Sensitivity analysis and calibration of the APSIM next-generation model under different irrigation and sowing density in wheat

The Agricultural Production Systems Simulator (APSIM) is currently one of the most widely-used crop and farming system models globally. With increasing challenges and demand for agricultural modeling, the APSIM Initiative is building the next generation of APSIM to improve its prediction accuracy and increase its applicability to a wider range of farming systems. The APSIM next-generation (APSIM NG) model implemented new phenology- and morphology-simulating mechanisms, introduced additional parameters, and allowed modelers to add custom parameters. These parameters cannot be directly measured and must be calibrated when the APSIM NG model is applied to a new environment and cultivar. Determining the relative importance of the parameters to the specific outputs can streamline the calibration of crop models for new cultivars, and a sensitivity analysis can quantify the influence of model input parameters on model outputs. In this study, we used the Extended Fourier Amplitude Sensitivity Test to perform a sensitivity analysis on the wheat module of the APSIM NG for the first time. We also calculated the main and total effect sensitivity indices of three outputs — yield, flowering day, and maturity day — to crop parameters under different irrigation and plant density treatments. We found that days to anthesis and physiological maturity were mostly sensitive to the parameters that determine the length of the reproductive stages (phyllochron, number of leaves the plant will produce when fully vernalized early and grown in long photoperiod, and photoperiod sensitivity), and yield was most sensitive to the cultivar parameters that determine the yield component (GrainsPerGramOfStem, MaximumPotentialGrainSize) and phyllochron. Irrigation and sowing density treatments affected the main effect and total effect sensitivity index of parameters to yield; however, it did not affect the order of parameters. Next, we calibrated the model against data from 2015 to 2018 from the Wuqiao Experimental Station of the China Agricultural University in Hebei Province. The data comprise four irrigation and plant density treatments. The calibrated APSIM NG model captured the Zadoks decimal growth scale and yield with acceptable accuracy. Across the treatments, the APSIM NG explained more than 98% of the variation in the growth scale. The root-mean-square error (RMSE) of the yield was 508 kg∙hm⁻², compared with the experimental data. This study provides guidelines for APSIM NG model calibration in the North China Plain, as well as guidance to simplify the APSIM NG model and improve its precision, especially when many parameters are used. For robust phenology and yield prediction with APSIM NG, more research on the environment, genotype, and management factors is suggested.