A Stochastic GM Parameterization of Mesoscale Eddy Transport

Correct representation of mesoscale eddies is required to accurately simulate many ocean features like heat transport, overturning, and the location and strength of major currents. The Gent-McWilliams (GM) framework for parameterizing mesoscale eddies is a foundation for successful representation of mesoscale eddy effects in models that do not resolve them. Traditionally, GM parameterizations are thought of as modeling the 'mean' effect of eddies, which leads to parameterizations with no variability. We demonstrate that the transport by mesoscale eddies is significantly variable: even when averaged over boxes of width ~85 km the transport displays significant non-Gaussian variability. To understand the impact of this variability we run an eddy-resolving double-gyre simulation in an idealized primitive-equation model and extract the large scale by a spatial low-pass filter; the large-scale state is saved every 10 days for 110 years. The large-scale variability induced by the variability in the mesoscale transport is significant, with typical temperature standard deviations on the order of 1 degree or higher in the subtropical thermocline. Non-eddy-resolving models with deterministic GM parameterizations display no (or very little) variability. We present a framework for non-Gaussian stochastic GM parameterization to account for variability in eddy transport. The framework is based on a stochastic model of the eddy field, with tracer transport a derived quantity. This stochastic model of the mesoscale eddy field is of interest in its own right, and we present preliminary results on its construction.