Ice Type Classification Using Polarimetric SAR Data in the West Antarctic

Abstract:

Since the distribution of sea ice and land ice in the Antarctic is closely related to the global climate system, there has been great effort in monitoring various types of ice in the area. In particular, the spatiotemporal distribution of various ice types (e.g., pack ice, fast ice, glacier, and iceberg) around the Wilkinson glacier in the West Antarctic has been rapidly changing due to on-going climate change. Remote sensing, especially high resolution Synthetic Aperture Radar (SAR), can be utilized to effectively monitor the spatial and temporal distribution of ice. The existing literature suggests that various polarimetric techniques can be used to classify ice types based on the magnitudes of cross-polarization ratios, correlation coefficients between HH and VV polarized signals, and entropy/alpha parameters. The distinctive difference of salinity in sea ice and land ice results in the different scattering patterns in the electromagnetic wave. However, land ice, land-fast ice and some types of icebergs are not distinguishable by the existing techniques. The purpose of this study was to investigate full-polarimetric RADARSAT-2 SAR data for detecting various types of ice focusing on land ice, land-fast ice, and icebergs using machine learning approaches—See5.0, random forest, and support vector machines. Very high resolution airborne SAR data were used for validation of the proposed approach.