Understanding heavy precipitation prediction bias in the Sichuan Basin based on the <scp>China Meteorological Administration</scp> operational mesoscale model

Abstract Heavy precipitation events (HPEs) are high‐impact weather phenomena, the prediction of which over complex terrain remains a substantial challenge for the scientific community. This work focused on the bias in prediction of HPEs over the Sichuan Basin (SCB) in China in the summers of 2019–2022. Three synoptic weather patterns (SWPs) were identified using a modified T‐mode principal components analysis of 37 HPEs with the western Pacific subtropical high (WPSH) located to the southeast (WPSH‐SE), east (WPSH‐E) and south (WPSH‐S) of SCB. We found that the position of the WPSH could substantially influence the occurrence of HPEs among the two major SWPs (WPSH‐SE, WPSH‐E). To further understand HPE prediction bias, we compared observed precipitation with that forecast by the operational numerical weather prediction (NWP) mesoscale model of the China Meteorological Administration (CMA‐MESO). Results revealed that overestimation of the precipitation area occurred in the morning, whereas underestimation of the precipitation amount occurred at night, with minimal differences between the two major SWPs. Additionally, the precipitation bias of both main SWPs was lower during the afternoon (0400–0800 UCT, i.e. 1200–1600 Beijing time). Further analysis revealed a strong relationship between the integrated water vapor flux deviation and the precipitation bias in the SCB. Major contributors to this deviation were identified as the overestimated wind speed and clockwise deviation in the wind direction in the boundary layer over the mountainous Yunnan–Guizhou Plateau, possibly attributable to lack of subgrid orographic drag. The findings of this study contribute to better comprehension of the relationship between HPEs and low‐level winds over complex terrain, thereby establishing a foundation for further improvement in the subgrid orographic and boundary layer schemes of operational NWP models.