Model simplification and development via reuse, sensitivity analysis and composition: A case study in crop modelling

Crop models, like many representations of environmental processes, tend to be over-parameterised. A redesign of the SUCROS family of crop models, largely driven by sensitivity analysis, is presented here. In particular, two new versions of WOFOST, the most widespread model from this family, were developed. The first (WOFOST-GT) reduces model complexity through the definition of functions driven by few parameters with biological meaning. The other (WOFOST-GT2) improves canopy representation and senescence. Each version was evaluated for rice and winter wheat. Results highlighted a similar accuracy for the three versions: the original one achieved mean normalized RMSE of 13.75% and 10.75% for winter wheat and rice; corresponding values for the new versions were 14.42% and 10.79% (WOFOST-GT), and 14.38% and 10.85% (WOFOST-GT2). The new versions were considerably less complex, (60% less parameters). These improvements, increasing model usability without compromising its sophistication, can be transferred to other models from the same family.