North Carolina innovators aim to revolutionize flood prediction
George Cheng explains aspects of the innovative, award-winning FloodGate tool. George Cheng photo
A team of students from the North Carolina School of Science and Mathematics has developed an artificial intelligence-based flood modeling system that could transform the way communities predict, and therefore respond, to extreme weather events.
The cutting-edge tool, called FloodGate, addresses limitations of traditional flood forecasting systems, offering a faster, more reliable and globally scalable approach to predicting floods.
Born in Georgia, George Cheng moved to North Carolina around the age of eight and was raised in Cary. While a student at the prestigious residential STEM high school, Cheng got the idea for a flood modeling tool after witnessing firsthand the devastating impact of flooding.
“Every single time heavy rainfall would happen, our dorms would get flooded,” he said. “A lot of students had to board up their rooms.”
Cheng also reflected on the experiences of both his family in China during the Yangtze River’s annual flooding as well as acquaintances in Delhi, India, where flooding killed more than a hundred people in July 2023.
“One of the projects that I was working on was trying to see if I could model flooding in the Delhi area to help out my friends and that eventually turned into FloodGate, which is basically using artificial intelligence machine learning models to try to predict flooding forecasts,” he said.
Related Items
HCA Middle School Shines at Regional Science Fair
Grab the lifejackets and the suntan oil
Stormwater overhaul plan eyes future flood fixes for Waynesville
More river gauges coming to Haywood
Traditional flood prediction models rely heavily on historical hydrological data and physics-based models, which, according to Cheng, have three major flaws.
Data scarcity remains a constraint. While places like New York City may have plentiful datasets — parameters like soil moisture, riverbed roughness and rainfall distribution — rural and less affluent areas usually don’t.
Static model architecture, as the name suggests, does not incorporate real-time data into flood prediction.
Computational latency affects both speed and precision.
“A lot of these current models are super, super slow and really, really inaccurate,” said Cheng. “So these are super impractical for emergency responses, like with flash floods.”
North Carolina currently utilizes a robust real-time tool called FIMAN, the Flood Inundation Mapping and Alert Network, which is generally well-regarded but has some drawbacks of its own. A network of more than 600 flood gauges provides flood stage data from across the state to a central processor that compiles and displays the information, as well as forecasts, on a mapping website.
The problem is, if there are no gauges, there’s no data. Currently there are 18 gauges in Haywood County, from Big East Fork in the south to Cataloochee in the north and from Newfound Road in the east to Dellwood in the west. During Helene, one was washed away. Miraculously, more weren’t.
The other problem is, they’re expensive.
Recently, the county purchased seven more gauges, two of which will be placed at new locations. The patent holder is the sole supplier of the equipment so the total cost, fully funded by a grant received from the Office of State Budget Management, was nearly $109,000 or more than $15,000 each.
While Cheng acknowledges the value of local flood monitoring systems like FIMAN, he believes FloodGate’s global reach sets it apart.
Unlike conventional models that can struggle with real-time adjustments, FloodGate adapts dynamically to changing weather conditions by incorporating live satellite data to provide hyper-local risk assessments tailored to specific regions.
“I think what really differentiates our tool and current tools is that not only does it provide relative accuracy, it takes out the need for gauges using satellite mapping and also AI modeling, basically trying to extrapolate the current data to what would happen if another flooding event would [happen] like this,” said Cheng. “For example, flooding in Anchorage, Alaska, is probably not going to be the same as somewhere where it’s below the river basins, like Louisiana.”
FloodGate also has the potential to aid emergency response teams by offering longer lead times and more precise risk assessments, including for Hurricane Helene.
“Machine learning models could have detected rainfall anomalies in the Smoky Mountain foothills 12 hours earlier than official forecasts [for Helene],” Cheng explained. “Everyone’s relying on these forecasts. Everyone’s trying to see whether they should evacuate, or they should try to stay in their home and not make a big fuss about it.”
Integrated inundation mapping could have also been used by response agencies, including FEMA, to locate survivors in areas that were cut off from the world when cellular and internet infrastructure failed.
The FloodGate tool is thought to hold so much promise that in 2024 Cheng and his teammates, Reichen Schaller, Shubhan Bhattacharya and Sumedh Kotrannavar, were awarded the $100,000 Earth Prize after winning an environmental sustainability competition for teenage students held by The Earth Foundation, a Swiss-based advocacy group that promotes young entrepreneurs with innovative ideas that focus on environmental challenges.
“FloodGate embodies everything The Earth Prize stands for—turning passion into action and real-world impact,” said Peter McGarry, founder of the Earth Foundation. “Their solution is innovative, practical and, most importantly, driven by firsthand experience of the very challenge they are solving. As flooding becomes an ever-greater threat, their technology has the potential to protect countless communities worldwide. It’s exactly this kind of bold, thoughtful innovation that makes me so excited about what young people can achieve.”
Currently, Cheng and his team are refining their model with mentorship from experts at Nanjing Technological University. Though still in development, they aim to launch a web-based application by late May, even though the team is still occupied with other pursuits.
“I know that we’ve done a lot of steady progress on it, but a lot of our members are just like super busy to trying to get into college right now. I think with Earth Prize’s mentorship and also the monetary support to train these models, I think we’re able to develop a really great model soon,” said Cheng, who is currently a student at the Massachusetts Institute of Technology.
Originally intending to major in bioengineering, Cheng shifted his focus to artificial intelligence and cognitive science at MIT, drawn by the potential of AI to help solve pressing global issues.
“If I were to, like, name myself, I’ll probably be like a jack of all trades,” he said. “I focus on a lot of different topic areas, from abortion care access to environmental flooding to diabetes to coronary heart disease detection. I think I just see a lot of different problems that my family, my friends, people close around me face and I really want to tackle them.”
