Machine learning unlocks secrets of early plate tectonics

Rock weathering and plate tectonics are essential for life: both regulate the Earth's surface temperature and provide vital nutrients for living organisms. However, how and when these vital processes began on Earth remains a mystery. And could these processes date back to Earth's early days, during the Hadean epoch, more than 4 billion years ago?

The biggest problem is that Earth hasn't preserved any rocks from that era, so geologists must rely solely on the oldest preserved Hadean material: the Jack Hills zircons from the Western Australian outback.Not surprisingly, the Jack Hills zircons have been both celebrated and questioned for whether they contain evidence of Earth's oldest sediments and signs of plate tectonics.

Geologists have been trying for years to determine whether the zircons in Jack Hills crystallized from sediment-bearing magmas, known as sediment-derived or “S-type” granites, but so far have had mixed results using traditional geochemical indicators.

Traditional discrimination diagrams using only a few trace elements have been used to argue that S-type granites were few or rare on the early Earth. Unfortunately, when phosphorus (P) content is low, there is overlap in the sources of different types of zircons. This is especially problematic because low-P zircons make up more than 95% of the oldest detrital zircons. Clearly, more information on trace elements in zircons is needed.

In a study published in the journal PNASA research team led by Professor Ross Mitchell from the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS) addressed this problem by applying a precise machine learning method to recognize Hadean S-type zircons. This new approach is particularly capable of identifying Hadean low-pressure S-type zircons and distinguishing S-type from non-S-type zircons.

Previous studies have argued that the early Earth had rare S-type granites, but this is the first to suggest that S-type zircons were abundant during the Hadean. They demonstrated, with a much higher accuracy of 96%, not only the presence but also the abundance of Jack Hills S-type zircons dating back to 4.24 billion years ago.

“This means that the weathering of the crust, the deposition of sedimentary rocks, and their burial and incorporation into magma sources could have begun more than 4 billion years ago,” Mitchell said.

The changes in Hadean S-type granites identified by machine learning are consistent with other geochemical signatures, including hafnium isotope cycling and elevated δ.¹⁸O values ​​consistently show a link between continental collision and mountain range formation.

By further sorting through detrital zircons across the globe through geological time, the researchers found that peaks in high S-type zircons have occurred repeatedly since 4 billion years ago, and surprisingly, they also found that these peaks coincide with the formation of supercontinents.

“This appears to be cyclical, with a cycle of about 600 million years that is surprisingly similar to that of supercontinents: the abundance of S-type zircons increases as the world's continents collide, then decreases during breakup,” said Jiang Jiren, lead author of the study.

This consistent pattern suggests that the process of subduction-driven plate movement and collision has been going on since almost the earliest times of the Earth – that is, subduction-driven plate tectonics has been in operation since the Hadean epoch.

“S-type granites kill two birds with one stone because they exhibit features of both surface weathering (producing sediments) and plate subduction (producing sediments into magma chambers),” Mitchell said. “With these two processes operating essentially from the beginning, Hadean Earth could have been habitable, and life could have arisen not only in hydrothermal vents but also in Darwin's hot pools.”

“The authors introduce a novel strategy in their machine learning approach that should serve as a foundation (and benchmark) for future studies of early Earth zircons,” said Bruce Watson, a geochemist at Rensselaer Polytechnic Institute who did some of the key work on the Jack Hills zircons but was not involved in the new study.

This study demonstrates the potential of machine learning to explore and discover knowledge that could not be found using traditional methods. Further research using AI-based knowledge mining should provide new insights into Earth's ancient history in the future.