Towards Improving Sea Ice Predictabiity: Evaluating Climate Models Against Satellite Sea Ice Observations

Abstract:

The last four decades have seen a remarkable decline in the spatial extent of the Arctic sea ice cover, presenting both challenges and opportunities to Arctic residents, government agencies and industry. After the record low extent in September 2007 effort has increased to improve seasonal, decadal-scale and longer-term predictions of the sea ice cover. Coupled global climate models (GCMs) consistently project that if greenhouse gas concentrations continue to rise, the eventual outcome will be a complete loss of the multiyear ice cover. However, confidence in these projections depends o HoHoweon the models ability to reproduce features of the present-day climate. Comparison between models participating in the World Climate Research Programme Coupled Model Intercomparison Project Phase 5 (CMIP5) and observations of sea ice extent and thickness show that (1) historical trends from 85% of the model ensemble members remain smaller than observed, and (2) spatial patterns of sea ice thickness are poorly represented in most models. Part of the explanation lies with a failure of models to represent details of the mean atmospheric circulation pattern that governs the transport and spatial distribution of sea ice. These results raise concerns regarding the ability of CMIP5 models to realistically represent the processes driving the decline of Arctic sea ice and to project the timing of when a seasonally ice-free Arctic may be realized.

On shorter time-scales, seasonal sea ice prediction has been challenged to predict the sea ice extent from Arctic conditions a few months to a year in advance. Efforts such as the Sea Ice Outlook (SIO) project, originally organized through the Study of Environmental Change (SEARCH) and now managed by the Sea Ice Prediction Network project (SIPN) synthesize predictions of the September sea ice extent based on a variety of approaches, including heuristic, statistical and dynamical modeling. Analysis of SIO contributions reveals that when the September sea ice extent is near the long-term trend, contributions tend to be accurate. Years when the observed extent departs from the trend have proven harder to predict. Predictability skill does not appear to be more accurate for dynamical models over statistical ones, nor is there a measurable improvement in skill as the summer progresses.