Extraction of Spectral Polarimetric Features With Weather Radar and Its Application in Vertical Wind Shear Identification

Dual-polarization weather radar (DPWR) is an effective tool for meteorological monitoring. DPWRs use spectral moment estimates to retrieve hydrometeor-based radar products and identify weather features through their spatiotemporal evolution. However, these product data cannot directly resolve the microphysical and dynamical characteristics of hydrometeors, which are crucial for meteorological monitoring. Spectral polarimetry, in which polarimetric variables are represented as functions of Doppler velocity, has recently been applied for qualitative analyses of hydrometeor microphysical growth, dynamical properties, and storm evolution in severe convection. In particular, previous studies have shown that small-scale vertical wind shear (VWS) in precipitation, a factor that poses a significant threat to aviation safety, can induce hydrometeor size sorting, resulting in a distinct slope in the spectral differential reflectivity (sZDR). Nevertheless, quantitative identification of VWS using spectral polarimetry remains limited. To address this issue, this study explores the application of spectral polarimetry for the quantitative detection of small-scale VWS. A feature extraction framework is developed to derive spectral polarimetric features within individual radar resolution volumes, including the slope and texture of sZDR, as well as higher-order moments (skewness and kurtosis) of the Doppler power spectrum. Then, the feasibility of spectral polarimetric features for characterizing VWS is verified through simulations. Furthermore, an eXtreme gradient boosting (XGBoost) model is employed to identify VWS using over 300 radar volume scans from convective events of different types observed on different days. The results show that incorporating spectral polarimetric features improves classification accuracy (ACC) by 9.55% compared to using conventional radar products alone. Finally, the proposed method is verified with specific cases, which show that the ACC can reach 80%. These results highlight the significant potential of spectral polarimetric features for enhancing VWS detection.