Emulating present and future simulations of melt rates at the base of Antarctic ice shelves with neural networks

Montag 08.01.2024, 11:00 → 12:00 Europe/Berlin

8A-002 - Hörsaal Ostufer / Lecture Hall East (GEOMAR - Standort Ostufer / GEOMAR - East Shore)

Speaker: Dr. Clara Burgard, LOCEAN, Paris, France

Abstract 
Melt rates at the base of Antarctic ice shelves are needed to drive projections of the Antarctic ice sheet mass loss and project freshwater input into the Southern Ocean. Current basal melt parameterisations struggle to link open ocean properties to ice-shelf basal melt rates for the range of current sub-shelf cavity geometries around Antarctica. In this talk, we will present a proof of concept exploring the potential of simple deep learning techniques to parameterise basal melt. We trained a simple neural network to emulate the behavior of circum-Antarctic cavity-resolving ocean simulations. The resulting melt rates are as good or better than melt rates inferred with traditional parameterisations on both the training and the independent testing dataset. Nevertheless, some limitations remain, and we will discuss how they can be further investigated and overcome.