AI-Enhanced Operational Ocean Forecasting in the Gulf of Mexico
This project puts machine learning inside the operational ocean forecasting workflow for the Gulf of Mexico, from vertical mixing to satellite data assimilation and the reconstruction of subsurface temperature and salinity.
Olmo Zavala-Romero Jose Miranda Eric P. ChassignetLuna HironJianguo Yuan
Goals
Improve the accuracy, efficiency, and utility of operational ocean forecasts in the Gulf of Mexico by integrating machine learning into core modeling workflows. This research line targets three bottlenecks in operational prediction: vertical mixing parameterizations, data assimilation of satellite observations, and reconstruction of subsurface fields and mesoscale eddy processes when in-situ data are sparse.
Methods
We combine physics-based ocean models with deep learning across complementary pathways. Neural networks augment the K-Profile Parameterization (KPP-DNN) in GOTM to improve vertical mixing estimates. Convolutional neural networks accelerate and enhance sea surface data assimilation in operational Gulf of Mexico model configurations. NeSPReSO uses satellite surface observations and PCA-based neural networks to generate synthetic temperature and salinity profiles for data-sparse regions. Separately, machine learning models fuse altimetry, SST, and chlorophyll-a to identify and predict Lagrangian coherent eddies that control tracer transport in the Gulf.
Key findings
KPP-DNN improves turbulent mixing estimates in GOTM relative to conventional KPP, providing a learned correction pathway for vertical physics in column models. CNN-based sea surface assimilation demonstrates that deep learning can be applied to operational-scale ocean models, not only simplified test cases. NeSPReSO reconstructs subsurface T/S profiles from satellite data, addressing a key constraint on forecast reliability in the Gulf of Mexico. ML eddy identification links satellite fields to Lagrangian coherent structures relevant to oil transport, Loop Current dynamics, and mesoscale monitoring — connecting process studies to operational observability.
Publications from this project
- The K-Profile Parameterization Augmented by Deep Neural Networks KPP-DNN in GOTM 2024
- Identifying and Predicting the Lagrangian Coherence of Eddies in the Gulf of Mexico Using Machine Learning and Satellite Observations 2025
- Neural Synthetic Profiles from Remote Sensing and Observations NeSPReSO - Reconstructing Temperature and Salinity Fields in the Gulf of Mexico 2025
- Convolutional Neural Networks for Sea Surface Data Assimilation in Operational Ocean Models - Gulf of Mexico Test Case 2025