Leveraging Multi-Scale Radar Data Assimilation to Investigate Multi-Scale Convection Events in Taiwan

Ensemble data assimilation applies error covariance localization to mitigate sampling errors and rank deficiency problem. To capture convective-scale signals while avoiding spurious correlations, the traditional approach uses an order of 10-km localization scale for all variables. However, the convection initialization and development are influenced by larger-scale environment flows, whose errors can affect the location and intensity of convection. Thus, achieving accurate synoptic-scale forcing and convective-scale information becomes essential to represent the multi-scale characteristics and cross-scale interactions. In this study, we reconstruct multi-scale convection events in Taiwan using a radar ensemble data assimilation system that integrates the Weather Research and Forecasting model and Local Ensemble Transform Kalman Filter (WRF-LETKF). This system assimilates radar reflectivity and radial wind data with a 15-min analysis interval at a 3-km horizontal grid-spacing. To extend the localization radius without introducing spurious correlations, we leverage the successive covariance localization method (SCL) as the rapid multi-scale correction technique for radial wind assimilation.

We explore the impact of this approach on analyses and short-term precipitation predictions for different multi-scale convection events over the complex terrain. The events involved synoptic-scale front, local circulation and topography effect, and produced heavy rainfall in Taiwan. The parameters of SCL are chosen to mimic the multi-scale characteristics that encompass meso-α (2000-200km), meso-β (200-20km), and meso-γ (20-2km) scales by adjusting localization radii based on varying data density from large to small scale. Additionally, variable localization is adopted when assimilate different types of observations. Compared to the convective-scale correction, multi-scale correction provides more accurate larger-scale wind patterns, which facilitates accurate convergence field for local convections and further enhances the moisture convergence in regions of heavy rainfall area. The verification of short-term precipitation predictions shows overall improvement in both deterministic and ensemble forecasts after applying the multi-scale correction method. Our results highlight the importance of the additional adjustment of the environmental wind on a 100 km scale in strong synoptic-scale forcing convection events. The performance of traditional convective-scale ensemble data assimilation suffers from the inaccuracy of environment flow.