Previous morphological studies of specific types of precipitating weather systems have typically relied on visual inspection of weather data. However, in order to analyze massive amounts of meteorological data, automated techniques for identification of weather systems are required. Precipitation threshold-based techniques have shown some level of success in this area, but they are not able to discriminate between different regions of atmospheric conditions within a contiguous area of precipitation greater than the given threshold. In order to separate various regions that have similar atmospheric conditions, such as convection along a cold front, stratiform rain/snow to the north of a connected warm front, etc., multivariate cluster analysis-based identification techniques appear to be quite promising. Marzban and Sandgathe (2006) demonstrated how cluster analysis could be used in QPF verification and in the identification of precipitating regions that are similar in terms of precipitation amount and spatial proximity. In this paper, additional meteorological variables will be included and multivariate cluster analysis techniques will be used to determine if regions can be successfully identified based on atmospheric conditions, spatial/temporal rainfall characteristics, and spatial proximity.

For this work, similarity measures based on the conditional distribution function (CDF) for each variable will be tested. In order to perform this task, the CDF for each variable (such as: CAPE, wind shear, temperature advection, etc.) will be determined by analyzing the North American Regional Reanalysis (NARR) data. NARR analyses will be used along with precipitation information from high-resolution analyses (Stage IV radar/gage 4km hourly U.S. domain) to determine if multivariate cluster analysis techniques can be used to enhance procedures for identification and characterization of precipitating weather systems.