Investigation of Radarsat-2 and Terrasar-X data for river ice classification

To date, monitoring of river ice using remote sensing has mainly focused on the use of mono-polarized and multi-polarized C-band radar data only. In this paper, Support Vector Machine (SVM) classifications using polarimetric parameters are tested to identify types of river ice. Classification algorithms are validated on the newly available C-band Radarsat-2 and X-band Terrasar-X data to investigate the potential of this new imagery, acquired in winter 2009. An electromagnetic model is improved to simulate the polarimet-ric response of a river ice cover to understand the interactions of the radar signal with the ice cover. At C-band, using dual-polarized data over mono-polarized data increases by 23.9% the final classification producer accuracy. Furthermore, the best producer accuracy is 91.6% when using dual-pol data at C-band, which stand for a gain of 2.2% compared to dual-pol data at X-band