DREE-RF: A Radar-Based Rainfall Energy Estimation Model Using Random Forest

Current radar techniques focus on rainfall observations, leaving a research gap in rainfall energy (E) involving the interaction of raindrops and land surface processes. E is defined as the accumulated kinetic energy per unit of rainfall and is a key parameter in the understanding process of the rainfall impact on the land surface. Utilizing the capability of dual-polarization radar to detect the rainfall microphysics characteristics, this study proposes the first computational model for estimating E from radar signals. The model investigates the mechanistic correlation between the radar dual-polarization parameters and E and finds that specific differential phase ( $K_{\text {DP}}$ ) and horizontal reflectivity ( $Z_{\text {H}}$ ) have the strongest correlation with E. Therefore, the study develops radar-based empirical regression and random forest (RF) models for E estimation, where RF models consider whether the sensitive $K_{\text {DP}}$ is available. The results show that the RF models improve the accuracy of estimating E and have a Pearson coefficient greater than or equal to 0.97 with station-measured E, and the spatially extensive capability of the models is further validated. In addition, both the traditional regression model (TRM) and RF-based radar data have underestimated daily E estimates compared to the disdrometer observations, with smaller BIAS and root mean square error (RMSE) and higher Pearson correlations for RF. This study contributes to enhancing the understanding of rainfall processes in the context of climate change and has great potential for applications in hydrological modeling, flood forecasting, and agricultural planning.