How Flickr Predicted Hurricane Sandy's Landfall
Modern mega-events like natural disasters or public rallies are unlike their historic predecessors for one big reason that excites sociologists: Now these famous moments are insanely well-documented by the people who witness them.
Take Hurricane Sandy. In addition to all that debris and damage, the storm left behind an extensive trail of tweets, Flickr photos, and even evolving Wikipedia edits. That data is now feeding the growing field of computation social science, more finely informing what we know about how vast groups of people behave during extraordinary situations.
As another great example of this, researchers at the University of Warwick, the University College London, and Boston University just published some striking data on Flickr use during Sandy. In the journal Scientific Reports, they describe examining photos uploaded to Flickr between October 20th and November 20th of 2012, sorting for the words "Hurricane," "Sandy," or "Hurricane Sandy" somewhere in the tags, title, or description text.
They tabulated the number of images by hour and found some startling symmetry between Flickr users and the behavior of the storm itself. The number of images increased as Sandy moved closer to the U.S. coast off of New Jersey. And then this happened:
"Quantifying the Digital Traces of Hurricane Sandy on Flickr" by T. Preis et al. in Scientific Reports.
Hurricane Sandy's landfall is measured against the atmospheric pressure in New Jersey, recorded at 62 stations around the state that day (as the severity of a hurricane increases, atmospheric pressure drops). The greatest quantity of Sandy photos were taken in the very same hour as the storm's landfall.
The implication? "We suggest," the researchers write, "that Flickr can be considered as a system of large scale real-time sensors documenting collective human attention."
It makes sense that more people would turn their attention to a storm like this the closer it gets (or the worse it appears). The authors again, lead by the Warwick Business School's Tobias Preis:
This would suggest that in cases where no external sensors were available, it may be possible to measure the number of Flickr photos relating to a topic to gauge the current level of this category of problems. A second alternative interpretation would be that users were well informed as to the expected time of landfall due to extensive media coverage, and that their attention to the problem increased as the anticipated climax of the disaster approached, leading to an increase in the numbers of photos taken. This would equally open the possibility that increases in Flickr photo counts with particular labels may reveal notable increases in attention to an issue, such that issues which have received less extensive media coverage but which may merit further investigation may be identified.
We can also imagine that in those moments shortly after landfall, many people within photographing range of the storm suddenly had other, more pressings things to do than photograph it (like conserve their cell phone batteries).
Of course, meteorologists and policy-makers don't need Flickr to tell them the extent and path of a catastrophic storm. But weather data can't necessarily tell us how people are responding to a disaster, or even whether most of them see it coming.