Lidar Maps Hidden Ancient Cities
Aircraft laser scanning sees the ground under forest; surveys in Mexico (2024) and Brazil (2026) show ancient settlement far denser than ground work found.
Open in the interactive tree →A plane fires laser pulses, and the few that reach the ground build a 3D terrain model with the trees removed, so mounds, walls and roads show up. It exposed the Khmer city of Mahendraparvata in Cambodia (announced 2013) and, in 2018, more than 60,000 structures in the Maya Biosphere Reserve in Guatemala. Archaeologists now also reuse lidar flown for other purposes: the Maya city Valeriana in Mexico was found in a 2013 forest-monitoring survey.
As of October 2026
On 29 July 2026 Nature published a lidar survey of about 1,740 square miles (roughly 4,500 km2) in Brazil that found nearly 400 previously undocumented earthworks of the Aquiry culture (600 BC to AD 850), on top of about 1,700 known; the authors project 23,000 to 30,000 in the region and 1.25 to 3 million people, a projection that assumes about 300 people per earthwork. The area is under 3% of Amazonia, and outside experts ask for more study of occupation sites. Valeriana (Antiquity, 29 October 2024) holds more than 6,500 structures on about 120 km2. Machine learning is starting to map such data: a March 2024 preprint reports an IoU of 0.84 for Maya platforms.
Open steps
- Automatic structure detection High AI leverageTrain detectors that find platforms, mounds and earthworks in lidar point clouds so surveys of thousands of km2 need no hand tracing.
- Check and date on the ground Low AI leverageAnomalies seen from the air must be visited and dated by excavation or radiocarbon; counts of structures do not show which were in use at the same time.
- From structures to people Medium AI leverageEstimate populations with stated uncertainty; the Aquiry estimate of 1.25 to 3 million rests on about 300 people per earthwork plus radiocarbon dates.
- Survey the unsurveyed 97 percent Low AI leverageThe Amazon survey covers under 3% of Amazonia; more flights, permits and open sharing of lidar flown for other purposes decide how fast the map fills.
Where AI could help
Medium AI leverage. Neural networks can find structures in huge terrain models, but flights, ground checks and dating by excavation still set the pace.
- Detect platforms, mounds and earthworks automatically in terabytes of point clouds
- Screen existing forest-monitoring lidar for unknown sites
- Rank anomalies so field crews visit the likeliest first
Shown so far
- A March 2024 arXiv preprint applied YOLOv8 to Maya lidar images and reports an IoU of 0.842 for platforms and 0.809 for annular structures. source
Prerequisites
Sources
More in Humanities & Culture · Present
- AI Reads Archives & Inscriptions2022-2026
- AI for Low-Resource Languages2022-2026
- AI Reads Herculaneum Scrolls2023
- Whole Herculaneum Scroll Read2025-2026