Skip to main content

Crop Model

Posted in


Crop modeling: A tool for agricultural research

Rajeew Kumar, Sumit Chaturevdi 
Department of Agronomy, G.B. Pant University of Agriculture and Technology, Pantnagar -263145

Changing markets, technological innovation and organizational progress in recent years have increased the intensity and scale of agricultural land use. Producer deal with very broad and rapidly expanding range of technology, in areas such as weed control, direct seeding and varieties selection, all of which are required to optimize productivity, protect the environment and maintain or improve the profitability. However, intensification in agriculture follows a pattern of declining marginal productivity and increasing complexity. In the past the main focus of agronomic research has been on crop production only but recently , profitable crop production, the quality of the environment has become an important issue that agricultural producer must address. So the agricultural manager requires strategies for optimize the profitability of the crop production while maintaining soil quality and minimizing environmental degradation. Solution of these new challenges requires consideration of how numerous components interact to effect plant growth. To achieve this goal, future agriculture research will require considerably with more effort and resource than present research activity.


Efficient crop production technology is based on a right decision at right time in a right way. Traditionally a crop production functions that are used in agricultural decision making were derived from conventional experienced base agronomic research, in which crop yield were related to some defined variable based on correlation and regression or regression analysis.  Crop yield were expressed as polynomial or exponential mathematical function of the defined variables, with regression coefficient. The application of correlation and regression analysis has provided some qualitative understanding of the variable and their interactions that were involved in cropping system and has contributed to the progress of agricultural sciences. However the quantitative information obtained from this type of analysis is very site specific. The information obtained can only be reliably applied to other site where climate, important .soil parameters and crop management are similar to those used in developing the original functions. Thus the quantitative application of regression crop based model for decision-making is severely limited. 


Agriculture models are, however, only crude representations of the real systems because of the incomplete knowledge resulting from the inherent complexity of the systems. Judicious use of such model is possible only if the user has a sound understanding of model structure, scope and limitation.  Crop modeling is a new discipline and back ground literature is scarce. So the purpose of this article is provide to some basic information about crop modeling.

Modelling in Agricultural Systems


Complexity of agricultural systems

Agricultural systems are characterized by having many organizational levels. From the individual components within a single plant , through constituent plants, to farms or a whole agricultural region or nation, lies a whole range of agricultural systems. Since the core of agriculture is concerned with plants, the level that is of main interest to the agricultural modeller is the plant. Reactions and interactions at the level of tissues and organs are combined to form a picture of the plant that is then extrapolated to the crop and their output.

Models in agriculture

Agricultural models are mathematical equations that represent the reactions that occur within the plant and the interactions between the plant and its environment. Owing to the complexity of the system and the incomplete status of present knowledge, it becomes impossible to completely represent the system in mathematical terms and hence, agricultural models images of the reality . Unlike in the fields of physics and engineering, universal models do not exist within the agricultural sector. Models are built for specific purposes and the level of complexity is accordingly adopted. Inevitably, different models are built for different subsystems and several models may be built to simulate a particular crop or a particular aspect of the production system.

Features of crop models

The main aim of constructing crop models is to obtain an estimate of the harvestable (economic) yield. According to the amount of data and knowledge that is available within a particular field, models with different levels of complexity are developed. The most pertinent aspects of crop models are described below.


Empirical model

Empirical models are direct descriptions of observed data and are generally expressed as regression equations (with one or a few factors) and are used to estimate the final yield. Examples of such models include the response of crop yield to fertiliser application, the relationship between leaf area and leaf size in a given plant species . the limitation of this model site specific, it can not use universally.


Mechanistic model

A mechanistic model is one that describes the behaviour of the system in terms of lower-level attributes. Hence, there is some mechanism, understanding or explanation at the lower levels. These models have the ability to mimic relevant physical, chemical or biological processes and to describe how and why a particular response results. 

Static and dynamic models

A static model is one that does not contain time as a variable even if the end-products of cropping systems are accumulated over time, e.g., the empirical models. In contrast dynamic models explicitly incorporate time as a variable and most dynamic models are first expressed as differential equations:


Deterministic and stochastic models

A deterministic model is one that makes definite predictions for quantities (e.g., animal liveweight, crop yield or rainfall) without any associated probability distribution, variance, or random element. However, variations due to inaccuracies in recorded data and to heterogeneity in the material being dealt with, are inherent to biological and agricultural systems. In certain cases, deterministic models may be adequate despite these inherent variations but in others they might prove to be unsatisfactory e.g. in rainfall prediction. The greater the uncertainty in the system, the more inadequate deterministic models becomes and in contrast to this stochastic models appears.


