Collaborative Research: CAIG: Inferring Hidden Urban Flood Controls Using Physics-Informed Multimodal Learning

Urban flooding increasingly threatens infrastructure and communities across the United States, where extreme rainfall interacts with dense built environments and aging stormwater systems. A major challenge is that cities often lack flood-depth information during storms, while many important conditions that control flood response are hidden or rarely measured. This project will investigate how reports, photographs, videos, and other unstructured evidence of flooding can help reveal hidden urban flood controls. The project will support flood-risk assessment and infrastructure planning, while training students, engaging city agencies, and broadening public understanding of urban flooding and artificial intelligence through shared data, software, and educational outreach. This project will develop a physics-informed multimodal artificial intelligence framework for inferring and explaining hidden system controls in urban flooding. Focusing on Greater New York, the project will establish flood datasets and physics-based simulation benchmarks; advance geospatial-aware language and vision technologies to estimate flood depth from unstructured evidence with quantified uncertainty; design inversion-enabled surrogate flood models that use sparse depth constraints to infer latent drainage and surface-flow controls; and analyze how inferred controls vary across storms, urban forms, and cities. Together, these efforts will establish a new modeling paradigm for inferring hidden urban flood controls from indirect flood observations. 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.