Résumé
Problem: Low crop yields in sub-Saharan Africa mainly result from low soil fertility and insufficient nutrient inputs. A key component of Integrated Soil Fertility Management (ISFM), namely combining inputs of mineral fertilizers and organic resources, presents an opportunity to boost yields and maintain soil organic carbon (SOC) stocks in the long run. Soil-crop models help to assess the performance of ISFM under contrasting soil, climate, and management combinations. Yet, to date, most soil-crop models have been calibrated and tested in temperate conditions. Objective: Our objective was to evaluate and compare the performance of two different soil-crop models, DayCent and STICS, to represent crop yields and SOC dynamics under contrasting organic resource amendments. Methods: We used a large dataset representing 3384 cropping situations (site x season x treatment) from four long-term experiments in Kenya. Each experiment included the same treatments with the addition of two quantities of low- to high-quality organic resource amendments (high vs low C/N ratio, respectively), with (+N) and without (-N) mineral nitrogen fertilizer. Each treatment included a cropped and uncropped subplot, allowing for a unique stepwise calibration of soil and crop parameters. Results: Both models represented SOC and yield dynamics with similar accuracy across sites and treatments. They reproduced SOC dynamics well (nRMSE below 30 %) in the two clayey soils sites but not in the two sandy soils. Yet, in most sites they reproduced well SOC differences between high (Farmyard manure, Thithonia and Calliandra) and low-quality (maize stover and sawdust) organic resources. Models reproduced the average yield across sites and treatments similarly. They reproduced the positive effects of high-quality organic resources and the addition of mineral N on maize yield well. Models had similar inaccuracy in reproducing yield and yield variability under poor-quality organic resources and -N treatments. Conclusion: The stepwise calibration approach used in this study enabled highlighting the models' strengths and weaknesses in soil and plant simulations. The results suggest that the two models have similar strengths and struggle with the same problems despite having different structures. Collecting detailed plant (leaf area index, plant N uptake) and soil (water, nitrogen dynamics) in-season data from long-term experiments will be critical to exploit the full model complexity and improve their accuracy for tropical conditions.