information unavailable except through
expensive field trials. We used the Decision Support System for Agrotechnology Transfer
(DSSAT) CERES-Maize crop simulation model to synthesize the available geo-referenced soil
and climate databases. Spatially interpolated climate surfaces, a growing season model
generated from those surfaces, and spatial soils layers were used as inputs into the CERES-
Maize sequence simulation model to simulate 30 years of yield for two maize cultivars in East
Africa. We were unable to acquire robust genetic coefficients for all tropical maize adaptation
zones nor reliable soil profiles appropriate, but our limited results suggest this approach will
serve to delineate adaptation zones.


Accurate identification and characterization of
production zones and potential production zones are
vital to agricultural research. Historically,
environmental characterization of agricultural areas
has been the subject of many research efforts
(Koppen and Geiger 1936; Thornthwaite 1948; FAO
1981) that integrated available data and expert
opinion to provide powerful interpretations of the
resource base for agricultural development. One goal
of these typically continental efforts was to map and
delineate zones of relative biophysical homogeneity
designed to communicate information useful for
planning agricultural and other human activities.
These methods, however, remain locked into their
historical roots since they provided static zones of
adaptation that end users could not modify to meet
specific needs.

New opportunities exist to greatly improve the
characterization mechanisms. Geographic information
systems (GIS) and interpolated climate surfaces expand
the scope of information readily available to
agriculturists (Corbett 1996). GIS and climate surfaces
linked with crop simulation models can provide detailed
spatial information on many aspects of germplasm and
crop management, particularly in relation to natural
resource concerns. GIS has provided researchers with
powerful tools for overcoming the limitations of
traditional agroecological models, which relied on static
zones, analog reproduction technology, and fixed crop-
environment relationships (Jones and Thornton 1996;
Corbett and O'Brien 1997; Corbett et al. 1998). The new
generation of more dynamic tools sought mechanisms to
use GIS technology and interpolated spatial data to
allow users to select boundary or 'discriminating'
criteria, with the output uniquely reflecting users'
interests, and enabled the characterization of agricultural
areas to be enhanced significantly and quickly.

CIMMYT. Published in White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing Countries: Proceedings of a
Workshop, CIMMYT, El Batan, Mexico, 4-6 May 1998. NRG GIS Series 01-02. Mexico, D.F.: CIMMYT.

Linking i .' iit Interpolated Climate Surfaces with Models

Crop simulation models represent a relatively untapped
source of analytic power for studying the interactions of
germplasm and management practices with the
environment. These tools enable analysis that has
previously not been possible, even using the more
dynamic approaches of classical GIS. When linked to
spatial data, however, a crop simulation model can
characterize the spatial extent of the adaptation zone for
a specific germplasm or management practice, while
providing detailed information unavailable through any
other mechanism except a massive (and prohibitively
expensive) field experiment.

For disaster mitigation, the creation of germplasm
specific zones with accompanying risk assessment and
scenario information is vital. For planning future
agricultural investments, such information is essential to
successful mitigation in light of factors such as climate
change and population increases. An enormous
investment has gone into the development of crop
simulation models because field trials are expensive and
time consuming. Crop simulation models can reduce the
number of field trials required for a particular spatial
location or site by narrowing down potential scenarios.
Application of crop simulation models across spatially
continuous surfaces (by simplifying inputs) to narrow
down the potential spatial domain of crop varieties is
equally valid as a more efficient means to determine
target environments.

We demonstrate the use of crop simulation models for
regional analysis in areas where data is scarce. Data
availability in the appropriate format can be one of the
greatest constraints in applied research activities (Jones
and Thornton 1996). To use existing data and
technologies, the premise is made that by simplifying
the inputs into crop simulation models, thus allowing the
application of these models spatially over large
geographic areas, we can generate spatially explicit
information of sufficient accuracy for indicative analysis
at the regional scale.

We used the Decision Support System for
Agrotechnology Transfer (DSSAT) CERES-Maize crop
simulation model to synthesize the available geo-
referenced soil and climate databases. Spatially
interpolated climate surfaces, a growing season model
generated from those surfaces, and spatial soils layers
were used as inputs into the CERES-Maize sequence
simulation model to assess the relative performance of
two maize cultivars in East Africa (Eritrea, Ethiopia,
Somalia, Kenya, Tanzania, Uganda, Rwanda, and
Burundi). Yields were simulated for both cultivars for a
single repetition over a 30-year sequence.

DSSAT Data Requirements

Due to detailed representation of processes in DSSAT-
compatible models such as CERES-Maize, detailed
inputs are required to achieve meaningful results. For the
DSSAT suite of models, the minimum data sets required
for model validation are: 1) daily weather data for the
duration of the experiment, 2) soil profile descriptions, 3)
management options used, 4) experimental data, and 5)
coefficients to characterize cultivars. Running the models
over large regions, however, requires judicious
simplification of the input data to reduce the number of
simulation runs.

