Deep Ocean Learning of Small Scale Turbulence

• Published in: Geophysical research letters 2022-08, Vol.49 (15), p.n/a
• Main Authors: Mashayek, Ali, Reynard, Nick, Zhai, Fangming, Srinivasan, Kaushik, Jelley, Adam, Naveira Garabato, Alberto, Caulfield, Colm P.
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial intelligence; Carbon; Climate models; diapycnal mixing; Hybrid systems; Learning algorithms; Machine learning; Meridional overturning circulation; Nutrients; Ocean circulation; Ocean currents; Ocean models; ocean turbulence; Oceanic turbulence; Oceans; Parameterization; Physics; Pollutants; Salinity; Tracers; Turbulence; Turbulence data; Turbulent mixing; Water circulation
• ISSN: 0094-8276; 1944-8007
• DOI: 10.1029/2022GL098039

Turbulent mixing at the sub‐meter scale is an essential component of the ocean's meridional overturning circulation and its associated global redistribution of heat, carbon, nutrients, pollutants, and other tracers. Whereas direct turbulence observations in the ocean interior are limited to a modest collection of field programs, basic information such as temperature, salinity, and depth is available globally. Here, we show that supervised machine learning algorithms can be trained on the existing turbulence data to develop skillful predictions of the key properties of turbulence from T, S, Z, and topographic data. This constitutes a promising first step toward a hybrid physics‐artificial intelligence approach to parameterization of turbulent mixing in ocean and climate models. Plain Language Summary Ocean turbulence plays an important role in sustaining the general ocean circulation and in the mixing of heat, carbon, nutrients, and other tracers within the ocean interior. Turbulent mixing is technically challenging to measure and is often inferred from measurable quantities using parameterizations that are based on numerous simplifying assumptions about the physics of turbulence. In this study, we show that artificial intelligence (more specifically, various machine learning algorithms) can be successfully employed to infer turbulent mixing from quantities measured routinely by global observational programs. Key Points Machine learning can be used to infer ocean turbulent mixing from basic seawater and geometric properties The machine learning models are trained based on limited available direct turbulence measurements The trained models can be applied to data from global observational programs, which do not sample turbulence directly