Crevasse locations and meltwater delivery to the bed in Pakitsoq, Greenland: Results from MimiNet, a new deep-learning model for crevasse detection

Naureen Khan , Kristin Poinar , Jason P. Briner, Sophie Nowicki, Jessica Mejia , Mary Verne

If surface crevasse fields deliver meltwater to the bed of the Greenland Ice Sheet, it would affect seasonal ice flow speeds and total mass balance. Whether they do is not currently known; some evidence suggests so, while specific field data suggest not. To address this gap, we develop MimiNet, a neural-network-based tool that identifies surface crevasse fields. We train MimiNet on Sentinel-1 scenes across a 629 km2 area in Pakitsoq, central-western Greenland, and use it to locate crevasse fields annually over 2015–2024. We find that the crevassed area varied from a minimum of 141±25 km2 in 2019 to a maximum of 183±27 km2 in 2016, with no overall trend over the ten-year study period. We find that seasonal ice flow speed anomalies in crevasse fields are significantly higher than those in moulin-drained areas in July, but that in all other months of the melt season there is no difference between the two regions. We therefore infer that crevasse fields in Pakitsoq deliver meltwater to the bed, but in a spatially isolated way that keeps the local subglacial drainage system in an inefficient state for the entire melt season, while surrounding moulin-drained areas transition in mid-summer to a more efficient state.