Grove School’s Naresh Devineni in Stony Brook AI collaboration to reveal hidden controls on urban flooding

Naresh Devineni, professor of civil engineering in The City College of New York’s Grove School of Engineering, is partnering with Stony Brook University experts on a National Science Foundation (NSF)-funded collaborative research project led by the SUNY institution, that combines multimodal AI with physics-informed flood modeling. 

Titled, “Collaborative Research: CAIG: Inferring Hidden Urban Flood Controls Using Physics-Informed Multimodal Learning,” the three-year study aims to better understand the hidden influences on urban flooding that can result in two nearby streets experiencing dramatically different flooding during the same storm. The answer may depend not only on rainfall and topography but also on aspects of urban drainage systems that are difficult to observe directly.

The NSF’s Collaborations in Artificial Intelligence and Geosciences (CAIG) program has provided a total of $886,293 in funding for the project, with Stony Brook receiving $537,060 and CCNY getting $349,233. 

The project addresses a fundamental challenge in urban flood modeling: some of the conditions that strongly influence where flooding occurs and how severe it becomes are rarely measured directly.

“Heavy rainfall is only part of the urban flooding story,” said Devineni, a water systems and hydroclimate expert. “We also need to understand why one street floods while another nearby stays dry. By combining the physics of how water moves through a city with AI that can learn from flood photos, videos and residents’ reports, we aim to uncover drainage limitations and other hidden factors that shape those differences. The goal is to help cities understand not just where flooding happens, but why—and use that knowledge to better target solutions.”

Assisting Devineni on the project at CCNY is his post-doctoral researcher Seon-Ho Kim. Together, they will develop the AI models based on language models at The City College. 

Last August, Devineni, Kim, Stony Brook’s Te Pei and Fatemeh Yavari co-authored the paper “Can Large Language Models Quantify Urban Floods from Multi-source Textual Information” in the journal NPJ Natural Hazards.
The paper led to the funding of the CCNY-Stony Brook collaboration. 

An assistant professor in the Department of Civil Engineering at Stony Brook, Te Pei serves as principal investigator, and Paola Cascante-Bonilla, assistant professor in the Department of Computer Science, serves as co-principal investigator. The project began Oct. 1 and will continue through Sept. 30, 2029.

“We have increasingly good information on rainfall, topography and where urban infrastructure is located, but we still do not always know how the system is actually performing during a storm,” Pei said. “Those difficult-to-observe conditions can make a major difference in where flooding occurs and how severe it becomes.”

Understanding those differences also requires observations of what actually happened during a storm. Yet detailed measurements of urban flooding, particularly water depth, are often sparse during fast-moving events.

“One of the main challenges with urban flooding is that detailed measurements are sparse, especially during fast-moving events,” Cascante-Bonilla said. “Our goal is to make better use of information that already exists, such as traffic cameras, street-level images and public reports, and convert it into flood observations that can complement physical models.”