Measurements on experimental silvoarable agroforestry plots (mixed tree-crop) showed exceptionally high productivity, with yields more than 30% higher compared to the separate crop and tree systems. We developed a 3-D process-based model (Hi-sAFe) and used it to (1) test if the spatiotemporal integration of resources sharing (light, water, nitrogen) between species could explain this productivity and (2) assess how agroforestry design or management may affect this productivity. We developed a novel method for the analysis of simulated data to identify and sort the processes determining yield by order of importance. The model was parameterized and calibrated on a hybrid walnut / durum wheat located on the Restinclières field (Hérault, France). After a detailed analysis of the functioning of this system, we explored by virtual experiments its response to design choices : (1) phenological traits of associated species, and (2) density and location of trees on the plot. The comprehensiveness of the Hi-sAFe model provides a new insight into the functioning of agroforestry systems. It clarifies the complexity of the real system. Our results put forward a deterministic interpretation of the performance of each species and of the total productivity of the crop-tree association (Land Equivalent Ratio). The analysis of virtual experiments offers practical ways for optimizing agroforestry systems.
