Hydrometeor classification for X-band dual polarization radar on-board civil aircrafts

• Main Authors: Roberto, N., Adirosi, A., Baldini, L., Facheris, L., Cuccoli, F., Lupidi, A., Garzelli, A.
• Format: Conference Proceeding
• Language: English
• Subjects: avionic radar; Classification algorithms; Dual polarization radar; Fuzzy logic; hydrometeor classification; Meteorological radar; Meteorology; Radar polarimetry; Support Vector Machine; Support vector machines
• ISSN: 2153-6996; 2153-7003
• DOI: 10.1109/IGARSS.2015.7326272

Polarimetric techniques applied to radars on-board civil aircraft can improve the estimation of risk zones due to dangerous weather during flight. Usually, X-band weather radars are installed at the nose of civil aircrafts. This band is affected by strong attenuation in case of intense precipitation (liquid or mixed phase). The current systems do not compensate backscattered power measurements for attenuation caused by propagation through precipitation, while dual-polarization radars are able to compensate effectively this source of error and to discriminate hydrometeors. In this work, a classification algorithm based on support vector machines (SVM) is proposed. Training is driven by the output of a Fuzzy Logic (FL) classification algorithm (typical classification approach used for ground-based weather radar). SVM high performance in terms of time processing and its flexibility of configuration using all type of inputs variables are important characteristics to be included in some avionic specific equipment, such as the Electronic Flight Bag (EFB). Two datasets have been used to test the SVM classification algorithm. The first dataset is composed of simulated radar polarimetric observations at X-band and the second one is composed of actual dual-polarization radar measurements collected during the Special Observation Period (SOP) 1.1 of HYdrological cycle in MEditerranean EXperiment (Hymex) campaign by the C-band Doppler dual-polarization weather radar (Polar 55C) installed at ISAC-CNR in Rome. Good performance are obtained for SVM classificator. The comparison with FL output shows a good agreement (up to 90%) both in qualitative comparison maps by maps and using a quantitative approach which metric is based on the confusion matrix.