Hydrometeor classification through statistical clustering of polarimetric radar measurements: a semi-supervised approach

Atmos. Meas. Tech., 9, 4425–4445, 2016 www.atmos-meas-tech.net/9/4425/2016/ doi:10.5194/amt-9-4425-2016 © Author(s) 2016. CC Attribution 3.0 License. Hydrometeor classification through statistical clustering of polarimetric radar measurements: a semi-supervised approach Nikola Besic1,2 , Jordi Figueras i Ventura2 , Jacopo Grazioli1 , Marco Gabella2 , Urs Germann2 , and Alexis Berne1 1 Environmental Remote Sensing Laboratory (LTE), École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland 2 Radar, Satellite and Nowcasting Department, MeteoSwiss, Locarno-Monti, Switzerland Correspondence to: Nikola Besic () Received: 29 March 2016 – Published in Atmos. Meas. Tech. Discuss.: 30 March 2016 Revised: 24 August 2016 – Accepted: 25 August 2016 – Published: 8 September 2016 Abstract. Polarimetric radar-based hydrometeor classification is the procedure of identifying different types of hydrometeors by exploiting polarimetric radar observations. The main drawback of the existing supervised classification methods, mostly based on fuzzy logic, is a significant dependency on a presumed electromagnetic behaviour of different hydrometeor types. Namely, the results of the classification largely rely upon the quality of scattering simulations. When it comes to the unsupervised approach, it lacks the constraints related to the hydrometeor microphysics. The idea of the proposed method is to compensate for these drawbacks by combining the two approaches in a way that microphysical hypotheses can, to a degree, adjust the content of the classes obtained statistically from the observations. This is done by means of an iterative approach, performed offline, which, in a statistical framework, examines clustered representative polarimetric observations by comparing them to the presumed polarimetric properties of each hydrometeor class. Aside from comparing, a routine alters the content of clusters by encouraging further statistical clustering in case of non-identification. By merging all identified clusters, the multi-dimensional polarimetric signatures of various hydrometeor types are obtained for each of the studied representative datasets, i.e. for each radar system of interest. These are depicted by sets of centroids which are then employed in operational labelling of different hydrometeors. The method has been applied on three C-band datasets, each acquired by different operational radar from the MeteoSwiss Rad4Alp network, as well as on two X-band datasets acquired by two research mobile radars. The results are discussed through a comparative analysis which includes a cor- responding supervised and unsupervised approach, emphasising the operational potential of the proposed method. 1 Introduction Radar-based hydrometeor classification, that is the proper identification of different types of hydrometeors from radar observations, is important for an improved understanding of atmospheric dynamics, an improved quantitative precipitation estimation (QPE), an improved verification and assimilation in numerical weather prediction models and operational nowcasting applications like aircraft or road safety (Bringi et al., 2007). The spread of polarimetry for weather radar has significantly changed the capability of radar systems to identify meteorological and non-meteorological echoes, as well as to identify different hydrometeor types under the radar umbrella (Bringi et al., 2007). The very first efforts to overcome the ambiguity arising from the overlap of measured reflectivity for different hydrometeors relied on a dual-polarisation parameter – the differential reflectivity (Seliga and Bringi, 1976; Hall et al., 1984). Since then, various methods, incorporating other Doppler dual-polarisation (called polarimetric hereafter) parameters, have been developed for the three frequency bands of major interest (S, C and X) (Bringi et al., 2007). Conceptually, all these methods can be categorised as supervised, unsupervised or semisupervised (Fig. 1). Being by far the largest (Chandrasekar et al., 2013), the first category encompasses mostly approaches based on Boolean logic decision tree, Bayesian theory and the most Published by Copernicus Publications on behalf of the European Geosciences Union. 4426 N. Besic et al.: Hydrometeor classification through statistical clustering of polarimetric radar measurements Figure 1. Schematic generalisation of hydrometeor classification methods. intuitive ones, based on fuzzy logic. The common ground for these seemingly different approaches is a necessity for a very reliable set of polarimetric signatures (Straka et al., 2000). These are obtained either by means of simulations (e.g. Dolan and Rutledge, 2009) or by additionally involving some empirical knowledge (e.g. Al-Sakka et al., 2013). Boolean logic decision tree methods are part of the earliest efforts to exploit radar polarimetry for the purpose of hydrometeor classification (Straka and Zrnic, 1993; El-Magd et al., 2000). However, assuming mutual exclusivity of polarimetric parameters for different hydrometeor types, these methods could not thoroughly exploit the potential of polarimetric measurements. Fuzzy logic has been considered to be the best way to make use of polarimetric signatures known a priori in distinguishing between different hydrometeors (Vivekanandan et al., 1999). Namely, inference combining matching scores of different parameters with overlapping membership functions precisely overcomes the mutual exclusivity limits of the Boolean logic decision tree and makes methods less susceptible to the potential presence of noise. Among the number of methods developed for S, C and X bands we can distinguish between those using reflectivity at horizontal polarisation (ZH ), differential reflectivity (ZDR ), specific differential phase shift (Kdp ) and correlation coefficient (ρhv ) (e.g. Dolan et al., 2013) and those using linear depolarisation ratio (LDR) (e.g. Straka et al., 2000). Also, we can discriminate between methods using temperature as an external parameter (e.g. Zrnic et al., 2001), rather than relative altitude with respect to the 0 ◦ C isotherm (e.g. Lim et al., 2005), as well as between those using two-, or more, dimensional membership functions (e.g. Marzano et al., 2007), rather than one-dimensional ones (e.g. Liu and Chandrasekar, 2000). As the most widespread approach in hydrometeor classification, fuzzy-logic classification methods have been subject of several validation campaigns. One of the most exAtmos. Meas. Tech., 9, 4425–4445, 2016 tensive, the Joint Polarization Experiment (Ryzhkov et al., 2005), demonstrated improved hail detection capabilities using ground measurements with hail-intercept vehicles. A Bayesian approach proposed by Marzano et al. (2010) is another representative supervised approach, in which each simulated class is characterized by its centre and covariance matrix. The labelling of the observations is done by means of Bayesian infer (...truncated)