Classification and snow line detection for glacial areas using the polarimetric SAR image

Snow cover and glaciers are sensitive indicators of the environment. The vast spatial coverage of remote sensing data, coupled with the tough conditions in areas of interest has made remote sensing a particularly useful tool in the field of glaciology. Compared to optical images, synthetic aperture radar (SAR) data are hardly influenced by clouds. This is important because glacial areas are usually under cloud cover. ngkemadi glacier in the Qinghai–Tibetan plateau was selected as the study area for this paper. We use polarimetric SAR (PolSAR) image for classification on and around the glacier. The contrast between ice and wet snow is remarkable, but it is difficult to distinguish the ice from the ground on SAR images due to similar backscatter characteristics in former research. In our study, we found that this distinction can be achieved by target decomposition. Support Vector Machines (SVMs) are performed to classify the glacier areas using the selected features. The glacial areas are classified into six parts: wet snow, ice, river outwash, soil land, rocky land and others. The PolSAR–Target decomposition–SVMs (PTS) method is proven to be efficient, with an overall classification accuracy of 91.1% and a kappa coefficient of 0.875. Moreover, 86.63% of the bare ice and 96.76% of the wet snow are correctly classified. The classification map acquired using the PTS method also helps to determine the snow line, which is an important concept in glaciology.