Machine Learning for Submesoscale Characterization, Ocean Prediction, and Exploration (ML-SCOPE)
Project Summary
Our long-term goal is to obtain machine-intelligent modeling systems that seamlessly integrate stochastic ocean dynamical models and their multi-fidelity representations with Bayesian and generative learning from data-model misfits, to construct improved ocean models with more accurate parameterizations and discover invariances or differential equations, over a range of spatial and temporal scales. We will use and vastly extend stochastic dynamic reduced-order methods, Bayesian GPs, adaptive DL schemes, and generative networks. Ultimately, our symbolic interpretation of ML models into emergent dynamical and constitutive relations would further compress knowledge relative to deep networks, thus extending outside the range of the training data.
Specific Objectives
- Learn from heterogeneous measured data sets, multi-resolution simulated fields from three ocean modeling systems, and data-assimilative simulations in several ocean regions and basins, and the global ocean
- Use and vastly extend hierarchical differential-equation-based Bayesian learning, stochastic dynamic reduced-order methods, data-driven closure models, Bayesian Gaussian Processes, adaptive DL schemes, and generative models and adversarial networks
- Obtain submesoscale ML super-parameterizations
- Develop ML for data assimilation and ML-based adaptive sampling to identify the most informative data for model learning
- Refine and incubate methods using idealized and semi-realistic test cases, and quantify and optimize their robustness, and verify ML results using novel metrics of success.
Co-Principal Investigators:
Pierre F.J. Lermusiaux (MIT)
Tamara Broderick (MIT)
Lawrence Carin (Duke University)
Eric P. Chassignet (Florida State University)
Mickaël D. Chekroun (UCLA)
Stefanie Jegelka (MIT)
James C. McWilliams (UCLA)
Tamay M. Özgökmen (RSMAS - University of Miami)