On Friday, September 14, Hurricane Florence made landfall along the North Carolina coast, prompting high winds and massive flooding throughout the region. All of that was in the forecast. But at the same time, disaster response agencies faced some vexing questions: What were the people warned to evacuate actually doing? And where exactly had they fled?
In what has become an increasingly common tactic, many organizations including Direct Relief, and the Red Cross, turned to Facebook data to provide some answers. Last year, Facebook launched Disaster Maps, an initiative through its Data for Good division that tracks the geographic coordinates of user signals to show trends in people’s movements during a crisis.
The result is a series of time-stamped images that can help emergency responders get a rough idea of how a situation is really unfolding, and where their services might be needed most. All of the data is aggregated in a way that makes it anonymous. It’s also collected automatically, as long as the social network users have their location services enabled.
At least a dozen nonprofit groups began using the service in 2017, including Direct Relief, which provides medical supplies to overtaxed hospitals and health centers in times of critical need. In 2017, for instance, the group used Disaster Maps data to help inform where they staged respiration mask depots for residents dealing with Southern California’s massive wildfires.
In the case of Hurricane Florence, Facebook began offering imagery several days ahead of landfall. To create it, the company first developed a baseline of what user activity historically looked like in the affected areas based on signal positions from three months prior. Then it began to color-code areas that were transmitting abnormal population density spikes (in blue) or lulls (in red) to highlight areas where folks seemed to be flocking or fleeing.
Andrew Schroeder, the director of research and analysis at Direct Relief, says Facebook shared the first maps with him on Sunday, September 9. The nonprofit has since received new snapshots every eight hours, which is especially important as relief efforts continue. “It lets us actually compare what Facebook is seeing with media reports of where people are evacuating,” Schroeder says. That helps because initial media reports can be vague. And so far, at least one trend has been surprising: Initially, Facebook data showed only the pace of the exodus where people were specifically taking shelter in larger cities like Raleigh and Charlotte.
In many big cities, Schroeder says there has been a flip in their usual activity. Whereas a blue cluster traditionally marks downtowns, more people seem to be congregating around the outer edges of these places versus in the center. “That’s probably because people are actually just staying home and they don’t go into the downtown area to seek services,” he says, noting that transportation routes may also have been disrupted.
For Direct Relief, the information could inform how aid is rendered in the days ahead. Many of the nonprofit group’s targeted health centers are located in downtown areas for obvious business reasons. Many actually re-supplied just before the storm because, as happens every storm season, Direct Relief pre-positioned “hurricane preparedness modules” with things like bandages, water purification supplies, antibiotics, and asthma medication at six locations throughout South Carolina, North Carolina, and Virginia.
Earlier this week, the group told Fast Company that it had since sent 11 shipments to five different health centers, and had five more shipments prepped to move. But some of the people clustered on the outer edge of these cities may not be able to get back downtown, or, if they arrived from other places, might be unsure where the nearest reliable medical care would be. Others may also be stranded there because they can’t yet return home. “Over the long term, if that persists, you do need to actually shift where you’re focusing the point of care,” Schroeder says. “It’s a real argument for mobile [units] going out to where people are, [and] for supporting health centers that are actually farther out along the periphery and making sure at least that we’re aware of the conditions that they’re facing.”
Direct Relief doesn’t just use Facebook data to inform such decisions. (It might, for instance, be especially misleading if, after a day or two, signals in places with no power blinked out.) The group also uses government data to pinpoint spots with large population densities and social vulnerabilities, and reports from another nonprofit called Humanity Road, which compiles all sorts of data about how crucial infrastructure is holding up, including power outages, 911 call routing, and cell sites.
“I don’t think that Facebook data is a slam-dunk on anything, but I think it’s good hypothesis formation stuff,” says Schroeder, who expects that the company will allow many responders continued access for at least the next couple of weeks. That’s important because as separate reports by Harvard and George Washington University on the tragically high death toll from Hurricane Maria in Puerto Rico have shown, much of the mortality that happens in events like this comes from the slow toll of being isolated or uprooted. People may not have access to important health services, and at the same time they can lack income because their workplace was wiped out or jobs aren’t transportable. In the meantime, serious medical conditions like diabetes or hypertension can remain untreated.
As Schroeder puts it, most mortality in Hurricane Maria was the result of “the impact on the system” of how traditional society works “and you’ll see the same thing with Hurricane Florence.” Facebook may be providing some early clues about how to render aid, but there’s plenty more work to be done. “It’s good to see that people left their homes,” he says. “We need to know how fast they come back. We need to know if there are neighborhoods where you see long term low population evidence that [means] whole areas have not necessarily seen recovery. And then that needs to be combined with other information to really figure out what the heck it means.”