Joint Assimilation of FY-3E Microwave Sounding and Ground-Based Microwave Radiometer Observations Incorporating AI-Based Cloudy-Sky Profile Retrievals over Complex Terrain on the Eastern Margin of the Tibetan Plateau

Forecasting convective precipitation over complex terrain requires accurate thermodynamic initialization. We developed a WRFDA 3D-Var framework combining nominally clear-sky FY-3E MWTS-3/MWHS-2 radiances, artificial-neural-network (ANN)-retrieved cloudy-sky thermodynamic profiles, and ground-based microwave radiometer (MWR) profiles. In a 5–15 July 2024 inner-domain sample, cloud routing assigned 66.1% of MWTS-3 and 66.9% of MWHS-2 fields of view to the cloudy-sky pathway. Independent radiosonde validation yielded RMSEs of 2.537 K for land temperature and 1.071 g kg−1 for bias-corrected land specific humidity, although a −1.748 K cold bias remained in temperature. Principal precipitation comparisons of four experiments (CTRL, FY, MWR, and FY-MWR) used a common CV7 background-error covariance configuration across nine daily forecasts over Chongqing during 7–15 July. During the principal 10–11 July event, MWR performed comparatively well at selected light-to-moderate precipitation thresholds, whereas FY-MWR had the highest TS at the highest 24 h thresholds. In pooled nine-day verification, CTRL retained the highest threat scores at 0.1–50 mm, while FY-MWR had the highest point estimate at 100 mm (0.083 versus 0.048 for CTRL); however, the paired bootstrap confidence interval for the FY-MWR-minus-CTRL difference included zero. Instability, convective inhibition, moisture-flux convergence, and vertical-motion diagnostics showed time-dependent relationships with precipitation differences. The results therefore indicate event- and intensity-dependent forecast responses rather than uniformly improved forecasts.