Skill of Seasonal Arctic Sea Ice Extent Predictions Using the North American Multimodel Ensemble

Abstract

Previous studies have outlined benefits of using multiple model platforms to make seasonal climate predictions. Here, reforecasts from five models included in the North American Multimodel Ensemble (NMME) project are utilized to determine skill in predicting Arctic sea ice extent (SIE) during 1982?2010. Overall, relative to the individual models, the multimodel average results in generally smaller biases and better correlations for predictions of total SIE and year-to-year (Y2Y), linearly, and quadratically detrended variability. Also notable is the increase in error for NMME predictions of total September SIE during the mid-1990s through 2000s. After 2000, observed September SIE is characterized by more significant negative trends and increased Y2Y variance, which suggests that recent sea ice loss is resulting in larger prediction errors. While this tendency is concerning, due to the possibility of models not accurately representing the changing trends in sea ice, the multimodel approach still shows promise in providing more skillful predictions of Arctic SIE over any individual model.