Editor's Note: This CNN series is, or was, sponsored by the country featured. CNN retains full editorial control over the subject matter, coverage and frequency of the articles and videos within the sponsorship, in accordance with our policies.
On the northern side of the Rub al-Khali, secrets lie buried in the sand.
The vast 650,000-square-kilometer desert on the Arabian Peninsula is known as "The Empty Quarter." And to most, aside from the waves of ochre-colored dunes, it looks empty.
But not for artificial intelligence.
Researchers from Khalifa University in Abu Dhabi have developed a high-tech solution to search large, dry areas for potential archaeological sites.
Traditionally, archaeologists have used ground surveys to detect potential sites of interest, but this can be time-consuming and difficult in rugged terrain like the desert. In recent years, remote sensing using optical satellite imagery from places like Google Earth has become more popular for surveying large areas for unusual features. But in the desert, sand and dust storms often obscure the ground in these images, while dune patterns can make it difficult to detect potential sites.
"We needed something that would guide us and focus our research," said Diana Francis, an atmospheric scientist and one of the project's principal investigators.
The team developed a machine learning algorithm to analyze images collected by synthetic aperture radar (SAR), a satellite imaging technique that uses radio waves to detect objects hidden beneath surfaces, such as vegetation, sand, soil and ice.
Neither technology is new: SAR imagery has been around since the 1980s, and machine learning is gaining traction in archaeology. But using the two together is a new application, Francis says, and to her knowledge, a first in archaeology.
She trained the algorithm using data from a site already known to archaeologists: Saruq Al-Hadid, a settlement with evidence of 5,000 years of activity that is still being uncovered in the desert outside Dubai.
"Once it was trained, it gave us an indication of other potential areas (nearby) that haven't been excavated yet," Francis says.
She adds that the technology is accurate to within 50 centimetres and can create 3D models of the expected structure, giving archaeologists a better idea of what lies beneath the ground.
Working with Dubai Culture, the government organization that manages the site, Francis and her team conducted a ground survey using ground-penetrating radar, which "mimicking what the satellite measured from space," she says.
Now, Dubai Culture plans to excavate the newly identified areas. Francis hopes the technique will reveal more buried archaeological treasures in the future.
Speeding up 'boring' work
The use of SAR images is not common in archaeology due to its cost and complexity.
But using it to identify buried sites is "really exciting," says Amy Hatton, a PhD student at the Max Planck Institute for Geoanthropology who is researching deep learning models to detect archaeological structures in northwestern Saudi Arabia.
Hatton notes that by using SAR images, which circumvent the problem of light scattering by dust particles, Francis and her team have solved technical details that make remote sensing in desert areas difficult.
Khalifa University is not alone in using artificial intelligence to detect potential locations.
Amina Jambajanstsan, another PhD candidate at the Max Planck Institute, is using machine learning to speed up the "tedious task" of searching through high-resolution drone and satellite imagery for potential sites of interest. Her project, which focuses on medieval burial sites in Mongolia - a country spanning more than 1.56 million square kilometers, nearly the size of Alaska - has uncovered thousands of potential sites that Jambajanstsan and her team say they could never have found on the ground.
Jambajanstsan says that while the cost and computational demands of SAR imagery may be a barrier to use for many researchers, the method is valuable for desert areas where other technologies struggle - and it's one she would consider for the Gobi Desert in southern Mongolia, where her "normal optical images" aren't producing results.
Man versus machine
Machine learning is increasingly being used in archaeology, although not all researchers are enthusiastic about it.
"There are two different belief systems," says Hugh Thomas, a lecturer in archaeology at the University of Sydney and co-director of the AlUla and Khaybar Prehistoric Excavation Project in Saudi Arabia. On the one hand, there are those who are looking for technological solutions such as AI to identify sites; on the other hand, there are those who believe you need a "trained archaeological eye" to identify structures, he explains.
While technology can help identify and monitor archaeological sites, especially those threatened by land-use change, climate change and looting, Thomas is wary of relying on it too much.
"I would like to use this kind of technology in areas where there may be no or very low probability of archaeological sites, so that researchers can focus more on other areas where we expect to find more archaeological sites," Thomas said.
Uncovering the past
The real test - and hopefully validation - of the technology will come next month, when excavations begin at the Saruq Al Hadid complex, of which an estimated 10% has been uncovered over an area of 6.2 square kilometers, according to Dubai Culture.
If archaeologists find the structures the algorithm predicted, Dubai Culture plans to use the technology to excavate more sites.
Francis and her team published a paper on their findings last year, and they continue to train the machine learning algorithm to improve its accuracy before deploying it more broadly.
"The idea is to export (the technology) to other areas, especially Saudi Arabia, Egypt and maybe the deserts in Africa," she says.
For more CNN news and newsletters, create an account at CNN.com
