A practical method for long-range forecasting of ice severity in the Beaufort Sea

Abstract

A linear regression model is used to forecast end of summer ice conditions in the Beaufort Sea with a few months lead-time. The model retains four sea ice and atmospheric parameters, where decreased spring total and winter multiyear ice concentrations, negative October East Atlantic phases, and positive March North Atlantic Oscillation phases are associated with lighter sea ice conditions. Monte Carlo simulations suggest that the results are not adversely affected by artificial skill, while Durbin-Watson and Variance Inflation Factor statistics imply the final model is statistically valid. Cross validation diagnostics indicate that variations in the four predictors are related to 85% of the variation in sea ice conditions, suggesting that a relatively simple ice-atmosphere statistical model can be used to forecast end of summer ice conditions in the Beaufort Sea.