UNC study sees flooding outside NC high-risk areas. Can AI reveal blind spots?

Amber Hazzard, The News & Observer (Raleigh)News & Observer

Using flood insurance claims isn’t enough to capture the true scale of building flooding and can mislead about future risks, UNC-Chapel Hill researchers say.

But adding many other factors, with local topography and building density included, produces a more accurate picture, they have concluded.

Doing that to estimate how many Eastern North Carolina buildings flooded during 78 events between 1996 and 2020, they counted twice the number of National Flood Insurance Program claims filed after the events.

They also estimated that 43% of flooded buildings were located beyond FEMA-designated high-risk zones, where flood insurance is mandatory.

During a time when flooding is regularly battering North Carolina, the scientists say they hope their research will be useful to help people better understand risks and protect their property.

“Not all places that flood are mapped or regulated from a regulatory standpoint,” Antonia Sebastian, an assistant professor in the Department of Earth, Marine and Environmental sciences at UNC-Chapel Hill, said in an interview.

Sebastian was among the authors of an article about the research published this week in the scientific journal Earth’s Future.

Using AI to better estimate past flooding

The research group used machine learning, an artificial intelligence process that finds and applies patterns in data, to model recreations of 78 previously unmapped flood events in Eastern North Carolina — including in Raleigh, Fayetteville, New Bern and Wilmington.

The model concluded that 90,000 buildings flooded at least once, with 35% of them being residential buildings that were outside FEMA’s designated high-risk areas. They estimate 9% of residential buildings that flooded more than once were outside of high-risk areas.

“It allows us to model a lot more floods a lot more quickly, and that gives us a better picture of previous flooding in the state and really how much repetitive flooding has happened,” said Helena Garcia, lead author and PhD candidate in the Environment, Ecology and Energy Program.

They used the information generated by their model to create a publicly available database called the Flood Extent Archive.

“The piece of this that’s really novel is that this is a very unique data set, that is this high spatial and temporal resolution, that’s going to enable us to answer a lot of questions we’ve had and haven’t been able to get answers to,” Sebastian said.

A better way to see flood risks?

Floodplains or areas next to rivers and streams are most at-risk for repeated flooding because they are flat and close to water. However, these characteristics also make floodplains an ideal place to put buildings, roads and airports.

In North Carolina, local governments are responsible for permitting and enforcing floodplain regulations to ensure safety, according to the North Carolina Division of Emergency Management.

Homeowners can further protect themselves by purchasing flood insurance. This is required for homeowners in floodplains with federally backed mortgages, Sebastian said.

FEMA offers a National Flood Insurance Program (NFIP) for these high-risk areas, which they define as having a 1% likelihood of flooding in any given year. This is also known as the 100-year flood estimate. For example, coastal areas like the outer banks are considered high-risk.

These estimates are partially based on historical flooding data, which is often insufficient when trying to characterize the area exposed and changes in exposure and vulnerability over time, according to the study.

Scientists commonly use river gauges that measure water levels and satellites to determine flooding. However, river gauges are few and far between and clouds may cover satellites, causing researchers to miss information about where it floods, Sebastian said.

”There is just very little information out there on how frequently it floods and what all floods when it floods,” Sebastian said. “There’s a lot of attention given to events like [Hurricanes] Florence and Matthew and now Helene… But there are a lot of events that occur sort of in an in-between space that matter for people.”

The new model helped capture these in-between events by using insurance data from the NFIP, showing individual buildings that filed flood claims, then using additional data to predict flooding in neighboring properties without NFIP claims, Sebastian said.

More Chantals in our future: Climate change will likely increase storms, flooding