Skill improvement of dynamical seasonal Arctic sea ice forecasts

We explore the error and improve the skill of the outcome from dynamical seasonal Arctic sea ice reforecasts using different bias correction and ensemble calibration methods. These reforecasts consist of a five-member ensemble from 1979 to 2012 using the general circulation model EC-Earth. The raw model reforecasts show large biases in Arctic sea ice area, mainly due to a differently simulated seasonal cycle and long term trend compared to observations. This translates very quickly (1-3 months) into large biases. We find that (heteroscedastic) extended logistic regressions are viable ensemble calibration methods, as the forecast skill is improved compared to standard bias correction methods. Analysis of regional skill of Arctic sea ice shows that the Northeast Passage and the Kara and Barents Sea are most predictable. These results show the importance of reducing model error and the potential for ensemble calibration in improving skill of seasonal forecasts of Arctic sea ice.