Researchers Develop AI-Powered Technology To Address Flooding Impacts

Funded by Ohio Sea Grant and the National Science Foundation, the team built a tool to quickly and accurately forecast flooding across the U.S. in real time and answer critical questions about water and energy

How can we predict the big floods of the future? What will be the full extent of their impacts?

When extreme storms hit cities or regions, they impact people and local economies through a complex series of infrastructure failures – events like water and power outages, transportation issues, and water contamination — that are often hard to anticipate.

To solve this problem, researchers funded by Ohio Sea Grant and the National Science Foundation harnessed AI technology to help address flooding and water-energy challenges in Northeast Ohio and across the country. Led by Dr. Yeghiazarian of the University of Cincinnati (UC), the team created new information networks to connect “siloed” infrastructure data and predict impacts.

“If the data about a very complex system like a city isn’t connected, then there’s no way to predict what happens when there’s a stress on the system,” said Yeghiazarian, a professor of chemical and environmental engineering at UC. “So, in order to predict the total impact of a flood, the first thing we need to do is connect the data that describes our very interconnected world – and that’s precisely what this technology does.”

The team addressed the devastating impacts of flooding in the U.S., Yeghiazarian said, a phenomenon that is only expected to worsen in the future as extreme precipitation events become increasingly frequent. To do that, the team studied how cities and regions function as complex networks made up of many smaller, interconnected networks.

“I wanted to reduce damage from flooding to people and to the environment by increasing the resilience of our infrastructure,” Yeghiazarian explained. “At the same time, this is a fascinating question of science and technology: how do you actually do it?”

Researchers began collecting data with a focus on flooding in Northeast Ohio, working alongside the Northeast Ohio Regional Sewer District. Over time, the project kept expanding, from Ohio, to the Ohio River Basin, to the continental U.S., relying on federal agencies for complex infrastructure data on buildings, roads, dams, energy sources, power grids, contamination sites, pollution regulations, and more.

This expansion was made possible thanks to the team capitalizing on powerful technology known as knowledge networks or knowledge graphs, a subcategory of artificial intelligence, Yeghiazarian said. Whereas large language models – the most popular form of AI – generate human-like text, knowledge networks connect and structure factual information. They consist of “nodes” representing entities and links representing the relationships among them.

“Knowledge networks are one tool in the spectrum of AI,” Yeghiazarian said. “The power of the network is that you can represent complex data from separate sources in a connected way.”

“You have to have a human in the loop here, because they need to tell the system what data to use – it’s not allowed to speculate. That’s why it’s so important to have uninterrupted access to large, high-quality data produced by state and federal governments. This data is critical for meaningful decisions and actions on the ground.” - Dr. Lilit Yeghiazarian

In total, the knowledge network Yeghiazarian’s team built incorporates more than 150 million critical assets, such as buildings, underground fuel storage tanks, and power infrastructure, across the country.

By using the network in conjunction with a large language model, researchers created a highly accurate, efficient tool to answer complex questions around the water and energy nexus.

“You have to have a human in the loop here, because they need to tell the system what data to use – it’s not allowed to speculate,” she continued. “That’s why it’s so important to have uninterrupted access to large, high-quality data produced by state and federal governments. This data is critical for meaningful decisions and actions on the ground.”

A startup, called Data2Action, was spun off at the University of Cincinnati to maintain this technology and to continue providing critical flood information to the American public. The tools, publicly available to demo at data2action.tech, offer real-time flood forecasting and analyses related to water and energy. The forecast shows the severity of flooding and how many structures are at risk in a given area over the next 16 hours. Users can view a map with live alerts and zoom in to see individual buildings impacted.

People can also use the technology to ask wide-ranging and complex infrastructure questions, including sample queries researchers developed through communications with stakeholders, federal agencies, utilities, and cities. For example, users could ask the tool to find all chemical contamination observations within 100 meters of a river and then pull census data from tracts downstream.

“Behind all of this is lots and lots of data that were primarily collected by the federal government, quality controlled and quality assured,” Yeghiazarian said. “These data are validated on multiple levels.”

The technology can run simulations of floods in a city, identifying how infrastructure failures can spread or “cascade,” leading to more failures and disruptions in areas that were not flooded. For example, a power failure that originated from a flooded electric facility can take out a water system that serves a certain number of hospitals, leaving them without drinking water.

“Predicting such cascading failures is critical because they disrupt essential services such as access to medical help, potable water, and transportation, and increase human exposure to heat or cold during prolonged power outages,” said Yeghiazarian.

“You can see that there are multiple cascading infrastructure failures that took place because of this one event,” she said of a flood simulation her team conducted focusing on the Cincinnati area.
From there, the team can even calculate how much money the regional economy would lose when infrastructure damage halts various activities, such as when people cannot take their kids to school and have to stay home. These estimates include disruptions in many different sectors, including agriculture, mining, utilities, construction, manufacturing, trade, transportation, information services, finance, insurance, real estate, education, healthcare and social services, arts and recreation, hospitality, and government.

“This is the level of detail that helps us understand the true impact of a flood on the region or the city,” Yeghiazarian said.

Various stakeholders – municipalities, state governments, nonprofits, health organizations, and industry – can use the tool to better understand the balance of water, energy, and potential contamination of the environment, she said. The real-time forecast could be used for “on the fly” emergency response and logistics in response to flood crises, and analyses could show who will be impacted. Used by decision makers, the technology could aid planning, reducing vulnerability and increasing resilience.

“This is the level of detail that helps us understand the true impact of a flood on the region or the city. It’s about making data-driven, evidence-based decisions to reduce damage and save lives.” - Dr. Lilit Yeghiazarian

“It’s about making data-driven, evidence-based decisions to reduce damage and save lives,” Yeghiazarian added.

The project, funded by Ohio Sea Grant and the National Science Foundation, relied on key partnerships. Data was sourced from the U.S. Geological Survey, the National Oceanic and Atmospheric Administration (NOAA), the Department of Energy, the U.S. Environmental Protection Agency, and the U.S. Army Corps of Engineers.

“I think this is a fantastic example of the huge impact that Ohio Sea Grant makes in giving awards to teams of researchers who partner with people in industry and then expand the work to larger scales,” she said.

For more information about this project, view her recent Freshwater Science webinar (https://www.youtube.com/watch?v=SDxZg7O8MUw). In addition, her team at Data2Action is looking for new partnerships, clients, and collaborators to adopt this technology, participate in continuous development, and take it to new levels.

Ohio Sea Grant is supported by The Ohio State University College of Food, Agricultural, and Environmental Sciences (CFAES) School of Environment and Natural Resources, Ohio State University Extension, and NOAA Sea Grant, a network of 34 Sea Grant programs nation-wide dedicated to the protection and sustainable use of marine and Great Lakes resources. Stone Laboratory is Ohio State’s island campus on Lake Erie and is the research, education, and outreach facility of Ohio Sea Grant and part of CFAES School of Environment and Natural Resources.