|Paper title||Automated mapping of ice sheet supraglacial hydrology using Machine Learning|
|Form of presentation||Poster|
The ice sheets of Greenland and Antarctica have been melting since at least 1990, suffering their highest mass loss rate between 2010 and 2019. With mass loss predicted to continue for at least several decades, even if global temperatures stabilize (IPCC, Sixth Assessment Report), mass loss from the ice sheets is predicted to be the prevailing contribution to global sea-level rise in coming years.
Supraglacial hydrology is the interconnected system of lakes and channels on the surface of ice sheets. This surface water is believed to play a substantial role in ice sheet mass balance by modulating the flow of grounded ice and weakening floating ice shelves to the point of collapse. Mapping the distributions and life cycle of such hydrological features is important in understanding their present and future contribution to global sea-level rise.
Using optical satellite imagery, supraglacial hydrological features can be easily identified by eye. However, given that there are many thousands of these features (~76,000 features identified across Antarctica in January 2017, for example), and they appear in many thousands of satellite images, accurate, automated approaches to mapping these features in such images are urgently needed. The standard approach to map these features often combines spectral thresholding (Normalised Difference Water Index, NDWI) with time-consuming manual corrections and quality control processes. Given the volume of the data now available, however, methods such as those that require manual post-processing are not feasible for repeat monitoring of surface hydrology at a continental scale. Here, we present results from ESA’s Polar+ 4D Greenland, 4D Antarctica and Digital Twin Antarctica projects, which increase the accuracy of supraglacial lake and channel delineation using Sentinel-2 and Landsat-7/8 imagery, while reducing the need for manual intervention. We use Machine Learning approaches, including a Random Forest algorithm trained to classify surface water from non-water features in a pixel-based classification.
Appropriate Machine Learning algorithms require comprehensive, accurate datasets. Because of a lack of in situ data, one of the few options we have available is to generate such datasets from satellite imagery. We, therefore, generate these datasets to carry out rigorous, systematic testing of the Machine Learning algorithm. Our methods are trained and validated over varied spatial and temporal (seasonally: within the melt-season, and yearly: between melt-seasons) conditions using data covering a range of glaciological and climatological environments. Our approach, designed for easy, efficient rollout over multiple melt-season, uses optical satellite imagery alone. The workflow, developed under Google Cloud Platform, which hosts the entire archive of Sentinel-2 and Landsat-8 data, allows for large-scale application over Greenlandic and Antarctic ice sheets and is intended for repeated use throughout the future melt-seasons. Ice sheets, a crucial component of the Earth System, impact global sea level, ocean circulation and biogeochemical processes. This study shows one example of how Machine Learning can automate historically user-intensive satellite processing pipelines within a Digital Twin, allowing for greater understanding and data-driven discovery of ice sheet processes.