Arctic regional changes revealed by clustering of sea-ice observations

Abstract

Understanding the evolution of Arctic sea-ice is crucial due to its climatic and socio-economic impacts. Usual descriptors (e.g., sea-ice extent, Marginal Ice Zone, sea-ice age, and ice-free duration) quantify changes but do not account for the full seasonal cycle. Here, using satellite observations of sea-ice concentration (SIC) over 1979-2023, we perform a k-means clustering of the Arctic sea-ice seasonal cycle, initializing with equal quantile separation and using Mahalanobis distance. Without providing prior information, this data-driven method shows that the Arctic is best described by four types of seasonal cycles: open-ocean (no ice year-round), permanent sea-ice (full coverage with a minimum of 70 % SIC), and two clusters showing ice-free conditions (SIC <0.15), namely partial and full winter freezing. The latter has larger SIC in winter, more abrupt melting and freezing periods, and a shorter ice-free season than the former. This reduction of dimension in the data suggests that the first date of retreat is a good indicator for ice-free conditions the following summer and the first date of advance a good indicator for fully ice cover conditions the following winter. We introduce the probability to belong to each four seasonal cycles as a descriptor to monitor Arctic sea-ice changes. The pan-Arctic probability to belong to the permanent sea-ice seasonal cycle has decreased by 3.1 % per decade which is compensated with an increase of probability to belong to the open-ocean cluster (1.6 % per decade), the full winter freezing cluster (1.1 % per decade) and the partial winter-freezing cluster (0.5 % per decade). Regionally, the permanent sea-ice retraction from the Pacific side is compensated by the full winter-freezing cluster while the open-ocean cluster expansion in the Atlantic side is lost by the partial winter-freezing cluster. The new classes of partial and full winter freezing are helpful for sea ice process understanding as it refines the classical MIZ category into two distinct sea-ice clusters. The trend is primarily controlled by the tendency of the more abrupt melting and growth seasonal cycle (full winter-freezing cluster) compared to the trend of the quasi-sinusoidal sea-ice seasonal cycle (partial winter-freezing cluster). Also, from the Beaufort to the Kara Seas, the southern parts have stabilized (experiencing a new typical seasonal cycle, corresponding to the full winter-freezing cluster) and the northern part have destabilized (losing their typical permanent sea-ice seasonal cycle). Therefore, this work provides a new way to describe Arctic regional changes using a statistical framework based on physical behaviours of sea-ice. Our study calls for a more latitudinal vision of the Arctic regions.