Probabilistic Precipitation Forecasting Using a Two-Parameter Model Based on Ensemble Model Output Statistics

Improving statistical post-processing methods to generate more reliable probabilistic precipitation forecast has great significance in hydrometeorological applications. In this study, the two-parameter generalized extreme value (GEV) distribution functions are used to model precipitation amounts and forecast ensembles of the same size as original ensemble precipitation forecasts are generated based on ensemble model output statistics (EMOS). Sensitive study shows that the distributions of forecast ensembles are mainly influenced by the location parameter and scale parameter of GEV. The proposed calibration method is applied to 24 h accumulated precipitation over Jiangsu Province of China from June to August 2017 using the ECMWF ensemble precipitation forecast. Verification results show that the calibrated ensemble precipitation forecast has less probabilistic forecast error and more probabilistic forecast skill for large and torrential rain events. Further study shows that the improved probabilistic forecast skill is mainly caused by the improved forecast reliability.