PlanetSense: Stepping in When Disaster Strikes

“If you don’t know the impact, you can’t mobilize the people,” Thakur said. “If a school is in the hurricane’s path, it’s not likely to make a good emergency shelter. We can compare pictures of the school before and after the hurricane. We can identify where the infrastructure is and whether it’s likely to be damaged or safe. We can help determine the locations of shelters and help pinpoint where more people will be required.”

Relying solely on confirmed reports from witnesses on the ground can take weeks or months to paint a full picture of a storm’s toll. That kind of timeline wasn’t an option when Dorian hit.

“People put their lives on this data,” Thakur said. “No one has six months to evaluate it. The first responders have to go now.”

Researchers used a ground-up approach to design what Thakur describes as a scalable, location-based intelligence-gathering platform. An early PlanetSense storm-chasing effort, tracking Hurricane Irma in 2017, took a turnaround time of about a week. By the time Dorian hit two years later, the team shrank that timetable to a 24- to 48-hour window, with the majority of the results still eyeballed by analysts.

“One of the challenges is to know how to trust this information,” Thakur said. “Is there a second line of evidence? We always have to verify.”

PlanetSense detects sudden bursts of online activity as they happen – photos, video, text – that can signal such events as hurricanes, wildfires or other events related to national security. The system can pick up keywords – fire, storm, flood – and spot clusters of similar posts from the same area.

“When we get that kind of information during one of these events, the system starts to spike,” Thakur said. “The whisper becomes a roar.”

Thakur and a team of 14 data analysts, geographers and others then start putting the pieces together to get a picture of what’s happening where.

“I’m a computer scientist, and I tend to think in very binary, zero-and-one terms,” Thakur said. “The human-dynamics geographers see the fuzziness between the zeroes and ones. We need both. The whole approach increases autonomy, observability, and confidence toward the decision-making process to advance the science of national security.”

The inspiration for PlanetSense came around 2014, when Thakur helped collect volunteered geographic information and other social-media data for a project that mapped relative occupancy at the University of Tennessee’s Neyland Stadium on football game days before and after kickoff.

“That’s how it evolved over time,” he said. “We wanted to know: Can we make this idea bigger, possibly covering the entire planet and sensing it every second? What really excited me to pursue this is primarily for the humanitarian impact. Even if I can make just a small difference with this work, it’s so impactful. This is the kind of computing you can use for the benefit of humankind.” 

Some of his grad-school classmates, former colleagues, and other friends hold jobs at the online economy’s biggest software companies. When they ask what he’s doing at ORNL, he smiles.

“I tell them: ‘I’m saving lives!’ ” he said. “You can’t put a price on that.”