A research team led by Amy Hatton has demonstrated that artificial intelligence can be used to identify large numbers of ancient stone structures in northern Saudi Arabia. Focusing on the southern edge of the Nefud Desert, the study explores a more efficient approach to mapping sites that are thousands of years old.
Across Arabia, people built many dry-stone features during the Holocene period. These include burial cairns, enclosures, camps and larger ritual structures. They are key to understanding how communities lived and moved through changing environments. Until now, archaeologists have mainly recorded them through field surveys or by manually tracing them on satellite images—both slow and labour-intensive approaches.
The new research, published in the Journal of Archaeological Science: Reports, explores whether computer models can take on part of this work. The team trained three deep learning systems—MA-Net, SegFormer and U-Net—using satellite imagery from the region. Each model was tested on how well it could recognise and outline stone features.
Results show clear differences between the systems. MA-Net produced the strongest overall scores, but SegFormer gave more consistent results across different structure types. The models performed better when they were fed higher-resolution images, with accuracy improving by about 20 percent.
The study covered around 2,500 square kilometres along the desert’s edge. Today, the area is dry and dominated by sand dunes, but conditions were wetter in the past. This made it more suitable for settlement, which helps explain the high number of archaeological remains.
Using automated detection could change how archaeologists approach large landscapes. Instead of mapping everything by hand, researchers can use AI to highlight likely features and then check them more closely. This saves time and allows work to be carried out over much larger areas.
There are also practical benefits for heritage protection. Many of these sites lie in remote regions and are difficult to monitor. Faster mapping makes it easier to record what exists and track changes over time. This is important in places where development or natural erosion may threaten archaeological remains.
The researchers make clear that this approach is not a substitute for fieldwork. On-the-ground survey and excavation remain necessary to verify results and collect detailed evidence. What the models offer is a way to narrow the search—pointing archaeologists toward areas most likely to contain structures and cutting down the time spent scanning imagery by hand.
The work sits within a broader shift toward using machine learning in archaeology. Although these techniques are already in use elsewhere, their application in Arabia has been limited. The results here show they can handle desert landscapes and large image datasets without major issues.
As higher-resolution satellite data becomes more widely available, the same approach could be extended across much larger areas. This would make it easier to trace how people organised and used these landscapes, and how those patterns shifted over long periods.
Header Image Credit : Shutterstock
Sources : Journal of Archaeological Science: Reports – https://doi.org/10.1016/j.jasrep.2026.105734
