Assessment of the Relative Quality of Hydrometeor Types Inferred from Polarimetric Radar Observations using Fuzzy Logic Techniques

A number of fuzzy logic based hydrometeor identification (HID) algorithms have been developed for use on polarimetric radar observations of precipitation at a variety of weather wavelengths. These fuzzy logic HID algorithms have been in use for nearly 20 years. Despite their wide use, absolute validation of these HID algorithms has been limited, especially in deep convection, due to the difficulty of obtaining in situ measurements of precipitation type in such harsh measurement conditions. Ongoing and proposed future efforts should continue to prioritize developing new in situ precipitation measurement capabilities in deep convection. However, the purpose of this study is to assess the relative quality of the microphysical information provided by an HID algorithm. More specifically, the discriminating power or uniqueness of the HID output is assessed by statistical evaluation of the crisp, singleton output relative to the entire fuzzy output set. This crisp output is typically the hydrometeor type with the maximum rule strength while the fuzzy set represents the range of potential hydrometeor type outcomes of varying rule strengths generated by the HID system at each storm location. For example, in addition to knowing the more typical, crisp output (or first ranked hydrometeor type), it is of interest to know the second ranked hydrometeor type in the fuzzy set at each radar grid point and the percent difference in rule strengths between the two. The percent difference in rule strengths between the top two ranked hydrometeor types in the fuzzy set at each grid point is related to the relative discriminating power of the HID. This and similar information from the fuzzy set could be useful for assessing the relative uncertainty in the crisp output of hydrometeor type at each storm location. This type of analysis could also provide additional insight into other hydrometeor types that might be present at each location. While the issues of overlapping polarimetric radar membership functions of various hydrometeor types and the potential non-uniqueness problem are widely recognized in the radar meteorology community, they are only rarely addressed in the analysis and interpretation of HID crisp output. This study extends recent analyses in Europe by analyzing the fuzzy set output of the Colorado State University (CSU) HID algorithm for S- and C-band polarimetric radar observations of deep convection in the United States. Particular attention is paid to mixed-phase and ice-phase precipitation regions of deep convection where less is known about the discriminating power of HID algorithms. Although absolute validation is not necessarily available for these deep convective cases, physical consistency with other observations (e.g., lightning) and expert human analyses will also be used to constrain the results.