Researchers are using convolutional neural networks to search satellite images to map ancient burial sites in parts of Russia, Mongolia, and China
Zach Zorich / New York Times :
Context & Ripple Effects
This story closes a loop that opened in 2016, when [[a:870137|Terrapattern launched as the first open-access neural-net reverse image search for satellite imagery]]. The intervening years built the toolkit: cheap, noninvasive survey methods like drones, lidar, and ground-penetrating radar reshaped fieldwork, as covered in earlier reporting on how these technologies are changing archaeology.
Applying convolutional neural networks to satellite images of Russia, Mongolia, and China extends a pattern already proven elsewhere — researchers recently used AI paired with drones to find 303 uncharted Nazca Lines in Peru, nearly doubling the mapped total, per that project's coverage. It matters because burial mounds across the Eurasian steppe are vast in number, spread over enormous terrain, and expensive to find by foot.
First-order effects
- Archaeologists working in Russia, Mongolia, and China gain a way to locate and map burial sites across huge steppe landscapes without surveying them on foot or excavating.
- The technique validates satellite imagery plus convolutional neural networks as a working archaeological instrument, not just a research demo.
Second-order effects
- Site-discovery work shifts toward desk-based analysis over large imagery archives first, with drones and ground-truthing deployed only where the models flag candidates — the same division of labor behind the Nazca Lines results.
- Other regional surveys face pressure to adopt model-assisted discovery: once one team maps a landscape this way, manual-only surveys look slow and costly by comparison.
Third-order effects
- If the pattern holds alongside the global vessel-traffic map built from 2,000TB of maritime imagery, neural-network analysis of satellite archives becomes a default method for inventorying anything visible from orbit — sites, ships, infrastructure — collapsing the cost of large-area survey across disciplines.
- Archaeology's center of gravity moves further from excavation toward remote detection, raising questions about who controls the imagery archives and models that determine which heritage gets found and documented.
The trend: Satellite imagery plus neural networks is becoming a general-purpose survey instrument, moving archaeology from walking the terrain to screening archives before anyone sets foot in the field.