Accurate prediction of winter surface air temperature (SAT) is crucial for proactive government responses to potential winter hazards. However, current state-of-the-art operational seasonal forecast systems still face challenges in accurately predicting winter SAT variations over China, with considerable model diversities in prediction skill. Here, we identified the critical physical processes that determine the prediction skill of winter SAT over China to better understand the sources of model diversity. The two leading EOF modes of interannual variability of winter SAT over China in observation are primarily driven by the strength of the Siberian High and the meridional displacement of the East Asian polar front jet (EAPJ). Additionally, sea ice conditions in the Barents–Kara region and the sea surface temperature (SST) in the Northwest Pacific also contribute to these leading modes. Although most models generally capture the spatial structure of the observed EOF patterns and their connections with critical atmospheric circulation drivers, substantial disparities exist in their ability to predict the principal components (PCs) associated with these modes. These disparities account for the differences in predictive skills for winter SAT across models. Among the prediction systems, those that more accurately predict the Siberian High index and Barents-Kara sea ice index (the EAPJ meridional displacement index and Northwest Pacific SST index) show the highest skill in forecasting PC1 (PC2), respectively. To improve prediction accuracy, we have developed a new statistical-dynamical method that combines model prediction results based on each model's ability to simulate and predict specific physical processes. This approach significantly enhances the skill of predicting winter SAT in China.

subseasonal prediction skill