Elucidating Driving Mechanisms in North Atlantic and Southern Ocean Dynamics: Physics-Informed and Trustworthy Machine Learning for Ocean Science
Séminaire de Maike Sonnewald (University of California, Davis)
The global ocean plays a central role in maintaining the health of our planet and regulating various critical factors, such as heat and carbon levels, biological productivity, and sea level. Despite its importance, there are still unresolved questions regarding the driving forces behind even major circulation features, which limits our ability to monitor and understand ongoing changes. In this presentation, I will discuss the basin-scale circulation in the North Atlantic and Southern Ocean. First I will use a physics-guided machine learning methodology to construct hypotheses regarding the balance of drivers in the global ocean. This approach provides a new way to understand the primitive equations. Looking at the Southern Ocean, I will reveal a new unifying framework to understand the gyre circulation, which is critical to the upwelling that is key to climate. Secondly, I will present a groundbreaking methodology to infer subsurface circulation by using a trustworthy neural network that reasons using geophysical fluid dynamics. When used on climate models, this methodology can detect changes in dynamics associated with the Atlantic Meridional Overturning Circulation and reveal differences in model physics that could explain the roots of climate projection uncertainties. Finally, I will discuss the problems of using machine learning as a black box and present solutions. Throughout this talk I will emphasize the use of data science for discovery, which opens doors to gain new knowledge.
Professor Maike Sonnewald is an Assistant Professor in the Computer Science Department and a CAMPOS faculty scholar, an Affiliate Assistant Professor at the University of Washington and an Affiliate Researcher at NOAA-GFDL. Focused on the ocean and climate, she uncovers underlying principles that govern ocean dynamics from small to global scales. She has pioneered methods for physics-informed machine learning models, automated regime identification, and trustworthy machine learning applications. Dr. Sonnewald’s core motivation is to deliver actionable information to decision-makers. Her focus is on the physical and biological ocean, and she is passionate about bringing the different branches of oceanography together. Her influence spans national and international policy, fundamental scientific discovery, and climate model development. Her work has been featured prominently in the NOAA Artificial Intelligence strategy (2021-2025) and informed the basis for New Zealand’s Marine Protected Area legislation. Her review articles include an invited contribution on machine learning applications in oceanography. She has given over 60 invited talks, including to the NOAA Research senior management, the Department of Energy, colloquia, and major conferences. She serves as Associate Editor at the journal Artificial Intelligence for the Earth System (American Meteorological Society) and is affiliated with the NOAA Geophysical Fluid Dynamics Laboratory.