Data assimilation for tropical cyclone-induced rainfall forecasting for central Viet Nam

Abstract

Data assimilation plays a particularly important role in numerical weather prediction (NWP) models. It ingests observational data into the initial fields of NWPs, improving initial conditions to better represent actual atmospheric states and consequently enhancing forecast accuracy. Globally, three-dimensional variational (3DVAR) and four-dimensional variational (4DVAR) assimilation methods have been widely applied, with 4DVAR considered the most advanced technique. This study focuses on data assimilation experiments using the WRFDA system with the WRF-ARW core, including cases without data assimilation as well as 3DVAR and 4DVAR schemes with different assimilation windows. Evaluation of the initial fields shows that both 3DVAR and 4DVAR effectively improve model initial conditions through observational data assimilation. Specifically, the RMSE values of the 3DVAR and 4DVAR analysis fields are consistently lower than those of the original model initial fields for all variables, including wind, temperature, moisture, and pressure. The results further indicate that 4DVAR provides greater improvements than 3DVAR under the same conditions, with generally smaller RMSE values. Verification using the probability of detection (POD) and false alarm ratio (FAR) demonstrates that the forecast skill of the 4DVAR scheme is superior to that of 3DVAR, particularly for heavy rainfall thresholds exceeding 50 mm, as clearly reflected in the 24-hour accumulated rainfall forecasts.