Data‐Driven Identification of Turbulent Oceanic Mixing From Observational Microstructure Data

Abstract

Characterizing how ocean turbulence transports heat is critically important for accurately parameterizing global circulation models. We present a novel data‐driven approach for identifying distinct regions of turbulent mixing within a microstructure data set that uses unsupervised machine learning to cluster fluid patches according to their background buoyancy frequency N, and turbulent dissipation rates of kinetic energy ϵ and thermal variance χ. Applied to data collected near the Velasco Reef in Palau, our clustering algorithm discovers spatial and temporal correlations between the mixing characteristics of a fluid patch and its depth, proximity to the reef and the background current. While much of the data set is characterized by the canonical mixing coefficient Γ = 0.2, elevated local mixing efficiencies are identified in regions containing large density fluxes derived from χ. Once applied to further datasets, unsupervised machine learning has the potential to advance community understanding of global patterns and local characteristics of turbulent mixing.

Document Details

Document Type
Pub Defense Publication
Publication Date
Dec 09, 2021
Source ID
10.1029/2021gl094978

Entities

People

  • Bethan Wynne‐Cattanach
  • Colm-Cille Caulfield
  • Gunnar Voet
  • Jennifer MacKinnon
  • Matthew H. Alford
  • Miles M. P. Couchman
  • Rich R. Kerswell

Organizations

  • Office of Naval Research
  • University of California, San Diego
  • University of Cambridge

Tags

Fields of Study

  • Environmental science

Readers

  • Coastal Oceanography
  • Fluid Mechanics and Fluid Dynamics.
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks