Adapting the EcPoint Post-Processing methodology for ensemble rainfall forecasts to CMA-GEPS numerical model, a Regional Calibration over the Yangtze River basin in eastern China

Accurate probabilistic rainfall forecasting remains a major challenge, particularly in regions characterized by complex topography and high climatic variability. Small errors arising from observations, data assimilation, and model configuration can grow nonlinearly, leading to rapid loss of forecast skill with time. Traditionally, rainfall forecasts have been generated at grid-based spatial resolutions however, Weather varies markedly within a grid box and therefore grid-based ensemble forecasts often struggle to capture sub-grid variability and localized extremes, limiting their usefulness for operational forecasting at specific points of interest. The EcPoint post-processing approach, developed at the European Centre for Medium-Range Weather Forecasts (ECMWF), addresses this limitation by tailoring ensemble forecasts to point-specific locations. This study adapts the approach to the China Meteorological Administration Global Ensemble Prediction System (CMA-GEPS) through regional calibration over the Yangtze River Basin covering provinces of Anhui, Jiangsu, and Zhejiang. The performance of the GEPS EcPoint was compared with both the Raw GEPS and the original ECMWF EcPoint system using probabilistic verification metrics. A case study of the 30 July 2022 Xuzhou flooding event was also studied to assess spatial rainfall representation.The study highlighted the added value of the post-processing method, five months of verification demonstrated that between the Raw ensemble and post-processed forecasts, EcPoint is more reliable and skillful markedly improving forecast performance across all rainfall thresholds and lead times, with significant reductions in CRPS and higher AUC values relative to the Raw ensemble. The Xuzhou case study further illustrated EcPoint’s ability to capture localized heavy rainfall with greater spatial accuracy and reducing false alarms compared to the Raw ensemble. EcPoint reasonably forecasts the spatial coverages of the rainfall event while the Raw GEPS presents a notable overestimation. The ECMWF EcPoint, which incorporates a broader range of predictors such as solar radiation, exhibited the higher skill and a stronger diurnal cycle. Nonetheless, the GEPS EcPoint demonstrated substantial value-added skill, confirming the robustness and transferability of the EcPoint methodology. Overall, the findings highlight EcPoint’s effectiveness as a model-independent, data-driven tool for enhancing ensemble rainfall forecasts. Its demonstrated adaptability across different ensemble systems underscores its potential for operational adoption, offering improved reliability for flood risk management and early warning applications.