GEORGIA: A Graph Neural Network Based EmulatOR for Glacial Isostatic Adjustment

Lin, Yucheng; Whitehouse, Pippa L.; Valentine, Andrew P.; Woodroffe, Sarah A.

Abstract

Glacial isostatic adjustment (GIA) modeling is not only useful for understanding past relative sea-level change but also for projecting future sea-level change due to ongoing land deformation. However, GIA model predictions are subject to a range of uncertainties, most notably due to uncertainty in the input ice history. An effective way to reduce this uncertainty is to perform data-model comparisons over a large ensemble of possible ice histories, but this is often impossible due to computational limitations. Here we address this problem by building a deep-learning-based GIA emulator that can mimic the behavior of a physics-based GIA model while being computationally cheap to evaluate. Assuming a single 1-D Earth rheology, our emulator shows 0.54 m mean absolute error on 150 out-of-sample testing data with <0.5 s emulation time. Using this emulator, two illustrative applications related to the calculation of barystatic sea level are provided for use by the sea-level community.

Citation

Lin, Y., Whitehouse, P. L., Valentine, A. P., & Woodroffe, S. A. (2023). GEORGIA: A Graph Neural Network Based EmulatOR for Glacial Isostatic Adjustment. Geophysical Research Letters, 50(18), https://doi.org/10.1029/2023gl103672

• Journal Article Type: Article
• Acceptance Date: Aug 9, 2023
• Online Publication Date: Sep 16, 2023
• Publication Date: Sep 28, 2023
• Deposit Date: Nov 1, 2023
• Journal: Geophysical Research Letters
• Print ISSN: 0094-8276
• Electronic ISSN: 1944-8007
• Publisher: Wiley
• Peer Reviewed: Peer Reviewed
• Volume: 50
• Issue: 18
• DOI: https://doi.org/10.1029/2023gl103672
• Keywords: General Earth and Planetary Sciences; Geophysics
• Public URL: https://durham-repository.worktribe.com/output/1873995
• Additional Information: Received: 2023-03-13; Accepted: 2023-08-09; Published: 2023-09-16