Max Trostel, UT-Austin

"Improving Ocean Models: Stochastic Parametrization and Adjoint Methods"

Abstract: Subgrid-scale parametrizations are used in fluid modeling as a way to capture the effect of smaller unresolved physical processes on the larger resolved physics. In ocean modeling, specifically, these parametrizations are used to approximate the effect of unresolved mesoscale eddies (scales of 10 to 100 km) on the larger scale motion of the ocean and distribution of tracers like temperature and salinity. Stochastic parametrization, where the parametrized tendencies at every point in space and time are drawn from a distribution rather than only assuming their mean values, has been shown to reduce model biases and produce more realistic variability in oceanic and atmospheric models. For example, the European Centre for Medium-Range Weather Forecasts (ECMWF) has successfully used stochastic parametrization to improve their weather forecasts for 25 years. Recent implementations in idealized ocean models have shown promise for this method in ocean eddy parametrization. In this talk I will discuss my work implementing stochastic parametrization in a toy model — the Lorenz 96 coupled system — and how we can use the adjoint state method (or equivalently, the Lagrange multiplier method) to tune our parametrization to better fit the data and produce more realistic variability. I show that this method can be applied effectively to stochastic parametrization in the toy model and discuss prospects for future applications in ocean modeling.