Stony Brook and City College Begin $886,293 AI Study of Urban Flooding

Stony Brook University and The City College of New York have begun a $886,293 research project aimed at explaining why nearby streets flood differently during the same storm. The promise is better understanding of hidden drainage conditions—not a proven forecasting system. Researchers will combine AI-derived observations with physics-based flood models, initially focusing on Greater New York.

Stony Brook described the new collaboration on October 6. It began October 1, 2026, and is scheduled to run through September 30, 2029. Funding comes through the National Science Foundation’s Collaborations in Artificial Intelligence and Geosciences program, with $537,060 allocated to Stony Brook and $349,233 to City College.

Turning street images into flood measurements

The first challenge is observing what happened. Detailed flood measurements, especially water depth, are often sparse during fast-moving storms, said co-principal investigator Paola Cascante-Bonilla. Her team plans to draw on traffic cameras, street-level images and public reports—sources that were not created as scientific sensors.

The proposed AI approach is multimodal: it combines visual, written and location-based information to estimate flood conditions. Cascante-Bonilla said the research will investigate water-depth estimates that also express uncertainty and recognize when the evidence is too weak for a confident prediction. Those observations are intended to complement physical models where conventional measurements are limited.

In particular, we are investigating how to produce useful depth estimates while also representing uncertainty and recognizing when the available evidence is not strong enough to make a confident prediction.

Paola Cascante-Bonilla, Stony Brook co-principal investigator

The mismatch is part of the investigation

Principal investigator Te Pei described a second information gap. Researchers have increasingly good data on rainfall, terrain and infrastructure locations, but do not always know how drainage systems perform during a storm. Those difficult-to-observe conditions can strongly affect where flooding occurs and how severe it becomes, he said.

The team will compare observed flooding with physics-based simulations. Pei said a mismatch between a model and street conditions can reveal information about how the real system functions. Rather than using AI only to predict flooding, the researchers want to connect surface observations to hidden urban properties and explain different responses under similar rainfall.

Testing across storms and neighborhoods

The project will combine flood-event datasets, simulations and AI to study different storms, neighborhoods and cities. Pei’s group contributes physical modeling and urban-hazard expertise; Cascante-Bonilla’s group brings computer vision and multimodal AI. Naresh Devineni leads the City College award, with his team contributing hydrology, extreme-precipitation and flood-risk expertise.

Stony Brook expects the work to support flood-risk assessment and infrastructure planning. The collaboration also includes interdisciplinary training for students in civil engineering, computer science, AI and hydrology.