An Unsupervised Deep Learning Approach for Monitoring the Snow Facies of the Greenland Ice Sheet with InSAR TanDEM-X Data

Abstract:
An integral part of monitoring ice sheets involves identifying snow facies, which are distinct layers or units within the snowpack. In this work, we explore interferometric synthetic aperture radar (InSAR) TanDEM-X data to monitor the Greenland Ice Sheet snow facies across a decade of observations. We propose a novel, fully unsupervised deep learning method utilizing InSAR features like backscatter, volume decorrelation, and geometric parameters. Furthermore, we address the challenges and caveats of unsupervised approaches for managing different bistatic InSAR acquisition geometries. The proposed approach showcases the potential of bistatic InSAR missions for a long-term monitoring of ice sheet dynamics.