1. Assessing the quality of short-term heavy rain forecasts of the IFS model in the Northern plains and midlands

Abstract

This study provides a detailed verification of the short-range (1 - 3 days) heavy rainfall forecast skill of the Integrated Forecasting System (IFS) operated by the European Centre for Medium-Range Weather Forecasts (ECMWF) over the Northern delta and midland region of Vietnam during 2014 - 2023. Using 24-hour accumulated rainfall observations from 39 surface stations and IFS forecasts at 24-, 48- and 72-hour lead times, a set of continuous error metrics (ME, MAE, RMSE, BIAS, correlation coefficient R) and categorical scores (POD, FAR, TS, BS) were computed for different rainfall thresholds (moderate, heavy and very heavy rainfall) and for distinct synoptic patterns. Statistical significance of correlation and skill scores was assessed at the 0.05 level. Results indicate that IFS provides reasonably accurate 24-hour rainfall forecasts at short lead times, with mean correlation coefficients around 0.7, RMSE of 18 - 25 mm, and BIAS close to unity, implying slightly under- or overestimated rainfall depending on the sub-region. Forecast quality, however, deteriorates markedly at 48 - 72-hour leads (R dropping to about 0.5 and RMSE exceeding 30 mm) and for very heavy rainfall events (>100 mm/24 h), where both missed events and false alarms remain substantial. Spatially, the model tends to underestimate rainfall over the midland areas and overestimate it along the coastal plains. Verification by synoptic pattern shows that rainfall associated with compressed low-level troughs and tropical cyclones is better captured than that related to cold surges with weak convergence or localized convective systems.