Spatially continuous geo-referenced weather and soil data
at the level of detail required by the DSSAT models are
not commonly available at the regional scale. Lack of
historical weather data is also a problem for single field
simulations, which has resulted in the development of
weather generators to fill gaps in data records. These
statistical models use stochastic techniques to generate
daily weather data from historical weather data and long
term monthly means. The DSSAT simulation models use
variations of the SIMMETEO (Geng et al. 1986) and
WGEN weather generators, which require a minimum
data set of monthly means for solar radiation, minimum
air temperature, maximum air temperature, precipitation,
and number of wet days (Hansen et al. 1994; Pickering
et al. 1994).1

1Jones and Thornton (1996) describe a proposal to develop DSSAT climate files and third-order Markov chain model parameters for South
America and Africa. The rainfall generator based on third-order Markov chain has been shown to better simulate year-to-year rainfall
variation in the tropics and requires 36 parameters calculated from historical records (Jones and Thornton 1993). The data is planned to be
released on CD-ROM and, once available, could be used relatively easily to replace current weather-generating techniques in the
application framework developed here.

Directions in Modeling Wheat and Maize for Developing Countries

Climate Surface Interpolation

Decades of effort to collect and collate historical
weather data from stations around the world have
culminated in the ability to generate continental
spatial climate surfaces (Jones and Thornton 1996).
Hutchinson (1991, 1995) developed a "Laplacian" or
thin-plate spline technique to interpolate climate
variable surfaces from long-term weather station
records. Using coefficients generated by Hutchinson
(Corbett 1996; Hutchinson and Corbett 1996), Corbett
and Kruska (1994) generated long-term monthly mean
minimum and maximum temperature, total
precipitation, and potential evapotranspiration grids at
a resolution of 3 arc minutes for the African continent.

Monthly climate surfaces for global radiation and
number of wet days, which are required for weather
generation, were interpolated for the East African
study area using ANUSPLIN and station data supplied
by the Food and Agricultural Organization of the
United Nations (FAO) database (FAO 1994).


In DSSAT, the soils file defines the soil profile
properties that are used in the soil water, nitrogen, and
root growth sections of the crop models (Jones et al.
1994). Generally the information is collated from a
combination of measurements in the field and from
soils databases. For spatial simulations at the regional
scale, digitized soil map classifications can be
associated with soil profile information, based on
available pedon data to provide the soil file data
(Thornton et al. 1996; Hoogenboom et al. 1993).

For the East African study area, spatial soils data were
provided by the World Soils Resources (WSR) group
of USDA. WSR used the 1:5,000,000 Soil Map of the
World in combination with interpolated climate
variables to convert the FAO classification system into
a USDA Taxonomy at the Great Group level of
classification. We used this classification and the soil
pedon database that accompanies DSSAT to select a
representative pedon for each soil classification. We
sought typicc" soil pedons and then selected the
representative pedon, based on completeness of the
soil pedon data within the DSSAT databases. Profiles
were generated using the DSSAT soil profile


Our methodology for linking the spatial data and simulation
models first involved reducing the gridded climate surfaces
to a more manageable number of climatic environments.
This was achieved by performing a cluster analysis on the
five climatic variable surfaces required for weather
generation plus evapotranspiration over a five-month
season defined by maximum precipitation to potential
evapotranspiration (P/PE) ratio (see the section "Growing
Season Model"). This resulted in what we call an "effective
environments" layer (Fig. 1). The number of climate
"scenarios" are thus reduced significantly whilst
maintaining the majority of the spatial variance. Means
were calculated for the five variables required for the
weather generator over each clustered region and were
output to DSSAT format climate files. The effective
environments layer was overlaid with the spatial soils layer
and the first month of the optimum season layer resulting in
a simulation layer (Fig. 1). An experiment file was
generated for each simulation zone from a template,
simulations run, and the simulated output variables mapped
back to the original zones.

Growing Season Model

To determine planting dates over the region, we used a five-
month growing season model to indicate the start month of
the planting window. This model avoids premature
automatic planting by the simulation model in bimodal
season regions.

The five-month optimum growing season was defined by
identifying the five consecutive months in which the mean
P/PE ratio is maximized. Water is a first-order limiting
factor for most of East Africa, and this simple model has
been found to be a reasonable identifier for the growing
season (Corbett et al. 1995). There are exceptions,
particularly in the Lake Region where the long rainy period
permits more flexibility in planting, but, even in those
locations, our model accurately identifies the principle first
month of the main maize planting season. This model did
not consider temperature information, although we
inspected climate graphs from a sample of sites to ensure
that our criteria selected the proper season for crop

The five-month optimum growing season was also used as
the temporal delimiter for the spatial cluster analysis
variables input for the effective environments definition.

Linking i ... tt, Interpolated Climate Surfaces with Models

Effective Environments

Ward's minimum variance algorithm was used to cluster
the climate data (SAS 1990), thus reducing the number
of climate scenarios. This was done since climate does
not necessarily vary significantly between the
interpolated cells, and our objective of discriminating
maize environments did not require such high resolution
(approximately 29 km2 cells). A second motive was
computational efficiency, with 76,000 cells and 30 years
of sequential maize growth simulation effectively
exceeding our computation capacity.

