Representing uncertainty in global climate models using stochastic sea ice parameterizations

This dissertation deals with the representation of uncertainties in sea ice modelling, especially within the sea ice dynamics. An important term of the momentum balance for computing the evolution of sea ice drift is the viscous-plastic sea ice rheology. It describes the deformation of sea ice under convergent drift. In this context, an ice strength parameter determines the internal ice strength, which counteracts plastic deformation and hence piling up of sea ice. Uncertainties in the choice of the parameter are simulated in this study by application of symmetric perturbation schemes. Temporal as well as spatial correlations are included in the generation of continuously applied stochastic perturbations. The parameter perturbations are implemented in an ocean-sea ice and in a coupled atmosphere-ocean-sea ice model. Results show that including these uncertainty estimates leads to a change in the mean sea ice distribution, especially in the Arctic. A randomly reduced ice strength parameter results in a relative acceleration of sea ice drift under convergence, which cannot be reverted by a randomly increased ice strength in the subsequent course of the simulation. This is caused by the highly nonlinear formulation of the sea ice rheology and results in a general acceleration of sea ice owing to the symmetric perturbations. As a result, the amount of thick, ridged sea ice in regions of predominantly convergent drift is increased. In the Arctic this increase accumulates slowly, but continuously over decades. Antarctic sea ice on the other hand exhibits relatively small changes in the mean sea ice distribution. In the coupled atmosphere-ocean-sea ice model ice strength perturbations lead to increased drift as well, although the impact on the sea ice thickness distribution is reduced. The reason are coupled feedback mechanisms, which counteract a general thickness increase. Finally, ensemble simulations are conducted with the coupled model in the context of sea ice predictions. Comparing ensembles with parameter perturbations and ensembles with atmospheric initial condition perturbations shows that the inclusion of model uncertainty leads to increased ensemble spread for the sea ice distribution of the central Arctic during the first weeks of the simulation. This has important implications for uncertainty estimations in data assimilation and forecasts for the polar regions.