Method-Based Multi-Objective Parameter Co-Calibration of Dryland Wheat Yield and Protein Content

Water and nitrogen availability jointly influence dryland wheat yield and grain protein formation; the ability of crop models to simulate the synergistic effects of these two factors depends on the accuracy of key parameters. However, the large number of model parameters and their complex interactions make parameter screening and calibration challenging. In this study, the model was calibrated using annual yield records for dryland wheat in Anding District, Dingxi City, and Gansu Province, from 2015 to 2019, along with field observation data (grain protein content) from Anjiagou Village, Fengxiang Town, and Anding District, collected during the same period. The model was then validated using independent data from two growing seasons (2020–2021) in the same study area. Based on the Agricultural Production sIMulator Next-Generation (APSIM NG) and using water–nitrogen response as the foundation, a multi-objective parameter calibration framework was developed with yield and grain protein content as calibration targets. The extended Fourier amplitude test (EFAST) was used to screen key model parameters, and the performance of the DREAM–zs algorithm, the Nelder–Mead simplex algorithm, and the multi-objective rime optimization algorithm (MORIME) in calibrating these parameters was compared and analyzed. All three algorithms reduced the simulation error in yield; for grain protein content, the Nelder–Mead and MORIME algorithms reduced the simulation error, while DREAM-zs did not. The simulation accuracy of the Nelder–Mead and MORIME algorithms was generally similar, but MORIME yielded a set of Pareto solutions that characterized the trade-off between the simulation accuracy of yield and grain protein content; therefore, its parameter solutions were selected for independent validation. The validation results showed that the RMSE for yield decreased from 248.45 kg hm−2 to 213.90 kg hm−2, and the NRMSE decreased from 12.60% to 10.85%; for grain protein content, the RMSE decreased from 0.280% to 0.250%, and the NRMSE decreased from 1.86% to 1.66%. Furthermore, the incorporation of parallel computing reduced MORIME’s runtime by 82.99%. The study demonstrates that MORIME with parallel computing can provide methodological support for the multi-objective, efficient optimization of crop model parameters.