Collaborative Research: CAIG: Improving the understanding of coastal groundwater and ocean exchange using machine learning

Rising seas, stronger storms, saltwater contamination of freshwater resources, and the degradation of coastal ecosystems increasingly challenge coastal communities. These pressures affect drinking water supplies, agriculture, infrastructure, local economies, and natural environments that millions of people rely on every day. This project seeks to improve understanding of these interconnected challenges by developing new tools for studying the exchange of water between coastal aquifers and the ocean. Using recent advances in artificial intelligence (AI) alongside established environmental modeling approaches, the project will enable the forecast of important coastal risks, including seawater intrusion into freshwater aquifers and submarine groundwater discharge, with greater speed and accessibility. These capabilities will support more informed decisions about water management, infrastructure planning, and long-term coastal resilience in vulnerable coastal regions. The project will also produce open-source software and visualization tools and provide interdisciplinary training opportunities for students and early-career researchers, helping prepare the next generation of scientists and engineers to address pressing environmental problems. This project will develop a next-generation computational framework for simulating the dynamic interactions between coastal groundwater systems and the ocean. The research combines neural operators and an open-source coupling framework to build novel, hybrid physics-aware surrogate models that emulate computationally intensive groundwater and ocean simulations at a fraction of the cost, while maintaining crucial conservation laws at the land-sea boundary through dynamic interface coupling. Furthermore, specialized neural network experts will be combined following a mixture-of-experts philosophy to capture phenomena ranging from long-term tidal and circulation patterns to short-lived events such as storm surges. These models will be integrated into a unified machine-learning framework, capable of representing coupled processes that operate across widely different spatial and temporal scales. The framework will be tested using benchmark applications based on Hawaiʻi and Texas coastal systems. Moreover, the differentiable nature of the proposed architecture will facilitate data assimilation and posterior uncertainty quantification using generative AI priors and multi-fidelity Bayesian inference methods. By enabling near-real-time simulation of seawater intrusion, submarine groundwater discharge, and related coastal processes, this project will advance both scientific machine learning and coastal geoscience while laying the groundwork for scalable AI-enabled digital twins of complex environmental systems. 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.