Simulation and optimizing models

Simulation models form a group of models that is designed for the purpose of imitating the behaviour of a system. They are mechanistic and in the majority of cases they are deterministic. Since they are designed to mimic the system at short time intervals (daily time-step), the aspect of variability related to daily change in weather and soil conditions is integrated. The short simulation time-step demands that a large amount of input data (climate parameters, soil characteristics and crop parameters) be available for the model to run. These models usually offer the possibility of specifying management options and they can be used to investigate a wide range of management strategies at low costs. Most crop models that are used to estimate crop yield fall within this category.


Optimizing models have the specific objective of devising the best option in terms of management inputs for practical operation of the system. For deriving solutions, they use decision rules that are consistent with some optimising algorithm. This forces some rigidity into their structure resulting in restrictions in representing stochastic and dynamic aspects of agricultural systems. Linear and non-linear programming were used initially at farm level for enterprise selection and resource allocation. Later, applications to assess long-term adjustments in agriculture, regional competition, transportation studies, integrated production and distribution systems as well as policy issues in the adoption of technology, industry re-structuring and natural resources have been developed.  Optimising models do not allow the incorporation of many biological details and may be poor representations of reality. Using the simulation approach to identify a restricted set of management options that are then evaluated with the optimising models has been reported as a useful option.

Some crop models reported in recent literature





Forage harvesting operation


Whole plant water flow




Model development and validation system


Irrigation scheduling model




Modelling framework for a range of crops


General weed model in row crops


Acacia spp.and Leucaena Spp.




Wheat & other crops


Crop (CERES crop modules) & economics




Potato & disease




Wheat & maize, Water and nutrient


Water and agrochemicals


Crop models


Rice, water


Rice, water




Rice, water


Pasture, water, lamb


Erosion Productivity Impact Calculator


Series of crop simulation models


Framework of crop simulation models including





Sugarcane, potential conditions


Sugarcane, potential & water stress conds., erosion


Sugarcane, potential & water stress conds


Sugarcane, potential growth, water and nitrogen stress


Kenaf, potential growth, water stress





The strengths of models in general include the abilities to:

  • Provide a framework for understanding a system and for investigating how manipulating it affects its various components
  • Evaluate long-term impact of particular interventions
  • Provide an analysis of the risks involved in adopting a particular strategy
  • Provide answers quicker and more cheaply than is possible with traditional experimentation

Model calibration

Calibration is adjustment of the system parameters so that simulation results reach a predetermined level, usually that of an observation. In many instances, even if a model is based on observed data, simulated values do not exactly comply with the observed data and minor adjustments have to be made for some parameters .


Model validation

The model validation stage involves the confirmation that the calibrated model closely represents the real situation. The procedure consists of a comparison of simulated output and observed data that have not been previously used in the calibration stage. Ideally, all mechanistic models should be validated both at the level of overall system output and at the level of internal components and processes. The latter is an important aspect because due to the occurrence of feedback loops in biological systems, good prediction of system's overall output could be attributed to compensating internal errors. However, validation of all the components is not possible due to lack of detailed datasets and the option of validating only the determinant ones is adopted. For example, in a soil-water-crop model, it is important to validate the extractable water and leaf area components since biomass accumulated is heavily dependent on these.


The methodology of model validation is still rudimentary. The main reason is that, unlike the case of disciplinary experiments, a large set of hypotheses is being tested simultaneously in a model. Furthermore, biological and agricultural models are reflections of systems for which the behavior of some components is not fully understood and differences between model output and real systems cannot be fully accounted for.


The validation of system simulation models at present is further complicated by the fact that field data are rarely so definite that validation can be conclusive. This results from the fact that model parameters and driving variables are derived from site-specific situations that ideally should be measurable and available. However, in practice, plant, soil and meteorological data are rarely precise and may come from nearby sites. At times, parameters that were not routinely measured may turn out to be important and they are then arbitrarily estimated. Measured parameters also vary due to inherent soil heterogeneity over relatively small distances and to variations arising from the effects of husbandry practices on soil properties. Crop data reflect soil heterogeneity as well as variation in environmental factors over the growing period. Finally, sampling errors also contribute to inaccuracies in the observed data. Validation procedures involve both qualitative and quantitative comparisons. Before starting the quantitative tests, it is advisable to qualitatively assess time-trends of simulated and observed data for both internal variables and systems outputs.

