Modelling in agroecology: from simple to complex models, and vice versa

The rise of agroecological cropping systems led to increase their complexity. Compared to conventional systems,they usually include richer communities of plants, e.g. mixed-cropping and agroforestry systems. In these systems, more diverse associated communities are involved in key processes including natural control of pest sand diseases or nutrient cycling. Modelling these complex agroecosystems implies taking into account network of interactions between plants, pest and diseases, regulating communities, water, nutrients and radiative resources. This complexity also includes the spatial and temporal organization of fields that tends to be more heterogeneous than in conventional systems; this is particularly the case in multi strata and agroforestry systems. This heterogeneity is also important at the level of the assemblages of cultivated and associated plants that hugely varies across the infinite possible combinations of plants. Agroecology also relies on more diverse plant species that did not always received an extensive effort of parametrization in current models. Most common models used to simulate agrosystems, e.g. crop models, rely on relatively simple hypotheses such as the homogeneity of the canopy that is unlikely to be valid in more complex agrosystems. Initially built in limited biotic stresses, such models rarely take into account the effect of pest and diseases nor their regulations, which are key processes when dealing with zero pesticide systems. The temptation for modelers in agroecology would be to increase models' complexity as agrosystems complexity increases.This presentation questions the type of models needed to address the issues related to the agroecological transition. I support the idea that modelling agroecological systems implies rethinking the models rather than just making existing ones more complex. Another change in modelling agroecological systems is in the use of models; it may not be as linear (parametrization, simulation, yield prediction) as usually done in many conventional agriculture approaches but may be done in order to address questions that are specific to agroecological systems. For instance, in agroecology we rather question: “what is the optimal plant community and its spatio-temporal organization that optimize its resilience to pest and diseases and to inter-annual variations?” than raw yield prediction. After briefly reviewing existing models in agroecology, I will present some successful examples of simple models that address key questions on complex agroecosystems (Poeydebat et al 2016). Bridging the statistical and process-based approaches is proposed as a relevant mean to address the issue of species diversification and of spatial heterogeneity. For instance, individual based models, which take the best of both approaches, can be applied to simulate tropical agroforestry systems. It illustrates how relatively simple models can represent complex canopies and pest regulation processes through i) statistical spatial relationship between plants and ii)the growth of each plant following resource-partitioning mechanisms. Finally, I will draw some perspectives ofhow interaction network models (Tixier et al 2013) can renew modelling approaches dedicated to highly complex assemblages of plants