Citizen Scientists Help Emergency Response in Caribbean

After natural disasters, citizen science platform Zooniverse puts the crowd to work helping emergency responders on the ground.

On September 12, approximately 500 satellite images of hurricane-battered Guadeloupe were uploaded to the citizen science website Zooniverse. In just two hours, volunteers had worked their way through all 500—comparing them to “before” photos, tagging them for things like flooding and road blockages, or simply noting that cloud cover obscured the land below. Every image was reviewed by 50 people to ensure nothing was missed. (After the initial batch, the requirement was lowered to just 30.) Since then, thousands of volunteers have poured over the ten image sets that have been uploaded—some with as many as 9,000 subjects—covering most of the Virgin Islands as well as Guadeloupe and Turks and Caicos. The results are used to create heat maps that emergency responders on the ground can use to inform their own aerial reconnaissance.

The project is led by the Planetary Response Network, a collaborative effort between several organizations, including Zooniverse, the University of Oxford Machine Learning group, and an NGO specializing in disaster response called Rescue Global. The network was first activated after the April 2015 earthquake in Nepal, which killed nearly 9,000 people, and then again in 2016 after an earthquake in Ecuador killed nearly 700 people. Last year, more than two thousand volunteers combed through 25,000 square kilometers of satellite imagery in a mere half day.

It’s hard to get an exact number of volunteers because you don’t have to create an account in order to contribute. “The full user counts (as of midnight last night [Tuesday]) are: 4,408 logged in and 5,195 not logged in, from 9,283 unique IP addresses,” Brooke Simmons, the project lead, wrote in an email to Civicist. “Even if every single registered volunteer has been classifying from both home and work IP addresses, there are still likely some people who have participated without ever signing in. Their classifications get counted too—even though we have no way to credit them for their work, or send them status updates.”

For the current response to Hurricane Irma and other tropical storms, the Planetary Response Network is using images from four sources: the European Space Agency (ESA), NASA, and two private companies, Planet and DigitalGlobe. The images vary in resolution quality, but Simmons has assured volunteers that even if landscape details are difficult to discern, their classifications are still helpful.

“Totally understand your feeling about resolution here,” Simmons wrote in response to a complaint that the images in one set were too pixelated to see detail. “In the diverse constellation of Earth-monitoring satellites available to draw data from, the data available in this first set of images is about middle-of-the-road in terms of resolution. Obviously we’d like to have better resolution, but for now this is what’s available, and we know from past experience that major damage like washed-out roads and destroyed ports are still possible to see here. And there is actually a benefit to the lower-resolution: together we can process the data very rapidly, so as long as we get some value out of the classifications it’s still worth it to deliver that to Rescue Global.”

The heat maps generated by the Planetary Response Network are really supposed to be quick and dirty initial assessments, but Simmons tells Civicist that in previous activations of the network, the heat maps showed areas that needed assistance that hadn’t been visible on other maps, and became a standalone source of supplementary information.

After trying a few classifications myself, I also wanted to know whether the photos were that blurry or if my eyesight was that bad.

“It can be very challenging and the uncertainty many people feel is quite natural,” Simmons responded. “We’ve done some work to study the accuracy of the classifications, and compared them with other assessments from other groups. You’re absolutely right that any one person can miss something or misidentify something; our strength is in having multiple people give their inputs on any given part of a satellite image. In combination, those inputs tend to be very accurate at recovering whatever information it’s possible to recover from the available imaging (which depends mostly on resolution).”

To help volunteers with their task, the Planetary Response Network created a field guide with visual examples of what blocked roads, flooding, or structural damage looks like. Volunteers can also ask questions about specific images or voice any concerns in the discussion forum, where collaborators or projects leads can publicly respond. This transparency helps everyone learn from each other, just as they can in the discussion pages for other citizen science projects hosted on Zooniverse.

The Planetary Response Network was first conceived of in 2013, according to Patrick Meier, an expert on the use of technology in humanitarian applications, who worked with Simmons and her colleagues to start the project. Zooniverse was by then already a veteran citizen science platform, with seven years of experience putting the crowd to work classifying everything from galaxies to whale tails. The platform was created on the premise that people want to contribute to projects they find worthy, like advancing scientific work. Turns out many people, as evidenced by the outpouring of volunteer support after Irma and Harvey, like to help when shit hits the fan, too.

And that participation adds up. “All the volunteers in total have done over 4 months’ worth of work—that is, if one person were to sit and classify for 24 hours a day, 7 days a week, no breaks of any kind, it would take then 4.2 months to put in the same effort as the volunteers,” Simmons said. “In a more humane comparison, it’s about 1.5 years of full-time employment from 1 person. In just over a week!”

The current set of images will be the last before Hurricane Maria sweeps over the region—the team is already processing post-Maria images—and then all the work will begin anew.