Impact of Ocean and Sea Ice Initialisation On Seasonal Prediction Skill in the Arctic

Abstract

Abstract There is a growing demand for skillful prediction systems in the Arctic. Using the Norwegian Climate Prediction Model that combines the fully coupled Norwegian Earth System Model and the Ensemble Kalman filter, we present a system that performs both, weakly coupled data assimilation (wCDA) when assimilating ocean hydrography (by updating the ocean alone) and strongly coupled data assimilation when assimilating sea ice concentration (SIC) (by jointly updating the sea ice and ocean). We assess the seasonal prediction skill of this version of the Norwegian Climate Prediction Model, the first climate prediction system using strongly coupled data assimilation, by performing retrospective predictions (hindcasts) for the period 1985 to 2010. To better understand the origins of the prediction skill of Arctic sea ice, we compare this version with a version that solely performs wCDA of ocean hydrography. The reanalysis that assimilates just ocean data exhibits skillful hydrography in the upper Arctic Ocean and features an improved sea ice state, such as improved summer SIC in the Barents Sea, or reduced biases in sea ice thickness. Skillful prediction of SIE up to 10–12 lead months are only found during winter in regions of a relatively deep ocean mixed layer outside the Arctic basin. Additional DA of SIC data notably further corrects the initial seaice state, confirming the applicability of the results of Kimmritz et al. (2018) in a historical setting. The resulting prediction skill of SIE is widely enhanced compared to predictions initialized through wCDA of only ocean data. Particularly high skill is found for July‐initialized autumn SIE predictions.