Five grids representing the five month sequence of the
optimum season for each of the long-term monthly
mean variables were generated for precipitation,
potential evapotranspiration, solar radiation, number of
rainy days, maximum temperature, and minimum
temperature. These grids reflect not the calendar month
(e.g., January, February, etc) but rather the "biological"
sequence of the growing season (e.g., month 1
precipitation, month 2 precipitation, etc.) for the five
months. The 30 grids were then ported to statistical
analysis software (SAS) for cluster analysis. Plots of R2
against the number of clusters indicated that
approximately 200 clusters would be sufficient to
represent the majority of the variance of the East
African data set.

Simulation Layer

The simulation layer was generated by simply
overlaying the effective environments, first month of
optimum season, and soil great groups layers. This layer
represents the zones of unique growing season, climate,
and soil characteristics, and can be used as the basis for
any subsequent simulation scenario or experiment file
configuration. For the East African region there were
1,212 unique combinations.

Genetic Coefficients

Two calibrated (but not verified) genetic coefficients,
MH-16 and Katumani, were available for maize in the
East African region at the time of this study. MH-16 is a
hybrid from Malawi and is bred from SR-52 which,
despite dating to the 1950s, is still one of the better
regional cultivars. This is a fairly long-season variety,
with shorter stature, and is capable of good yields.
Katumani is a composite that has not performed as well
as MH-16 in trials. The genetic coefficients for these
cultivars are preliminary but were considered the best
option for testing our methodology. These two cultivars
and others are currently being calibrated for East Africa.
The coefficients assumed for MH-16 and Katumani are
given in Table 1.

Simulation Runs

A simple maize-fallow crop rotation with the following
planting details, which are considered to be standard for
East Africa, were used (P. Thornton, personal
communication, 1997):

Plant population (PPOD + PPOE):
Row spacing (PLRS):
Planting depth (PLDP):

3.9 plants/m2
75 cm
5 cm

Default initial conditions with default amounts of nitrate
in the soil profile were used. No residues, fertilizers,
chemicals, or tillage were included in the model runs
(Phil Thornton, personal communication, 1997). The
planting window is set individually for each run using
simulation controls and is determined by the first month
of the optimum season surface. The window is then
"open" for three months in which the crop will be
planted if planting conditions are met. The default
planting condition requirements were used.

Table 1. Genetic coefficients assumed for two leading cultivars from eastern and southern Africa, for use
in a simulation using the CERES-Maize model.
Variety # Cultivar Ecotype P1 P2 P5 G2 G3 Phint
CM0001 KATUMANI SA0001 172.0 0.50 999.0 398.0 6.27 75.00
CM0005 MH-16 SA0001 245.3 0.28 843.0 417.3 7.87 75.00

Directions in Modeling Wheat and Maize for Developing Countries

Harvest date for the maize crop was set to occur
automatically at maturity. Harvest for the fallow period
was set to occur one day prior to the maize planting date.
To limit the number of unnecessary simulation runs, we
excluded zones where the five-month optimum season
rainfall was less than 300 mm. Some simulation zones
were also excluded due to lack of soil profile data.

An ArcView application interface was developed to
integrate the various data sets, run the crop simulations,
and map output. Much of our initial efforts to integrate
DSSAT and ArcView were based on the AEGIS/WIN
application developed by Engel et al. (1995) and adapted
for gridded surfaces.

When using sequence simulation, it is highly
recommended to carry out at least 10 replicates of each
sequence to obtain relatively stable estimates of means
and variances. A replicate is the repetition of the same
experiment run with a different sequence of weather
conditions (Thornton et al. 1994). Due to the preliminary
nature of the genetic coefficients, however, initial runs
were carried out for 30 years and 1 replicate. As such, the
output results for each year were not analyzed
individually, as would be possible with 10 repetitions, but
were meaned over the 30-year period. The total means
were then mapped to the original simulation zones.


The simplest comparison between the two maize cultivars
was to overlay the 30-year mean yield at harvest maps to
indicate where each variety performs best (Fig. 2).
Overall, the results suggest higher mean yields from cv.
Katumani (dark gray regions). The results correctly
identify Katumani as the preferred variety for both the
Machakos and Kitui districts of eastern Kenya. Katumani
was specifically developed for this area, and its center of
origin is the Katumani research station, just south of
Machakos town.

Beyond the literal translation of the simulated yields, our
method allows the systematic description of an area for
its potential yield with respect to specific cultivars. For
this study, we were unable to acquire robust genetic
coefficients that would represent the spectrum of broad
tropical maize adaptation zones, as described by
CIMMYT mega-environments (highland transitional,
midaltitude, dryland, and coastal or lowland). Nor were

we able to obtain reliable soil profiles appropriate for the
CERES-Maize model. However, with more reliable data,
simulation of a representative variety of each of the
aforementioned mega-environments could be carried out.
A map could then be created that outlines the mega-
environment or adaptation zone by virtue of initially the
highest simulated mean yield. Our more limited results
are encouraging: this approach, given a representative set
of genetic coefficients, will work to create a map which
will delineate adaptation zones.

