The regional data assimilation system based on the Ensemble Kalman filter (EnKF) with the flow-dependent error statistics is expected to better use observations, which leads to better severe weather prediction. To capture the heavy precipitation during the Mei-Yu season in Taiwan, the key elements include representing the prevailing southwesterly flow and the local convergence over coastal area of Taiwan. High-resolution simulation and assimilation are required to improve a multi-scale convective system, including the topography effect. Due to sampling errors, EnKF with high-resolution grids applies a shorter localization scale to limit the influence of the observation. However, applying a shorter localization scale on observations in the high-resolution grids might cut off the large-scale moisture transport, which is essential to storm development. In this study, we applied the dual-localization method proposed by Kondo and Miyoshi (2013) to highlight the importance of multi-scale correction for predicting such kind of multi-scale events with multi-resolution nesting model grids.
Results suggest that the dual-localization scheme helps to better use the flow-dependent error statistics of the regional EnKF system and optimize the corrections from observations like radio sondes and GPS Radio Occultation data. Compared with the prediction using the single-scale localization, the synoptic-scale moisture correction is now better preserved and the intensity and location of heavy precipitation prediction are significantly improved.