Global sensitivity analysis of a generic agent-based approach to model diverse behaviour of farmers

We present a global sensitivity analysis of a generic agent-based modelling approach called FARMIND (FARM Interaction and Decision-making). The purpose of the model is to simulate agricultural production decisions. FARMIND reflects the heterogeneity in farmers decision-making processes based on standard and behavioural economic concepts. The key functionality of the model is to consider farmers’ individual characteristics such as attitudes and risk preferences and the farmers’ social network in the decision-making process. The model allows to inject diverse behaviour into existing bio-economic simulation models and is intended to improve the understanding and explanation of farmer’s decision-making trough hypothesis testing. Applications of the model can be used to test and evaluate environmental impacts of farmers’ production decisions e.g. under climate or policy changes. To evaluate model sensitivity, we linked our agent-based framework to a bio-economic weed control model for silage maize production and applied three consecutive analyses: Morris screening, standardized regression coefficients and Sobol’ method. The Morris's elementary effects screening allowed to identify the relative importance of the input parameters. This, in turn, allowed selecting the more important parameters for global sensitivity analyses. The results showed that agents’ decisions in FARMIND are in accordance with the underlying theoretical expectations. We also found that the contributions of the individual parameters are additive and a larger part of the model variation can be explained by linear combinations of the input parameters. The global sensitivity analysis, however, also showed that our approach might be prone to equifinality, i.e., the generation of the same model output under different conditions. With respect to the purpose of the model, FARMIND successfully diversified weed control decisions of farmers. This presents an opportunity to align the simulation of farmer’s decisions with existing bio-economic models of agricultural systems and a detailed representation of environmental externalities from agricultural production.