C004-0012
Detection of Grounding Line Migration in Antarctica Using InSAR and Machine Learning Methods
Detection of Grounding Line Migration in Antarctica Using InSAR and Machine Learning Methods
Monday, 7 December 2020
Poster
Abstract:
Despite significant progress in process understanding, numerical simulations, and observations of ice sheet contributions to sea-level rise (SLR), the predictive capability of ice sheet models remains limited, hence the contributions of ice sheets to sea level is highly uncertain and likely underestimated. Improving our understanding of grounding line dynamics is a key point to reduce the uncertainty. The grounding line is the transition boundary between grounded ice and floating ice. Using interferometric synthetic-aperture radar methods, we have been measuring grounding line position and migration precisely over large areas, but the mapping exercise is a huge challenge given the large increase in satellite data and the sheer size of the Antarctic continent. We do not have a procedure to do the mapping automatically. The digitization has traditionally been done manually from human experts based on the interferometry data. Here, we apply a machine learning (ML) algorithm to automatically extract grounding lines from Sentinel-1a/b data. The method is tested on the Ross Ice Shelf and Sulzberger Ice Shelf, Antarctica by comparing manual delineations with the ML automated procedure. The test results demonstrate that the ML method is not only reliable but also comparable in precision to that achieved by human experts. The ML algorithm provides an error estimate for the delineation and when used repeatedly provides information on the grounding zone, i.e. the zone over which the grounding line migrates back and forth with changes in oceanic tide. We also employed the ML algorithm on differential interferometric data from the CSK constellation, which is a different radar system, without re-training the ML technique. Finally, we applied the approach to the entire data set of differential interferometry data from Sentinel-1a/b acquired in Antarctica during 2017-2018. The results provide a quantum leap in our automated characterization of grounding line in Antarctica.