|What Are Crop Simulation Models?|
Agricultural production managers, natural resource managers, and strategic decision makers require accurate, timely, and cost-effective information to maintain a quality food and fiber supply for the nation and the world. The Crop Systems and Global Change Laboratory conducts research to develop crop simulators for predicting growth, development, and yield of agricultural crops.
Crop simulators are computer programs that mimic the growth and development of crops. Data on weather, soil, and crop management are processed to predict crop yield, maturity date, efficiency of fertilizers and other elements of crop production. The calculations in the crop models are based on the existing knowledge of the physics, physiology and ecology of crop responses to the environment.
CSGCL has developed crop simulation models that answers questions involving global climate change, precision agriculture, soil hydraulic properties and plant physiology.
A methodology for parameterizing, validating, and comparing models has been developed and illustrated with an example of leaf photosynthesis models. Many models of leaf photosynthesis have been proposed but there have been few attempts to compare them and determine their adequacy for various purposes. The methodology includes the following steps. (1) A modified Marquardt algorithm was shown to be the most efficient one for fitting the experimental data to models and for parameterizing the models. (2) The F-test was algorithmized for estimating the quantitative accuracy of the models. (3) The residuals analysis and the autocorrelation of errors test (for the models with a single input variable) provided the qualitative assessment of the models. (4) The statistical test by Williams and Kloote was shown to be efficient for comparing models.
Comprehensive crop models require parameters for soil hydraulic properties as input data. These properties include the soil moisture release curve and hydraulic conductivity as a function of soil matric potential. These data are difficult and expensive to measure in the field so methods to estimate these from more easily measured soil properties are potentially very useful. We applied new data modeling techniques including artificial neural networks and fractal geometry to estimate soil moisture release curves from soil texture, and achieved a significant improvement in accuracy. A neural network is a non-linear procedure to elucidate relationships among data. Evaluation studies on the use of estimated soil hydraulic properties in GLYCIM showed that the accuracy as compared to using measured values improves with the temporal scale of the model application. These results are important for extending the applicability of crop models for large scale estimates.