Proper Arctic sea-ice forecasting for the melt season is still a major challenge because of the recent lack of reliable pan-Arctic summer sea-ice thickness (SIT) data. A new summer CryoSat-2 SIT observation data set based on an artificial intelligence algorithm may alleviate this situation. We assess the impact of this new data set on the initialization of sea-ice forecasts in the melt seasons of 2015 and 2016 in a coupled sea ice-ocean model with data assimilation. We find that the assimilation of the summer CryoSat-2 SIT observations can reduce the summer ice-edge forecast error. Further, adding SIT observations to an established forecast system with sea-ice concentration assimilation leads to more realistic short-term summer ice-edge forecasts in the Arctic Pacific sector. The long-term Arctic-wide SIT prediction is also improved. In spite of remaining uncertainties, summer CryoSat-2 SIT observations have the potential to improve Arctic sea-ice forecast on multiple time scales.

We have equipped the unstructured-mesh global sea-ice and ocean model FESOM2 with a set of physical parameterizations derived from the single-column sea-ice model Icepack. The update has substantially broadened the range of physical processes that can be represented by the model. The new features are directly implemented on the unstructured FESOM2 mesh, and thereby benefit from the flexibility that comes with it in terms of spatial resolution. A subset of the parameter space of three model configurations, with increasing complexity, has been calibrated with an iterative Green’s function optimization method to test fairly the impact of the model update on the sea-ice representation. Furthermore, to explore the sensitivity of the results to different atmospheric forcings, each model configuration was calibrated separately for the NCEP-CFSR/CFSv2 and ERA5 forcings. The results suggest that a complex model formulation leads to a better agreement between modeled and the observed sea-ice concentration and snow thickness, while differences are smaller for sea-ice thickness and drift speed. However, the choice of the atmospheric forcing also impacts the agreement of FESOM2 simulations and observations, with NCEP-CFSR/CFSv2 being particularly beneficial for the simulated sea-ice concentration and ERA5 for sea-ice drift speed. In this respect, our results indicate that the parameter calibration can better compensate for differences among atmospheric forcings in a simpler model (i.e. sea-ice has no heat capacity) than in more energy consistent formulations with a prognostic ice thickness distribution.