Soybean multi‐model sensitivity analysis for prediction of seed nitrogen, biological N fixation, and N cycling

There is limited work on model evaluation for prediction of seed protein and N2-fixation in soybean production. The objectives of the study were to calibrate soybean models under baseline conditions with data from 5 locations worldwide and to quantify model uncertainty under variable atmospheric [CO2], Temperature, Water, and N fertilizer (CTWN sensitivity analysis). Model error due to a systematic over-prediction of total aboveground and seed N content (kg N ha-1), and seed protein (%) was reduced in most models after full calibration. Optimization of critical seed N concentration to minimize error in seed protein (%) may reduce bias but not improve predictability across environments -> need of more multi-environment data with same genotypes. N2-fixation showed the expected response: increased due to greater photosynthesis under high CO2, and decreased under increasing N fertilizer applications. However, the rate of response was widely variable across models. Differences in how models predict N demand for seed growth may introduce high uncertainty in model responses to CO2, Temperature, and N fertilizer.