The ability to forecast a major flooding event like Hurricane Florence has improved significantly. But understanding how such a storm will interact with the built environment and affect people living in a specific area is still quite limited.
The factors involved in predicting flood scenarios are changing faster than tools that help people prepare and adapt. For example, we know baseline sea levels made higher by climate change will mean bigger storm surges from hurricanes.
But human activity has intensified disasters in other ways, too. Population has exploded along U.S. coastlines, exposing far more people and infrastructure to related threats. Stephen Strader, an assistant professor of geography and the environment at Villanova University, calls this the “expanding bull’s-eye effect.”
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
“Societal growth is the biggest influence on disasters because it increases the potential for loss,” Strader says. It also means more of the environment has been paved over by human habitation, changing the hydrology of major floods in ways that are not well measured or communicated.
Craig Fugate thinks a lot about the interplay of these issues. He led the Federal Emergency Management Agency (FEMA) from 2009 to 2017. And as a Floridian he sees the lingering effects of previous hurricanes, as well as how rising seas are making flooding more routine—and displacing certain populations in the process.
In his post-FEMA career, Fugate is now Chief Emergency Manager of a start-up called One Concern, which he says is using big data and machine learning to help communities and businesses better prepare for threats such as flooding—and not just single, extreme events like Florence, but for the more routine problems climate change will bring.
One Concern says its flooding platform, which the company plans to release by the end of this year, will help predict inundation levels up to five days in advance of a storm—at a block-by-block level. This kind of resolution could make it easier to prepare for storms and adapt to future threats with much greater specificity. Fugate told Scientific American about the project this summer.
[An edited transcript of the interview follows.]
After your time at FEMA and seeing the 30,000-foot view of disaster response in America, what drew you to an AI start-up?
I want to change outcomes. How do we prepare for future risk when all of our data for FEMA maps is looking backwards? What I was fascinated by with the [One Concern] flood model was that it’s being designed to be a response tool. It takes the rainfall estimates and instead of just saying, “You’re going to have flash flooding,” it predicts that this neighborhood is going to see three to five feet of flooding and that neighborhood might see up to 20 feet. We can also know which areas have had heavy impacts right away—decision makers don’t have to wait for information to come in before activating response.
What really sold it to me is how much data is involved, and how we can see things at high-resolution, and quickly. We can run various scenarios in the days before a storm arrives and understand when and how systems would fail. Using AI lowers the threshold to do the “what-ifs.”
What also got me interested is we did a lot of, quote–unquote, “mitigation projects” in the New Jersey area after [Superstorm] Sandy. And a lot of those projects [such as elevating structures] were based upon looking at a 100-year cost-benefit analysis; our assumption is that they will perform better in the next storm. But we could never really ask the question, “How much better will it do?”
The term “resilience” gets tossed around a lot when we talk about preparing our coastal communities for sea level rise—in some cases by moving people permanently out of harm’s way with buyouts. How can these models make resilience less abstract?
If we’re always waiting until after disasters to move people, then that’s going to be the most painful and disruptive way to do it. How do we get these tools in front of officials who are responsible for changing building codes and land use in places vulnerable to sea level rise?
A town could use this tool to identify a no-growth zone—an area where they won’t issue any new building permits. Then they say, “If a home gets destroyed in a storm, you can’t rebuild.” And then you can redirect those resources to a new area, to make it an attractive place to live. We have carrots and sticks to reduce the amount of people living in the no-growth zone by giving opportunities to relocate to safer locations. We build time into this process.
If we can make the risks more definitive, we can at least start to get more control over our destiny—versus it being inflicted upon us each time a storm hits, and we lose more ground we never get back. It won’t be easy and it won’t be cheap, but it’s a strategy.
If the value of many of these coastal places is the beach itself, what happens when preserving the sand becomes untenable?
Beaches come and go as a natural process. Re-nourishment (dredging offshore sand and pumping it onto eroded beaches) is an artificial process. I was once down in Virginia Beach where they were doing a huge re-nourishment project that was just about finished. Then a big nor’easter sat off the coast for about three days. Afterwards I went outside, and what was basically 200 yards of beach was now, you know—the ocean was butting up against the retaining wall leading to the hotel pool.
Imagine Congress saying that the Army Corps of Engineers budget will not go toward re-nourishment anymore as a maintenance issue to preserve beaches that we’re losing to sea level rise. Are local governments ready to take on that role? Generally the answer is no.
How do the tools of artificial intelligence and big data fundamentally change how we prepare for flooding in a climate-changed future?
Right nowit’s hard to give local officials any visualization tools to see the what-ifs at high enough resolution. Scientists say, “We are going to see these types of climate impacts within these ranges.” But a local planner wants to know, “What does that mean for building a new road?” And when officials look at the high end of the range for sea level rise, they say, “This is absurd. Why am I planning if all of this will be gone?”
If we depopulate a residential area over 10 years and turn that block into a dune line, what does that look like under all these different sea level and storm scenarios, so we can know when that dune system would no longer offer protection? You have to run scenarios to see whether investing in that really changes the flooding and damage outcome in the future. It’s a process that lends itself to using AI.
Using artificial intelligence to create adaptation scenarios can be fraught. There’s danger in relying too heavily on data, and in the biases that can be programmed given what data is included in—and excluded from—the models. How do you avoid creating mitigation strategies that make vulnerable populations even more so?
This model is not a tool for affluent communities. It’s a tool for vulnerable communities, and the original idea was in saving lives during the response phase of a disaster.
Just knowing the built environment does not tell us how a population is going to fare in a flood. A lot of the socioeconomics—from social networks to coping mechanisms to financial resilience—are a better predictor.
We are doing a lot of work so that we are not creating situations where the unintended consequences of AI could be to essentially force people out of homes; or build a model that’s so biased toward affluence that we eliminate affordability; or that we don’t look at preexisting social structures. It’s not just the geoscience—it’s people.
As we start migrating away from sea level rise, one of the things we have to be cautious about is that we don’t do it on the backs of those that don’t have a voice, that we don’t disproportionally put the burden to move on low-income areas—especially if it means that that space is being redeveloped for more affluent communities. We don’t want to be picking the winners and losers in climate change based upon economic factors.
It’s no secret that our National Flood Insurance Program is ill-equipped to deal with the scope of flooding we’re experiencing today and will be in the future. Reforming it is necessary, but it’s not the whole answer, right?
I do think we need to stop providing [federal] flood insurance to new construction in flood zones. Go to the private market. If they will write it—have a nice day. If not, don’t rebuild.
But the double-edged sword of letting market forces drive mitigation is that it will displace low-income families in low-lying communities. In some coastal areas market forces are already pricing risk out of the range of people who are least able to afford it. Which means they are getting priced out of housing markets within reasonable distances to their work and schools and community.
These populations are also the least able to manage and survive the financial impacts of recurring floods that sea level rise will bring. So if the government offers buyouts, it gives them the opportunity to move. But you can’t be callous about it; you first have to target affordable housing with good schools and good infrastructure, so people have somewhere to go in their community.
There’s very few places you can go that don’t have any kind of disasters. But the question is: How fast can you bounce back? This is about taking a step back and not just looking at disaster programs but looking at how communities are planning their futures. If they’re going to start displacing their populations due to climate change, how are they going to build affordable communities for the workforce necessary for their economy? Or are they increasingly creating the climate haves and the climate have-nots?