To weathermen, birds are a nuisance — “clutter,” is their preferred euphemism. That’s because, on a radar screen, a flock of sandhill cranes heading out to forage can look an awful lot like a surge of nasty weather. Luckily, there’s a way to keep them from mucking up the data. Automated precipitation-detection systems do their best to identify these signals, sifting them out of the final report so we don’t open our umbrellas every time a flock of seagulls wings by.
But in the field of ornithology, these trimmings are the main course. Wildlife biologists at the U.S. Geological Survey are applying principles of artificial intelligence to programs that sort radar data, teaching them to recognize birds and bats while ignoring everything else. In the process, they are proving that radar can think as well as see. And depending on how they use their smarts, these systems could help energy companies chose where to build wind turbines, track avian responses to climate change, or steer low-flying military planes around bird strikes.
The National Weather Service operates 159 Doppler radar sites around the U.S. and overseas, usually referred to as NEXRAD. It also provides various algorithms that use the data to detect and analyze rain, tornados, cyclones, and mesocyclones. So far, however, they haven’t put together a program to track birds, even though the data is all there and plenty of people would use it.
Without the proper tools, most of the data just sits there. “It’s one of the largest databases of biological information that no one ever looks at,” says Richard Sojda, a wildlife biologist at the Northern Rocky Mountain Science Center in Montana.
A few years ago, Sojda proposed a new way to use the data from NEXRAD — teach the system to zero in on avian activity, the same way meteorologists focus on precipitation. The first step in doing this is to collect enough data to train an algorithm how to spot birds on a radar screen. This requires a lot of labor, and the project was stalled when Sojda was re-assigned to an administrative post.
However, Sojda sees an opportunity to pick up the speed, with a little help from the Department of Homeland Security. The U.S. agency has begun flying drones along the Canadian border for security purposes, very near where Sojda conducts his research. He hopes to pull data from their radar system and use it to build an automated bird-detecting program. “We’d be able to use that as additional training for our algorithms,” he says.
Janet Ruth, an ecologist and colleague at the Fort Collins Science Center in New Mexico already uses data from NEXRAD to study the migration patterns of birds along the border with Mexico. But she does it at human speed. Experts on her project look at each smear on a radar screen and determine whether it’s a flock of birds or sheets of rain.
A flock of European starlings in California. Courtesy Flickr user etgeek.
“This is one of the reasons that these analyses are so expensive and labor intensive. They require a lot of time from someone with experience in visually examining radar output. We hope that that will change in the future,” says Ruth.
With an efficient way of analyzing NEXRAD data, Ruth and her colleagues would suddenly have access to decades of radar recordings. Research of this kind could benefit diverse industries. Mapping out migration patterns could reduce the number of birds that get killed by the spinning blades of wind turbines. But, if data from radar were analyzed in real-time, it could also help pilots to steer around flocks, saving human lives as well as birds.
* A note on the headline: it was changed after being published to reflect that the algorithms do not yet exist.
Top image: European starlings in California. Courtesy Flickr user etgeek.
Morgen E. Peck is a contributor to IEEE Spectrum, Innovation News Daily and other publications. Her last article for Txchnologist questioned the claim that cloud computing is a green technology.