An Ensemble Sensitivity Analysis for Optimizing Physical Schemes in Summer Heavy Rainfall Predictions Over DPR Korea

Abstract

Extreme summer rainfall events pose a major hydrological hazard in the Democratic People's Republic of Korea (DPRK), where majority of annual heavy precipitation occurs during the July–August monsoon season. Accurate prediction of these events is essential for flood early warning and disaster risk reduction, yet remains challenging due to complex terrain and uncertainties in model physics. This study addresses a critical gap by systematically evaluating the sensitivity of the Weather Research and Forecasting (WRF) model to combinations of physical parameterization schemes for simulating 15 major summer heavy rainfall events between 2011 and 2022. Using a dense observational network of 130 rain gauges and a multi-event ensemble approach, 16 physics configurations were tested—spanning four microphysics, two cumulus, and two planetary boundary layer schemes—and their performance was compared against operational Global Forecast System (GFS) forecasts. The Lin microphysics, Kain–Fritsch cumulus, and YSU planetary boundary layer combination (Lin–KF–YSU) consistently outperformed all others, achieving the highest spatial correlation (0.68), lowest root-mean-square error (10.2 mm), and best threat score (0.38)—a statistically significant improvement over GFS. While all simulations showed some underestimation of peak intensities above 400 mm/day, likely due to unresolved microphysical and terrain effects, the optimal configuration captured event timing, spatial structure, and rainfall totals more reliably across diverse synoptic conditions. These results demonstrate that regionally tuned, convection-permitting WRF simulations offer substantial added value over global models for extreme rainfall prediction in complex terrain, while maintaining or improving performance for standard meteorological variables. For operational forecasting in monsoon-affected regions like the DPRK, adopting such optimized configurations can meaningfully reduce false alarms and missed events within the context of extreme rainfall warning systems—enhancing public safety and resilience. This work underscores the importance of localized model validation and the potential for high-resolution numerical weather prediction to support effective climate adaptation in vulnerable areas.