Archaeologists are using AI and US spy satellite imagery from the cold war to find ancient underground aqueducts that helped humans survive in the desert.
Most of the ancient underground aqueducts that enabled humans to settle in the world’s hottest and driest regions have been lost over time. Now, archaeologists are rediscovering them by using artificial intelligence to analyse spy satellite images taken during the cold war.
 |
| Holes at the top of this image are vertical shafts to underground aqueducts called qanats Nazarij Buławka et al |
The oldest known underground aqueducts that are found across much of North Africa and the Middle East are called qanats and are up to 3000 years old. They were designed to carry water from highland or mountain areas to irrigate crops in desert regions, and were comprised of dozens of kilometres of channels and vertical shafts used for maintenance and airflow.
“These systems were extremely innovative,” says Hector Orengo at the Catalan Institute of Classical Archaeology in Spain. “They allowed people to live in areas where it would have been unthinkable before.”
Orengo and his colleagues trained an AI model to look through black-and-white satellite images for lines of holes that represent the openings of vertical access shafts leading down to the underground aqueducts. They trained the AI on satellite imagery taken by US cold war spy satellites that operated between 1959 and 1986, and created additional artificial images of the underground aqueducts to provide more training examples that could boost the AI’s performance.
They then tested the AI on satellite images from places known to have qanats: Afghanistan’s Maiwand region, Iran’s Gorgan plain and the town of Rissani in eastern Morocco. More than 88 per cent of the AI’s predictions of qanat locations proved correct, and it also correctly identified more than 62 per cent of all the underground aqueducts in the area.
It is unclear whether this approach can accurately find subterranean aqueducts in more complex landscapes beyond desert areas, says Mehrnoush Soroush at the University of Chicago. But she praised the research for demonstrating the AI technique on a broad scale and emphasised that it need not be 100 per cent accurate.
“As long as we can look at a very large region and identify areas that we want to look at more closely, we have gained a lot of success,” says Soroush.
Journal reference
Journal of Archaeological Science DOI: 10.1016/j.jas.2024.106053