Abstract:

Based on precipitation data observed at 291 national meteorological stations from June to September in 2019, we verified the precipitation forecasts from eight numerical models as well as the precipitation grid forecast released from the National Meteorological Center.We also studied the clear or rain forecasts based on the eliminating false alarm ratio (FAR) method.The results showed that the European Center for Medium-range Weather Forecast (ECMWF) model has the best precipitation forecast performance, and the Japan Model has the highest threat score (TS) among all the models.Some mesoscale models show good potential forecast on the local and short-time heavy rainfall events, among which the GRAPES_MESO model and the RMAPS Dongbei-3km model have better performance.The 24- and 3-h occurrence frequency of the heavy rainfall events predicted by the global model is 30% and 60% lower than the real situations, respectively, and that predicted by the mesoscale models is 30% higher and 20% lower than the real conditions, respectively.The false alarm ratio (FAR) for small precipitation events is high, leading to a high false alarm ratio for clear or rain events among the all models.The clear or rain forecast accuracy can be improved significantly using the method of eliminating FAR of small precipitation events.Based on this method, the ECMWF model has the best clear or rain forecast performance among the all models.The optimized eliminating thresholds for 24- and 3-h precipitation forecast are 1.0 mm and 0.8 mm, respectively, and the clear or rain accuracy can be promoted by 15.58% for the 24-h forecast, and by 10%-30% for the 3-h forecast.

Key words: Heavy rainfall, Precipitation forecast, Verification, Eliminating false alarm ratio of precipitation