Stepwise correction of ECMWF ensemble forecasts of severe rainfall in China based on segmented hierarchical clustering

Abstract

Ensemble forecast plays a vital role in numerical weather prediction. Hence, effectively extracting useful information from ensemble members to improve precipitation forecasting skills has always been an important issue. Using the ensemble forecast data on precipitation from the ECMWF-GEPS (Global Ensemble Prediction System), we propose a stepwise correction method, based on segmented hierarchical clustering (SHC), for forecast of daily precipitation. This method employs a segmented correction scheme, thereby generating more probabilistic forecast information and improving forecasts. Validations of the SHC method have been performed by comparison with two other methods, namely the ensemble-mean (EM) method and the direct hierarchical clustering (HC) method. Our results showed that deterministic forecast via SHC improved the ability to forecast heavy precipitation in short- and medium-range forecast timeframes. Therefore, SHC performed better than either EM or HC by effectively extending lead time to impending severe rainfall by 2–3 days relative to the other two methods. SHC also demonstrated better performance than the other methods through continuous forecast verification in summer 2021, and even had better effects in the forecast of multiple heavy-precipitation cases, including the Zhengzhou extreme rainfall on 20 July 2021. Overall, the SHC method has great potential for improving ensemble rainfall forecasts in the current operational system.