Global sensitivity analysis of APSIM-Wheat parameters in different climate zones and yield levels

Abstract: Abstract: Uncertainties of crop model are mainly from model structure, parameters' sensitivity and input error. It is essential to quantify the parameters' sensitivity and the results' uncertainties of crop model for model calibration and application. Furthermore, uncertainty and sensitivity analysis can improve the reliability of model prediction. There are 2 categories of parameter sensitivity analysis methods, i.e. the local sensitivity and the global sensitivity. Local sensitivity analysis is called derivative-based or one-at-a-time method, which changes only one parameter at a time around a basis point while keeping other parameters constant. It cannot detect the interactions among the parameters and suffers some shortcomings such as a heavy dependence on the input parameters and instability for non-linear models. Global sensitivity analysis is a better method for exploring the entire multi-dimensional parameters simultaneously. It can quantify the influence of single parameter and the interactions among different parameters. Several global sensitivity methods including Morris, variance-based, linear regression, FAST (Fourier amplitude sensitivity test) and EFAST (extended Fourier amplitude sensitivity test) are widely used in the parameter analysis. EFAST is robust and has lower computational cost than the others. Previous studies of sensitivity analysis focused on a single site. However, the performance of crop model is variable in different climatic zones due to the heterogeneity of climate and soil characteristics. In this study, we collected crop experimental data in the locations of Luancheng, Hebei Province Changwu in the Chinese Loess Plateau, Yanting in the Sichuan Basin and Wulanwusu in Xinjiang Autonomous Region. And then using the global sensitivity analysis method i.e. EFAST, we analyzed the sensitivity and uncertainty of crop model (APSIM-Wheat) brought by the cultivar, soil and biological parameters in different climate zones and yield levels (i.e. potential, rainfed and actual yield). We found that: 1) The most sensitive parameters for anthesis and maturity date were successively: accumulated temperature in the early flowering season, accumulated temperature from emergence to jointing, vernalization index, photoperiod factor, and accumulated temperature in grain filling stage; 2) The most sensitive parameters for yield were successively: vernalization index, accumulated temperature from emergence to jointing, grain mass per stem at the beginning of grain filling, potential daily grain filling rate, photoperiod factor, maximum grain mass and radiation use efficiency; the most sensitive parameters for evapotranspiration (ET) were successively: vernalization index, accumulated temperature from emergence to jointing, photoperiod factor and accumulated temperature in the early flowering season; 3) The parameter sensitivity for anthesis, maturity, yield and ET in different yield levels was almost coincident, which indicated that the simulation of APSIM-Wheat was not influenced by different yield levels; 4) The parameter sensitivity for anthesis and maturity in different climatic zones was almost the same, while that for yield and ET was different. The sensitivity difference in different climatic zones for yield and ET warns the model users to use the model carefully in different locations. The results indicate that we should calibrate the more sensitive parameters for phenology and then calibrate the yield and ET during the APSIM-Wheat calibration.