Artificial intelligence (AI) can help reconstruct the ocean interior, predict waves and forecast the El Niño Southern Oscillation (ENSO) — but can it do so in ways that remain consistent with ocean physics? The answer is yes, according to a recent paper available online now and slated for upcoming publication in Ocean-Land-Atmosphere Research .
The paper draws on discussions and representative studies presented at The Fifth Forum on Artificial Intelligence Oceanography, held earlier this year in Jinan, Shandong Province, China. The forum brought together more than 300 experts, scholars and students from more than 80 institutions.
A clear consensus emerged from the talks and discussions at the forum, according to the authors: the next stage of AI oceanography will increasingly rely on models that combine the flexibility of learning algorithms with the governing principles of ocean dynamics. The article emphasizes that AI will be most valuable when it learns in ways that are consistent with ocean dynamics. Physics-constrained approaches incorporate physical knowledge, dynamical equations, process-based relationships into AI learning and evaluation. These approaches can improve reliability, interpretability and performance when observations are sparse or environmental conditions differ from the training data.
In practical terms, historical observations and model simulations provide the data from which AI learns patterns, while incorporating physical relationships can help predictions remain more reliable when observations are sparse or environmental conditions change.
Drawing on the forum, the authors highlighted three complementary classes of oceanographic problems: reconstructing subsurface ocean states from surface information, long range climate prediction and operational wave forecasting. The studies presented at the forum and summarized in the paper showed that incorporating physical knowledge, or learning physically meaningful relationships among ocean and climate processes, can improve the reliability, interpretability and usefulness of AI-based predictions.
“The next step is to develop AI systems that combine physical constraints, uncertainty estimation and real ocean observations more systematically,” the authors noted. “Our ultimate goal is to make AI a trustworthy complement to numerical ocean models — one that can support faster and more reliable ocean forecasting, climate prediction, marine hazard warning and scientific understanding.”
Co-authors of the paper include corresponding author Changming Dong and Huarong Xie, School of Marine Sciences, Nanjing University of Information Science and Technology; Guangliang Liu with the State Key Laboratory of Physical Oceanography and Qilu University of Technology (Shandong Academy of Sciences); Xiang Gong, School of Mathematics and Physics, Qingdao University of Science and Technology; Wenfang Lu, School of Marine Sciences, Sun Yat-Sen University, and Southern Marine Science and Engineering Guangdong Laboratory; Guangjun Xu, College of Electronic and Information Engineering, Guangdong Ocean University; and Xueming Zhu, Southern Marine Science and Engineering Guangdong Laboratory.
#
Ocean-Land-Atmosphere Research
Commentary/editorial
Not applicable
Physics-Constrained AI Models for a More Predictable Ocean
30-Jul-2026
No conflicts of interest to declare.