I-Corps: Translation Potential of Explainable Artificial Intelligence (AI) for Satellite-Derived Nearshore Bathymetry and River Profiles

This I-Corps project is based on the development of explainable artificial intelligence (AI) that converts satellite imagery into high-resolution maps of underwater depth (bathymetry) for coastlines, rivers, and lakes. Nearly half of U.S. waters remain unmapped to modern standards and ship- and aircraft-based surveys are too costly and slow to keep pace with changing conditions. This technology produces rapid, repeatable, and affordable depth maps from routinely collected satellite data, including automated updates after major storms, each accompanied by calibrated confidence estimates showing where the data can be trusted. Applications include flood and storm forecasting, water-supply and reservoir management, safe navigation and charting, coastal engineering, and fisheries and habitat assessment. This technology may serve government agencies, engineering firms, and environmental data providers to improve public safety and community resilience. This I-Corps project utilizes experiential learning coupled with first-hand investigation of the industry ecosystem to assess the translation potential of an explainable, uncertainty-aware machine learning framework for satellite-derived bathymetry across coastal, river, and lake systems. The framework combines physics-informed neural networks embedding radiative-transfer attenuation constraints with interpretable architectures. It maintains sub-meter accuracy in turbid, optically complex waters where satellite-derived methods have historically degraded. The same architecture maps lake bathymetry, validated at dedicated lake sites, while river depth and velocity are derived from satellite altimetry and model-derived daily discharge data. Unlike approaches that produce point estimates alone, the framework returns calibrated uncertainty bounds that flag where predictions are reliable and guide targeted refinement. This capability is essential when data products feed operational hazard, engineering, and navigation models. An agentic model-design framework transfers model structures to new domains and incorporates expert and community knowledge into later modeling stages, enabling scalable coverage without site-by-site redevelopment. 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.