AI tool that scans grains of sand opens window into recent and distant past

A scanning electron microscope reveals the shape and texture of quartz sand grains from the Mississippi River. The grains in this photo are about 200 micrometers long. Courtesy of Michael Hasson/Stanford University
Stanford University researchers have developed an artificial intelligence-based tool called SandAI that can reveal the history of quartz sand grains going back hundreds of millions of years, allowing researchers to determine with a high degree of accuracy whether wind, rivers, waves, or glacial movements caused the sand grains to form and deposit.
The tool offers researchers a unique window into geological and archaeological studies, especially for times and environments where few clues such as fossils remain. SandAI’s approach, called microtextural analysis, can also aid in modern forensic investigations of illegal sand mining and related issues.
“Working with sediments that have not been disturbed or transformed is almost like stepping into a time machine, allowing us to see exactly what was on the Earth’s surface hundreds of millions of years ago. SandAI adds even more detail to the information we can extract from it,” said Michael Hasson, a doctoral student with Matthieu Lapotre, assistant professor of Earth and Planetary Sciences at Stanford’s Doerr School of Sustainability.
Hasson is lead author of a new study demonstrating the tool, published in the Proceedings of the National Academy of Sciences.
Telltale signs
Historically, microstructural analysis has been done by hand and eye, using magnifying glasses and microscopes to try to make inferences about the history of sand grains.
Modern science has validated this approach, revealing that transport mechanisms do indeed leave characteristic signatures: particles that have traveled farther often appear rounder, for example, as sharp edges are blunted, and waves and wind also leave characteristic wear patterns.
However, traditional microtextural analysis is highly subjective, time-consuming, and fragmented across different studies. Thanks to new tools that harness the power of machine learning to scrutinize microscopic images of sand grains in detail, microtextural analysis can become much more quantitative and objective, potentially useful for a wide range of applications. It also allows for a more complete assessment, as each individual grain of sand is analyzed, rather than lumping multiple grains together into one category.
“Instead of a human looking at the texture of a grain of sand and judging one texture from another, we use machine learning to make micro-texture analysis more objective and rigorous,” said Lapotre, lead author of the paper. “Our tool opens the door to micro-texture analysis applications that were previously unavailable.”
Sand is the world’s most used resource after water and is essential in the construction industry. Materials like concrete, mortar, and some plasters require fine-grained sand for proper adhesion and stability. But measuring sand’s origins and ensuring it’s ethically and legally sourced can be difficult, so researchers hope SandAI can enhance traceability. For example, SandAI could help forensic investigators crack down on illegal sand mining and dredging.
Tool training
To build SandAI, the researchers employed a neural network that “learns” in a way that resembles the human brain, in which correct answers strengthen connections between artificial neurons, or nodes, in the program, allowing the computer to learn from its mistakes.


The SandAI neural network is trained on modern quartz sand to help uncover history engraved in ancient rocks. Here, ancient ripples formed by flowing water in Oman are being remodeled by modern wind-blown sediments. Credit: Mathieu Lapôtre/Stanford University
With the help of collaborators from around the world, Hasson collected hundreds of scanning electron microscope images of sand grains representing materials from the most common terrestrial environments: fluvial (rivers and streams), aeolian (wind-transported deposits such as sand dunes), glaciers, and coasts.
“We wanted the method to work not only across geological time, but across all landforms on Earth,” Hasson says. “For example, the wind-blown dune classes were designed to include wet and dry, large and small. The classes needed to be as diverse as possible.”
SandAI analyzed this set of images and trained itself to predict the history of grains of sand based on features that a human researcher could never discern. The tool naturally made errors and then iteratively improved. Once SandAI reached a stable prediction accuracy of 90%, the researchers introduced new samples that the model had never seen before.
SandAI performed well, using images of sandstone from well-characterized environments dating back to the Jurassic Period, about 200 million years before present, to accurately resolve grain transport history.
New Science and Applications
The researchers then tested the tool on images of sand grains from Norway that date back more than 600 million years to the Cryogenian period. This period is better known as “Snowball Earth,” a time when ice sheets are thought to have covered the entire planet before plants and animals evolved. The origins of the sample in question, the Bravica Formation, are a matter of debate, with different research groups coming to differing conclusions.
“With this cryogenian sample, we wanted to see how far we could take Sun AI, not just to validate that the tools work, but to actually use them to power new scientific research,” Hasson said.
Interestingly, SandAI inferred that the ancient sand grains formed and were deposited as part of wind-blown dunes, which is consistent with the manual microstructural study. Additionally, because the tool analyzes individual grains rather than lumping multiple grains together into one category, other details became apparent.
While the primary features do indicate wind transport, secondary features that would be missed by hand indicate glacial sand. Taken together, these signals suggest a range of sand dunes somewhere near a glacier, just as would be expected during the Snowball Earth period.
To further evaluate these findings, Hasson and his colleagues looked for possible modern-day counterparts to this Cryogenian geological condition. They ran SandAI through grains of sand blown by the wind from Antarctica, and sure enough, they got the same results.
“These SandAI findings suggest that Antarctica is indeed a modern version of the environment represented by the Bravica Formation,” Hasson said. “This is very strong evidence that the signal we get from the cryogenian deposits is not just a coincidence.”
The researchers have made SandAI available to anyone online. They plan to continue developing it based on user feedback and look forward to seeing the tool applied in a variety of situations.
“It’s amazing to me that we can now draw detailed conclusions about geological deposits that we never knew existed,” Hasson said. “I’m excited to see what else SandAI can do.”
More information: Michael Hasson et al., “Automated determination of transport and depositional environments of sand and sandstone,” Proceedings of the National Academy of Sciences of the United States of America (2024). DOI: 10.1073/pnas.2407655121
Courtesy of Stanford University
Citation: AI tool that scans sand grains opens window into recent and distant past (September 16, 2024) Retrieved September 17, 2024 from https://phys.org/news/2024-09-ai-tool-scanning-sand-grains.html
This document is subject to copyright. It may not be reproduced without written permission, except for fair dealing for the purposes of personal study or research. The content is provided for informational purposes only.