Open Access
- 28 August 2024
- other
- Published by Copernicus GmbH
- Vol. 2024, 1-19
Abstract
The ice thickness of the world’s glaciers is mostly unmeasured and physics-based models to reconstruct ice thickness can not always deliver accurate estimates. In this study, we use deep learning paired with physical knowledge to generate ice thickness estimates for all glaciers of Spitsbergen, Barentsøya, and Edgeøya in Svalbard. We incorporate mass conservation and other physically derived conditions into a neural network to predict plausible ice thicknesses even for glaciers without any in situ ice thickness measurements. With a glacier-wise cross validation scheme we evaluate the performance of the physics-informed neural network. The results of the experiments let us identify several challenges and opportunities that affect the model’s performance in a real-world setting.Keywords
This publication has 17 references indexed in Scilit:
- A Machine Learning Framework to Automate the Classification of Surge‐Type Glaciers in SvalbardJournal of Geophysical Research: Earth Surface, 2022
- Inversion of a Stokes glacier flow model emulated by deep learningJournal of Glaciology, 2022
- Ice velocity and thickness of the world’s glaciersNature Geoscience, 2022
- Deep learning speeds up ice flow modelling by several orders of magnitudeJournal of Glaciology, 2021
- Physics-informed machine learningNature Reviews Physics, 2021
- Accelerated global glacier mass loss in the early twenty-first centuryNature, 2021
- A consensus estimate for the ice thickness distribution of all glaciers on EarthNature Geoscience, 2019
- Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equationsJournal of Computational Physics, 2019
- A mass conservation approach for mapping glacier ice thicknessGeophysical Research Letters, 2011
- The creep of polycrystalline iceProceedings of the Royal Society of London. A. Mathematical and Physical Sciences, 1955