Understanding Arctic Sea Ice Variability and its Patterns: A Comparative Clustering Analysis

• Main Authors: E. Bianco, M. Sangiorgio, D. Iovino, P. Ruggieri, S. Materia, A. Castelletti, S. Masina
• Other Authors: E. Bianco, M. Sangiorgio, D. Iovino, P. Ruggieri, S. Materia, A. Castelletti, S. Masina, Bianco, E., Sangiorgio, M., Iovino, D., Ruggieri, P., Materia, S., Castelletti, A., Masina, S.
• Format: Conference Object
• Language: English
• Published: 2022
• Subjects: Arctic; Arctic Ocean; ice pack; Sea ice
• Online Access: http://hdl.handle.net/11311/1216923

The multi-decadal record of Arctic Ocean observations reveals an unequivocal downward trend in sea ice concentration, as well as a rapid transition towards a thinner ice pack. On sub-seasonal time scales, however, frequent departures from the linear trend indicate a strong response to internal variability of the atmosphere and the ocean, which exert both a thermodynamic forcing on sea ice, regulating its seasonal cycle through heat transport and exchange and a dynamic forcing, driving ice motion within the Arctic Ocean and outflow via its main gateways. The growing need for accurate sea ice predictions on monthly to seasonal timescales, especially in the context of increasingly frequent episodes of abrupt ice loss, requires a better understanding of the dominant modes of sea ice variability and its drivers. To address this gap, we first apply Hierarchical Agglomerative Clustering (HAC) to linearly detrended sea ice concentration anomalies from the ECMWF ERA-5 reanalysis record (1979-2018) to reduce the time and space-varying anomaly signal to a set of recurrent and physically robust sea ice patterns. Secondly, we use reanalysis data from ERA-5 and the CMCC Global Ocean Physical Reanalysis System (C-GLORS) to identify the leading mechanisms of atmospheric and oceanic variability that act as drivers for the identified sea ice clusters at different lead times within the time frame of sub-seasonal prediction. Results suggest that sea ice variability during northern hemispheric minima (September) and maxima (JFM) can be summarized into four key patterns, each characterized by a representative spatial configuration of sea ice concentration anomalies depending on the dominant atmospheric and oceanic predictors. This study highlights the potential for applying machine learning methods to disentangle the complex interactions between Arctic sea ice and its drivers, drawing links between the key components of the northern hemispheric climate system.