Using the CERES-Maize model in a semi-arid Mediterranean environment. Validation of three revised versions

The original version (V0) of the CERES-Maize model predicted quite accurately the productivity of wellwatered maize grown in a semiarid Mediterranean environment. The relative difference (D%) between simulated and observed data was lower than 15%. However, under soil water shortage, the simulations provided by V0 were not satisfactory in terms of leaf area index, LAI (D%>40), biomass (D%>25), grain yield (D%>20). The aim of this paper is to verify whether the poor performance of CERES-Maize derives either from a non-correct simulation of the leaf area, from an extremely severe effect of the water stress coefficient, or from the joint effect of both. The soundness of these hypotheses has been tested by comparing three different revisions of CERES-Maize model (V1, including the modification to submodel for LAI simulation; V2 with the revised function describing the water stress estimate; V3 combining both revisions) with the observed data and with the original V0 model. The revised models were validated at two levels: the final data at the end of the growing season and their dynamics on a daily basis. Revisions V1 or V2 improved simulation of some CERES output variables, particularly, LAI in V1 and grain yield in V2. However, only the combined version V3 shows the best performance (D%<13 for grain yield, biomass and LAI) of the CERES-Maize model under Mediterranean conditions, characterised by a changing soil water availability. This supports the third hypothesis according to which the goodness of model-fitting under the Mediterranean conditions can be increased only by both revisions, in leaf surface and in water stress coefficient. The generalisation of the two sub-models included in the V3 version, concerning the LAI and the stress function calculations, is also discussed.