Mapping sub-ice geomorphology and constraining Greenland Ice Sheet history using machine learning by Dr Guy Paxman, Durham University
Loss of ice from Greenland is one of the largest contributors to anthropogenic global sea level rise. To help inform global policy decisions, an overarching objective of the glaciological community is to develop numerical models of ice-sheet behaviour that are capable of robustly projecting the future evolution of the Greenland Ice Sheet.
|An Institute for Data Science and Artificial Intelligence seminar|
|Date||22 November 2023|
|Time||12:30 to 14:00|
|Place||Building:One Kolade Teaching Room|
To evaluate these models, we can use geological records to constrain how the Greenland Ice Sheet behaved during past warmer climates that act as analogues for projected 21st Century warming. One such archive is the landscape covered by the ice sheet itself, which likely contains geomorphological records that can provide valuable insights into past ice-sheet behaviour. Subglacial topography is mapped using ice-penetrating radar surveys; as of 2023, more than 800,000 line-km of radar data have been acquired across the Greenland Ice Sheet, but the potential of these data is largely untapped, and the long-term history of the ice sheet remains highly uncertain.
In this talk I will describe the use of automated and machine-learning-based techniques to examine spatial patterns in Greenlandic topography in a time-efficient and objective manner that is not feasible using manual mapping techniques. I will describe the identification of more than 5000 cross-sections of subglacial valleys in the radar dataset, and the use of random forest to classify these valleys based on their morphological similarity to those formed by glacial or fluvial incision elsewhere in the Northern Hemisphere. This analysis reveals intricate valley networks in the subglacial highlands of Greenland with morphologies that are indicative of substantial glacial modification of an inherited fluvial landscape, but which were not incised beneath the modern ice sheet. Geomorphological mapping is integrated with simple ice-sheet modelling experiments to infer that these valleys were incised several million years ago, at a time when Greenland’s climate was warmer than present and ice coverage was much more restricted. Our results suggest that Greenland likely contained less than 10% of its modern ice volume at this time, with global sea levels at least 7 metres higher as a result. These new constraints on early Greenland Ice Sheet extent can be used to improve our understanding of global climate and the state of the cryosphere in a warmer-than-present world.
Delivery: Hybrid. To register for the seminar, click here. Details will be provided nearer the time.
Queries: Please contact IDSAI.
Building:One Kolade Teaching Room