Evaluating the phase evolution of CMIP GCMs for agricultural climate-change impact assessments in China

Abstract

The performance of general circulation models (GCMs) in the Coupled Model Intercomparison Project (CMIP) critically determines the reliability of climate-change impact assessments and has continuously progressed (e.g., from CMIP3, CMIP5 to CMIP6). It remains unclear whether this progression enhances the reliability in evaluating the effects of climate change on agricultural systems at a daily resolution, particularly concerning crop production. To address this question, the study selected AquaCrop as a crop model for large-scale agricultural impact assessment due to its compatibility, robustness, and simplicity. Subsequently, the study coupled AquaCrop with multiple GCMs from different CMIP phases: 9 from CMIP3, 14 from CMIP5, and 15 from CMIP6, and attributed GCM-driven crop yield simulations to GCM biases over China. According to the modeling results, the progression enhanced the simulation performance for daily precipitation and temperature. The impacts of CMIPs on assessment results exhibited variability across temporal scales and crop types, further modulated by water management practices. Overall, crop simulations across three CMIP phases revealed a reduction in cold and water stresses, a shortened growing period (particularly evident in CMIP6), and an underestimation of yields. The evolution of CMIP phases increased spatial-temporal correlations for maize (0.61 to 0.81), wheat (0.68 to 0.77), and rice (0.63 to 0.77), without significantly reducing yield biases. Yield biases in early growth period were primarily influenced by daily temperature fluctuations, while biases in latter growth period were correlated with precipitation and maximum temperature. Irrigation mitigated the crop model's sensitivity to precise daily precipitation data compared to rainfed systems. This comprehensive analysis suggests, when evaluating climate change impacts on agriculture—at least for Chinese crops—CMIP6 better captured regional and temporal yield distributions than earlier phases, despite potentially underestimating yields and growth periods in certain regions.