Abstract This study proposes a new scheme combining ensemble blending and dual‐localization of ensemble error covariance in an ensemble‐variational (EnVar) framework to better account for both the multi‐scale characteristics of the background and background error covariance for high‐resolution data assimilation and forecasting. The ensemble blending introduces large‐scale analysis information from a 30‐member global model's ensemble into the regional model's background and ensemble‐based background error covariance using a low‐pass Raymond tangent implicit filter. The dual‐localization in EnVar also employs two different localization radii at meso‐ α and meso‐ β scales respectively to determine multiscale analysis increments through the scale‐blending background error covariance. We evaluate the impact of the new scheme on the assimilation of multi‐source observations including all‐sky Fengyun‐4A (FY‐4A) Advanced Geostationary Radiation Imager infrared radiances and on forecasting two convective events in the Yangtze‐Huaihe region influenced by different weather systems using a 4‐km model. Results show that while ensemble blending or dual‐localization in the EnVar framework alone provides improvements although the former contribution more than the latter, the proposed scheme combining them yields the best performance in the simulation of the ambient environment, moisture transport, and thermodynamic vertical structure. Notably, the proposed scheme enhances weak and low‐level updrafts and provides more accurate convection regions, improving precipitation forecasted patterns and fraction skill scores. In addition, the proposed scheme demonstrates greater efficacy in convections driven by multiple interacting weather systems across scales than a single local weather system.
Wang et al. (Fri,) studied this question.