· 4 min read
“You need to find the asteroid before the asteroid finds you.”
That’s a prime directive for NASA, which has been working to catalog near-Earth objects for over two decades, Davide Farnocchia, a navigation engineer at NASA’s Jet Propulsion Laboratory, told Emerging Tech Brew.
During that time, the organization has discovered more than 20,000 asteroids—including nearly 2,000 potentially hazardous ones, per a 2019 report from NASA. Since 2016, the job has been entrusted to NASA’s Planetary Defense Coordination Office.
Currently, the main goal is to discover near-earth objects larger than 140 meters in size (think: about one and one-third football fields). There are an estimated 25,000 of these potentially destructive objects out there, and as of NASA’s 2019 report, the PDCO had at least an estimated 16,000 left to find.
To do that, they need data from astronomers all over the world—and software to process it all.
“The data we’re using now is much higher quality than it used to be in the past, [and] telescopes are much more sensitive—they do a much better job of measuring the position of asteroids, even when they’re faint,” Farnocchia, who helped develop one of NASA’s primary software tools for flagging asteroids, said.
Emerging Tech Brew spoke with Farnocchia about the tech that powers its detection process.
This interview has been edited for length and clarity.
Can you paint a picture of how, exactly, that whole process works?
Sometimes, you can discover a small object—let’s say a few meters in size—that could be headed for impact in the next day or so. And so you want to realize that as quickly as possible.
The most recent example was an object called 2022EB5. It was discovered by an astronomer in Hungary only two hours prior to reaching the Earth. It was only two meters in size. This astronomer submitted the data to the Minor Planet Center, which then posted the data on the confirmation page right away—and we picked up the data and figured out it was going to hit two hours later.
Bigger objects would be discovered ahead of time, but smaller objects like this are a good test run for us. They’re much smaller than what we’re really concerned about, but the fact that we could discover it, and figure out the impact trajectory, was a pretty good validation of the system. We only had two hours of time, and we pinpointed the impact location…In this case, we figured out he was going to hit in the Norwegian Sea, so there wasn’t any damage to the ground.
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
Talk to us about how the software processes incoming information.
Once you get started, you only have a handful of data. The data in this case will tell you how far the object is—or whether the object is getting closer to the earth or is moving away from the earth. So in this case, we use an approach that is called systematic ranging, where we scan a grid of possible values for the distance to the object and the radial velocity, so whether it’s coming closer or going away from the earth
The algorithm is really based on understanding physics—so, given the position of the object at a given time, to project its position into the future…You have to plug in the gravity from the sun, the gravity from the planets, the gravity from other asteroids, relativity, and all these kinds of accelerations. And that essentially tells you how an object is going to move in time…Once you’ve got a trajectory, you want to see whether the object could come close to the earth—that’s the first concern.
How about the role of machine learning?
I can see an application for machine learning when you deal with data collected by telescopes—essentially, telescopes collect a lot of data, and you need to identify the detections that go together. Especially if you take four images of the same patch in the sky, and you’re going to see a lot of light sources there, and so you have to figure out which ones belong to the same object. And as telescopes become more and more sensitive and get to observe fainter and fainter sources, you might need algorithms that can effectively link detections in different images in an efficient manner.
When it comes, on the other hand, to what we do on our side with trajectory modeling and impact-probability calculations, we have a very good understanding of how the physics of these things work. And so we don’t really need to rely on something like machine learning—direct modeling and physical knowledge are usually better than machine learning, if you can afford that.