Optimization of C- and L-band synthetic aperture radar for all-season rift detection: A case study of the Larsen C Ice Shelf
Date
2025
Authors
McDougall, Kali Ann
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
More than half of ice mass loss from Antarctica occurs through calving of large tabular icebergs along rifts at the outer margins of ice shelves, which can lead to ice shelf destabilization and collapse. Synthetic aperture radar (SAR) sensors provide the greatest potential utility for the study of calving-related mechanisms by offering year-round, all-weather imaging and penetration of surface snow. To fully utilize SAR for rift detection, the constraints on fracture detectability posed by surface melt, and radar frequency and polarization, must be characterized. We examined dual-pol (HH and HV) Sentinel-1 C-band frequency (5.4 GHz) and PALSAR-2/SAOCOM L-band frequency (1.2 GHz) SAR images of the Gipps Ice Rise rift system on the Larsen C Ice Shelf during the 2020-2021 melt year. Rift geometry was characterized using the ATL06 land ice height product from ICESat-2, and surface melt was identified using a fixed threshold applied to Advanced Scatterometer (ASCAT) imagery. A Kolmogorov-Smirnov test was performed on ice type classes to determine their spectral separability in SAR images throughout the melt year. To further evaluate the performance of different SAR configurations on rift detection, pixel-based and image object-based Random Forest classifications were tested during late winter and late melt conditions. Overall, rifts that are filled with mélange are difficult to discriminate from the surrounding firn across seasons. Greater consistency in rift detection is found using L-band frequency compared to C-band, with enhanced L-band capability in winter in the presence of a thick ice mélange layer. In general, HV polarization provides greater separability between ice types compared to HH polarization and improves the detection of rifts through most of the melt year, apart from the late melt stage. Lastly, an object-based approach is superior to a pixel-based approach for the application of machine learning to automate rift detection using SAR. Optimizing the full potential of C- and L-band SAR for rift detection, and the development of an object-based machine learning detection method, will lay the groundwork for future automated rift detection and calving studies. Here a bottom-up assessment of SAR-based rift detection is presented, which provides a useful foundation for future algorithm development and the implementation of a standardized rift detection method.
Description
Keywords
remote sensing, cryosphere, antarctica, synthetic aperture radar, ice shelves