Machine‐learning approaches for characterizing the raindrop size distributions in Western Pacific tropical cyclones

Abstract Long‐term observations of raindrop size distributions (RSDs) in the WPTCs are leveraged to enhance estimates of rainfall rate ( R ) and slope parameter. The radar reflectivity–rainfall ( Z – R ) relationship for the WPTCs exhibits a distinct dependence on the mass‐weighted mean diameter ( D m ). Machine‐learning techniques, including Random Forest (RF), XGBoost, and Decision Tree, were applied to estimate R , with RF demonstrating superior performance. Additionally, a hybrid moment‐based approach was employed to elucidate the interrelationships among the gamma distribution parameters (slope, shape, and intercept), revealing strong predictability for the slope and intercept. These parameters showed a close association with liquid water content. Compared to linear regression, the RF method yielded more accurate estimates of the slope parameter. These results indicate that RF enhances both rainfall rate and slope parameter estimation, thereby improving precipitation forecasts and cloud modeling for the WPTCs.