ReDDDoT Phase 2: Responsible Multi-Modal AI Systems for Multi-Hazard Resilience and Situational Awareness

Coastal storms and climate change, aging infrastructure, and rapid urbanization pose increasing risks to coastal communities. Institutions charged with supporting communities before, during, and following storm events require reliable and timely information on current and forecasted hydrometeorological conditions and infrastructure impacts, including roadway access and potential natural hazards-triggered technological incidents. Recent focus groups and structured interviews with emergency response organizations have revealed both the lack of integrated information sources and the lack of trusted, timely, and scientifically sound technology available to support situational awareness of compound hazard events and their anticipated impacts on infrastructure, a deficiency that hampers decision-making and response efforts. This project will design, develop, and deploy OpenSafe.AI, a framework that advances communities’ ability to reliably sense current conditions and forecast potential hazards and infrastructure impacts. This information is critical to inform response and recovery actions targeted at public health and safety and enhanced community resilience to coastal storm events. Working in concert with emergency response agencies in the Houston-Galveston area, we will not only iteratively design and tailor such a system, but probe transferability and scalability, design robustness, and data and model equity across diverse communities, including those that are under-resourced and under-served. This project combines expertise in hazard and infrastructure resilience modeling, user-centered design, and responsible AI to revolutionize intelligent systems for situational awareness and scenario exploration under multiple compound coastal hazards. This convergent research will address the technical, theoretical, and methodological gaps in responsibly designing and developing situational awareness tools to support emergency response actions and risk mitigation interventions during tropical cyclones and coastal storm events. With an overarching user-centered design approach, it will pioneer responsible design strategies to enable (1) equitable and fair, (2) reliable and safe, (3) human-centered applications of AI in the disaster resilience domain. Along the way, the team will develop multi-modal foundation models that gain insights from a combination of physics-based, data-driven, and human-in-the-loop sources, and will advance methods to detect and largely overcome systemic bias, paucity of real-time data, and equity issues in models and data to promote equitable and fair situational awareness. As a result, the OpenSafe.AI framework pursues estimates of near-real time conditions and short-range forecasts (e.g., hours to days in advance) of multi-hazards (e.g., wind, wave, compound flooding) and their impacts on the built environment (e.g., damage hampering access to critical facilities or yielding hazardous material spills), thereby affording practical, timely, and equitable situational awareness. 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.