Highest mean yield is not the only criterion that can be
used to delineate zones. Given the power of a simulation
environment, it will be possible to build a database of
simulation results so that the highest mean yield can be
identified, as well as the variability in mean yield.
Nitrogen uptake maps could also be used as N application
indicators. Adaptation zone delineation would just begin
with highest mean yield. Beyond that, zones focused on
risk assessment could be created. For example, we could
further evaluate the midaltitude adaptation zone as
follows: calculate the mean yield during the driest 25% of
years and compare that yield to the mean yield of a
dryland variety. Those areas of the midaltitude zone in
which a dryland variety attained a higher yield than the
"correct" variety might be targeted for further
socioeconomic analysis. Farmers with little cash or the
most risk averse might elect to grow the variety that
yields more in the driest of years, rather than attempt to
attain a higher mean over the long run. This kind of
analysis helps to target research on a different aspects of
the issues surrounding germplasm adoption: risk and
resource access.


Actual meteorological data exists in sufficient detail for
some locations that a risk assessment analysis could use
either simulated or actual weather data. Risk assessment
offers a valuable addition to the characteristics of any
germplasm adaptation zone. Whether simulated or based
on actual meteorological records, the ability to estimate
variability over space and in time of the yield of a crop
variety is a potentially powerful decision-making asset to
both agricultural research and agricultural and economic
development efforts. At a minimum, this methodology
describes an opportunity to improve our characterization
and assessment capabilities using spatial data and crop
simulation models.

Linking i ,iit Interpolated Climate Surfaces with Models


Corbett, J.D., and R.L. Kruska. 1994. Africa monthly climate
surfaces, v. 0. Three arc-min resolution. Based on climate
coefficients from CRES, Canberra, Australia. Data for
mean long term normal minimum temperature, maximum
temperature, and precipitation. Nairobi: ICRAF/ILRAD.
CD-ROM publication.
Corbett, J.D., R.F O'Brien, R.J. Kruska, and E.I. Muchugu.
1995. Agricultural Environments of the Greater Horn of
Africa A Database and Map Set for Disaster Mitigation. 9
pp. text, 31 maps, and database.
Corbett, J.D. 1996. Dynamic crop environment classification
using interpolated climate surfaces. In M.F Goodchild,
L.T. Steyaert, B.O. Parks, M. Crane, C. Johnston, D.
Maidment, and S. Glendinning (eds.), GIS and
Environmental Modeling: Progress and Research Issues,
1995. Published for the Second International Conference/
Workshop Integrating GIS and Environmental Modeling,
Breckenridge, Colorado, 27-30 September, 1993.
Corbett, J.D., and RF. O'Brien. 1997. The Spatial
Characterization Tool Africa v 1.0. Texas Agricultural
Experiment Station, TexasA&M University, Blackland
Research Center Report No. 97-03. Documentation and
CD-ROM. Temple, TX: Texas A&M University Press.
Corbett, J.D., S.N. Collis, and R.F O'Brien. 1998. Almanac
Characterization Tool forAngola, Sierra Leone, and
Liberia. A Resource Base for ( I,/ ..'.. '.:, ,: the
Agricultural and Natural Environments Including a Digital
Library. Texas Agricultural Experiment Station, Texas A&M
University, BlacklandResearch Center Report No. 98-02.
USAID OFDA. Documentation and CD-ROM. Temple,
TX: Texas A&M University Press.
Engel, T., J.W Jones, and G. Hoogenboom. 1995. AEGIS/
WIN A windows interface combining GIS and crop
simulation models. Presented at 1995 Annual Winter
Meeting sponsored by the American Society of
Agricultural Engineers, Chicago, Illinois.
Food and Agricultural Organization of the United Nations
(FAO). 1981 \ i,. l,. ,. -1. andResults for South and
CentralAmerica, Vol 3. Report on the Agro Ecological
Zones Project. Rome, Italy: FAO.
Food and Agricultural Organization of the United Nations
(FAO). 1994. CLIMWATfor CROP WAT A Climatic
Database for Irrigation Planning and Management. Rome,
Italy: FAO.
Geng, S., FWT. Penning De Vries, and I. Supit. 1986. A
simple method for generating daily rainfall data.
Agricultural and Forest Meteorology 36:363-376.
Hansen, J.W, N.B. Pickering, J.W Jones, C. Wells, H. Chan,
and D.C. Godwin. 1994. Weatherman. In G.Y Tsuji, G.
Uehara, and S. Balas (eds.), DSSAT Version 3 Manual,
Volume 3. Honolulu, Hawaii: University of Hawaii.

