The Impact of Model Uncertainty on Seasonal Sea Ice Forecasts

Abstract:

In the Sea Ice Outlook prediction system, we find that dynamical models not only show poor skill in forecasting observed September sea ice extent, but are also equally unsuccessful at predicting each other, indicating a large divergence in model physics and/or initial conditions. Motivated by this, we have performed two sets of experiments with SIO models that include both fully coupled global models, and global/regional ice-ocean models. One is an initial condition perturbation experiment, whereby we apply a fixed -1m perturbation to the sea-ice thickness initial conditions of a past SIO, and find that the response varies significantly across models. In the second experiment, we perform two sets of simulations. The control set is initialized with a climatological May 1 sea ice thickness in the central Arctic, while the experiment set is initialized with May 1 2015 sea ice thickness. This allows us to investigate both model uncertainty, and the forecast signal provided by initialized ice thickness. While the models show agreement in certain aspects of the forecast, we find significant differences in forecast uncertainty between fully coupled and ice-ocean models, that the 2015 sea ice thickness information provides a weak but consistent positive thickness response, and that model uncertainty has a strong regional signal. The results from both experiments suggest that different model physics across the Sea Ice Outlook make a significant contribution to model uncertainty and forecast skill degradation, yet provide a guide on how to interpret and improve such forecasts.