Unsupervised classification of vertical profiles of dual polarization radar variables

Abstract. Vertical profiles of polarimetric radar variables can be used to identify fingerprints of snow growth processes. In order to systematically study such manifestations of precipitation processes, we have developed an unsupervised classification method. The method is based on k-means clustering of vertical profiles of polarimetric radar variables, namely reflectivity, differential reflectivity and specific differential phase. For rain events, the classification is applied to radar profiles truncated at the melting layer top. For the snowfall cases, the surface air temperature is used as an additional input parameter. The proposed unsupervised classification was applied to 3.5 years of data collected by the Finnish Meteorological Institute Ikaalinen radar. The vertical profiles of radar variables were computed above the University of Helsinki Hyytiala station, located 64 km east of the radar. Using these data, we show that the profiles of radar variables can be grouped into 10 and 16 classes for rainfall and snowfall events respectively. These classes seem to capture most important snow growth and ice cloud processes. Using this classification, main features of the precipitation formation processes, as observed in Finland, are presented.