Hoogenboom, G., H. Lal, and D.D. Gresham. 1993. Spatial
yield prediction. Presented at 1993 International Winter
Meeting sponsored by the American Society of
Agricultural Engineers, Chicago, Illinois.
Hutchinson, M.F 1991. The application of thin plate
smoothing splines to continent-wide data assimilation. In
J.D. Jasper (ed.), DataAssimilation Systems, BMRC
Research Report No. 27, 104-113. Melbourne, Australia:
Bureau (ci kf I,, ,1 ._4,.
Hutchinson, M.F 1995. Interpolating mean rainfall using
thin plate smoothing splines. International Journal ofGIS
Hutchinson, M.F, and J.D. Corbett. 1996. Spatial
interpolation of climate data using thin plate smoothing
splines. Paper prepared for the FAO Expert Consultation
on the Coordination and Harmonization of Databases and
Software for Agroclimatic Applications, Rome, 29
November-3 December, 1993. Rome, Italy: FAO.
Jones, J.W, L.A. Hunt, G. Hoogenboom, D.C. Godwin, U.
Singh, G.Y. Tsuji, N.B. Pickering, PK. Thornton, WT.
Bowen, K.J. Boote, and J.T. Ritchie. 1994. Inputs and
outputs. In G.Y. Tsuji, G. Uehara, and S. Balas (eds.),
DSSAT Version 3 Manual, Volume 2. Honolulu, Hawaii:
University of Hawaii.
Jones, PG., and PK. Thornton. 1993. A rainfall generator
for agricultural applications in the tropics. Agricultural
and Forest Meteorology 63:1-19.
Jones, PG., and PK. Thornton. 1996. Continental-Scale
Climate Databases forAgriculturalApplications. 1996-
1997 work plan submitted to the Rockefeller Foundation.
Koppen, W, and R. Geiger. 1936. Handbuch de
Klimatologie. Berlin: Gebr. Borntraeger.
Pickering, N.B., J.W Hansen, J.W Jones, C.M. Wells, VK.
Chan, and D.C. Godwin. 1994. Weatherman: a utility for
managing and generating daily weather data. Agronomy
Journal 86:332-337.
SAS Institute. 1990. SAS/STAT User Guide. Version 6,
Fourth Edition, Volume 1. Cary, NC: SAS Institute.
Thornthwaite, C.W 1948. An approach to the rational
classification of climate. Geographical Review 38:55-94.
Thornton, PK., PW Wilkens, G. Hoogenboom, and J.W
Jones. 1994. Sequence analysis. In G.Y. Tsuji, G. Uehara,
and S. Balas (eds.), DSSAT Version 3 Manual, Volume 2.
Honolulu, Hawaii: University of Hawaii.
Thornton, PK., WT. Bowen, A.C. Ravelo, PW Wilkens, G.
Farmer, J. Brock, and J.E. Brink. 1996. Estimating millet
production for famine early warning: an application of
crop simulation modeling using satellite and ground-
based data in Burkina Faso. Draft paper submitted for

Directions in Modeling Wheat and Maize for Developing Countries

Long term monthly climate
Potential Evapotranspiration
Solar Radiation
Number of Rainy Days
Maximum Temperature
Minimum Temperature

Wards minimum variance
clustering algorithm
(resulting in 200 zones).

Genetic Coefficients

Management Options

DSSAT Sequence Analysis
+ Simulation

Yield Layer:
Showing relative
performance of crop
varieties (see Fig. 2)

Figure 1: Methodology for linking spatially interpolated climate surfaces with crop growth
simulation models in East Africa.

Simulation Layer

growing season:
5 month maximum
precipitation to potential
evapotranspiration ratio.

Soil Profiles:
"Typic" soil profiles
from the DSSAT
soils database.

First Month of Optimum
Growing Season

Great Groups

Effective Environments

Linking j .-. u.i Interpolated Climate Surfaces with Models



\ Ethiopia


Rwanda -. Indian Ocean


"-' I East Africa Katu mani > Mh 16 (kg/ha) 30yr 1 rep
Burundi r Talnzail 1-4472
'' I I NoData
." --. East Africa Mh16 > Katumani (kg/ha) 30yr 1 rep
I No Data

Figure 2. Highest 30 year mean yield at harvest of two maize varieties in East Africa, 1) Katumani (dark
gray), a dryland cultivar MH16 (light gray), a midaltitude cultivar.

Directions in Modeling Wheat and Maize for Developing Countries

Work Group Outputs

Compiled by J. W White
Natural Resources Group, CIMMYT, Mexico


Following the formal presentations, possible
modifications to models and other modeling-related
issues were reviewed to set priorities and resolve a few
issues relating to software management (Tables 1 and 2).
The discussions were reopened for brief periods
throughout the rest of the workshop.

DSSAT Version Control

The multiple pre-releases of DSSAT version 3.1 created
confusion over versions that different researchers were
using. Naming conventions were proposed to allow
upgrading of software on a consistent basis. The next
official version of DSSAT was to be version 3.5, which
was released in late 1998.

