Methodological comparison of calibration procedures for durum wheat parameters in the STICS model

Parameter estimation is an important preliminary step before using a crop model to evaluate the effect of different management strategies. This study compared three procedures for estimating parameters of the STICS soil–crop model. Two procedures estimate parameters using information on both dynamic and end-of-season variables. One of those procedures estimates parameters in successive estimation steps, whereas the other estimates many parameters simultaneously in a single step. The third procedure uses only information on end-of-season variables. In all cases, the parameter values were constrained to lie within the same bounds, which represent physiologically reasonable values. Data from 235 and 138 experimental plots were used for parameter estimation and independent model evaluation, respectively.Mean squared error of adjustment and of prediction were analysed to evaluate the performance of the different procedures. An agronomic analysis was conducted to evaluate the capacity of the model to respond to varying fertilisation strategies.The results indicate that using information from dynamic variables is not useful if the only objective is to improve model prediction of end-of season variables. However, to improve model prediction of both dynamic and end-of-season variables, all information should be used. In this case, better results were obtained when all dynamic variables were used in a single estimation step.This study provides some methodological guidelines on running parameter estimation for complex crop models. Our results also indicate that parameter estimation can highlight problems of process representation in a crop model.

► Calibration is effective when bias makes a large contribution to MSE.
► Dynamic variables are not necessary to improve end-of season variables prediction.
► One step estimation is more efficient than stepwise estimation.
► Parameter estimation can highlight problems of crop model process representation.
► Methodological guidelines on running parameter estimation for crop models.