Simon et al. (2025) Arctic regional changes revealed by clustering of sea-ice observations

Identification

• Journal: ˜The œcryosphere
• Year: 2025
• Date: 2025-12-08
• Authors: Amélie Simon, Pierre Tandeo, Florian Sévellec, Camille Lique
• DOI: 10.5194/tc-19-6639-2025

Research Groups

• IMT Atlantique, Lab-STICC, UMR CNRS 6285, Brest, France
• Univ Brest CNRS Ifremer IRD, Laboratoire d'Océanographie Physique et Spatiale (LOPS), Brest, France
• ODYSSEY Team-Project, INRIA CNRS, Brest, France

Short Summary

This study applies k-means clustering to satellite sea-ice concentration data (1979–2023) to identify four distinct Arctic sea-ice seasonal cycle types, revealing a significant decline in permanent sea-ice (3.1% per decade) compensated by increases in open-ocean and seasonal ice types. The research introduces a probabilistic framework to monitor these regional changes and identifies areas of stability, stabilization, and destabilization in the Arctic sea-ice regimes.

Objective

• To identify and characterize spatio-temporal regions of the Arctic based on statistically distinct sea-ice concentration seasonal cycles using a data-driven clustering method, and to describe how these regions and their boundaries have evolved over time.

Study Configuration

• Spatial Scale: Arctic Ocean, specifically oceanic grid cells above 55° N.
• Temporal Scale: January 1979 to December 2023 (45 years).

Methodology and Data

• Models used: k-means clustering algorithm (unsupervised machine learning), fuzzy k-means clustering, Mahalanobis distance (for clustering), Euclidean distance (for probability calculation), Silhouette coefficient (for optimal cluster number), linear regression analysis.
• Data sources: National Snow and Ice Data Center (NSIDC) Climate Data Record (CDR) product (Version 4) of gridded Sea-Ice Concentration (SIC) fields from passive microwave satellite measurements, daily data aggregated to 5-day mean values, 25 km polar stereographic projection. GEBCO 2024 Grid for bathymetric data.

Main Results

• The Arctic sea-ice seasonal cycle is best described by four clusters: open-ocean (no ice year-round), permanent sea-ice (full coverage with a minimum of 70% SIC), partial winter-freezing (quasi-sinusoidal, ~70% SIC in March, ice-free early August to mid-October), and full winter-freezing (100% SIC mid-November to April, almost ice-free by mid-September, with more abrupt changes).
• The first date of retreat is a good indicator for ice-free conditions the following summer (e.g., early July retreat has ~70% chance of belonging to the full winter-freezing cluster).
• The first date of advance is a good indicator for fully ice-covered conditions the following winter (e.g., early September advance has ~95% chance of belonging to the full winter-freezing cluster).
• The pan-Arctic probability of belonging to the permanent sea-ice cluster decreased by 3.1% per decade. This loss is compensated by an increase in probability for the open-ocean (1.6% per decade), full winter-freezing (1.1% per decade), and partial winter-freezing (0.5% per decade) clusters.
• Regionally, permanent sea-ice retraction from the Pacific side is compensated by the full winter-freezing cluster, while open-ocean expansion in the Atlantic side is compensated by loss of the partial winter-freezing cluster.
• The full winter-freezing cluster is more prevalent in coastal areas, supporting the hypothesis that coastlines drive asymmetric seasonal cycles.
• Four regimes (stable, unstable, destabilization, stabilization) were identified, showing that southern parts of the Beaufort to Kara Seas are stabilizing to the full winter-freezing cluster, while northern parts are destabilizing from the permanent sea-ice cluster.

Contributions

• Introduces a novel data-driven, unsupervised machine learning method (k-means clustering with Mahalanobis distance and quantile-based initialization) to regionalize the Arctic based on the full sea-ice seasonal cycle, avoiding arbitrary thresholds.
• Identifies four distinct types of Arctic sea-ice seasonal cycles, refining the classical Marginal Ice Zone (MIZ) category into two distinct, physically meaningful sea-ice clusters.
• Proposes the probability of belonging to each seasonal cycle type as a new, robust descriptor for monitoring Arctic sea-ice changes.
• Develops a new diagnostic to quantify regime stability and transition of Arctic sea ice, revealing a more latitudinal vision of regional changes and their propagation over time.
• Demonstrates that the first dates of sea-ice retreat and advance can serve as effective indicators for subsequent ice-free or fully ice-covered conditions.

Funding

• French government grant managed by the Agence Nationale de la Recherche under the France 2030 program (project CLIMArcTIC, reference ANR-22-POCE-0005).

Citation

@article{Simon2025Arctic,
  author = {Simon, Amélie and Tandeo, Pierre and Sévellec, Florian and Lique, Camille},
  title = {Arctic regional changes revealed by clustering of sea-ice observations},
  journal = {˜The œcryosphere},
  year = {2025},
  doi = {10.5194/tc-19-6639-2025},
  url = {https://doi.org/10.5194/tc-19-6639-2025}
}