Modelling the North Atlantic: How parameterizations affect model biases and uncertainties

In this presentation, we discuss the role of a variety of parameterizations for simulating ocean dynamics in the North Atlantic and how they contribute to biases and model uncertainties. Their effect is analyzed via a range of diagnostics and model setups.

Many of the crucial processes in the ocean still need to be parameterized in state-of-the-art global ocean and climate models. Among those processes are mesoscale ocean eddies and mixed layer dynamics which cannot be fully resolved in most multidecadal simulations. However, they play a crucial role in setting the dynamic and hydrographic conditions in the North Atlantic and the global oceans. Increasing resolution tends to improve some of the long-standing ocean biases, but is very costly and makes it difficult to disentangle which specific processes or boundary conditions are driving certain improvements.

A consequence of imperfect process parameterizations are systematic errors resulting in large model biases. Furthermore, they can lead to inaccurate representation of the chaotic evolution of the ocean system, leading to insufficient representations of forecast uncertainties via ensemble simulations. In the North Atlantic, both of these consequences play a large role, leading to strong model biases and a general underdispersion of ensemble forecasts.

Classical biases of ocean models at so called eddy-permitting resolution, where mesoscale eddies are barely resolved, are related to overdissipation of kinetic energy and enhanced diffusion of tracers. We introduce a set of parameterizations that tackle the overdissipation of kinetic energy via specific viscosity schemes, including schemes that reinject some of the overdissipated energy back into the system. A combination of such schemes reduces classical ocean biases such as the North Atlantic cold bias by enhancing eddy activity and improving the path of mean currents such as the Gulf Stream. In addition, we demonstrate how stochastic methods can be used to account for parameterization uncertainties in the North Atlantic, quantifying the role of parameterization errors in ocean and climate simulations. These new schemes come at a small additional computational cost, especially compared to higher resolution simulations, and provide a means of understanding the origin of model biases and uncertainties.