About a million years ago, an asteroid smacked into the normally tranquil surface of Mars. The impact released a fountain of debris, and some of the rocky fragments pierced the sky, escaping the planet’s gravity to journey through the dark.
Some of the rocks eventually found their way to Earth and survived the plunge through our planet’s atmosphere to thud into the surface–including a hefty 15-pound shard that crashed into Morocco in 2011. Now known to scientists as the depleted shergottites, this collection of more than a dozen space rocks makes up an intriguing portion of the 317 known Martian meteorites—the only material from Mars we have on Earth.
Determining what part of Mars these meteorites came from is a critical part of piecing together the planet’s history—but it’s proven to be a major scientific challenge. Now, with the assistance of a crater-counting machine learning program, a team of researchers studying the depleted shergottites may have finally cracked the case: They concluded that these geologic projectiles came from a single crater atop Tharsis, the largest volcanic feature in the solar system.
This ancient volcanic behemoth on Mars is adorned with thousands of individual volcanoes and extends three times the area of the continental United States. It was built over billions of years by countless magma injections and lava flows. It is so heavy that, as it formed, it effectively tipped the planet over by 20 degrees.
If these meteorites do come from Tharsis, as the analysis published in Nature Communications suggests, then scientists have their hands on meteorites that can help identify the infernal forces that fueled the construction of this world-tipping edifice.
“This could really change the game about how we understand Mars,” says Luke Daly, a meteorite expert at the University of Glasgow who was not involved with the study.
Most Martian meteorites are in a category called the shergottites, named after the Indian town of Sherghati where one was seen falling from the heavens in 1865. The shergottites are all volcanic rocks with similar compositions, but a handful of them, the depleted shergottites, possess a strange chemical signature.
On Mars, certain elements such as neodymium and lanthanum don’t like to bond with minerals in the mantle, the solid-but-squidgy part of the planet below the crust. The depleted shergottites are lacking in these elements—hence the name “depleted”—suggesting they are from Mars’s mantle.
But how did these rocks get close enough to the surface to be ejected in an impact? On Earth, mantle rock can work its way to the surface in two ways: when two tectonic plates move apart and permit the mantle to ascend, or when a fountain of superhot mantle matter known as a plume rises from the deep. Mars doesn’t appear to have ever had plate tectonics, so a mantle plume is the most likely scenario.
If these spacefaring volcanic rocks all came from a single impact, then it must have been quite powerful, leaving a crater at least two miles across and potentially much bigger. And the crater would have to be about 1.1 million years old, as cosmic rays that bombarded and altered the meteorites’ surfaces over time reveal how long they were traveling through space after the impact.
Even with these clues, however, tracing these bits of Martian rock back to their original location has proven extremely difficult. They are like individual jigsaw pieces separated from the rest of the puzzle: Without knowing what their original environment looked like, it is almost impossible to place them in a specific part of the planet.
“As geologists, we record loads of information about where we collect rock samples from, because context matters,” says áine O’Brien, a doctoral student studying Martian meteorites at the University of Glasgow who was not involved with the study. “With Martian meteorites, because we don’t know the context, we have to make a very well educated guess at what happened to it to form it.”
And to make that educated guess, scientists turned to a new tool in planetary science: machine learning.
One crater among millions
The only way to definitively determine the age of a planet’s surface is to take a physical sample and study its radioactive compounds. But until NASA and the European Space Agency’s Mars Sample Return campaign brings some pristine Martian rocks back to Earth in the 2030s, researchers need to rely on a technique to estimate surface ages known as crater counting.
On Earth, strong winds, flowing water, erupting lava, and a cornucopia of living things speedily erase craters from old impacts. Not so on Mars, a geologically comatose world with weak winds and no surface water. There, sizable craters remain intact for hundreds of millions or even billions of years. Assuming the rate of impacts over time is known, a surface on Mars with more craters would be older than one with fewer craters.
Scientists can use other tricks to deduce a crater’s age. “When an asteroid impacts the surface, a bunch of debris will be ejected,” says Anthony Lagain, a planetary geologist at Curtin University and the new study’s lead author. The bits that fall back to Mars impact the surface and make small, secondary craters around the original primary crater. Even on Mars, these secondary craters are eroded by wind within a few million years, so any large crater surrounded by secondary craters must have been made very recently in the planet’s history.
