Comprehensive global gridded crop model improvements reduce the uncertainty of extreme climate impact assessment

Extreme climate events like drought and heatwave are increasingly co-occurring and considerably threaten global food security. Global gridded crop models (GGCMs) are widely used to assess the impacts of climate extremes on crop yields; however, in which way and to what extent the uncertainty of assessment can be reduced remains largely unknown. Here, we jointly improve the CERES-Wheat model from model inputs, structure, and parameterization at 10-km resolution to reduce the uncertainties globally. The improved model parameterization remarkably increase the model explanatory power of observed global wheat yield losses from drought, heatwave, and their compounds during 1981-2015 by 25% to 60% compared to the multi-model ensemble (MME) approach. Improved temperature response functions for key physiological processes particularly contribute to a better representation of wheat response to heatwave by 20%. Taking 2003 European drought and heatwave events as examples, the improved model is capable of closely replicating the observed yield declines (> 90%), whereas most of the existing GGCMs fail to show any impact and MME merely explains < 25% of the reported influences. Our findings provide the first evidence for comprehensively constraining crop model uncertainty in extreme climate impact assessment, benefiting the accurate understanding of climate risk and the design of effective adaptation strategies.