First cropping system model based on expert-knowledge parameterization

Abstract

Models are promising tools to support the design of cropping systems toward sustainable agriculture. Process-based deterministic models are predominantly used, whereas most of them involve a limited range of crop techniques, and are unsuited to organic agriculture. Moreover, their parameterization and local adaptation require a large amount of experimental data. We thus designed a model simulating the yields of successive crops, taking into account the effects of most crop techniques embedded in a cropping system, and suited for both conventional and organic farming. This model was designed assuming that its parameterization, mostly based on expert-knowledge elicitation, could enlarge the range of environmental conditions and crop techniques considered. The PerSyst model involves three types of parameters based on expert knowledge: (i) reference yields reached in the most common cropping system conditions, (ii) yield change due to crop sequence variation, and (iii) yield change due to variation in crop management. These parameters are stochastic to report yield variability across climatic years. The model was parameterized through an original expert elicitation method—combining individual interviews and collective validation—on three case studies, including one in organic farming. Model accuracy was assessed for two long-term experiments. Parameters such as yield change due to crop sequence and to crop management were close among case studies, highlighting possibilities to compensate for a local lack of knowledge. Moreover, simulated yields in both experiments showed great consistency with observed yields, with average relative root-mean-square error of prediction of 15% for winter wheat and faba bean for example. For the first time, thanks to expert-knowledge parametrization, we built a cropping system model, considering all techniques, which could be easily tailored to a diversity of conditions, both in conventional and organic farming. Lastly, advantages and limits of the PerSyst model to assess innovative cropping systems were discussed.