Unsupervised Classification of Vertical Profiles of Dual-Polarization Radar Variables

The size, shape and degree of riming of a growing snow particle depend mainly on ambient temperature, supersaturation and presence of liquid water. As these ambient conditions are not constant through a vertical column, the significance of different growth processes vary along the paths of growing snow particles. Although, with remote sensing techniques we cannot directly measure which growth processes are taking place in a vertical column, dual polarization radar measurements of vertical columns can be viewed as containing fingerprints of these processes. Currently there are studies of individual features in the profiles, e.g. enhanced values of specific differential phase (K_dp) or differential reflectivity (Z_dr), but no attempts have been made so far to create objective documentation of such profiles.

In the present study, an unsupervised classification method is applied for vertical profiles of dual polarization radar observations derived from RHI scans in southern Finland. The proposed classification method is based on k-means clustering of dimensionally reduced combination of vertical profiles of equivalent reflectivity factor (Z_e), Z_dr and K_dp, and on ground temperature.

The aim of this study is to characterize different types of vertical profiles of dual polarization radar variables in winter precipitation and to link them to snowfall properties such as liquid water equivalent (LWE) precipitation intensity, ensemble mean density and degree of riming, as well as to other properties of the vertical column such as liquid water path (LWP). By employing the unsupervised classification method over the complete tropospheric profiles of polarimetric radar measurements we avoid making priori assumptions on the features of the snow process fingerprints these profiles contain. Instead we try to characterize the variety of observed profiles.

In this investigation we look at vertical profiles over Hyytiälä forestry station in Juupajoki, Finland using Ikaalinen weather radar which is located 64 km west from the station. Measurements have been performed during the Biogenic Aerosols – Effects on Clouds and Climate field campaign in early 2014, and the winter of 2014/2015. The training material for the classification currently consists of 15 snow cases, generally with full temporal extent of the event including profiles in which falling particles have not reached the ground.

With the proposed method we obtain a classification scheme of 20 separate classes, which explain roughly 88% of the observed variance. As we are looking at temporal evolution of events, the majority of classes represent situations with precipitation rates close to zero. In 7 out of the 20 classes, the LWE precipitation intensity reaches at least 1 mm h⁻¹. The connection between high vertical extent of reflectivity profiles and high precipitation rate is found to be strong. The profiles of high precipitation intensity also generally show a clear K_dp signal. Among situations with high precipitation intensity, the lowest LWP values were connected with low temperature and vice versa.