Applying Multiple Crop Models for Assessing Climate Change Impact: The Devil Is in the Detail.

One way to address uncertainty in assessing climate change impacts on crop production is to employ multiple crop models. While using multiple models is a standard approach amongst climate modellers, there is little experience with it amongst the agricultural modelling community. Here we describe modelling climate change impacts on maize and sugarcane production in Africa with APSIM and DSSAT cropping systems models. The two models appear similar in many ways and have quite extensive applications for these crops in the region. Thus, their joint application should be straightforward with the modelling teams being able to draw on each other’s experiences (e.g. as represented in model parameter values). This similarity was true for the ‘tipping bucket’ soil water modules, and in this study there was reasonable interchangeability between parameter values. With the crop modules, while some parameters had similar names their definition was somewhat different in each model. So cultivar parameters could not necessarily be ‘ported’ directly between models and some cultivars were parameterised from primary data. The soil C and N cycling modules in the two models have different structures and so need to be independently parameterised, although this need was not obvious to crop modellers who didn’t have a background in soil C modelling. Both models have powerful representation of farm management. However, when drawn upon to ‘manage’ irrigation in long-term simulations of sugarcane production, it proved very difficult to get precisely the same irrigation amounts and timing applied in the models and, as a result, yields differed markedly. Finally, the climate change responses predicted for maize yield in 47 farms in Kenya differed, with DSSAT generally predicting yield decreases whereas APSIM predicted both decreases and increases. Clearly, multiple-model assessment is not straight forward. Substantial effort and understanding of the models involved is needed to successfully complete these assessments.