Correcting the precipitation ensemble forecasts using frequency matching based on cluster analysis over eastern China

Both the Frequency Matching Method (FMM) and cluster analysis are widely recognized as standard bias correction techniques for precipitation forecasts. Therefore, how to effectively integrate cluster analysis with FMM to improve precipitation forecasting skills emerges as a critical scientific issue. To improve the forecasting skill of heavy rainfall, this study leverages forecasts from the European Centre for Medium-Range Weather Forecasts Global Ensemble Prediction System during the summers from 2020 to 2024. In this study, ablation experiments are designed based on the clustering results of historical precipitation characteristics and a FMM approach that anchors frequency. The Clustering-Ensemble sampling-anchoring frequency FMM (Cluster-FMM in short) significantly improves the (probability) forecasting skill of heavy rainfall events. Notably, the 100 mm Fractions Skill Score increases by 30%—80%. Two extreme precipitation cases ("23·7" in North China and "24·7" in Hunan Province) show that the Cluster-FMM results effectively correct the dry biases in ensemble forecasts of heavy precipitation. The ablation experiments also confirm that the clustering analysis, ensemble sampling and anchoring frequency collectively contribute to the enhancement of forecasting accuracy for heavy precipitation events. For operational systems, the Cluster-FMM provides a new perspective for improving accuracy of ensemble precipitation forecasts.