ICCS Seminar: David Kamm

The global ocean circulation is both driver and outcome of fluid interactions spanning scales from millimetres to thousands of kilometres. Among these are mesoscale eddies with horizontal scales of approximately 10-100 km, which form the ocean's dominant reservoir of kinetic energy and play a central role in the transfer of momentum and the redistribution of heat and other tracers. Current state-of-the-art climate models only partially capture these mesoscale processes due to limited computational resources and rely on their accurate parameterisation. Recent studies have addressed this issue by using machine learning tools to develop new, data-driven parameterisations for mesoscale eddies, promising improvements in the predictive skill of climate models. Integrating such parameterisations into production climate models, however, presents significant software engineering and computational challenges. Machine learning workflows are typically developed within GPU-oriented software ecosystems, whereas most climate models are large CPU-oriented codes designed for massively parallel execution across thousands of cores, with distinct performance constraints. In the presented work, we implement and evaluate two complementary data-driven parameterisation approaches within an idealised ocean model configuration, and compare their scientific performance together with the practical trade-offs involved in their implementation within climate models.

How to attend

Join on Zoom or in-person at MR2, Centre for Mathematical Sciences, Wilberforce Rd, Cambridge CB3 0WA

About ICCS Seminars

This talk is part of a series of ICCS Seminars. Members of the VESRI teams, ICCS and the wider research community present their work, research papers or concepts significant to their field.

Watch previous seminar recordings