Skip to main content
U.S. flag

An official website of the United States government

A multi-model comparison of September Arctic sea ice seasonal prediction skill

Presentation Date
Friday, December 16, 2022 at 9:13am - Friday, December 16, 2022 at 9:24am
McCormick Place - S504abc



Many earlier studies have documented individual prediction systems that skillfully forecast September sea ice extent (SIE) 1-3 months in advance, however, these studies are difficult to compare due to the different skill metrics used and time periods considered. These published skill estimates are also generally higher than the prediction skill revealed by retrospective analyses of real-time predictions submitted to the Sea Ice Outlook (SIO) over the period 2008-2021. In order to directly compare prediction skill across systems and resolve this apparent discrepancy with SIO skill, SIO Contributors have assembled a novel multi-model dataset of retrospective seasonal predictions of September Arctic sea ice. The dataset includes predictions from 17 statistical models and 16 dynamical models, spanning a minimum period of 2001-2020, with SIO initialization dates of June 1, July 1, August 1, and September 1.

We find that most statistical and dynamical models skillfully predict detrended Pan-Arctic SIE, and that detrended anomaly correlation coefficients of 0.5, 0.7, 0.8, and 0.9, respectively, could be expected at SIO lead times. Regional SIE predictions are found to be generally less skillful than Pan-Arctic predictions, and show comparatively better performance in the Alaskan and Siberian regions than the Canadian and Atlantic sectors. The skill of dynamical and statistical models is generally comparable. We find that models consistently struggle to predict extreme sea ice years, such as 1996, 2007, and 2012. Overall, this analysis shows that skillful operational predictions of September SIE are likely possible at least three months in advance. The SIO Community anticipates this new retrospective prediction dataset will provide future opportunities to understand mechanisms of prediction skill, the importance of model formulation, and sources of forecast error.

Funding Program Area(s)