NOTE: This tool will be updated with a link to the hosted tool and/or source code as soon as it is available. Thank you for your patience during our transition.
Crevasses on the Greenland Ice Sheet are a major factor in the ice sheet’s hydrology, transporting supraglacial meltwater from the surface to the bed, which affects basal sliding velocities. Greenland is expected to experience increased melting overall, as well as more melting in the inland parts of the ice sheet, as the climate keeps warming. Subglacial hydrology models require crevasse location to pinpoint the locations of meltwater inputs to the bed, and thus to more accurately calculate basal sliding velocities. This study aims to locate the current crevasses and crevasse fields in Pakitsoq, central western Greenland. We use semantic segmentation and a fully convolutional U-Net based deep learning approach through Keras, an open-source deep learning library built with the Tensorflow machine learning framework. We developed this workflow on Ghub, a computing gateway that provides access to data sets, analysis tools, and super-computing resources for ice sheet science. We trained MimiNet on multiple 10 km × 10 km Sentinel-1 SAR HH median of January 2020 scenes across a 630 km2 area of the Pakitsoq region. We developed MimiNet as a classification tool that distinguishes surface crevasses from bare ice, supraglacial streams, and lakes on the ice sheet.
Sponsored By
NSF grant Ghub to Briner et al.; NASA grant Moulin formation to Poinar & Andrews