Abstract:The European Centre for Medium Range Weather Forecasts (ECMWF) high resolution deterministic forecasts (HDet) and probability forecasts from the ensemble prediction system (EPS) are evaluated and compared for the heavy rainfall from May to September 2012 in Huaihe River Basin, China. For the EPS, the daily precipitation extreme forecast index (EFI) and the calibrated probability calculated by the modified Bayesian model averaging (MBMA) are used. Relative operating characteristic (ROC) verification which is independent of forecast bias is used. The TS vs. false alarm rate (FAR) and the relative economic value are measured. The results indicate that the MBMA has the best performance, followed by the HDet and then the EFI. For the forecast within lead times of 2 days, the HDet is close to MBMA. As the lead times prolong, the advantage of MBMA relative to HDet keeps increasing. Under different criteria, the heavy precipitation forecast thresholds for the three methods are estimated. It is shown that the MBMA is the best and the EFI is the worst. However, the advantage of MBMA is accompanied by higher forecast bias. If the forecast bias is set to be 1.37, which is equal to the subjective forecast of National Meteorological Centre (NMC) forecasters, the MBMA is almost similar to HDet.