Test versions of models for development should be
based on version 3.5. However, they should be identified
as 3.6xx, where "xx" suffix identifies the developer
(e.g., "3.6PG" for Peter Grace). A facility needs to be
included in the code to allow input of this suffix.

Specific Modifications for CERES
Version 3.5

Variable phyllochron interval in CERES-Maize
In many situations, the number of maize leaves is
overestimated. Although the phyllochron interval
(PHINT) appeared in the cultivar file, the value was
actually "hard wired" at 75 degree-days. This has
subsequently been corrected.

Cultivar-specific parameters
The lists of cultivar-specific parameters in CERES (e.g.,
MZCER980.CUL) provide no indication of how reliable
the values are. Ideally, the number of observations, types

of data, and method used should all be indicated. Tony
Hunt strongly urged that the list of cultivars be
shortened to include only the most reliable coefficients.
Five generic cultivars should also be provided.

Externalization of nitrogen mineralization
parameters in to a SOIL.PAR file
To facilitate modeling of nitrogen mineralization in
CERES and CROPGRO, a new soil parameter file,
SOILN980.PAR, was created (Table 3). This file
externalizes many of the coefficients needed for
simulating the decomposition of soil organic matter (one
pool) and organic matter added as residue or manure
(three pools). If the file does not exist in the data
directory, it is created using default values upon the first
run of the model.

Definitions of the parameters are:

DMINR: Potential decomposition rate of SOM pool.
Default value is 0.8300E-04 per day
RTCNR: C/N ratio of initial root residue.
Default = 40.0.
DSNCV: Depth to which soil C (SCDD) and total N
(SNDD) values are integrated for output to
Default value is 20.0 cm.
RE001: First three values are the potential
decomposition of the carbohydrate, cellulose,
and lignin pools; the next three values are the
relative of carbohydrate, cellulose, and in the
residue or manure dry matter.
Default = 0.2000, 0.0500, 0.0095, 0.2000,
0.7000, 0.1000.
Up to nine different residue or manure types
can be defined.

Reducing the thickness of deeper soil layers
To handle tile drains, Bill Bachelor defined a soil layer
15cm-thick at the approximate depth of the drain.

CIMMYT. Published in White, J.W., and P.R. Grace (eds.). 2001. Directions in Modeling Wheat and Maize for Developing Countries: Proceedings ofa
Workshop, CIMMYT, El Batan, Mexico, 4-6 May 1998. NRG GIS Series 01-02. Mexico, D.F.: CIMMYT.

Work Group Output

Unfortunately, testing showed that introducing this layer
produced unexpected changes in model outputs.

Proposed changes for CERES v3.6x
A series of changes were suggested for subsequent
releases. These included:
* Improve thermal time calculations.
* Account for mass of dead leaves and their subsequent
incorporation into soil organic matter.
* Improve modeling of grain number in maize (based
on approach of Andre Du Toit).
* For wheat, Zadok stages should be output along with
standard phenology stages.
* Model effects of conservation tillage, tile drainage,
and runoff/erosion.
* Model response to soil phosphorus.
* Generate solar radiation if actual data is erroneous or
not available.
* Include new genetic coefficients for winterkill,
vernalization, and prolificacy.
* Improve the water balance routines both for root
uptake and estimates of potential evaporation.

Various software problems that should be rectified were
also noted in the version 3.5 or 3.6 releases:

* WINGRAF default scaling for water content and
harvest index.
* WINGRAF loss of plotted lines on large monitors.
* Sequencing memory problems; need to increase the
number of individual phases.
* A and T files to handle replications.
* Data conflicts in A and T files.

Documentation and Modularization
of Codes

It was recognized that documentation of models is still
very problematic. Many researchers assume that Jones
and Kiniry (1986) is still an accurate description of the
CERES models. Much more effort is needed on the
updating of documentation. Reports should be sent to
the Hawaii for posting on the list server. Links to
modeling sites such as Michigan State University and
University of Florida can also be included.

People identified to provide leadership and quality
checking of documentation were:
* CERES-Maize: Joe Ritchie
* CERES-Wheat: Tony Hunt
* CROPGRO: Gerrit Hoogenboom

It was noted that modular programming would
facilitate publication of algorithms. A draft Spanish
version of DSSAT 3 documentation is available
through CIP (Walter Bowen).

Further Discussion

Several additional topics were touched upon. The
need to assemble quality data sets collected through
the IBSNAT project and other sources, e.g., GCTE,
arose several times. It is easy to criticize models, but
the models are only as good as the data sets used to
develop routines and test model performance.

No progress is being made on incorporating
intercropping. Modularization might make it easier to
handle more than one crop in a single model, but an
approach similar to that used by APSIM, where the
soil is the central resource, would still seem necessary.

The ICASA file standards should be modified to
handle a wider range of treatments. One set of data
includes 25 treatments including GA3, boric aid, and
urea solutions.


