Scientists Discover 27,500 Asteroids in Old Telescope Images Using A.I.
While most of the team’s new finds are located in the main asteroid belt, about 100 are near-Earth asteroids that pass close to our planet’s orbit
:focal(1920x1097:1921x1098)/https://tf-cmsv2-smithsonianmag-media.s3.amazonaws.com/filer_public/48/ed/48ed9748-7021-489a-a4c9-daf2d57d3cd1/angledview.png)
About 4.6 billion years ago, our solar system formed from a giant cloud of gas and dust. As this cloud collapsed under its gravity, likely prompted by a nearby supernova explosion, it flattened into a swirling disk of debris around the sun. Gradually, these particles began sticking together into larger bodies—some became big enough to form planets, while other rocky or metallic fragments remained as smaller, irregularly shaped objects, which we call asteroids.
Today, NASA estimates these space rocks number 1,351,400, with most orbiting the sun in the main asteroid belt between Mars and Jupiter. But now, a team of researchers working with artificial intelligence is poised to rapidly increase that total by discovering asteroids that have been overlooked.
Last week, scientists with the Asteroid Institute and the University of Washington announced they have identified 27,500 new asteroid candidates—including about 100 near-Earth asteroids that pass close to our planet’s orbit—in 2024 alone. All other observatories worldwide have found about 2,300 asteroids this year.
The research team made these discoveries with the help of a machine learning algorithm called Tracklet-less Heliocentric Orbit Recovery, or THOR.
“We don’t own a telescope. We don’t operate a telescope,” Ed Lu, executive director of the Asteroid Institute and co-founder of the B612 Foundation, said during a discussion about the project last month. “We’re doing this from a data science perspective.”
Astronomers typically discover new asteroids by repeatedly observing sections of the sky through telescope images captured every few hours each night. Asteroids, which appear as moving points of light, are then flagged, verified and monitored, Space.com’s Sharmila Kuthunur explains.
THOR, however, analyzed archival images of the sky, picking out more than 1.7 billion dots of light across more than 400,000 snapshots captured by the National Optical-Infrared Astronomy Research Laboratory (NOIRLab).
The THOR algorithm connects a specific dot of light in one image to a different dot in another image and determines whether the two points represent the same celestial object, often an asteroid. THOR can link observations made at arbitrary intervals, as long as five to six images in a 15- to 30-day window are available—and it can identify an object captured in data from different telescopes on different nights.
“This actually is a tremendous amount of discovery done just by scouring the images that are there already from other telescopes,” Massimo Mascaro, technical director of Google Cloud’s Office of the CTO, said at the event last month.
“We’ve taken an existing dataset… and what we realized is that there are many, many tens of thousands of asteroids in those datasets that nobody has noticed before,” Lu, who is also a former NASA astronaut, added. “And you can unlock them if you have enough computation.”
To access that computation, the Asteroid Institute partnered with Google Cloud’s Office of the CTO, which helped scale computing, manage datasets and store images and data points, per a statement from B612. The THOR algorithm runs on a cloud-based, open-source astrodynamics platform called Asteroid Discovery Analysis and Mapping (ADAM) on Google Cloud.
THOR’s method for identifying asteroids across images on different days was previously impossible. But cloud computing allows the calculations to occur in approximately five weeks, the New York Times’ Kenneth Chang reports.
The NOIRLab dataset, called NSC DR2, is the first of many the Asteroid Institute plans to scan. Researchers are prepared to “scale this work and scan new datasets more efficiently and effectively as they become available,” per the group’s statement. “Asteroid Institute’s goal is to automate this process to the benefit of the astronomical community and space industry.”
None of the asteroids discovered were found to be on a collision course with Earth. But the new technology could help identify asteroids that may be potentially hazardous, helping astronomers keep tabs on space rocks for planetary defense.
“The real issue that we have to solve is finding and tracking asteroids, because you can’t keep an asteroid from hitting the Earth if you don’t know it exists and you don’t know where it is,” Lu said during the discussion last month.
“The most problematic object is the one you don’t know about,” Amy Mainzer, a planetary scientist at the University of Arizona, told Live Science’s Brandon Specktor in November. “If we can know what’s out there, then we can have a much better estimate of the true risk.”