We have studied the skies for centuries, but we have only found two objects known to come from another star system. The first interstellar object to be confirmed was 1I/2017 U1, more commonly known as ?Oumuamua. It was discovered with the Panoramic Survey Telescope and Rapid Response System (Pan-STARRS) and stood out because of its large proper motion. Because ?Oumuamua swept through the inner solar system, it was relatively easy to distinguish. The second interstellar object, 2I/Borisov, stood out because it entered the inner solar system from well above the orbital plane. But while we have only discovered two alien visitors so far, astronomers think interstellar objects are common. It’s estimated that several of them visit our solar system each year, and there may be thousands within the orbit of Neptune on any given day. They just don’t stand out, so we don’t notice them. But that could soon change.
The Vera C. Rubin Observatory is scheduled to come online in 2025. Unlike many large telescopes, Rubin Observatory isn’t designed to focus on specific targets in the sky. Its mirror can capture a patch of sky seven Moons wide in a single image. It will capture more than a petabyte of data every night, capturing images of solar system bodies every few days. This will allow astronomers to track even faint and slow-moving bodies with precision. The orbit of any interstellar object will stand out clearly. IF astronomers can find them. Which is where a new study comes in.
With so much data being gathered, there is no way to go through the data by hand. Some things, such as supernovae and variable stars, will be easy to distinguish, but interstellar bodies in the outer solar system will pose a particular challenge. In any given image, they will appear as a common asteroid or comet. It’s only after months or years of tracking that their unique orbits will reveal their true origins.
So the authors of this new work propose using machine learning. To demonstrate how this would work, the team created a database of simulated solar system bodies. Some of them were given regular orbits, while others were given interstellar paths. Based on this data, they trained algorithms to distinguish the two. They found that some machine learning methods worked better than others. In this case, the Random Forest approach, where one classifies decision trees statistically, and the Gradient Boosting method, which prioritizes “weak learners” to strengthen them, seem to work the best. The more commonly known Neural Network method was less effective.
Overall, the team found that machine learning can detect interstellar objects with great efficiency, and the number of false positives should be small enough that they could be effectively managed. While the approach won’t find all the interstellar bodies in our solar system, it should be able to find hundreds of them within the first year of Rubin’s operation. And that will give us plenty of data to better understand these enigmatic visitors.
Reference: Cloete, Richard, Peter Vereš, and Abraham Loeb. “Machine learning methods for automated interstellar object classification with LSST.” Astronomy & Astrophysics 691 (2024): A338.