Inadequate predictions of model outputs may require "re-fitting" of the regression curves or fine-tuning of one or more internal variables. This exercise should be undertaken with care because arbitrary changes may lead to changes in model structure that may limit the use of the model as a predictive tool. In some cases, it is best to seek more reliable data through further experimentation than embarking on extensive modification of model parameters to achieve an acceptable fit to doubtful data. This decision relies on the modeller's expertise and rigour as well as on human resources and time available to invest in fine-tuning model predictions.


Models are developed by agricultural scientists but the user-group includes the latter as well as breeders, agronomists, extension workers, policy-makers and farmers. As different users possess varying degrees of expertise in the modelling field, misuse of models may occur. Since crop models are not universal, the user has to choose the most appropriate model according to his objectives. Even when a judicious choice is made, it is important that aspects of model limitations be borne in mind such that modelling studies are put in the proper perspective and successful applications are achieved.

Misperceptions and limitations of models

Agricultural systems are characterised by high levels of interaction between the components that are not completely understood. Models are, therefore, crude representations of reality. Wherever knowledge is lacking, the modeller usually adopts a simplified equation to describe an extensive subsystem. Simplifications are adopted according to the model purpose and / or the developer's views, and therefore constitute some degree of subjectivity. Models that do not result from strong interdisciplinary collaboration are often good in the area of the developer's expertise but are weak in other areas.  Model quality is related to the quality of scientific data used in model development, calibration and validation.


When a model is applied in a new situation (e.g., switching a new variety ), the calibration and validation steps are crucial for correct simulations. The need for model verification arises because all processes are not fully understood and even the best mechanistic model still contains some empirism making parameter adjustments vital in a new situation. Model performance is limited to the quality of input data. It is common in cropping systems to have large volumes of data relating to the above-ground crop growth and development, but data relating to root growth and soil characteristics are generally not as extensive. Using approximations may lead to erroneous results.


Most simulation models require that meteorological data be reliable and complete. Meteorological sites may not fully represent the weather at a chosen location. In some cases, data may be available for only one (usually rainfall) or a few (rainfall and temperature) parameters but data for solar radiation, which is important in the estimation of photosynthesis and biomass accumulation, may not be available. In such cases, the user would rely on generated data. At times, records may be incomplete and gaps have to be filled. Using approximations would have an impact on model performance.


Model users need to understand the structure of the chosen model, its assumptions, its limitations and its requirements before any application is initiated, e.g, using a model like QCANE, developed for cane growth under non-limiting conditions, would lead to erroneous output and analysis if it is used to simulate under water or nitrogen stress conditions.   At times, model developers may raise the expectations of model users beyond model capabilities. Users, therefore, need to judiciously assess model capabilities and limitations before it is adopted for application and decision-making purposes. Generally, crop models are developed by crop scientists and if interdisciplinary collaboration is not strong, the coding may not be well-structured and model documentation may be poor. This makes alteration and adaptation to simulate new situations difficult, specially for users with limited expertise.  Finally, using a model for an objective for which it had not been designed or using a model in a situation that is drastically different from that for which it had been developed would lead to model failure.

Model uses

The above points may give the impression that crop modelling has a bleak future but recent literature confirms the contrary. Simulation modelling is increasingly being applied in research, teaching, farm and resource management, policy analysis and production forecasts. These model can be applied into three areas, namely, research tools, crop system management tools, and policy analysis tools. A summary of some specific applications within the different groups follows:

As research tools

Research understanding: Model development ensures the integration of research understanding acquired through discreet disciplinary research and allows the identification of the major factors that drive the system and can highlight areas where knowledge is insufficient. Thus, adopting a modelling approach could contribute towards more targeted and efficient research planning. For example, changing the plant density in a sugar beet model resulted in model failure. This failure stimulated studies that gave additional information concerning biomass partitioning in the sugar beet.


Integration of knowledge across disciplines: Adoption of a modular approach in model coding allows the scientist to pursue his discipline-oriented research in an independent manner and at a later stage to integrate the acquired knowledge into a model. For example, the modular aspect of the APSIM software allows the integration of knowledge across crops as well as across disciplines for a particular crop. Adoption of a modular framework also allows for the integration of basic research that is carried out in different regions, countries and continents. This ensures a reduction of research costs (e.g., through a reduction in duplication of research) as well as the collaboration between researchers at an international level.


