Environmental Conditions Impacting Late-Summer Sea Ice Concentration Predictability in the Northwest Passage

Abstract

The Northwest Passage, located in the Canadian Archipelago, is a conglomerate of sea routes that connect the eastern and western sides of the North American continent. Historically, sea ice in the region has made it difficult for ships not strengthened for ice breaking to utilize the Northwest Passage. With rising temperatures, sea ice in the Arctic has decreased and is likely to continue doing so. During some years in the late summer, the Northwest Passage has so little sea ice that it is essentially open to all forms of maritime traffic. In 2013 and 2016, the Northwest Passage had anomalously high and low August sea ice concentration (SIC), respectively. Other years with high and low average SICs are identified and various thermodynamic and dynamic environmental conditions (Two-meter temperature, zonal and meridional winds, sea ice thickness, and SIC anomalies prior to August) are composited (averaged over the high and low SIC years for each month) to determine which are likely to have impacts on SIC. Composites show differences between these two samples of high and low August SIC, especially for sea ice thickness, prior SIC, and temperature early on in the year. Correlating anomalies of these variables accumulated over a number of months directly to the August SIC anomalies for the years 1982 through 2020 shows where there are areas of strong, significant correlations. Based on composite and correlation analyses it can be deduced that strongly anomalous years of high and low SIC in the NWP in August do not differentiate themselves until the spring. A convolutional neural network is created to assist in predicting when and where sea ice concentration anomalies will occur within the Northwest Passage. The model’s accuracy is largely dependent on the predictors used, and the temporal range that these predictors cover. When variables related to the radiative/heat flux at the surface are included in the training data, models based on early-year data are able to increase their average August SIC prediction accuracy in the Northwest Passage by about 5%.