Belgian researchers successfully use machine learning to find meteorites in Antarctica

by Ole Ellekrog
01/23/2025

1
A meteorite as it looks when it's found in the ice of Antarctica. Photo: ULTIMO / INTERNATIONAL POLAR FOUNDATION
A meteorite as it looks when it’s found in the ice of Antarctica. Photo: ULTIMO / INTERNATIONAL POLAR FOUNDATION

Its vast ice sheet makes Antarctica the best place on Earth to look for meteorites. Recently, a Belgian team returned with 115 new space rocks after testing a new method for locating them.

A gang of researchers are on the move.

They drive slowly through the icy landscape near the Belgica Mountains in Antarctica’s Queen Maud Land. They drive on snowmobiles in a V-shaped formation, looking around carefully, searching the snowy surface for rocks and looking out for dangerous crevasses.

If their vehicles had been different and their location another, you might mistake them for a branch of the Hells Angels. But these scientists are not looking to disturb public order; instead they are searching their surroundings for an extraterrestrial kind of treasure: they are looking for meteorites.

One of them, Steven Goderis of the Vrije Universiteit in Brussels, is back in Antarctica for the sixth time. Because of his experience in identifying both meteorites and crevasses, he usually takes up a position on the extreme left or the extreme right of the formation, guiding the less experienced researchers in front of him by walkie-talkie.

And there are good reasons that the researchers have come to the inhospitable environment of Antarctica to search for objects that are not, seemingly, a particularly Antarctic phenomenon.

“Meteorites fall more or less equally across the globe with about 5,000 falls larger than a kilogram hitting the Earth every year. But in Antarctica the meteorites end up in the ice, which makes them much easier to find,” said Steven Goderis, who recently returned from the ongoing 2024-2025 Belgian Antarctic Research Expedition (BELARE).

“In fact, about 60 percent of the more than 80,000 known meteorites have been found in Antarctica,” he told Polar Journal AG. 

In this photo, the meteorite researchers did not form that v-shape that they apparently did during most of the search. Photo: ULTIMO / INTERNATIONAL POLAR FOUNDATION
In this photo, the meteorite researchers did not form that v-shape that they apparently did during most of the search. Photo: ULTIMO / INTERNATIONAL POLAR FOUNDATION

Finding fallen stars with machine learning

The systematic search for meteorites in Antarctica has been going on for decades. In that time, scientists have gradually discovered patterns in the way the meteorites are distributed across the continent. Wind, ice flows, and mountain ranges, for instance, are natural phenomena that make certain areas more ripe for meteorite detection.

Steven Goderis and colleagues’ mission this year took place near the Belgica Mountains in East Antarctica. This particular part of the continent was chosen for practical reasons as it lay relatively close to the Belgian research station Princess Elisabeth.

But another, more high-tech reason was also behind: a self-taught algorithm. A goal of this year’s mission was, for the first time, to test the utility of a new machine learning model. The model, which was developed by Veronica Tollenaar et al in 2022, is based on a number of parameters identified from previous meteorite findings.

The parameters included ‘surface temperature’, ’surface slope’, and ‘ice thickness’. By compounding these parameters a map of Antarctica was created. The map, which can be viewed online, is a patchwork of promising icefields, typically located inland and at the foot of the continent’s many mountains and so-called nunataks.

One among the many promising areas was just north of the Belgica Mountains.

“Most of the meteorites we found were in this ice field here,” Steven Goderis said, pointing it out on the map: “So there is definite value in the predictive power of the model.”

He did, however, see some room for improvement in the machine learning model.

“I will also say that once you are at a certain level of detail, other types of information need to be taken into account. Information such as local bedrock and crevasse-zones, for instance. These could possibly be included in future refined models,” he said.

Locating the meteorites

As will now be obvious, the V-shaped searching described above was far from random. The method has been developed through years of experience, and usually involves an experienced guide driving at the front. The researchers take up the following positions while they look for rocks on the surface of the blue ice.

Steven Goderis recalls other visits to Antarctica during these last 15 years, when almost no meteorites were found. But on this year’s mission, using both the algorithm-generated map and the refined searching method, a total of 115 were located.

In some fields, anything that is not ice or snow can be assumed to be a meteorite, but in other areas regular local rocks can be scattered on the ice surface, making the search more difficult. Typically, the researchers will find around five meteorites per day, but on one successful day this year a total of 27 meteorites were located. 

“We’ll find a new meteorite every hour, but sometimes you’ll have three hours of nothing before we find a new one. So, it is always exciting when a new one is found,” Steven Goderis said.

“When we find a new one, we will note down the coordinates, take pictures, and put the fragments in a bag. The new meteorites are then also given a field name, which in our case is the initials of the person who found it plus the year, month, date, and respective number on that day,” he said.

Determining the age of Antarctica’s ice

Steven Goderis is happy with the outcome of this year’s mission. 115 meteorites is a substantial number when only 80,000 have been found and recorded globally. But the scientific implications of the newly found meteorites are difficult to predict.

All known meteorites are recorded in a global database where scientists from across the world can locate and research them more thoroughly. Since the new rocks still need to be classified and studied, any one of them could turn out to be special, Steven Goderis explains.

To illustrate his point, he mentions a meteorite known as ASUKA12236. It was originally found in 2012, but only in recent years, through a thorough study, was it discovered to be one of the most primitive meteorites ever found, giving scientists important hints about the building blocks of the Earth and other planets.

“There are multiple scientific goals in searching for meteorites. They teach us both about what’s out there in the Solar System but also about Earth itself and how it formed and evolved,” he said.

“Currently, we are also very interested in how long they have been on Earth before we found them, and in the area around the Belgica Mountains, we have dated the meteorites through cosmic ray exposure dating.”

“This showed us that some of the meteorites have been lying there for several hundred thousand years. This is also important for determining the age of Antarctica’s ice in this area, as it reveals its minimum age,” Steven Goderis said.

Ole Ellekrog, Polar Journal AG

More on the topic: