Improved Response to Disasters and Outbreaks by Tracking Population Movements with Mobile Phone Network DataA Post-Earthquake Geospatial Study in Haiti

Imagine trying to steer a response to one of the century's worst urban disasters while flying blind. That was Haiti in January 2010. The ground shook, the capital crumbled, and within weeks, a cholera outbreak loomed. Relief teams needed to know fast where people had gone and in what numbers. The usual tools—eyewitness reports, camp registrations, rough counts from satellite images—were slow, patchy, and often biased. It's like trying to pour water into a moving cup when you can't see the cup. So here's the simple, powerful idea Bengtsson and colleagues put on the table: use the exhaust of the mobile network—anonymized traces of where Subscriber Identity Module, or SIM, cards connect to towers—to estimate how people move. No names, no personal content. Just time-stamped tower connections that, at scale, sketch a population in motion. If this worked, responders could replace guesswork with maps and numbers that update in hours, not weeks. They partnered with Digicel, Haiti's largest carrier, and pulled the kind of dataset you almost never see in disasters: millions of SIMs with hundreds of millions of tower connections across two windows—one centered on the January earthquake, the other on the October cholera outbreak. From that universe, they focused on 1.9 million SIMs that were active both before the quake and in the final month of follow-up. That inclusion rule matters. It filters out SIMs that suddenly went dark, giving a stable panel you can track over time rather than a shifting cast of devices. How do you turn SIM pings into locations? By treating each phone tower as a proxy for where that SIM was when it placed a call. Coverage areas vary—a few square kilometers in dense cities, up to a hundred in rural stretches—so this is a coarse map, not GPS. But it's consistent. And with enough SIMs, you can see the river of people even if you can't see each drop. There's one more piece you need: a way to scale SIMs to people. On the day of the earthquake, the team counted eight hundred nine thousand five hundred seventy-three SIMs in Port-au-Prince. Census data put the city at about two point six million residents, which means roughly thirty-one percent of people had a Digicel SIM active that day. Flip that fraction over, and you get a working conversion: one SIM stands in for about three point two people. It's an assumption—mobile use isn't uniform—but it gives you a denominator. And because the study also reports results in raw SIM counts, you can see the estimate and the data side by side. With that machinery in place, the headline result comes quickly and hits hard. By day nineteen after the quake, about one hundred ninety-seven thousand four hundred eighty-four SIMs that had been in the capital on January twelfth were now outside it. Scale that, and you get roughly six hundred thirty thousand people leaving. At the same time, about thirty-eight thousand seven hundred twenty-nine SIMs moved the other direction into the city, corresponding to around one hundred twenty thousand arrivals. Netting those flows, you land on a striking number: a quarter of a million SIMs worth of movement, or about five hundred ten thousand people gone from Port-au-Prince—roughly twenty percent of the pre-quake population. The population of the capital bottomed out around that nineteenth day, then began inching back as some people returned. Where did everyone go? Almost everywhere, but not evenly. The SIMs painted a map of the exodus fanning out across Haiti with hotspots that jumped off the page. Les Cayes absorbed about twenty-eight thousand people at the peak of outflow. Leogane, closer to the epicenter, took in around twenty-three thousand. Saint-Marc saw about twenty-two thousand. That uneven geography matters. Relief isn't just about how much; it's about where and when. If you can see the bulges as they form, you can move water, food, and shelter into the right valleys. Now, it's fair to ask: is this real movement or an artifact of who has phones? Bengtsson and colleagues didn't just trust the SIMs; they checked them against other sources. The broad magnitude of departures—half a million people leaving the capital—lined up with the Haitian National Civil Protection Agency's total count at the time. But the SIM-based map of where those people went looked very different from the official department-by-department distribution that agencies were using. Instead, the pattern from the SIMs mirrored a retrospective United Nations Population Fund household survey, which canvassed about two thousand five hundred households with an average size of four point nine people. So the totals converged, but the spatial story split: official counts put the emphasis differently across departments, while both the survey and the SIMs agreed on the hotspots. When two independent methods with different biases land on the same geographic pattern, you pay attention. The study didn't stop at disaster displacement. In October, when cholera emerged near Saint-Marc, the team asked a sharper question: as cases appear here, where are potentially exposed people going now? They defined an outbreak area around the city—several communes and communal sections—and identified the active SIMs within it. In that zone, the SIM-to-population ratio was lower than in the capital, about twenty-one percent, which is another way of saying the phone lens was a bit dimmer there. Even so, it was bright enough to see movement begin. Over the first eight days, an average of three thousand six hundred seventy-six SIMs left the outbreak area each day. That's two point seven percent of the local SIMs on the move, day after day. And, again, the flows weren't uniform. Many headed toward the communes that ring Port-au-Prince, others to urban centers north and northeast of Saint-Marc. The southwest, by