Jones, C.A., and J. Kiniry. 1986. CERES-N Maize: A
Simulation Model of Maize Growth and
Development. Temple, TX: Texas A&M University

Directions in Modeling Wheat and Maize for Developing Countries

Table 1. Summary of activities during the model development and testing phase of the workshop.

Acti vity Responsible 1 Date completed/comments

DSSA T v3.5
Agreed that the next release of DSSAT and all All During the workshop,
components should be version 3.5 to remove confusion GH coordinated naming
over intermediate r eleases of v version 3.1. revisions f or all
Calculation of degree-days is now calculated on Tbase JR, GH, BB, PW 7 May 1998.
and Topt defined in the species file. This allowed Requires recalibration of
simplifying code among species. cultivars.

CERES v3.5
Make PHINT (phyllochron interval) a variable input GH, PW 7 May 1998.
for all crops. This fixed the over-prediction of leaf Requires recalibration of
number for CERES-Maize. cultivars.
Determine whether soil, air, or crown temperature is BB, JR, PW 7 May 1998.
used to control phenology in early stages of Requires recalibration of
development using a switch based on leaf number. cultivars.
For maize, sorghum, and millet, the switch is leaf number.
For wheat and barley, I-stage 1 is used.
Calculation of degree-days is now calculated on Tbase JR, GH, BB, PW 7 May 1998.
and Topt defined in the species file. This allowed Requires recalibration of
simplifying code among species. cultivars.
Modify and test algorithms for winterkill. PW, TH 7 May 1998.
Is killing off plants, but yield
effect is less than expected.
For lower soil layers, automatically divide the profile BB Not implemented. Found to
into 15-cm layers rather than 30-cm layers. affect yields as much as 500 kg/ha.
Externalization of N mineralization parameters in to WB, PW, PG 6 May 1998.
a SOLN980.PAR file.
Recalibrate cultivars based on the above changes. TH, JR, PW, GH 8 May 1998.
Shorten the list of cultivars to include reliable
coefficients only. Include five generic cultivars in list.
Maize list JR, BB
Wheat list TH, JW
Assemble final code and data sets. PW, GH 6 May 1998.

CERES v3.6xx
Add genetic coefficient for winter kill (independent JR
of vernalization) in wheat.
Add genetic coefficient for tolerance to high densities JR
(prolificacy) in maize.
Zadoks stages output along with standard phenology stages. TH
Solar radiation to be calculated if actual data is erroneous JR
or not available.

Work Group Output

Improved thermal time calculations. JR
Water balance modifications: root uptake and potential
Conservation tillage. PG, PW, .
Output and transfer of dead leaves. GH, DH
Improve handling of grain number based on linear JR, GE
relation between crop growth in critical pre-grain set
phase and grains/m2.
Tile drainage. BB

Software e and data management
Endorsed the idea of repeating these meetings on an
annual basis, with possible briefer, intermediate meetings
(e.g., in conjunction with the ASA meetings).

Version control and naming con ventions
Agreed that the next DSSAT release should be version 3.5. All 5 May 1998.
Agreed that modifications based on v3.5 models should All 5 May 1998.
be identified as 3.6xx, where the "xx" suffix is a
two-letter code to identify the investigator or project.

Programming issues
Agreed to evaluate the modular approach. It has clear All
advantages for model maintenance, revision and
improvement, and for documentation. However,
implications for run time need to be evaluated.
Agreed that models should move to a 32-bit operating All
system but try to maintain 16-bit functionality for the
next two to three years. Digital Fortran is the preferred
Endorsed the Michigan meeting's recommendation of All
a Windows-based user interface for model applications,
but more specifics are needed for standards on icons,
menu bars, screen layouts, etc.

The need for better model documentation was again cited.
The ICASA www site offers one possible access point.

1 BB Bill Batchelor; DH = Dewi Hartkamp; GE = Greg Edmeades; GH = Gerrit Hoogenboom; JR = Joe Ritchie; JW = JeffWhite; PG = Peter
Grace; TH = Tony Hunt; PW = Paul Wilkens.

Directions in Modeling Wheat and Maize for Developing Countries

Table 2. Summary of data sets prepared or partially prepared during the workshop.

Data set Responsible

CIMMYT trials at Tlaltizapan and Poza Rica, Mexico G. Edmeades, W. Bowen
CIMMYT historic wheat cultivar series T. Hunt
Punjab Agricultural University planting dates x cultivars x years J. White

Velvet bean (Mucuna) D. Hartkamp, G. Hoogenboom

Table 3. Example of the file SOILN980.PAR.


