| 摘 要： |
Climate change and the increasing frequency of climate extremes associated with warming have been the most important climatic stressors for maize production. However, crop -model based assessments of the major determinants of yield variability at regional scale under future climate conditions are still underrepresented. In this study, we simulated maize yield in Northeast China at a reference period (1986-2005), and two future periods (2030 s: 2020-2039, 2050 s: 2040-2059) of Shared Socioeconomic Pathways (SSPs) of SSP1-2.6, SSP2-4.5 and SSP5-8.5 using Agricultural Production sIMulator (APSIM). We first characterized the variations of maize yield under climate change based on the simulations, and further investigated the key determinants of yield variability using machine learning techniques. The results suggest that maize yield would decrease by 14.8 % to 19.6 % (depending on the climate scenarios) compared to the reference period without adaption. Random forest performed best in explaining yield variability in a suite of machine learning models (extreme gradient boosting, gradient boosting, classification and regression tree, ridge and lasso regression and random forest), with a mean R2 of 0.77, a mean RMSE of 1239.2 kg/ha, a mean MAE of 885.1 kg/ha. Extreme climate indicators show a greater ability to explain yield variability in over half agro-ecological regions under higher warming levels such as SSP5-8.5. Cumulative precipitation (CPR) and compound drought and hot days (CDH) during growing seasons are the most important mean and extreme climate indicators affecting yield variability, respectively. Moreover, the effect sizes of CPR and CDH on yield variability are 1898 kg/ha and 5596 kg/ha, respectively. Therefore, CDH will be a key constraint on maize production. Future adaptive measures such as irrigation, breeding hot- or drought -tolerant cultivars should be implemented to enhance the resilience of maize crops in the face of climate change. |
在线留言
联系我们
所长信箱