A Statistical Approach to Calibration of the Phenology Sub-Model of Apsim Rice : An Application and General Lessons.

Parameter estimation for the phenology sub-model of a crop model is of importance in itself, since phenology is a major difference between varieties and therefore a major reason that model calibration is required for each new variety. It can also serve as a relatively simple example of the general principles involved in crop model calibration, from which we can draw more generally applicable lessons about calibration.

In this study we estimated the phenology parameters for APSIM rice, for four varieties from Sri Lanka. In this model the cardinal temperatures are fixed; the six parameters to estimate are thermal times to each of four development stages and two parameters related to photoperiod sensitivity. Since plant growth does not affect development, one can separate  the phenology sub-model from the rest of APSIM rice.  We use standard statistical software to calculate the least squares parameter estimates, and the standard deviations of the estimators.

This study illustrates major general principles about calibration of crop models.  First, there are two levels of information that are used.  Information from detailed studies is used to fix the parameters that represent the cardinal temperatures of the model, while field data is used to estimate the remaining parameters. Secondly, the possibility of parameter estimation is directly related to the data available for estimation. In the present case one of the model parameters is not estimable, and in fact we argue that it will probably never be estimable from the data likely to be available. The two photoperiod sensitivity  parameters are very poorly estimated, which could lead to very poor predictions.

A general conclusion is that the calibration step should be considered when the model is developed. The model parameters should be estimable using commonly available data.