Critical Role of Mid-Tropospheric Southwesterly Steering in a Failed Extreme Rainfall Simulation over Northeastern Qinghai–Xizang Plateau

Accurate precipitation forecasting over the Qinghai–Xizang Plateau (QXP) remains a major challenge in numerical weather prediction. During 3–4 September 2024, extreme rainfall struck the Hehuang Valley in the northeastern QXP, with hourly precipitation exceeding 60 mm and daily records broken. Both global and regional models failed to capture the intensity and location of this event, motivating a diagnostic investigation into sources of the simulation error. This study uses a 3-km convection-permitting WRF simulation verified against surface observations, satellite precipitation, ERA5, and radar data. Mesoscale convective systems (MCSs) are tracked by using the PyFLEXTRKR algorithm. Three sets of sensitivity experiments are conducted: wind nudging toward ERA5 to isolate the role of winds from low-level moisture transport, a 31-member WRF ensemble to identify bias sources, and an initial wind-field replacement between a poorly-performing and a better-performing ensemble member. The model primarily failed to reproduce the observed northeastward-moving MCS, linked to a deficient mid-tropospheric southwesterly flow. Successful ensemble members systematically exhibited a stronger southwesterly, with a more eastward-extended westerly trough and a deepened mesoscale low. The initial wind-field replacement experiment confirmed that correcting the initial large-scale wind field recovered the southwesterly steering flow and significantly improved the precipitation simulation. By contrast, artificially enhancing the southeasterly moisture transport via wind nudging toward ERA5 only marginally improved the rainfall and did not correct the MCS track, demonstrating that the southeasterly wind bias was not the primary error source. The enhanced southwesterly flow steered the MCS into the Hehuang Valley, where orographic and frontal lifting, along with abundant moisture, jointly intensified convection. The simulation failure is mainly attributable to a deficient mid-tropospheric southwesterly steering flow originating from errors in the initial large-scale wind field rather than to an underestimation of southeasterly moisture transport. Accurate representation of this steering flow is therefore essential for forecasting extreme precipitation in complex terrain. The findings highlight the value of ensemble-based sensitivity diagnostics and suggest that targeted assimilation of upper-air wind observations could improve operational precipitation forecasts over the northeastern QXP.