Impacts of Stochastic Parametrizations of Ocean Mixing on Seasonal to Decadal Timescales

The ocean component of most global climate models does not adequately resolve eddies. As a consequence, deterministic parametrizations are used to simulate the mean impact of the unresolved subgrid-scale eddy mixing and convection on the resolved large scale flow. However, this approach does not capture subgrid-scale variability and underestimates eddy-mean flow interactions. Non-linearities in the prognostic equations may respond susceptible to the inclusion of such variability and affect the model mean state as well as simulated timescales of large scale variability.

One approach to incorporate unresolved variability is the notion of stochastic parametrizations. These have been successfully applied in atmospheric models used for weather and seasonal forecasts, increasing reliability of ensemble simulations as well as reducing model biases. We follow a similar approach for the ocean component of a climate model and implement stochastic perturbation schemes for parametrizations of vertical mixing, deep convection, and eddy induced isopycnal mixing. Thereby we are reintroducing some of the missing subgrid-scale variability as well as providing a measure of model uncertainty for probabilistic ensemble forecasts. The impact of these schemes is analyzed on seasonal to decadal timescales, regarding changes to the simulated ocean mean state due to non-linear rectification and the generated ensemble spread, both in uncoupled ocean-only as well as coupled simulations. While all three schemes impact the model variability and mean state in regions of strong eddy activity, the three perturbation schemes complement each other in other regions of the ocean such as convectively active areas.

The proposed stochastic parametrizations are a first step to seamlessly analyze the impact of incorporating measures of ocean model uncertainty and subgrid-scale variability in coupled global models on timescales from seasons to decades.