Assessing the Predictability of the Beaufort Sea Minimum Ice Extent in a Changing Arctic Climate Regime

Understanding the climatic drivers of changes in sea ice extent in the Arctic has become increasingly important as record minima in the September sea ice extent continue to be reached. This research therefore addresses the question of which synoptic scale climatological features are most important in affecting changes in sea ice extent in the Beaufort Sea. First, three measures of sea ice extent—the Barnett Severity Index, the Beaufort Sea minimum sea ice extent, and the Arctic-wide minimum sea ice extent—are compared to assess their degree of agreement and consistency using goodness of fit techniques. Secondly, a number of atmospheric predictor variables are analyzed using a composite approach to identify the most relevant predictors of sea ice in the region. Thirdly, monthly statistical forecast models are created based on multiple regressions and classification and regression trees (CART) to predict the minimum sea ice extent beginning in October of the previous year. Many differing measures have been used to quantify sea ice conditions in the Beaufort Sea, although no study has assessed these measures for consistency. When compared, all three measures indicate the same level of agreement according to the goodness of fit tests. This indicates that the choice of measure can be determined based on the specific application, as no measure outperforms another. In addition to differing measures of sea ice extent, differing predictor variables have been utilized to predict summer sea ice conditions. This study assesses all potentially relevant predictor variables and indicates that upper atmospheric air temperatures at 850 hPa, 700 hPa, and 500 hPa, monthly mean surface air temperatures, freezing degree days, thawing degree days, sea level pressure, total ice concentration, and multiyear ice concentration showed the strongest relationships with sea ice. Various teleconnection patterns including the Arctic Oscillation, the North Atlantic Oscillation, and the Pacific-North American pattern also showed strong relationships with these variables and are therefore believed also have some predictive utility. Finally, monthly multiple linear regression and CART models are created to predict the September sea ice extent using a number of climatic predictor variables. The results of these models suggest that antecedent sea ice conditions (total and multiyear ice concentration) and surface air temperature are the most important variables in predicting summer sea ice extent. The potential predictive power of the forecasts increases as predictions are made closer to the September minimum sea ice extent, with the most precise predictions made during July. This research confirms previous studies and provides a useful compilation of the state of the knowledge on the drivers of sea ice changes in the Beaufort Sea