Artificial intelligence has mastered tasks ranging from writing essays to identifying diseases. Now it is helping scientists decode one of the oldest mysteries in the night sky: what meteors are actually made of.
In a new study, researchers at Lowell Observatory used machine learning to analyze more than 28,000 meteor events and develop a sophisticated new system for classifying the rocky debris that streaks through Earth’s atmosphere. The approach allows scientists to identify subtle physical differences among meteoroids, the tiny fragments of asteroids and comets that create shooting stars, with a level of precision that traditional methods could not achieve.
The breakthrough highlights a growing trend in modern science. As observatories and research networks generate enormous amounts of data, scientists are increasingly turning to AI to uncover patterns that were previously invisible. In this case, AI is helping astronomers better understand the origins and composition of some of the solar system’s oldest material.
“Meteors have been observed for centuries, but only recently have we had datasets large and detailed enough to apply modern machine-learning methods,” said Samantha Hemmelgarn, lead author of the study. “This allows us to extract physical information that was previously hidden in the data.”
The research draws on observations collected in 2023 by the Lowell Observatory Cameras for All-Sky Meteor Surveillance, or LO-CAMS, network, which is part of the Global Meteor Network. Each event was described using 13 directly measured properties, including speed, brightness, duration, altitude and atmospheric density.
For decades, astronomers have relied on classification systems that use only a handful of measurements to estimate a meteoroid’s composition and origin. While effective, those approaches didn’t fully capture the diversity of material entering Earth’s atmosphere from across the solar system.
“Our goal was to move beyond traditional classification schemes,” said Nick Moskovitz, a co-author of the study. “Modern meteor networks capture a wealth of observational information, and we wanted a framework that could fully take advantage of that.”
Rather than forcing meteors into predefined categories, the researchers used machine-learning algorithms to search for natural patterns within the data. The result was a classification system built from the observations themselves rather than assumptions about what scientists expected to find.
The analysis revealed three key factors that govern how meteoroids behave as they plunge through Earth’s atmosphere. The first relates to how the object travels through the atmosphere, including its speed and trajectory. The second involves activation — how easily the meteoroid heats up and begins to glow. The third concerns how its size and shape influence the way it breaks apart during atmospheric entry.
One finding was particularly revealing.
“One of the most exciting results was how clearly the activation behavior separated asteroidal material from cometary material,” Hemmelgarn said. “That tells us we’re capturing something fundamentally physical, not just statistical patterns.”
The distinction matters because meteoroids originate from vastly different sources. Some are fragments created by collisions between asteroids. Others are remnants shed by comets as they travel around the Sun, releasing dust and debris along their paths. Understanding those differences helps scientists reconstruct the history of the solar system and better understand the materials that helped form planets billions of years ago.
Based on the machine-learning analysis, the team developed a new classification framework called Hclass. The system ranks meteoroids according to their “hardness,” ranging from dense, iron-rich material commonly associated with asteroids to fragile, porous material characteristic of cometary debris.
Unlike traditional classification systems, Hclass can be adjusted to provide either broad categories or highly detailed subdivisions within meteoroid populations. Researchers say that flexibility offers a more realistic picture of the wide range of materials entering Earth’s atmosphere.
“Hclass gives us a more nuanced view of meteoroid composition,” Hemmelgarn said. “It bridges the gap between classical meteor theory and the realities of modern, large-scale observations.”
To test the new framework, the researchers applied it to several well-known meteor showers and compared the results with what scientists already know about their origins. The classifications aligned with expectations based on whether the showers were produced by comets or asteroids, while also revealing previously unseen differences within those meteor populations.
The researchers believe the system is particularly valuable because it is designed to scale. Its mathematical framework can be applied to a single meteor observation or expanded to analyze millions of events as global meteor-monitoring networks continue to grow.
“This work shows that machine learning isn’t just about handling big data,” Moskovitz said. “It’s about turning [that] data into physical understanding of where this material comes from and how our solar system works.”
AI is often associated with chatbots, image generators and consumer technology, but it is becoming an indispensable scientific tool, helping researchers uncover patterns hidden within massive datasets and transforming raw information into new knowledge. In this case, AI is turning ordinary observations of shooting stars into a powerful tool for understanding cosmic history.
