Raindrop Size Distribution Retrieval Model for X-Band Dual-Polarization Radar in China Incorporating Various Climatic and Geographical Elements

The X-band dual-polarization radar can provide raindrop size distribution (DSD) data at a continuous spatiotemporal scale; however, the traditional empirical-formula-based DSD retrieval methods for X-band radar are limited by high uncertainty and poor generality such as differences due to rainfall type, seasons, and geographical environment. This study identified strong correlations between polarimetric radar variables and various climatic and geographical elements. Two random-forest-based DSD retrieval models, one using only X-band radar parameters as the model input (RF-DSD) and the other containing multielement inputs as well (RF-M-DSD), were established using datasets from 18 sites in China. This study also evaluated the proposed models using different error indices over three test sites. Compared with the traditional DSD retrieval model for X-band dual-polarization radar, RF-DSD and RF-M-DSD showed improved predictive ability for all DSD parameters. The root-mean-square error (RMSE) values for $D_{0}$ (median volume diameter parameter), ${\text {log}}_{10}N_{w}$ (normalized intercept parameter), and $\mu $ (shape parameter) were 27%, 55%, and 39% lower, respectively, for RF-DSD, and 48%, 68%, and 43% lower, respectively, for RF-M-DSD. Finally, the DSD retrieval models were applied on one CLA-12A X-band radar in the Jiangsu Province. The DSD retrieval results suggest that the proposed models perform well on radar observation data with low uncertainty especially for $D_{0}$ and ${\text {log}}_{10}N_{w}$ . Summary reveals that the RF-M-DSD performs superior and has stable DSD retrieval capabilities across different geographical areas, which can provide reliable radar-DSD retrieval data for areas with scarce disdrometer data.