A Cluster-Based Method for Hydrometeor Classification Using Polarimetric Variables. Part II: Classification

Two new algorithms for hydrometeor classification using polarimetric radar observations are developed based on prototypes derived by applying clustering techniques (Part I of this two-part paper). Each prototype is defined as a probability distribution of the polarimetric variables and ambient temperature corresponding to a hydrometeor type. The first algorithm is a maximum prototype likelihood classifier that uses all prototypes attributed to the different hydrometeor types in Part I. The hydrometeor type is assigned as the prototype with the highest likelihood when comparing the polarimetric variables and temperature with each prototype. The second algorithm is a Bayesian classifier that uses the probability density functions (PDFs) as derived from the prototype set associated with the identical hydrometeor type. The posteriori probability in the Bayesian method is calculated from a combination of the PDFs and the prior probability, the maximum of which corresponds to the most likely hydrometeor type. The respective merits of the two techniques are discussed. The two classifiers are applied to CP-2 S-band radar observations of two hailstorms that occurred between 16 and 20 November 2008, including the so-called Gap storm, which produced a devastating microburst and large hail at the ground. Results from the classifiers are compared with those derived using the well-established National Center for Atmospheric Research fuzzy logic classifier. In general, good agreement is found, yielding overall confidence in the robustness of the new classifiers. However, large differences are found for the melting ice and ice crystal categories, which will need to be studied further.