Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of GraphSearch.
Radar-based hydrometeor classification typically comes down to determining the dominant type of hydrometeor populating a given radar sampling volume. In this paper we address the subsequent problem of inferring the secondary hydrometeor types present in a volume - the issue of hydrometeor de-mixing. The present study relies on the semi-supervised hydrometeor classification proposed by Besic et al. (2016) but nevertheless results in solutions and conclusions of a more general character and applicability. In the first part, oriented towards synthesis, a bin-based de-mixing approach is proposed, inspired by the conventional coherent and linear decomposition methods widely employed across different remote-sensing disciplines. Intrinsically related to the concept of entropy, introduced in the context of the radar hydrometeor classification in Besic et al. (2016), the proposed method, based on the hypothesis of the reduced random interferences of backscattered signals, estimates the proportions of different hydrometeor types in a given radar sampling volume, without considering the neighboring spatial context. Plausibility and performances of the method are evaluated using C-and X-band radar measurements, compared with hydrometeor properties derived from a Multi-Angle Snowflake Camera instrument. In the second part, we examine the influence of the potential residual random interference contribution in the backscattering from different hydrometeors populating a radar sampling volume. This part consists in adapting and testing the techniques commonly used in conventional incoherent decomposition methods to the context of weather radar polarimetry. The impact of the residual incoherency is found to be limited, justifying the hypothesis of the reduced random interferences even in a case of mixed volumes and confirming the applicability of the proposed bin-based approach, which essentially relies on first-order statistics.
Anne-Claire Marie Billault--Roux
Alexis Berne, Josué Etienne Gehring, Etienne Gabriel Henri Vignon, Noémie Planat