Abstract

This study presents an experimental evaluation of heavy rainfall forecasting using the Weather Research and Forecasting (WRF) model over the southeastern area of Vinh Long Province (formerly Tra Vinh Province) through six representative rainfall events during the period 2021–2023. The selected events represent typical synoptic patterns responsible for heavy rainfall in Southern Viet Nam, including the Intertropical Convergence Zone or tropical cyclones over the East Sea associated with enhanced southwest monsoon activity, the equatorial trough combined with upper-level easterly disturbances, and strong southward cold-air intrusions. The WRF model was configured with four nested domains, with the innermost domain having a horizontal resolution of 1 km to better simulate convective rainfall. Initial and boundary conditions were provided by the Global Forecast System (GFS) and dynamically downscaled to high spatial resolution over the study area. Model outputs were verified at eight rain-gauge stations using statistical error metrics, including Mean Error (ME), Mean Absolute Error (MAE), and Root Mean Square Error (RMSE), together with binary forecast skill indices such as Probability of Detection (POD), False Alarm Ratio (FAR), and Critical Success Index (CSI) for rainfall thresholds of 16–25 mm, 25–50 mm, and greater than 50 mm.

The results indicate that the WRF model demonstrates relatively good forecast skill for moderate rainfall thresholds of 16–25 mm, particularly at the 24-hour forecast lead time, with several stations achieving relatively high POD and CSI values. However, forecast skill decreases significantly with increasing forecast lead time and rainfall intensity thresholds. For rainfall thresholds of 25–50 mm and above 50 mm, POD and CSI values decrease markedly, especially at the 72-hour forecast lead time, indicating limitations of the model in forecasting heavy and extreme rainfall events. Forecast errors also differ between coastal and inland stations. Overall, the study confirms the potential of the WRF model for heavy rainfall forecasting in Southern Viet Nam and highlights the need for further studies on post-processing correction approaches, including machine learning and artificial intelligence techniques, to improve extreme rainfall forecasting performance in future applications.