Sea level rise has the potential to cause catastrophic damage to coastal regions, with broad implications across economic, geopolitical, and geophysical systems. However, current projections of future sea level rise are deeply uncertain, largely due to the simplistic nature of ice sheet models. In particular, the model of ice flow, the primary process by which ice sheets eject ice, is an empirical relationship calibrated on only a few datapoints. This project will develop a neural ice flow model, built on principles of AI and trained on high-fidelity models of ice microphysics, which will capture complexities of ice flow physics within ice sheet models. The researchers will apply this novel model to produce updated projections of ice mass loss from the Antarctic Ice Sheet. This initial work will pave the way for robust AI representations of ice physics and improved certainty in SLR projections used by climate decisionmakers.
“A major challenge with modeling ice sheets is the multi-scale behavior they can exhibit. Microphysical processes occurring within the ice – such as changes to ice crystals that are a few millimeters big – can impact ice sheet dynamics on scales of kilometers. Our project aims to build a more physically-realistic ice flow model by explicitly resolving these microphysical processes and using machine learning to couple them to large-scale ice sheet dynamics. Ultimately, this new ice flow model should capture complexities within ice sheets that current models cannot, and much of that complexity is driving the changes to ice sheets we see today.”
Meghana Ranganathan, Assistant Professor, Geophysical Sciences, University of Chicago
