Upcoming Event: CSEM Student Forum

Parameterizing Subrigid Vertical Turbulence in the Arctic Ocean using Machine Learning

Ben Longaker, Oden Institute

Abstract

The Ocean Surface Boundary Layer (OSBL) governs the transfer of momentum, heat, and tracer fluxes from the atmosphere to the interior ocean. It is important to model the OSBL accurately, which is done by parameterizing a subgrid vertical turbulent mixing term that appears after using Reynolds averaging to remove subgrid scales from the underlying equations. Common parameterizations include KPP, which works well in some ocean domains but not well for others. These parameterizations tend to break down in the arctic ocean, where additional physics are present due to the melting and freezing of sea ice in the OSBL. Hence, more complicated parameterizations are required for arctic ocean models.

In this talk, I will discuss my summer work at Los Alamos National Laboratory, where we used machine learning to build such a parameterization for the vertical turbulent mixing term in the arctic ocean. Since observations of the arctic ocean are sparse, we train our models on data from Large Eddy Simulations (LES) with inputs that closely model the arctic ocean. Comparing to unseen LES data, our best models achieve 15.7% relative error on predicting temperature fluxes and 13.3% relative error on predicting salinity fluxes. These preliminary results indicate that using machine learning to model turbulent fluxes is a promising approach.

Biography

Ben started the CSEM PhD program in Fall 2024. Before attending the Oden Institute, he was a student at Colorado School of Mines, where he studied Computational and Applied Mathematics. His research interests are developing and analyzing algorithms for solving differential equations.