Unsupervised Learning Reveals Geography of Global Ocean Dynamical Regions

Area of Science:

• Oceanography
• Fluid Dynamics
• Climate Science

Background:

• Understanding ocean circulation is crucial for climate modeling and predicting oceanic changes.
• Previous studies often relied on regional analyses or simplified dynamic frameworks.

Purpose of the Study:

• To objectively identify and classify dynamically similar regions across the global ocean.
• To characterize the dominant physical balances governing ocean dynamics in different regions.

Main Methods:

• Utilized a 20-year mean from the Estimating the Circulation and Climate of the Ocean (ECCO) state estimate at 1° resolution.
• Applied an unsupervised machine learning algorithm (K-means) to cluster the standardized barotropic vorticity (BV) equation.
• Analyzed the dominant terms in the BV equation for each identified cluster.

Main Results:

• Identified five distinct ocean dynamic regimes based on BV framework.
• Cluster 1 (43% of ocean) shows surface/bottom stress torque balanced by bottom pressure and nonlinear torque.
• Cluster 2 (24.8%) features beta effect balancing bottom pressure torque; Cluster 3 (14.6%) is Quasi-Sverdrupian (beta effect balanced by wind/bottom stress).
• Cluster 4 (6.9%) exhibits baroclinic dynamics, and Cluster 5 is prominent in the Southern Ocean.
• Nonlinear regions, particularly near rough topography and western boundaries, indicate limitations of the BV approach.

Conclusions:

• The study objectively delineates global ocean regions based on barotropic vorticity dynamics.
• The findings provide a new framework for understanding regional ocean dynamics and improving climate models.
• Identified areas where the barotropic vorticity framework is insufficient, highlighting the need for more complex models in specific regions.