Poster

IAPSO

JPCM08 - Impacts of climate change on the ocean

Classification of Seal-CTD profiles using machine learning approaches in the Ross Sea, Antarctica

Recent advancements in miniaturized seal conductivity-temperature-depth (CTD) tagging have greatly improved the monitoring of marine environmental and behavioral data during the harsh Antarctic winter. This study applied machine learning methods to analyze CTD profiles from 64 adult Weddell seals (Leptonychotes weddellii) tagged in February from 2021 to 2023. The objective was to investigate how these profiles represent oceanographic conditions in the Ross Sea and their linkage with seal foraging behavior. Temperature and salinity from each profile were linearly interpolated at consistent depth intervals from 1 to 400 meters. Principal component analysis (PCA) was performed separately on the interpolated temperature and salinity datasets, extracting the first three principal components based on explained variance. This resulted in a total of six PCA-derived variables per profile. These variables were then used as inputs for a Gaussian Mixture Model, which classified the profiles into four distinct clusters. Each cluster displayed unique temperature and salinity characteristics with predominant spatiotemporal patterns: Warm surface Cluster (near Terra Nova Bay, predominant in February), Mild surface Cluster (near Terra Nova Bay, predominant in February?April), Cold Cluster (near Terra Nova Bay, predominant in May?July), and Warm subsurface Cluster (near the shelf break, predominant in April?May). Prey capture attempts, measured as events per dive, were highest in the Warm surface Cluster (mean: 3.16 events/dive) and lowest in the Mild surface Cluster (mean: 2.91 events/dive). These findings demonstrate that machine learning-based classification of seal-CTD profiles can provide valuable insights into the environmental conditions of the Ross Sea and their implications for seal ecology and oceanographic monitoring.