Scale dependence and parameter sensitivity of the EPIC model in the agro-pastoral transitional zone of north China Check

Under the background of rapid population growth, achieving accurate assessments of crop yields is critical to national policy adjustment and food security. Among all the evaluation methods for crop yields, the Environmental Policy Integrated Climate (EPIC) model is considered to be an effective tool. Since published, it has been widely applied on local, regional, and global scales. However, in its practical applications, the impact of scale variation on the simulation results has not received full consideration. The scale variation can introduce many uncertain effects in the simulation results. Furthermore, as a basis for parameter calibration, parameters sensitivity analyses were always been ignored at large scale studies. In this study, we used the GIS-based EPIC (GEPIC) model to assess the simulation accuracies at different scales and local sensitivity analysis was adopted to analyze the sensitivity of the crop parameters in each grid first time. The results showed that EPIC model has scale-dependence and the smaller the scale is, the more accurate the results are. The most sensitive parameter of maize and soybeans is the energy conversion to biomass factor (WA) in the agro-pastoral transitional zone of north China, whereas the harvest index (HI) is the most sensitive parameter for the three other crops of rice, spring wheat, and winter wheat. For the same crop, parameter sensitivity is heterogeneous in space due to varied input variables. In addition, we find that parameters sensitivities (S-i) vary as the parameters and the variations in S(i )for the five parameters are different. Excluding the S-i of HI for spring wheat which has the greatest change and the variation is nearly 0.5, the S-i of other parameters change slightly. In the application of GEPIC model, we should consider the effects of scale variation on crop yields, particular for areas with high heterogeneity of cropland patches. In addition, this study provides a reference for setting the input parameters of the EPIC model.