Sea Ice Initialization and Its Impact on Winter Seasonal Prediction Skill over the Northern Hemisphere in Coupled Forecast System

Abstract

Recent advancements in coupled models and Arctic sea ice satellite observations have prompted research on sea ice initialization. To assess its impact on winter surface air temperature (SAT) seasonal prediction skill, three initialization methods based on nudging are evaluated using the Community Earth System Model, version 2 (CESM2). The methods include 1) generating ocean/sea ice initial conditions (ICs) solely from atmospheric forcing (Exctrl), 2) building upon Exctrl by directly nudging sea ice concentration to observation and thickness to reanalysis data to produce improved ICs (Exicenudge), and 3) further enhancing Exicenudge by applying additional atmospheric forcing to adjust model balance (Exbalance). The retrospective predictions are initialized on 21 October for 24 years from 1993 to 2016. The anomaly correlation coefficients from the retrospective predictions are 0.27, 0.15, and 0.45 for northern Eurasia and 0.23, 0.27, and 0.39 for southern Eurasia in Exctrl, Exicenudge, and Exbalance, respectively. The Exbalance demonstrates the highest prediction skill, with notable improvements in areas associated with the warm Arctic-cold Eurasia pattern. The Exbalance accurately simulates the SAT distribution, which is characterized by the Barents Oscillation, and effectively captures the polar vortex, a crucial factor in determining Arctic temperatures. The enhanced prediction skill in Exbalance can be attributed to improved SST bias of ICs and better-balanced sea ice ICs with the atmosphere, significantly reducing the strong warm bias within the Arctic Ocean compared to Exicenudge. Altogether, this study highlights that when model bias is substantial, maintaining model balance is more critical than assimilating sea ice conditions that closely match observations for improving seasonal prediction skill.