Sub-seasonal prediction of Arctic sea ice concentration using time series forecasting technique

Recently, many studies have highlighted the importance of the ability to predict the Arctic sea ice concentration in the sub-seasonal time scales. Notably, the Arctic sea ice concentration has a potential for skillful predictions through their long-term trend memory. Based on the long-term memory of Arctic sea ice concentration, we evaluate the predictability of Arctic sea ice concentration (SIC) by applying a time-series analysis technique of the Prophet model on sub-seasonal time scales. A Prophet is a recently introduced method as a statistical approach inspired by the nature of time series forecasted at Facebook and has not been applied to the prediction of Arctic SIC before. Sub-seasonal prediction skills of Arctic SIC in the Prophet model were compared with the NCEP Climate Forecast System Reforecast (CFS-Reforecast) model as a dynamical approach and verified with the satellite observation during wintertime from 2000 to 2018 for 1 to 8 weeks lead times. The result shows that the Prophet model exhibits much better skill than the NCEP CFS-Reforecast model in the climatology prediction except for the 1 to 3 weeks lead times, as the Prophet model has mainly the ability to capture the long-term trend. In the anomaly prediction, however, the NCEP CFS-Reforecast model is superior to the Prophet model in the prediction of sub-seasonal time scales, as the NCEP CFS-Reforecast captures more effectively the sub-seasonal transition of the underlying dynamical system. Therefore, even if the Prophet model has shown a useful skill in predicting the climatological Arctic SIC, there is still a need to improve the accuracy and robustness of the predictions in an anomalous Arctic SIC. Further, we suggest that the bias correction method is needed to improve the forecast skill of Arctic SIC using the time-series analysis technique, and it will be critical to advance the field of the Arctic SIC forecasting on the sub-seasonal time scales.