AI tool for scanning grains of sand opens windows into recent times and the distant past

Stanford researchers have developed an artificial intelligence-based tool called SandAI that can reveal the history of quartz sand grains over hundreds of millions of years. Using SandAI, researchers can determine with great precision whether wind, rivers, waves, or glacial movements formed and deposited sand grains.

The tool offers researchers a unique window into the past for geological and archaeological studies, particularly for eras and environments where few other clues, such as fossils, have been preserved over time. SandAI’s approach, called microtextural analysis, can also aid in modern forensic investigations into illegal sand mining and related issues.

“Working on sedimentary deposits that haven’t been disturbed or deformed is like being in a time machine: you see exactly what was on the Earth’s surface, even hundreds of millions of years ago. SandAI adds an extra layer of detail to the information we can extract from it,” said Michael Hasson, a doctoral student with Mathieu Lapôtre, assistant professor of Earth and planetary sciences at the Stanford Doerr School of Sustainability.

Hasson is the lead author of a new study demonstrating the tool, published in Proceedings of the National Academy of Sciences.

Telltale Signatures

Historically, microtextural analysis has been done by hand and by eye, using magnifying glasses and microscopes to try to draw conclusions about the history of sand grains.

Modern science has validated this approach, showing that transport mechanisms do indeed convey telltale signatures: for example, grains that have traveled farther often appear rounder because their sharp corners have been blunted; waves and wind also leave distinctive abrasion patterns.

However, traditional microtextural analysis is highly subjective, time-consuming, and scattered across different studies. With the new tool, which harnesses the power of machine learning to examine microscopic images of sand grains in depth, microtextural analysis can now be much more quantitative, objective, and potentially useful in a wide range of applications. It also analyzes sand grains individually instead of grouping multiple grains into a single category, providing a more comprehensive assessment.

“Instead of relying on a human to determine the texture of sand grains, we use machine learning to make microtexture analysis more objective and rigorous,” said Lapôtre, lead author of the study. “Our tool opens doors to applications of microtexture analysis that were not previously available.”

Sand is the most used resource in the world, after water, and is essential in the construction industry. Materials such as concrete, mortar and some plasters require angular sand for proper adhesion and stability. However, it is difficult to assess the origin of sand to ensure ethical and legal sourcing. Researchers therefore hope that SandAI can strengthen traceability. For example, SandAI could help forensic investigators combat illegal sand mining and dredging.

Train the tool

To create SandAI, the researchers used a neural network that “learns” in a way similar to the human brain, where correct responses strengthen connections between the program’s artificial neurons, or nodes, allowing the computer to learn from its mistakes.

With the help of collaborators around the world, Hasson has collected hundreds of scanning electron microscope images of sand grains, representing materials from the most common terrestrial environments: fluvial (rivers and streams), aeolian (wind-blown sediments, such as sand dunes), glacial and beach.

“We wanted this method to work across geological time, but also across the entire geography of the Earth,” Hasson said. “So, for example, the windblown dune class was designed to include wet and dry examples, large and small. We needed the classes to be as diverse as possible.”

SandAI analyzed this set of images to train itself to predict the history of sand grains based on features that human researchers could never discern. The tool naturally made mistakes and then iteratively improved. Once SandAI reached a robust prediction accuracy of 90%, the researchers introduced new samples that the model had not seen before.

With images of sandstone from well-characterized environments ranging from the present day to about 200 million years ago in the Jurassic era, SandAI performed well, correctly elucidating the grain transport history.

New sciences and applications

The researchers then tested the tool with images of grains of sand collected in Norway that were more than 600 million years old, during the Cryogenic Era. Better known as the “Snowball Earth,” this period corresponds to a time when ice sheets would have covered the entire planet, before plants and animals appeared. The origin of the sample in question, called the Bråvika Member, has been disputed, with several research groups reaching different conclusions.

“With this cryogenic sample, we were able to see how far we could push SandAI and actually use it to do new research rather than just verifying that the tool worked,” Hasson said.

Interestingly, SandAI assumed that the ancient sand grains were shaped and deposited as part of a windblown sand dune, which is consistent with some manual microtexture studies. Additionally, because the tool analyzes sand grains individually, rather than lumping multiple grains into a single category, other details emerged.

While the dominant signature did indeed indicate wind transport, a secondary signature that manual techniques would likely have missed indicated the presence of glacial sand. Together, these signals paint a picture of sand dunes located somewhere near a glacier, as would be expected during the Snowball Earth period.

To further evaluate these results, Hasson and his colleagues looked for a potential modern analogue of this cryogenic geological scene. The researchers ran windblown sand grains from Antarctica through SandAI and, as expected, arrived at the same result.

“These SandAI results suggest that Antarctica is indeed a good modern analogue of the environment represented by the Bråvika Member,” Hasson said. “They are very strong evidence that the signal we received from the cryogenic deposits is not just a fluke.”

The researchers have made SandAI available online for everyone to use. They plan to continue developing it based on user feedback and are eager to see the tool applied in a variety of settings.

“I find it simply mind-blowing that we can now offer detailed conclusions about geological deposits that were not previously known,” Hasson said. “We are excited to see what else SandAI can do.”