As Superstorm Sandy threatened the East Coast of the United States in fall of 2012, people stocked up on bread and batteries. They boarded their windows. And they tweeted. During the height of the storm and its impact, between October 26 and November 10, people sent more than 20 million hurricane-related tweets.
Now, scientists have taken almost 10 million of those tweets and used them to show where Sandy hit. They also found that tweets about the storm’s damage were associated with how much financial support ended up getting handed around. The results show that social media can be used to track the damage from natural disasters. And they suggest that eventually, these social networks might be able to contribute to disaster response.
These days, if an earthquake happens miles away, you may read about it on Twitter before you even feel the tremor. In some cases, earthquake information actually travels faster via tweet than from scientific instruments. In 2009, the Defense Advanced Research Project Agency held the DARPA Network Challenge — a contest designed to see how social media and the Internet could be used to solve time-sensitive problems. Manuel Cebrian, a computational social scientist now based at the Commonwealth Scientific and Industrial Research Organization in Clayton, Australia, won that challenge with his colleagues by using social media to hunt down 10 red weather balloons released across the United States.
Since then, Cebrian and his colleagues have been applying social networks to other, more disastrous events.
First, they tried to see if they could predict outbreaks of contagious disease. But soon, they turned their minds to large-scale disasters. Hurricane Sandy was the perfect target. It was a disaster big enough that everyone cared about it. The storm, which made landfall in New Jersey on October 29, 2012, killed at least 147 people. It also caused $50 billion in damages and left 8.5 million people without power.
The disaster was large — and came at a time when Twitter had become “a pervasive technology that everyone was using,” Cebrian explains. This made Sandy a good storm to test the powers of Twitter.
A storm of social media
The scientists analyzed 9.7 million tweets from 2.2 million users posted between October 15 and November 12, 2012. They restricted their search to tweets where they could determine the location of the user. Cebrian and his group compared “hurricane-related” messages using keywords such as “Frankenstorm,” “Sandy,” and “hurricane” to tweets that contained vague, non-Sandy-related terms such as “weather.”
By comparing Twitter use in 50 cities in the United States, the researchers found that the closer the city was to the hurricane’s path, the more storm-related tweets were sent in the period immediately before, during and after Sandy’s big day, the researchers report March 11 in Science Advances.
Showing that tweets could track a storm’s path was “the main goal,” Cebrian explains, but then the team moved on to a more difficult question: Whether tweets were also associated with material storm damage. “It was kind of a long shot for us,” Cebrian admits. But Superstorm Sandy again had good data to offer. Using the publicly available data from the Federal Emergency Management Agency and private insurance companies on how much money was spent on storm recovery, the scientists were able to show the volume of tweets about Sandy in New Jersey and New York was associated with how much money people in the area received for damages. Cebrian and his group were even able to apply their methods to 12 other natural disasters, including the South Napa earthquake and the Oso mudslide, and see similar results.
“We were surprised there was any correlation at all” between tweet volume and damage, Cebrian notes. “These are two very different processes.” On one hand, you have Twitter, something many people do when distracted, “you are looking at a million things and talking to your mom and then you tweet,” Cebrian says. “On the other hand, to know how much money goes into your house…they have to take pictures and assessment and get the numbers approved. You are looking at a well calibrated process.”
The study is much larger than many previous studies of social media and disaster, says Julia Skinner, who studies social media, political unrest and natural disasters at Kennesaw State University in Georgia. “It combines a lot of different ways that researchers have considered social media and disasters into one approach,” she explains.
The study also showed that people retweet less in the path of the hurricane, which suggests there is “more new information coming from areas close to the hurricane,” says James Bagrow, who studies social media and disaster response at the University of Vermont in Burlington. “This isn’t ‘cowboy science,’” he says. “They did a good job of supporting the decisions they made along the way.”
To tweet or not to tweet
Though Cebrian has done several analyses of Twitter, he himself doesn’t tweet. “I’ve never used Twitter,” he admits. “I’m not very good with this technology.”
But it has still proven to be a rich resource to mine. “Twitter is this weird mix of a social network where people check on their friends, but also an information network where agencies try to reach out to people,” Cebrian says.
Twitter is also useful because it’s public, Skinner explains. “Even if you don’t have an account, you can see what people are saying about a disasters, while on a platform like Facebook or Snapchat, people’s privacy settings might prevent you from seeing what they have to say.”
But not everyone tweets, and that means the results will end up skewed. “Twitter has great penetration in America,” Cebrian notes. But other countries might not be so well represented. “We can’t claim universality,” he says. It also depends on the natural disaster. People may be very excited about issues that are not yet local. “Twitter can be very good if you want to raise awareness, but it might not be a good tool today to see where Zika is traveling,” Cebrian notes.
So far, the associations between tweet and destruction was still only moderate. “This is the beginning,” Cebrian notes. He hopes that down the line, companies such as Twitter and Google might want to help do further research. But he also says, in the end, the emergency aid groups have many methods, and “this is only another tool.”