• Researchers in Spain looking to develop a fresh approach to finding long-lost aqueducts have turned to U.S. Cold War-era satellite imagery.
• By mining the black-and-white images from the HEXAGON satellite program, the team developed machine learning models that have proven adept at picking out ancient aqueducts in varied terrain.
• The use of underground aqueducts is common throughout the world, most significantly in the Middle East, but also across the globe from Chile to China.

---

When you’re on the hunt for 3,000-year-old underground aqueducts, the best new tool to find them has actually been around for quite some time: a U.S. Cold War-era satellite.

The unlikely repurposing of the satellites is thanks to a research team in Spain, as the group has been using the high-resolution black-and-white imagery of the HEXAGON series of satellites, also known as Big Bird, from the United States to hone in on ancient underground aqueducts, known throughout the Middle East as qanats. For further aid, the group created a machine learning model for improved accuracy.

A line of regularly spaced ventilation shafts make these qanat systems visible from above, according to a paper published published in the Journal of Archaeological Science, making satellite imagery valuable in detecting their locations. By feeding a wealth of these images, coupled with artificial intelligence-generated images for even greater variety, into a machine learning model, the team created a global model for detecting qanats, no matter the environmental type, according to the paper.

“Our approach allows us to create a model that can be used in different areas,” they wrote, adding that “our model can be used for mapping large areas and detailed case studies.”

To back up that claim, the team used the HEXAGON imagery from areas in Afghanistan, Iran, and Morocco, all regions where the ancient qanats systems have largely been mapped, to build their model. They wrote that their approach provided improved precision in predicting the location of qanat shafts, the first case study for successfully detecting qanats in different landscapes using different types of satellite imagery.

“Using real, augmented, and artificial data allowed us to generalize the representation of qanats into lineal groups of circular features,” the authors wrote, adding that their system eliminated most of the isolated and clustered false positives that surface in existing systems that then require additional research, resulting in an 88-percent success rate in identifying the underground systems.

The HEXAGON KH-9 reconnaissance satellites were the largest and last U.S. intelligence satellites to return photographic film to Earth, according to the National Museum of the U.S. Air Force. Used from 1971 through 1986, the 18 HEXAGON missions imaged 877 million square miles of the Earth’s surface.

The Lockheed Corp. built the HEXAGON satellites, which feature two separate film cameras that worked together to produce stereo images. The “optical bar cameras” located on the bottom of the satellite spun on their axes, taking overlapping images to form a panoramic picture. Objects smaller than two feet across could be imaged from up to 100 miles of altitude.

The search for qanats is part of a hunt for historical significance, with some qanats systems placed in the UNESCO World Heritage directory. A type of ancient hydraulic structure designed for water distribution in arid environments, the aqueducts feature subterranean channels transporting water from highland or mountain areas. While some are still in use today, the oldest known systems are roughly 3,000 years old.

“These systems were extremely innovative,” said Hector Orengo, one of the study’s authors from the Catalan Institute of Classical Archaeology in Spain, according to New Scientist. “They allowed people to live in areas where it would have been unthinkable before.”

The team chose Afghanistan’s Maiwand region, Iran’s Gorgan Plain, and an area west of the town of Rissani in eastern Morocco to show off how the new “object detection model for mapping qanat systems” can function in different environment types.

The model waded through the satellite imagery and AI-produced images to learn the lines of holes that represented the vertical access shaft spacing, with some of these systems running miles in length. The varied looks provided by the two sources improved the model’s training.

“As long as we can look at a very large region and identify areas that we want to look at more closely,” said Mehrnoush Soroush of the University of Chicago, who was not involved in the study, according to New Scientist, “we have gained a lot of success.”