Comparison of parameter estimation methods for crop models

Crop models are important tools in agronomic research, a major use being to make predictions. A proper parameter estimation method is necessary to ensure accurate predictions. Until now studies have focused on the application of a particular estimation method and few comparisons of different methods are available. In this paper, we compare several parameter estimation methods, related, on the one hand, to model selection, and on the other, to ridge regression based on an analogy to a Bayesian approach. The different methods are applied to a simplified crop model derived from the STICS model, using simulated data. The criteria for comparison are prediction error and errors in the parameter estimates. Among the methods of model comparison a version of the Schwarz criterion, corrected for small samples and with maximum and minimum bounds for each parameter, is the preferred method. Ridge regression is found to be superior to this best method of model selection.