Towards Automated Generation of Ice-Sheet Wide Supraglacial Meltwater Depth Measurements from ICESat-2 Data, Using High-Throughput Computing

Seasonal surface melting on the Greenland and Antarctic ice sheets forms complex supraglacial hydrological systems that can rapidly evolve in response to regional climatic drivers. Due to their remote, inaccessible location and vast extent, few direct meltwater depth measurements exist. To infer ice-sheet wide estimates for each melt season, we rely on models estimating light attenuation in water from passive optical satellite imagery. Lacking ground truth data, the parameters in such models remain poorly constrained and difficult to reliably validate. NASA's Ice, Cloud and land Elevation Satellite-2 (ICESat-2) now makes it possible to obtain accurate spaceborne measurements of water depths requiring few, simple assumptions. Its laser altimeter system uses green light that penetrates clear water and its sensor allows for single-photon-sensitive detection. Over shallow melt lakes, it obtains a double reflection: one from the water surface and a second from the underlying lake bed. We present an algorithm that automatically detects supraglacial lakes in the ICESat-2 photon cloud data product (ATL03) based on the flatness of the upper reflective surface and the presence of a lower reflection from the lake bed. For each detected lake, the two surfaces are delineated by fitting a straight line through the surface and applying a density-weighted robust LOESS fit to the remaining photons. The elevation difference is corrected for the refractive index of water to obtain a depth measurement. Our algorithm can be applied to a single raw or subsetted ATL03 data granule, so generating meltwater depth data sets across large regions or long time scales is highly parallelizable. To aggregate depth estimates across large amounts of granules in ICESat-2's 300-terabyte data catalog, we use the Open Science Grid's Open Science Pool for distributed High-Throughput Computing, using idle resources across universities and other organizations on heterogeneous hardware. This makes it possible to extract meltwater lake depths from ICESat-2 across both ice sheets in an efficient, cost-effective and reproducible manner. A comprehensive data set of these depth measurements can be used to tune and validate imagery-based algorithms, moving us closer towards accurately quantifying meltwater volumes across the ice sheets.