Crop Model Calibration: A Statistical Perspective
- Daniel Wallach *
It is common practice to calibrate crop models. This involves estimating some of the model parameters to give a better fit of the model to data. The purpose of this study was to investigate the statistical treatment of crop model calibration, which has not previously been done and is important to better understand the properties of the calibrated model. We considered the case where there is only a single type of variable measured, in the limit of a very large amount of data, and where estimation is based on least squares. We supposed that the individual process models that make up a crop model are such that, for the true parameter values, model error has expectation zero and is independent of the explanatory variables. We show that even in this case, the crop model does not have these properties (i.e., it is misspecified). Based on the structure of crop models and a simulation study, we argue that misspecification is often quite important for crop models. Under misspecification, calibration still tends to minimize prediction error but only for the variable and sampled population used for calibration. There is no assurance that calibration improves prediction for other variables or other populations.Please view the pdf by using the Full Text (PDF) link under 'View' to the left.
Copyright © 2011. . Copyright © 2011 by the American Society of Agronomy, Inc.