A spring barrier for regional predictions of summer Arctic sea ice

The striking changes undergoing Arctic sea ice have created a pressing need for accurate seasonal predictions of regional sea ice extent (SIE). For summer Arctic SIE, in particular, predictability assessments have revealed a significant drop in prediction skill for forecasts initialized prior to May, possibly limiting accurate forecasts for end users. However, this "spring predictability barrier" has only been documented in a few fully-coupled general circulation models (GCMs). In this work, we present evidence of a late spring prediction skill barrier that is remarkably universal across GCMs. We first show that sea ice volume (SIV) is a skillful linear predictor of regional summer SIE in the Arctic and has similar skill to that identified from a perfect model experiment. Using this linear regression model, we then assess regional SIE predictability across an ensemble of GCMs participating in phase 5 of the Coupled Model Intercomparison Project, finding that the marginal seas of the central Arctic ocean in each GCM displays a spring predictability barrier structure. Further analysis using a sea ice mass budget shows that the physical mechanism causing this barrier may result from the combination of unpredictable wintertime mass variations driven by mass transport convergence and predictable mass variations which are activated by the ice-albedo feedback beginning in May. We discuss the implications of a prediction barrier for current and future observational networks and seasonal prediction systems.