Building the next generation of ocean-climate model with machine learning

Speaker

Alistair J. Adcroft

Atmospheric & Oceanic Sciences

Princeton University

Abstract

Building the next generation of ocean-climate model with machine learning

Machine learning and GPU-based computing are fundamentally reshaping our ability to model the ocean's central role in regulating Earth’s climate. While traditional models have made significant advances, they still face major challenges in simulating the interplay of fine-scale turbulence and long-term climate dynamics within computational constraints. Hybrid approaches that integrate data-driven techniques with physically-grounded models offer a powerful path forward.

This talk will explore this new frontier, demonstrating how we can build more accurate and efficient ocean-climate simulations. We will consider two key areas where machine learning is making a transformative impact: data-driven parameterization to represent unresolved sub-grid processes that were previously major sources of uncertainty; and accelerated simulation that leverages surrogates to speed up computationally expensive components of the models. Drawing lessons from decades of development in traditional numerical solvers, we will ask whether the new techniques satisfy the conservation and accuracy requirements of both global and coastal ocean applications. Ultimately, these advancements are not just technical; they pave the way for more democratized climate prediction, open science, and actionable decision support.

Bio

Alistair Adcroft is a computational oceanographer whose work sits at the intersection of climate science, applied mathematics, and high-performance computing. He earned his Ph.D. in Physics from Imperial College London in 1995, where his thesis introduced a finite volume formulation for ocean models. Following a UCAR Ocean Modeling Postdoctoral Fellowship at UCLA, he joined MIT, where he spent seven years as a research scientist and later principal research scientist. At MIT, he was a lead developer of the MITgcm, a widely adopted general circulation model used for both large-scale simulations and process studies, including non-hydrostatic applications.

Since 2003, Adcroft has been a Research Oceanographer in the Program in Atmospheric and Oceanic Sciences at Princeton University, collaborating closely with NOAA’s Geophysical Fluid Dynamics Laboratory (GFDL). He contributed to the development of the ESM2G and ESM2M Earth system models and led the team that developed GFDL’s OM4 ocean-ice model. He’s the co-creator of the sixth version of the Molular Ocean Model (MOM6) which is now the primary ocean circulation model at NOAA and other centers around the US and the world. His research focuses on numerical methods, model formulation, and the parameterization of sub-grid processes, with applications ranging from global scales to sub-mesoscale eddies and ice-ocean interactions. Adcroft has also worked on models of sea ice, tabular icebergs and collapsing ice shelves, all coupled to the ocean for climate studies. His recent work integrates machine learning into ocean modeling, including the development of hybrid parameterizations and AI-based emulators for climate-scale simulations. He co-leads the M2LInES project, a multi-institutional effort to apply machine learning to coupled Earth system modeling.

Adcroft serves on several advisory boards, including the Community Earth System Model Scientific Steering Committee and chaired the World Climate Research Programme’s CLIVAR Ocean Model Development Panel. In recognition of his contributions to ocean modeling, he received the 2021 Ocean Sciences Award from the American Geophysical Union.