Evaluation of IFS Model for Tropical Cyclone Forecasting in Vietnam: Track, Intensity, and Rainfall (2018–2022)

Abstract

This study evaluates the forecasting performance of the Integrated Forecasting System (IFS) model developed by the European Centre for Medium-Range Weather Forecasts (ECMWF) in predicting tropical cyclones (TCs) affecting Vietnam during the period 2018–2022. The assessment focuses on TC track, intensity, and rainfall forecasts, employing best-track data from the Regional Specialized Meteorological Center (RSMC) Tokyo and rainfall observations collected from 186 meteorological stations across Vietnam. The results show that the mean direct position error (DPE) increases with forecast lead time, from 71 km at 24 hours to 118 km at 48 hours and 183 km at 72 hours. A statistically significant downward trend is observed at the 48-hour forecast horizon, indicating improvement in track accuracy. Overall, these errors remain within the acceptable operational thresholds for Vietnam, confirming the model’s applicability for routine forecasting. In terms of TC intensity, forecast errors show limited variation across different lead times, with the mean absolute error (MAE) of V_max averaging 9 knots (4.63 m/s) and the root mean square error (RMSE) of V_max averaging 12 knots (6.17 m/s). Both values are consistent with national standards; however, a few cases exhibited substantial deviations, particularly for rapidly intensifying cyclones. For TC-induced rainfall, the IFS model demonstrates the highest skill in forecasting light rainfall (<16 mm) across all lead times, whereas predictive skill declines notably for heavier rainfall categories. At a 72-hour lead time, both the threat score (TS) and the probability of detection (POD) drop below 0.10 for rainfall exceeding 50 mm, underscoring considerable limitations in forecasting heavy and very heavy precipitation events.