A Bayesian Approach to Identify Non-hydrometeor Echoes Using Single-Polarization Radar Data

Polarimetric radar observations provide information regarding hydrometeor’s shape, size, and phase as well as an improved skill in differentiating radar returns of precipitation from those of non-precipitation. Radar data quality control (QC) of WSR-88D radars has been greatly improved with the upgraded of polarimetric capability. However, comparing to single polarization variables, the polarimetric measurements are more sensitive and vulnerable to corruptions, which has negative impacts on radar QC process, hydrometeor classification, quantitative precipitation estimation (QPE), flash flood warning, and aviation products in the Multi-Radar-Multi-Sensor (MRMS) system. In order to enhance the robustness of the radar QC and mitigate the negative impacts from corrupted polarimetric variables, an automatic system was developed, which applies a Bayes based approach as a supplement using the reflectivity and Doppler variables only. The contamination of dual polarization data occurs intermittently at low frequency, and the Bayes method’s efficiency and effectiveness fit it a good complement. The Bayes method is trained with a supervised learning process, where the training example is real-time updated using the un-contaminated data. This work demonstrates severe negative impacts on dual polarization processes including radar data QC, hydrometeor classification, severe weather forecasting, and etc., from corrupted dual-polarization measurements. The performance of the proposed approach was demonstrated using several precipitation events.