CAREER: Drain or Retain? Strategic Assessment of Integrated Flood Infrastructure Portfolios with Physics-Informed ML

The objective of this Faculty Early Career Development Program (CAREER) project is to support research on how combinations of flood and stormwater infrastructure perform as a system under realistic storm conditions. Flooding causes more economic damage than any other US natural hazard. Communities invest in stormwater drainage, detention basins, and river modifications, but these projects are typically designed one at a time using simplified design storms that do not capture how risks to homes, roads, and critical services vary from storm to storm, or how infrastructure interactions redistribute those risks. A project that reduces flooding in one area during one storm can worsen it elsewhere during another. Using Greater Houston -- where recent floods have motivated over $2.5 billion in adaptation -- as a testbed, this project seeks to develop methods for evaluating how these projects perform together across a wide range of storms. The project promotes education and workforce development through open-source teaching modules, a Teaching Fellows program that trains professionals to evaluate computational flood risk models, and a Vertically Integrated Project that engages undergraduates in research. All code, datasets, and teaching materials will be released as open-source. The project develops physics-informed machine learning emulators trained on detailed hydrodynamic flood simulations to evaluate how large ensembles of storms -- varying in intensity, duration, spatial pattern, and movement -- interact with spatially distributed infrastructure configurations to produce flood hazards and risks. The research addresses three questions: (1) what spatial patterns of flood risk emerge from variability between and within storms, and how well do design storm methods capture these patterns; (2) what storm characteristics drive different types of infrastructure failure, and can those thresholds be predicted; and (3) when and how do combinations of infrastructure provide benefits beyond what individual projects achieve. To address these questions, the project constructs probabilistic storm catalogs from historical and synthetic records, couples them with multi-scale flood models that represent infrastructure operations, and trains physics-informed machine learning models to efficiently simulate thousands of storm-infrastructure combinations. 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. NSF Award ID: 2543303 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: James Doss-Gollin | Institution: William Marsh Rice University, HOUSTON, TX | Award Amount: $599,581 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2543303 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2543303.html