Wet snow-cover mapping by C- and L-band polarimetric SAR

The main focus of this paper is discrimination between snow and bare ground for snow cover mapping using airborne EMISAR multipolarization SAR data. A series of three experiments investigating (i) the effect of the local incidence angle; (ii) the optimal subset of polarimetric features for best accuracy; and (iii) the optimal classifier was performed. The backscatter values vary with local incidence angle. When this was corrected the backscatter values were less dependent on the incidence angle. Nevertheless, the classification accuracy was not significantly better using angle-corrected data in comparison to non-corrected data. A number of studies have proposed and compared different features derived from multipolarization measurements. The aim of the second experiment was to find an optimal subset of such features. The highest classification accuracies were obtained for a one-dimensional feature vector consisting of C-band span, VH, or HV polarization. Several supervised and unsupervised classification methods were studied. It turned out that, for this data set, clustering gave as good results as the supervised classifiers and the more complicated unsupervised classifiers studied