A population density map of France derived from more than 1 billion cellphone call records shows that people congregate in urban areas during working periods (indicated by orange spikes), and head for coastlines during holidays (indicated by the blue spik

A population density map of France derived from more than 1 billion cellphone call records shows that people congregate in urban areas during working periods (indicated by orange spikes), and head for coastlines during holidays (indicated by the blue spik

Catherine Linard

Taking the census, with cellphones

If you want to figure out how many people live in a particular part of your country, you could spend years conducting home visits and mailing out questionnaires. But a new study describes a quicker way. Scientists have figured out how to map populations using cellphone records—an approach that doesn’t just reveal who lives where, but also where they go every day.

“This is the first time people have provided statistical evidence that population data produced from cellphone records are of really good quality,” says applied mathematician Renaud Lambiotte of the University of Namur in Belgium, who was not involved in the study.

Ninety-six percent of the world’s people have active cellphone subscriptions. In developed countries, the number of mobile phone subscribers has surpassed the total population as some individuals own more than one phone, and subscription rates continue to rise in developing countries, reaching as high as 90%. That’s great news for census scientists, because they can locate the calls by identifying the cellphone towers that send and receive them and use call density around the phone towers to estimate the local population density.

As part of WorldPop, an open-source project mapping detailed population information from countries around the world, a team of researchers led by geographer Catherine Linard of the Université Libre de Bruxelles and data scientist Pierre Deville of the Université Catholique de Louvain in Belgium used mobile phone data to estimate population density in France and Portugal. For each country, they obtained aggregate, anonymized call records from major carriers containing more than a billion calls. The call records in Portugal came from 2 million users, covering about 20% of the population. Every call record indicated the originating and receiving phone towers, the timing of the call, and a user identifier—information collected and stored by network providers for billing purposes. Records in France came from 17 million users, about 30% of the population, and contained only the day of the call and the phone tower locations, due to differences in the carrier’s policy.

Using the call records, the researchers developed a model to estimate population density around every phone tower from call density, taking into account variations in phone usage between high-coverage and low-coverage areas. The results showed clear trends in population dynamics across weeks and seasons that traditional survey-based censuses can’t reflect, the team reports online today in Proceedings of the National Academy of Sciences. During the holiday season, populations in cities dropped sharply, while tourist sites such as coastlines and Disneyland Paris boomed. During the week, people traveled to cities for work on weekdays and back to rural areas on weekends.

The researchers also compared their results to population density data gathered through remote sensing technologies, a widely used method that relies on satellite imaging to gather detailed information on population settlement patterns and estimate population counts. They found that the two methods are comparable in accuracy when checked against actual survey-based census data, but estimates from mobile phone data can provide more timely information, down to the hours.

Still, the method isn’t perfect, Linard says. To apply it in other countries, the team would need to adjust the model to account for diverse mobile phone usage patterns. Some countries may prefer texting to calls, for example, while others may have many residents who are too poor to own a cellphone. Rather than replacing census surveys, the method would be most effective when combined with technologies like remote sensing, she says.

The study shows the merit as well as limitation of big data, says statistician Tom Louis, chief scientist at the U.S. Census Bureau at Suitland, Maryland, who was not involved with the work. Though the information is timely, it is not yet accurate enough for official use, he says. “Big data can be very valuable, but at least at this point in our history, it needs the validation of traditional surveys to show that it works.”

But for low-income countries, where census data are likely outdated and unreliable, mobile phone records present an easy and efficient alternative, Linard says. In the Democratic Republic of the Congo, for example, the most recent census took place in 1984. In contrast, about 70% of the people subscribe to mobile phones.

With the ongoing Ebola outbreak, cellphone records could provide a valuable tool for tracking population movements, says co-author Andrew Tatem, a geographer at the University of Southampton in United Kingdom who leads the WorldPop project. His organization has used a different model to estimate population flow across affected West African countries, based on cellphone data from Senegal and Ivory Coast. Better access to up-to-date data from the affected countries could lead to more accurate information on population movement that would help governments coordinate responses to the outbreak, he says.