Mechanisms of Surface Meltwater Ponding and Drainage on the Greenland Ice Sheet Revealed Using SkySat Imagery and Deep Learning

Abstract

Surface meltwater impacts Greenland Ice Sheet mass balance indirectly by reducing albedo and promoting hydrofracture. However, fully understanding both processes requires accurate mapping of small-scale features such as ponds, channels, and moulins that govern meltwater formation and drainage. Here we investigate surface water dynamics at high spatial (∼1 m) and temporal resolution by applying deep learning to high-resolution imagery from the SkySat constellation. We develop a U-Net model that robustly classifies surface meltwater with higher accuracy than a conventional thresholding approach. Our mapping reveals that small water features (<0.015 km2) account for a substantial fraction of surface water area in the western Greenland Ice Sheet ablation zone, especially during May (67%) and August (38%). However, we find that seasonal variability in surface water area is primarily driven by the filling and draining of 12 large supraglacial lakes. The high spatial resolution of the SkySat imagery reveals that much of this variability can be attributed to the development of narrow supraglacial channels that facilitate the drainage of upstream lakes into a single downstream lake. When the downstream lake rapidly drains, we observe synchronous lake drainage across our study site between 13–16 June. This cascading drainage event explains how lakes drain even when they are situated in compressive ice flow regimes and provides an alternative mechanism for synchronous lake drainages typically attributed to transmission of stress perturbations. Our study demonstrates that deep learning applied to high-resolution satellite imagery can provide valuable insights into supraglacial hydrology.

Plain Language Summary
Meltwater produced by the Greenland Ice Sheet is a major contributor to sea-level rise. But, before it flows into the ocean, meltwater can further increase ice sheet mass loss by (a) reducing surface albedo, which increases solar energy absorption, and (b) weakening glaciers and ice shelves, which enhances iceberg production. Understanding both processes requires detailed maps of small water features, such as ponds, streams, and moulins that are difficult to detect with standard satellite imagery. In this study, we applied a deep learning model (U-Net) to map surface meltwater in high-resolution SkySat images. We found that small water features account for a substantial fraction of surface water area, especially in May and August. However, most seasonal variation in meltwater coverage is driven by the filling and draining of 12 large supraglacial lakes. The high-resolution SkySat images reveal that many of these lakes drain simultaneously through a previously undocumented mechanism that begins when several upstream lakes drain into a single downstream lake via narrow supraglacial channels. When the downstream lake drains rapidly, it triggers drainage of the entire connected system. This work shows that combining deep learning with high-resolution satellite images can enhance our understanding of ice sheet meltwater dynamics.