Abstract

In this study, we investigated the use of synthetic aperture radar (SAR) polarimetry (Pol-SAR) and a supervised classification technique, support vector machine (SVM), for the classification of bare soil, ice, and snow, over the Ortles–Cevedale massif, (Eastern Italian Alps). We analyzed the importance of topographic correction on the backscattering and polarimetric SAR signature and the advantage of quad-pol with respect to dual-pol data. When backscattering values only are employed, the incidence angle used as input feature of the SVM classifier assures the best classification accuracy, 9.9% higher than the accuracy obtained with cosine corrected ${\gamma ^0}$ backscattering. The introduction of polarimetric features and decomposition parameters (such as Cloude–Pottier or Touzi decomposition parameters) increases the classification accuracy by 5.2% with respect to the backscattering case. The simulation of RADARSAT-2 data as Sentinel-1 like for dual-pol data shows a decrease of accuracy equal to 7.8% with respect to the fully polarimetric case (93.5%). The first Sentinel-1 image acquired on our test area was also employed for classification. We then tested the capability of C-band SAR to detect accumulation and ablation zones of the glaciers under the winter dry snow by setting up a multi-incidence angle and fully polarimetric SVM classifier, exploiting ascending and descending RADARSAT-2 data. In this case, the accuracy increased by 14.7% combining different geometric acquisitions (88.9%) with respect to the single geometry case. Finally, from the resulting classification maps, we extracted the snowline altitude for a sample of three glaciers, using both optical and SAR data, comparing the different products.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call