contrast, received very few. For a waterborne disease, those pathways are not trivia. They're the routes along which exposure and fear can travel. Here's the part that made responders lean forward. The processing wasn't academic-slow; it was operationally fast. Analyses were produced within about twelve hours of receiving the network data and turned into short reports that highlighted where substantial outflows were headed. In the early outbreak days, Digicel pushed hygiene and prevention messages—first by text, later also by voice—into areas of concern. When you're trying to get chlorine tablets and clean water where they'll actually be used, a half-day turnaround on population movement is the difference between waiting for a weekly bulletin and acting this afternoon. Under the hood, the pipeline was standard but disciplined. The data sat in a relational database. Analysts built daily location lists per SIM—today you connected here, yesterday there—and rolled those up to the administrative areas that matter for decision-making: departments, communes, even communal sections. Geographic visualization came from off-the-shelf mapping tools. None of the tooling was exotic. The novelty was the marriage of a huge, passive data stream with a clear operational question and a clock. Still, you should hear the caveats loudly, because the authors did. Phones aren't people. Children, the elderly, the very poor, and many women have lower mobile use in Haiti. If those groups move differently, a SIM-weighted measure will misrepresent the true picture. Some people carry more than one SIM or swap them in and out, which can make a single person look like several or temporarily hide them from view. Tower density shapes precision: in cities, you can localize movement to a few blocks; in rural areas, a tower can cover tens of square kilometers, blurring the map. Power outages and broken infrastructure can keep phones off even if their owners are on the move. And none of this counts phones that were lost in the rubble. These are not small footnotes; they're the boundaries of what this method can claim. Privacy and ethics weren't afterthoughts. The team received only anonymized records. No names, no numbers tied to individuals. The analysis aggregated flows up to areas that match how aid is organized, and the work went through regional ethical review. That matters not only for the people behind the data but also for trust. In crises, the legitimacy of a tool can be as important as its accuracy. So what's the value proposition, net of all the complexity? It's speed plus structure. When time is elastic, surveys and site visits are fine. When time collapses, a living baseline of how many people are likely in a place today and where they came from last week can reset the whole response. Bengtsson and colleagues showed that with a preexisting relationship to a carrier, analysts can stand up a movement map across an entire country in days and refresh it in hours. They also showed that you can stitch those relative movements to census data to estimate area-specific population sizes—the denominators that make needs assessments real instead of rhetorical. The comparison exercise—SIM data versus official counts versus a household survey—also hints at a deeper lesson. Different methods carry different biases, but when two independent views converge on the same geographic pattern, your confidence grows. In Haiti, the SIM-derived redistribution matched the United Nations Population Fund survey more closely than the official department breakdown. That's not an indictment of any one source. It's a nudge toward blending them. Let the fast, rough signal from the network guide where to send enumerators, and let careful surveys ground truth and calibrate the model as you go. It's healthy to ask what would break this approach. Low mobile penetration will. Sparse towers will. Prolonged power loss will. And in places where phone ownership is itself highly patterned by gender or wealth, you risk building an exquisitely precise map of the wrong population. The authors are explicit about those limits and about the practical steps to improve things in advance: establish data-sharing protocols with operators before disaster strikes, bake privacy in from the start, and pair the network lens with whatever local data—census, clinic registers, camp lists—you can trust. There's a temptation, hearing all this, to leap to a grand generalization: that mobile data can solve situational awareness in any crisis. It can't. What it can do, when the conditions line up, is compress the time between question and answer. In the Haiti earthquake, that meant quantifying a net outflow of about five hundred ten thousand people—one in five residents of the capital—within weeks, and seeing where they landed. In the cholera outbreak, it meant spotting a daily two point seven percent outflow from the outbreak area and mapping its preferred routes within hours. Those are not nice-to-haves. Those are the bones of an agile response. If you're a planner listening to this on your commute, the quiet homework is obvious. Ask today, not during the next emergency, what legal and technical rails would let your team receive anonymized, aggregate movement data from carriers in a crisis. Decide who needs to see what, and how quickly. And decide what other data you'll marry it to, so that when a map tells you Saint-Marc is sending people north and into Port-au-Prince's orbit, you know which water points to refill tomorrow morning. That's the progression here. From a crisis of visibility to a method that makes people visible in the aggregate. From slow, static counts to dynamic flows. From arguing about how many to asking where and when. Bengtsson and colleagues didn't promise perfection. They delivered something humbler and, in practice, more useful: a fast, privacy-preserving way to see the shape of displacement and to move help along the same paths people take.