Revealing the Unexploited Potential of Dual‐Polarization Radar in Rainfall Kinetic Energy Prediction

Abstract Accurately predicting rainfall kinetic energy (RKE) is critical for assessing soil erosion and mitigating related environmental hazards. Traditional methods rely on rainfall intensity ( R ) as a proxy, oversimplifying the complex microphysical processes of raindrop formation. In contrast, our study introduces a physically based framework that leverages dual‐polarization radar (DPR) variables—capturing key raindrop properties—to enhance RKE prediction through methods ranging from statistical regression to advanced machine learning. Statistical regression models with DPR variables show superior accuracy over traditional R ‐based methods in RKE prediction. Meanwhile, machine learning models, developed through four algorithms to create 12 advanced models, surpass linear DPR models by better handling extreme conditions. Among these, the Local Cascade Ensemble model utilizing DPR variables and simple environmental conditions (LCE‐DPRS) stands out for its balance of effectiveness and ease of use, making it the recommended approach for RKE estimation. Specifically, this model achieves reductions exceeding 70% in both root mean square error (RMSE) and mean absolute error compared to traditional rainfall R ‐based methods. Additionally, 1 month of X‐band dual‐polarization radar observations was validated using in situ raindrop size distributions measured by disdrometers. The LCE‐DPRS model demonstrated effective high‐resolution, real‐time spatiotemporal predictions, significantly reducing errors during intense rainfall events. This study establishes a new benchmark for leveraging radar technology in hydrological forecasting.