Development of statistical seasonal prediction models of Arctic Sea Ice concentration using CERES absorbed solar radiation

Abstract

Statistical seasonal prediction models for the Arctic sea ice concentration (SIC) were developed for the late summer (August-October) when the downward trend is dramatic. The absorbed solar radiation (ASR) at the top of the atmosphere in June has a significant seasonal leading role on the SIC. Based on the lagged ASR-SIC relationship, two simple statistical models were established: the Markovian stochastic and the linear regression models. Crossvalidated hindcasts of SIC from 1979 to 2014 by the two models were compared with each other and observation. The hindcasts showed general agreement between the models as they share a common predictor, ASR in June and the observed SIC was well reproduced, especially over the relatively thin-ice regions (of one- or multi-year sea ice). The robust predictability confirms the functional role of ASR in the prediction of SIC. In particular, the SIC prediction in October was quite promising probably due to the pronounced icealbedo feedback. The temporal correlation coefficients between the predicted SIC and the observed SIC were 0.79 and 0.82 by the Markovian and regression models, respectively. Small differences were observed between the two models; the regression model performed slightly better in August and September in terms of temporal correlation coefficients. Meanwhile, the prediction skills of the Markovian model in October were higher in the north of Chukchi, the East Siberian, and the Laptev Seas. A strong non-linear relationship between ASR in June and SIC in October in these areas would have increased the predictability of the Markovian model.