Improvement in experiment documentation and data organization: Simulation model development, testing and application demand the use of a large amount of technical and observational data supplied in given units and in a particular order. Data handling forces the modeller to resort to formal data organisation and database systems. The systematic organisation of data enhances the efficiency of data manipulation in other research areas (e.g., productivity analysis, change in soil fertility status over time)


Genetic improvement: As simulation models become more detailed and mechanistic, they can mimic the system more closely. More precise information can be obtained regarding the impact of different genetic traits on economic yields and these can be integrated in genetic improvement programs, e.g., the NTKenaf model. Researchers used the modelling approach to design crop ideotypes for specific environments.


Yield analysis: When a model with a sound physiological background is adopted, it is possible to extrapolate to other environments. The use of several simulation models to assess climatically-determined yield in various crops . The CANEGRO model has been used along the same lines in the South African sugar industry. Through the modelling approach, quantification of yield reductions caused by non-climatic causes (e.g., delayed sowing, soil fertility, pests and diseases) becomes possible. Almost all simulation models have been used for such purposes. Simulation models have also been reported as useful in separating yield gain into components due to changing weather trends, genetic improvements and improved technology.

As crop system management tools

Cultural and input management: Management decisions regarding cultural practices and inputs have a major impact on yield. Simulation models, that allow the specification of management options, offer a relatively inexpensive means of evaluating a large number of strategies that would rapidly become too expensive if the traditional experimentation approach were to be adopted. Many publications are available describing the use of simulation models with respect to cultural management (planting and harvest date, irrigation, spacing, selection of variety type) and input application (water and fertiliser).


Risks assessment and investment support: Using a combination of simulated yields and gross margins, economic risks and weather-related variability can be assessed. These data can then be used as an investment decision support tool.  


Site-specific farming: Profit maximisation may be achieved by managing farms as sets of sub-units and providing the required inputs at the optimum level to match variation in soil properties across the farm. Such an endeavour is attainable by coupling simulation models with geographic information systems (GIS) to produce maps of predicted yield over the farm. But, one of the prerequisites is a systematic characterisation of units that may prove costly.

As policy analysis tools

Best management practices: Models having chemical leaching or erosion components can be used to determine the best practices over the long-term. The EPIC model has been used to evaluate erosion risks due to cropping practices and tillage.  


Yield forecasting: Yield forecasting for industries over large areas is important to the producer (harvesting and transport), the processing agent (milling period) as well as the marketing agency. The technique uses weather records together with forecast data to estimate yield across the industry. 


Introduction of a new crop: Agricultural research is linked to the prevailing cropping system in a particular region. Hence, data concerning the growth and development of a new crop in that region would be lacking. Developing a simulation model based on scientific data collected elsewhere and a few datasets collected in the new environment helps in the assessment of temporal variability in yield using long-term climatic data. Running the simulations with meteorological data in a balanced network of locations also helps in locating the industry.


Global climate change and crop production: Increased levels of CO2 and other greenhouse gases are contributing to global warming with associated changes in rainfall pattern. Assessing the effects of these changes on crop yield is important at the producer as well as at the government level for planning purposes. 



Crop/soil simulation models basically applied in three sections (1) tools for research, (2) tools for decision-making, and (3) tools for education, training and technology-transfer. The greatest use of crop/soil models so far has been by the research community, as models are primarily tools for organizing knowledge gained in experimentation. However, there is an urgent need to make the use of models in research more relevant to problems in the real world, and find effective means of dissemination of results from work using models to potential beneficiaries.   Nevertheless, crop models can be used for a wide range of applications. As research tools, model development and application can contribute to identify gaps in our knowledge, thus enabling more efficient and targeted research planning. Models that are based on sound physiological data are capable of supporting extrapolation to alternative cropping cycles and locations, thus permitting the quantification of temporal and spatial variability. Over a relatively short time span and at comparatively low costs, the modeller can investigate a large number of management strategies that would not be possible using traditional methodologies. Despite some limitations, the modelling approach remains the best means of assessing the effects of future global climate change, thus helping in the formulation of national policies for mitigation purposes. Other policy issues, like yield forecasting, industry planning, operations management, consequences of management decisions on environmental issues, are also well supported by modelling.



Formatted and uploaded by Priyanka Shukla
Your rating: None Average: 4.5 (4 votes)

Please note that this is the opinion of the author and is Not Certified by ICAR or any of its authorised agents.