Collaborative research: caig: characterization of 3D flow structure in transient headland eddies using model-integrated adaptive sampling with multiple auvs

Pacific Lutheran University received a $118,725 Project Grant from the National Science Foundation's (NSF) Geosciences program (CFDA 47.050) awarded on October 1, 2025, with a completion date of September 30, 2028. This collaborative research project develops an innovative observing system that integrates artificial intelligence (AI) and machine learning with multiple autonomous underwater vehicles (AUVs) to characterize the three-dimensional structure of tidally-driven coastal headland eddies. The primary deliverable is a learning-based digital twin trained on hydrodynamic models that enables real-time adaptive sampling capabilities aboard AUVs while maintaining physical consistency in model outputs.

The research will produce in-situ oceanographic data and advanced modeling frameworks that characterize convergence streaks within eddies and diagnose the mechanisms driving their evolution. Key products include communication-aware coordinated adaptive sampling strategies designed to minimize data transmission requirements over low-bandwidth connections, as well as a novel framework combining traditional hydrodynamic modeling with AI methods for autonomous adaptive sampling. The project's outcomes will enhance understanding of how transient coastal eddies influence the aggregation and dispersal of marine constituents, including coral larvae and pollutants, thereby advancing coastal ocean process knowledge through fundamental research in ocean sciences.

Abstract

This project will develop a novel observing system that uses artificial intelligence (AI) to coordinate multiple autonomous underwater vehicles (auvs). The system will be used to characterize the three-dimensional structure of tidally-driven coastal headland eddies, investigate spatiotemporal variability of the small-scale flow convergence within the eddies, and understand the underlying mechanisms and their influence on the dispersal of marine materials. This approach is designed to overcome computational barriers of traditional models to adaptive sampling aboard auvs. The learning-based digital twin will be trained with a hydrodynamic model to approximate full model data assimilation for rapid short-horizon forecasts. This work will establish a framework for combining traditional hydrodynamic modeling with novel AI and machine learning methods to achieve a deeper understanding of key dynamics of coastal ocean processes using autonomous adaptive sampling. The strategically obtained in-situ data will be used with the hydrodynamic model to identify smaller-scale convergence streaks in the eddy, characterize their three-dimensional structure, and diagnose the mechanisms that drive their evolution. The result will further our ability to understand the contribution of transient eddies in aggregation or dispersal of marine constituents, such as coral larvae and pollutants. The constraints imposed by real-time use onboard auvs necessitate new learning-based models that approximate traditional geophysical models, while maintaining guarantees on the physical consistency of the model output. Finally, to maximally leverage the predictions of the digital twin, we will also develop communication-aware coordinated adaptive sampling strategies that minimize the amount of data that the auvs need to transmit over a low-bandwidth underwater acoustic modem. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the foundation's intellectual merit and broader impacts review criteria.- subawards are not planned for this award.