Gaussian Process and ice sheet modeling

A fundamental aim of the Physical Sciences is to be able to make useful predictions about the world around us. Predictions can be made by leveraging relationships between the different quantities that describe the state of a physical system which are often inferred empirically via observation of the system in different states. In the Earth Sciences, the physical systems we wish to model are often highly complex and, as such, require high-performance computer technology in order to solve numerical implementations of the physical relationships that we have derived.

A Gaussian Process (GP) is a valuable machine learning tool that can be used to predict the relationship between data points, a problem known as regression, and make predictions in a fraction of the time taken to run most numerical models.  RISeR PhD student, Oliver Pollard, has been using GP to emulate regional ice-sheet volumes in order to understand the uncertainty in our MIS 6 Eurasian ice-sheet models, and thus the effect this uncertainty may have on piecing together LIG sea-level records.

You can read more about Oliver's work in a summary he wrote for the Leeds Institute for Data Analytics website, and in the RISeR paper resulting from Oliver's PhD: Pollard, O.G., Barlow, N.L.M., Gregoire, L., Gomez, N., Cartelle, V., Ely, J.C. and Astfalck, L.C., 2023. Quantifying the Uncertainty in the Eurasian Ice-Sheet Geometry at the Penultimate Glacial Maximum (Marine Isotope Stage 6). The Cryosphere Discussions, pp.1-31. DOI: https://doi.org/10.5194/tc-2023-5.