! Model parameter file which externalizes many of the coefficients needed for simulating the decomposition of
!soil organic matter (one pool) and organic matter added as residue or manure (three pools). If SOILN980.PAR
! does not exist in the data directory, it is created upon the first run of the model. Definitions follow:

! DMINR: Potential decomposition rate of SOM pool.
! Default value is .8300E-04 per day.
!RTCNR: C/N ratio of initial root residue.
Default = 40.0.
! DSNCV: Depth to which soil C (SCDD) and total N (SNDD)
values are integrated for output to CARBON.OUT.
Default value is 20.0 cm.
!RE001: First three values are the potential decomposition
rates of the carbohydrate, cellulose, and
lignin pools; next three values are the relative
proportions of carbohydrate, cellulose, and
lignin in the residue or manure dry matter.
Defaults = 0.2000, 0.0500, 0.0095, 0.2000, 0.7000, 0.1000.
!Up to nine different residue or manure types can be defined


DS DMINR 0.8300E-04
DS RE001 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE002 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE003 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE004 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE005 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE006 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE007 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE008 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000
DS RE009 0.2000 0.0500 0.0095 0.2000 0.7000 0.1000

Contact Information

Contact Information for Participants

William D. Bachelor
Assistant Professor
Agricultural and Biosystems Engineering
Iowa State University
Ames IA 50011
Phone: +1 (515) 294 9906
Fax: +1 (515) 294 2552
Email: bbatch@iastate.edu

Hector Barreto
Phone: +504 232-2502
Email: hbarreto@hondutel.hn

Bruno Basso
Dept. of Crop and Soil Sciences
Michigan State University
East Lansing MI 48824
Phone: +1 (517) 353 4705
Fax: +1 (517) 355 0270
Email: Basso@pibot.msu.edu

Marcus Bergman
Applied Statistician
Witteveen+Bos Consulting Engineers
P.O. Box 233
The Netherlands
Phone: +31 570 697911
Fax: +31 570 697344

Kirsten de Beurs
Postbus 2176
3800 CD Amersfoort
tel: 033 4637433
fax: 033 4637340
E-mail: postmaster@neo.nl

Ken Boote
Agronomy Dept.
University of Florida
Gainesville Fl 32611-0500
Phone: +1(352) 392 1811
Fax: +1 (352) 392 1840
Email: kjb@gnv.ifas. ufl.edu

Walter T. Bowen
P.O. Box 1558
12 Lima
Phone: +51 (1) 349 5783
Fax: +51 (1) 349 5638
Email: w.bowen@cgnet.com

Stewart Collis
Mud Springs Geographers, Inc.
18 S Main
Suite 718
Temple, TX 76501
Tel: (254) 778-5150
Fax: (254) 642-1158
Email: MudSprings@MudSprings.com

Gregory O. Edmeades
Research Fellow
Pioneer Hi-Bred International, Inc.
P.O. Box 609
Waimea HI 96796
Phone: +1 (808) 338-8300
Fax: +1 (808) 338-8325
Email: greg.edmeades@pioneer.com

Directions in Modeling Wheat and Maize for Developing Countries

Peter Grace
Manager, Natural Resource Management Systems
Sinclair Knight Merz
369 Ann Street
Brisbane QLD 4004
Telephone: +61 7 3244 7100
Facsimile: +61 7 3244 7300
Email: pgrace@skm.com.au

Agnes Dewi Hartkamp
Policy Officer Research
Product Organisation Grains, Seeds and Pulses
Stadhouderplantsoen 12
2517 JL The Hague
The Netherlands
Phone/fax: +31 30 2510726

Gerrit Hoogenboom
Department of Biological and Agricultural
University of Georgia
Georgia Experiment Station
Griffin GA 30223-1797

L. Anthony Hunt
Plant Agriculture Department
University of Guelph
Guelph ON N1G 2W1
Phone: +1 (519) 824 4120 Ext. 3595
Fax: +1 (519) 763 8933
Email: thunt@crop.uoguelph.ca

Joost G.F. Lieshout
Wageningen Information Services (WIS)
Oude Houtensepad 18
3582 CW Utrecht
The Netherlands
Phone/fax: +31 30 2510726
Email: j.lieshout@wisint.org

Matthew Reynolds
Wheat Program
Lisboa 27, Col, Juarez
Apdo. Postal 6-641
06600 Mexico, D.F.
Phone: +52 5804 2004
Fax: +52 5804 7558/9
Email: mreynolds@cgiar.org

Joe Ritchie
217 Camino Principal
Belton, Texas 76513
Phone (254) 933 7742
Email: ritchie@msu.edu

Serena Stornaiuolo
Dept. of Crop and Soil Sciences
Michigan State University
East Lansing MI 48824
Phone: +1 (517) 347 6684
+1 (517) 353 4705

Jeffrey White
GIS/Natural Resources Group
Lisboa 27, Col. Juarez
Apdo. Postal 6-641
06600 Mexico, D.F.
Phone: +52 5804 2004 Ext. 2207
Fax: +52 5804 7558/9
Email: jwhite@cgiar.org

Paul W. Wilkens
International Fertilizer Development Center
P.O. Box 2040
Muscle Shoals AL 35662
Phone: +1(256) 381 6600
Fax: +1 (256) 381 7408
Email: pwilkens@ifdc.org

University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs