Evaluating the accuracy of WRF multi-physics ensemble forecasts for heavy rainfall events in the Democratic People’s Republic of Korea

This study evaluates multi-physics WRF ensemble forecasts for extreme rainfall in the Democratic People’s Republic of Korea (DPR Korea), where complex topography and monsoonal dynamics drive localized heavy precipitation. Focusing on two 24-h events (2018 and 2020) exceeding 200 mm/day, we analyze 15 ensemble members with diverse physics configurations and introduce a probability matching (PM) method to address spatial displacement and intensity biases. Simulations using 3-km resolution reveal that PM outperforms simple averaging (SM) and deterministic members by preserving rainfall extremes and spatial coherence. PM reduced centroid displacement errors to 18 km (vs. 49 km in SM) and achieved a 72% area overlap with observations, while SM underestimated peak rainfall by 38% (208 mm vs. 310 mm observed). The PM method’s MAE (12.3–22.8 mm) and RMSE (18.7–24.3 mm) were 30–50% lower than SM, with higher TS (0.38 vs. 0.25) and ETS (0.42 vs. 0.33) at 30 mm/3 h thresholds. Superior performance stemmed from PM’s statistical alignment of ensemble intensity distributions and mitigation of double-penalty errors, particularly in resolving terrain-forced convection and low-level jets. Top-performing members (e.g., Member 7) combined graupel-aware microphysics and scale-aware convection schemes but still exhibited 20–30% underestimation. Overall, our analysis underscores the critical role of model selection and statistical calibration in accurately predicting precipitation patterns. The results indicate the value of advanced statistical methods, such as PM, for improving forecasting of extreme weather events, supporting the need for enhanced meteorological practices to bolster community resilience against flooding and water-related disasters.