“In order to get a better idea of ages, you need to get to smaller and smaller craters,” says Gretchen Benedix, an astrogeologist at Curtin University and co-author of the study. Smaller impacts are more common than larger ones, so you can use minor differences in the number of smaller craters across two surfaces to work out more detailed timelines.
To figure out if a crater was exactly 1.1 million years old, the team had to catalog Mars’s small craters and use them to precisely date the surface. Doing this manually would have been torturous. Instead, they fed orbital imagery of Mars into a machine learning program and trained it to find craters less than two-thirds of a mile long.
It quickly found about 90 million, says Kosta Servis, a data scientist at Curtin University and co-author of the study. With that timeline of craters in hand, the team was able to start narrowing down the possible origins of the depleted shergottites.
Shards of a volcanic titan
After sifting through the data, the team identified 19 large craters in volcanic regions on Mars that were surrounded by multiple secondary craters—a sign that these planetary scars could be as young as the 1.1-million-year-old crater they sought. Using the catalog of 90 million small craters, the researchers were then able to precisely date the blankets of debris radiating from the larger craters, which revealed more accurate estimates of their ages.
Some of the craters were about the right age, but that wasn’t enough. The formation age of the surrounding terrain had to match the ages of the minerals found in the meteorites as well. To check, the team once again used its crater catalog to date the volcanic plains.
Out of those 19 craters, just two were excavated from youthful volcanic deposits by an impact event 1.1 million years ago: crater 09-00015 and Tooting crater. The latter (named after a district in London) looks to have been formed by a powerful oblique impact—the kind of collision that would propel a lot of Martian meteorites into space.
“Tooting crater has a special type of multi-layered ejecta deposit that suggests there was ice or water around at the time of the impact,” says Peter Grindrod, a planetary scientist at London’s Natural History Museum who was not involved with the study. Impact simulations show that ice and water can generate more debris, plenty of which can escape into space if given enough momentum.
With all this evidence, the team identified the 19-mile-long Tooting crater as the prime suspect for the source of the depleted shergottites. “It’s a really well put together argument,” Daly says. “Everything seems to fit.”
The scientists have not completely ruled out crater 09-00015, but the important thing is that both craters “lie in the Tharsis region, where a vast hotspot, or superplume, has long been thought to have produced the massive bulge on the surface of Mars,” Grindrod says. Regardless of which specific crater the meteorites came from, they can tell us about the history of the largest volcanic region on Mars.
Crater counting has previously revealed that some of Tharsis’s features were made over 3.7 billion years ago, but the younger depleted shergottite meteorites are just a few hundred million years old. That suggests the Tharsis superplume is almost as old as Mars itself, and it continued producing magma long after many other volcanic centers on the planet died out.
Like Earth’s plumes, Mars’s mantle plumes helped shape the evolution of the planet’s surface, erupting enormous volumes of atmosphere-altering gases while dramatically changing its topography. The Tharsis superplume may have had a near-continual influence on the red planet’s development.
Mars’s days of frequent and prolific eruptions are long gone. But Tharsis’s prolonged volcanism bolsters the notion that even small planets, those that should have lost their internal heat eons ago, can remain volcanically active for far longer than anyone originally suspected.
Decoding the craters of other worlds
Buoyed by their discovery, Lagain’s team is hoping to identify the source craters of other Martian meteorites—including some of the very oldest, which could reveal more about Mars’s waterlogged past.
But future success, as well as this study’s implications, depend on whether the machine learning program properly counted its craters. Crater counting is rife with difficulties: the rate of impacts over time is estimated, for example, and small circular structures on Mars that resemble craters could potentially fool a computer program.
Machine learning “is a really inventive way of trying to tackle this problem,” says Lauren Jozwiak, a planetary volcanologist at Johns Hopkins University Applied Physics Laboratory not involved with the study. “Boy, I hope this method works,” she says, because if it does, “it would be really cool to take this and apply it to other planets.”
The study’s authors concur. “Mars is cool,” Benedix says. “But this algorithm and this methodology isn’t just applicable to Mars. It’s going to the moon. It’s going to Mercury.”
If machine learning really has solved this long-standing meteorite mystery, it opens the door to all sorts of undreamt-of possibilities. “We are arguably only just starting to see the implications of machine learning in planetary science,” Grindrod says.