CAREER: Open-source Probabilistic AI Infrastructure for Dynamic Cascading Disaster Impacts Modeling and Assessment

No-notice and short-notice natural hazards, such as earthquakes and hurricanes, often cause cascading impacts. Timely and accurate modeling and assessment of hazard impacts are critical for rapid disaster response and long-term recovery and for enhancing disaster resilience. Recent developments in remote sensing technologies continue to bring massive data on disaster impacts.

However, it remains challenging for end users like emergency management agencies and researchers to directly and accurately decipher the location, severity, and uncertainty of disaster impacts from these datasets, mainly due to the highly complex and uncertain causal dependencies in the cascading physical-digital disaster processes. This project aims to overcome this barrier by pioneering the development of next-generation scalable, efficient, and interpretable probabilistic AI infrastructure to enable joint modeling and assessment of cascading disaster impacts and to advance STEM education and emergency management practice through integrated research, education, training, and outreach framework.

It aligns with NSF's commitment to promoting the progress of science and facilitating breakthroughs in AI and resilience. The infrastructure can potentially improve emergency management and enhance community resilience across the U.S. and worldwide by providing near real-time spatial probability estimates of multiple disaster impacts. It can advance the knowledge of complex causal dependencies in compound hazard scenarios.

It also may enable the advancement of multi-modal, multi-resolution learning, causal Bayesian networks, causal dependency modeling, and probabilistic inference. Multiple education and outreach activities are integrated into this research, including co-creating computation- and art-infused high school teaching modules with high school art and STEM teachers, engaging diverse students in research activities, improving curriculum design, and convening professional training sessions.

These activities will foster interest in computational and interdisciplinary science, raise awareness of natural hazard impacts, promote diversity and inclusion, and facilitate workforce training.

The technical objective of this project is to develop a next-generation scalable, efficient, and transparent causality-informed probabilistic AI infrastructure that supports the automatic construction and inference of various complex disaster-type Bayesian networks and thus enables joint modeling and assessment of cascading disaster impacts. This cyberinfrastructure will advance modeling, integration, and inference of complex causal Bayesian networks for dynamic cascading disaster impacts assessment and enhancing disaster resilience by enabling the convergence of geophysical hazard/infrastructure damage models, causal representation, multi-resolution data assimilation, deep probabilistic graphical modeling and inference, and domain-specific probabilistic programming.

Specifically, the project introduces the following cyberinfrastructure innovations. First, a novel causal Bayesian network formulation and modeling framework will be developed to allow flexible, interpretable, and structured probabilistic representations of various disasters, environment factors, multi-resolution multi-modal sensing observation, and complex causal dependencies.

A novel variational inference framework will be developed to enable fast inference over various causal Bayesian network structures for large-scale cascading disaster impact assessment from multi-modal multi-resolution observations. The framework will be further extended to enable online updating with newly arrived sparse ground truth disaster impact reports.

The developed modeling, inference, and preprocessing modules will be evaluated and integrated into an efficient GPU-accelerated probabilistic programming software infrastructure with user instructions made available to the research and emergency response community.

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.