A huge haul of possible planets has been hiding in NASA data, waiting for scientists to look more closely at faint, flickering stars.
In a new study, researchers reanalyzed the first year of observations from NASA’s Transiting Exoplanet Survey Satellite (TESS). They found 11,554 planet candidates. Of these, 10,091 had not been flagged before, 1,052 were already known TESS candidates, and 411 appeared only once as they crossed in front of their stars.
Fresh Tracks
TESS hunts for planets by watching stars dim.
When a planet passes in front of its star, it blocks a tiny fraction of the light. Astronomers call this a transit. TESS keeps an eye out for these dips, using the transit method.
But stars are messy and they can flicker for many reasons. Another star may eclipse its companion, or a nearby object could contaminate the signal. The spacecraft itself can leave patterns in the data. Turning a dip in starlight into a planet candidate is not an easy task; turning a candidate into a confirmed planet is even harder. Most TESS searches have favored bright stars, where follow-up observations are easier.
“Instead of looking at only the bright stars, which has been done previously, we expanded our search for planets to include fainter stars,” Joshua Roth, a graduate researcher at Princeton University and lead author of the study, told IFLScience.
That decision opened up a huge trove of overlooked data. The team examined more than 83 million records of starlight from TESS’s first year, including many stars too faint to be searched closely before. But they didn’t do it by hand.
Planetspotting
The researchers built a semi-automated pipeline that used machine learning to sort through the data. Their main tool was a random forest classifier—a system that lets many decision trees vote on whether a signal looks like a planet, an eclipsing binary star system, or noise.
×
“This just gives us a much larger base of stars that we can search for these planets,” Roth said. “We developed a semi-automated pipeline that incorporates some machine learning to go through tons of this data and find planets. And we found about 10,000 new planet candidates.”
The team trained separate models for brighter and fainter stars because faint stars carry different kinds of noise. The software learned from known TESS candidates, known eclipsing binaries, and artificial planet signals added to real light curves. After automated cuts, the researchers manually inspected about 50,000 possible transit signals. Every published candidate passed human vetting.
The new list is a map of promising targets. Some candidates may turn out to be false alarms. TESS pixels cover large patches of sky, so light from nearby stars can blend together. The team also removed many likely eclipsing binaries and known contaminants, but they caution that follow-up work remains essential.
The scale is still remarkable. Humanity has confirmed more than 6,000 exoplanets in about three decades. This study adds more than 10,000 new possibilities from one year of TESS data.
Most of the candidates appear to be large worlds. The study classifies 97.7% as gas giants, with smaller numbers resembling Neptunes, sub-Neptunes and super-Earths. It also includes 66 ultra-short-period candidates, possible worlds that circle their stars in less than a day.
One Confirmed Case
To test their pipeline, the researchers followed up on one target: TIC 183374187. They wanted to see if this really was a planet.
Using the Planet Finder Spectrograph on the 6.5-meter Magellan Clay Telescope in Chile, they measured the star’s wobble and confirmed the candidate as TIC 183374187 b, a hot Jupiter. The planet orbits every 5.059 days, has about 0.56 times Jupiter’s mass and measures about 1.25 Jupiter radii.
Its host star is old, metal-poor and likely belongs to the Milky Way’s thick disk, an ancient population of stars. Hot Jupiters like this one puzzle astronomers because giant planets are thought to form farther from their stars, where there is enough cold gas and ice to build them. Yet these worlds end up roasting in tight orbits that last only days. Finding one around such an old star gives astronomers another system to test how giant planets form, survive and move inward over billions of years.
Year Two
The first-year search mainly covered TESS’s southern-sky observations, so the catalog is in its infancy. Roth and his colleagues have already started working through TESS’s second year of observations, where the spacecraft looked at a different half of the sky and revisited some stars at later times.
Some planets orbit too slowly to show up clearly in one 27-day TESS observing window. A single dip can suggest a planet, but repeat dips are needed to measure its orbit. By linking observations taken months apart, the team can recover longer-period candidates and reject signals that appear only once because of noise or contamination.
They are also changing the search itself. The next version of the pipeline is designed to look across multiple observing windows instead of treating each short TESS stare mostly on its own. That should improve the signal for faint planets and may help uncover smaller worlds that were too subtle to stand out in the first pass.
The work now shifts from discovery to triage. Astronomers must decide which candidates deserve telescope time, which are likely impostors and which may answer bigger questions about how planets form around different kinds of stars.
The study was published in The Astrophysical Journal Supplement Series.