Implementation and Evaluation of a Physics-Informed Machine Learning Parameterization for Ocean Surface Boundary Layer Turbulence in HYCOM
Abstract
Turbulence in the ocean surface boundary layer, the near-surface ocean directly impacted by the atmosphere, plays an important rolein the ocean and its coupling with the atmosphere. It modulates the temperature, material concentrations, current in the near-surface ocean, and the ocean-atmosphere exchange of heat and material. Ocean surface boundary layer turbulence cannot be resolved in ocean circulation models, and its mixing effect is represented by parameterizations in those models. The parameterization of ocean surface boundary layer turbulence is one of the remaining challenges in ocean circulation models, leading to biases in the simulation of sea surface temperature and mixed layer depth. The accurate simulation of sea surface temperature and mixed layer depth is crucial to the prediction of the physical states of the upper ocean, the exchange between surface and interior oceans, the coupling between the ocean and the atmosphere, and underwater acoustic propagation.The overarching objective of the proposed project is to improve theHYbrid Coordinate Ocean Model (HYCOM) in simulating sea surface temperature and mixed layer depth by implementing a new physics-informed data-driven machine learning algorithm for the parameterization of vertical mixing in the ocean surface boundary layer. This new parameterization has the potential to assuage current biases in the simulated upper-ocean states and to improve forecasts of sea surface temperature and mixed layer depth. Our preliminary study is the first one that developed and assessed a data-driven parameterization for vertical mixing in the ocean surface boundary layer and its proposed implementation in a regional HYCOM configuration is quite novel and will be the first attempt to resolve a long-standing issue in ocean circulation models using state of the art physics-informed machine learning models. Research activities to achieve the objective are (1) generation of training data using a state-of-the-art large eddy simulation model that resolves ocean surface boundary layer turbulence; (2) training of the machine learning algorithm; (3) implementation of the machine learning algorithm in the HYCOM codes, and (4) the assessment of the HYCOM model with the machine learning based mixing parameterization using a configuration for the Gulf of Mexico. The assessment will be performed on the HYCOM with and without data assimilation. The updated HYCOM codes and the framework to train the machine-learning algorithm could be applied to other ocean regions. The updated HYCOM codes will be sent to the HYCOM GitHub repository.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 29, 2023
- Source ID
- N000142312553
Entities
People
- Jun‐Hong Liang
Organizations
- Louisiana State University
- Office of Naval Research
- United States Navy