Precipitation Classification Using the Lowest Tilt Radar Data with a Support Vector Machine Method

A precipitation classification approach using support vector machine method is developed and tested by multiple precipitation events. Different from some existing classification approaches utilizing a whole volume radar data, the new developed approach uses the lowest tilt unblocked polarimetric radar data to classify precipitation echoes into stratiform or convective types. In this approach, the lowest tilt radar variables of reflectivity, differential reflectivity, standard deviation of reflectivity, and the separation index derived from drop size distribution data are utilized as the inputs of the proposed approach. A support vector machine approach is then used in the classification, where the feature vector and weight vector in the support vector machine is trained using well classified precipitation types. The proposed approach is trained and tested on a C-band polarimetric radar located in Taiwan (RCMK) with stratiform, convective, and mixture precipitation events, and the results are compared with two existing approaches. In the evaluation, the classification results from Multi-radar/Multi-Sensor system are used as the reference.