20 May 2021

20 May 2021

Review status: this preprint is currently under review for the journal OS.

Defining Southern Ocean fronts using unsupervised classification

Simon D. A. Thomas1,2, Daniel C. Jones1, Anita Faul1, Erik Mackie1,3, and Etienne Pauthenet4 Simon D. A. Thomas et al.
  • 1British Antarctic Survey, NERC, UKRI, UK
  • 2Department of Physics, University of Cambridge, UK
  • 3Cambridge Zero, University of Cambridge, UK
  • 4Sorbonne Universités, UPMC Université, LOCEAN-IPSL, Paris, France

Abstract. Oceanographic fronts are transitions between thermohaline structures with different characteristics. Such transitions are ubiquitous, and their locations and properties affect how the ocean operates as part of the global climate system. In the Southern Ocean, fronts have classically been defined using a small number of continuous, circumpolar features in sea surface height or dynamic height. Modern observational and theoretical developments are challenging and expanding this traditional framework to accommodate a more complex view of fronts. Here we present a complementary new approach for calculating fronts using an unsupervised classification method called Gaussian mixture modelling and a novel inter-class parameter called the I-metric. The I-metric approach produces a probabilistic view of front location, emphasising the fact that the boundaries between water masses are not uniformly sharp across the entire Southern Ocean. The I-metric approach uses thermohaline information from a range of depth levels, making it more general than approaches that only use near-surface properties. We train the statistical model on data from an observationally-constrained state estimate for more uniform spatial and temporal coverage. The probabilistic boundaries appear to be relatively sharp in the open ocean and somewhat diffuse near large topographic features, possibly highlighting the importance of topographically-induced mixing. For comparison with a more localised method, we use edge detection in principal component space and correlate the edges with surface velocities. The I-metric approach may prove to be a useful method for inter-model comparison, as it uses the thermohaline structure of those models instead of tracking somewhat ad-hoc values of sea surface height and/or dynamic height, which can vary considerably between models. In addition, the general I-metric approach allows front definitions to shift with changing temperature and salinity structures, which may be useful for characterising fronts in a changing climate.

Simon D. A. Thomas et al.

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on os-2021-40', Anonymous Referee #1, 24 Jul 2021
    • AC1: 'Reply on RC1', Simon Thomas, 11 Sep 2021
  • RC2: 'Comment on os-2021-40', Anonymous Referee #2, 26 Jul 2021
    • AC2: 'Reply on RC2', Simon Thomas, 11 Sep 2021
  • RC3: 'Comment on os-2021-40', Anonymous Referee #3, 14 Aug 2021
    • AC3: 'Reply on RC3', Simon Thomas, 11 Sep 2021

Simon D. A. Thomas et al.

Model code and software

so-wise/so-fronts: Initial submission to Ocean Sciences (Version v-0.1) Simon D. A. Thomas. (2021, May 6)

Simon D. A. Thomas et al.


Total article views: 747 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
544 186 17 747 11 14
  • HTML: 544
  • PDF: 186
  • XML: 17
  • Total: 747
  • BibTeX: 11
  • EndNote: 14
Views and downloads (calculated since 20 May 2021)
Cumulative views and downloads (calculated since 20 May 2021)

Viewed (geographical distribution)

Total article views: 691 (including HTML, PDF, and XML) Thereof 691 with geography defined and 0 with unknown origin.
Country # Views %
  • 1


Latest update: 17 Sep 2021
Short summary
We propose a simple probabilistic method for highlighting fronts in the Southern Ocean with model data from below the mixed layer. We compare it with an image processing method that provides a more localised view of fronts that effectively highlights currents. We suggest that our new probabilistic method has multiple advantages, including not being reliant on surface data.