To better understand how continents were formed and how the Earth’s climate has evolved over hundreds of millions of years, geologists analyse patterns on sand grains, using scanning electron microscopes to observe their texture down to the micrometre scale. This process “enables us to determine how grains of sand were transported, because the textures reveal information on collisions encountered by the grains, which we use to determine if they were carried by air or water etc.,” explains Mathieu Lapôtre, an assistant professor at Stanford, who leads the Earth & Planetary Surface Processes research group. “It’s labour-intensive work and unfortunately interpretations may vary from one geologist to another,” adds Lapôtre. To expedite this analysis, the French researcher has co-developed an artificial intelligence tool christened SandAI which will not only be of assistance to geologists but could also open up new approaches for geoarchaeologists and forensic scientists.
The model achieved 90% accuracy, proving that AI can indeed be used to determine the origin of rocks
Determining the origins of rocks with AI
“We collected sand samples with known transport environments from dunes, rivers, beaches and glacial deposits and took photos of the grains under a microscope, which we fed into the AI along with a simple indication of their origin. At no point did we teach the model to recognise specific textures.” SandAI then underwent an initial test in which it was instructed to analyse modern sands from sites it not yet seen. “In a second test, in which the model was supplied with images of several sandstones from well-characterised environments ranging from the current day back to the Jurassic period, the model achieved 90% accuracy, proving that AI can indeed be used to determine the origin of rocks.”
Finally, the researchers presented SandAI with samples from a site in Norway dating back to the Cryogenian period — a time before the emergence of fish or plant life —which are of contested depositional origin. “Different groups of geologists have proposed different interpretations. However, the AI enabled us to determine that the original grain transport environments were 84% aeolian and 16% glacial. AI is therefore useful in contexts where there are few paleoenvironmental indicators, and also for analysing drill cores when prospecting for energy resources.
Scope for wider use in geoarchaeology and forensic science
“The tool could also be put to other uses, for example in the field of geoarchaeology. Migration involving the transport of materials, has always been a part of human history ever since we emerged on Earth. SandAI will better enable us to better determine sources of materials and the places or rivers where they may have been.” The same applies to forensic science, for example, it could be used to determine if sand used in construction is from illegal sources. ‘There are a large number of operations illegally mining sand, the most widely used material on earth, which cause major environmental damage, particularly coastal erosion.” SandAI could help regulatory bodies to conduct more investigations.
People we quote
Dr Lapôtre
Dr Lapôtre is an Assistant Professor in Earth & Planetary Sciences at Stanford University. His research focuses on the physics behind sedimentary and geomorphic processes that shape planetary surfaces (including Earth's), with the goal to unravel what landforms and rocks tell us about past hydrology, climate, and habitability. He earned a BS in Geophysics, a MS in Environmental Science & Engineering, and a MS in Geophysical Engineering from the Ecole et Observatoire des Sciences de la Terre, Strasbourg, France, and a MS in Planetary Science and PhD in Geology from the California Institute of Technology. During his PhD, he was part of the team that operated NASA's Curiosity rover on Mars.
Read more
Read more
An AI tool for scanning grains of sand reveals volumes about the past (en anglais)
