A modelling chain combining soft and hard models to assess a bundle of ecosystem services provided by a diversity of cereal-legume intercrops

Cereal-legume intercropping is known to improve the sustainability of crop production. However, it remains uncommon on commercial farms in Europe due to a number of socio-technical lock-ins and the many practical issues raised when integrating intercrops in cropping systems (e.g. which species, cultivars, sowing densities). Crop modelling is an option to explore integration scenarios and support farmers' decisions. However, available crop models are not able to simulate bundles of ecosystem services provided by a large diversity of binary cereallegume intercropping scenarios. To address this challenge, we developed a hybrid modelling chain that combines process-based, statistical and knowledge-based models to benefit from the strengths of these three different modelling approaches. The chain (i) simulates potential biomass of the sole cereal and legume crops independently using the crop model STICS; (ii) uses statistical interaction models built in R to convert potential biomass in sole cropping into attainable biomass in intercropping by considering competition effects among species, using a field trial database; (iii) converts attainable biomass into actual biomass by considering pest damage using a knowledge-based multi-attribute DEXi model, and also assesses control of pests (i.e. weeds, insects and diseases); and (iv) uses another set of multi-attribute models to assess five additional ecosystem services (i.e. cereal and legume grain yields, cereal protein content, nitrogen supply to the following crop and impact on soil structure) from the actual biomass of the intercrop at harvest and/or cropping system features. The chain was calibrated for grain cereal-legume intercrops sown simultaneously in a random pattern under low-input French conditions. We used an expert-based approach to assess the performances of each model and evaluate the accuracy of the entire modelling chain. In 18 simulated scenarios, 79% of the predicted levels of ecosystem services were consistent with experts' opinion. Predictions were more accurate for intercropping scenarios that included species from the trial database used to build linear interaction models (relative RMSE of 27-31%) but remained satisfactory for other intercropped species (relative RMSE of 32-37%). This is the first modelling chain able to assess bundles of ecosystem services provided by multiple cereal-legume intercrops in function of their cropping system contexts. This chain is intended to be included in an educational tool that is used face to face with farmers or students to design cropping systems that include intercrops.

Automated summary

This study developed a hybrid modeling chain that combines process-based, statistical and knowledge-based models to assess the ecosystem services provided by multiple cereal-legume intercrops. The chain demonstrated high accuracy and can be used as an educational tool for farmers and students to design intercropping systems.