AI agents identify Maya site in Mexico using public NASA lidar data
An author used multiple AI agents to analyse lidar strips and locate a Maya site not yet in the official Mexican register, though archaeologists had previously mapped it.
Key points
- The agents processed 1,200 square kilometres of jungle lidar data to find structures.
- The site was already mapped in a 2025 paper but absent from the official INAH register.
- The process involved 48 agents filtering images without a specialised classifier model.
An update published on 11 November clarifies that the Maya site identified by AI agents had already been mapped in a paper by Estrada-Belli et al. in 2025.reported The agents missed this existing literature during their review, meaning they rediscovered a known site rather than making a new discovery.AI tools helped find an ancient Maya site, but archaeologists had already found it.Multimodal agents processed NASA lidar data to identify a Maya site, revealing gaps in literature retrieval capabilities. The location was not yet listed in the register of Mexico's National Institute of Anthropology and History, INAH.
How we got here
The author was travelling in Peru and discussing history with Gemini when learning about Hiram Bingham, who found Machu Picchu in 1911 while searching for Vilcabamba.reported This prompted a thought experiment about what Bingham could achieve with modern tools. The author then brainstormed with Gemini, ChatGPT, Claude and Grok. All four agents suggested using lidar technology, which fires laser pulses through canopy gaps to map the ground beneath, unlike standard satellite photos.
What is new
The author abandoned plans for Peru due to poor lidar coverage and high survey costs, instead turning to the Maya region, which has better public lidar data. After asking the agents to analyse the data, the author woke to reports of villages and regular shapes resembling structures.reported The agents converted lidar strips covering approximately 1,200 square kilometres of jungle into images for review.reported
What to watch
The workflow involved 48 agents examining images sequentially, acting like patient archaeologists.reported They looked for pyramids, platforms and plazas while filtering out natural features such as hills, sinkholes and roads. The author noted that no special model or classifier was used; the agents literally looked at the pictures following a precise guide. The entire process, from initial idea to an official reply from the Mexican government, took less than 72 hours.
Availability and limits
For developers, this demonstrates that current multimodal agents can process large-scale geospatial data without custom-trained classifiers. The capability relies on public NASA lidar data and standard agent orchestration. However, the output requires verification against existing academic literature, as the agents failed to identify the prior 2025 mapping.reported The utility is limited by the completeness of public datasets and the agents' ability to cross-reference scholarly databases.
Why it matters
- It shows agents can perform visual inspection of geospatial imagery without bespoke computer vision models.
- The workflow highlights a gap in agent capabilities regarding literature review and prior art detection.
- Public lidar data remains a viable resource for automated archaeological or geographical analysis.
- The speed of the workflow suggests potential for rapid hypothesis testing in field research.
Topics: agents, lidar, archaeology, multimodal
Updated
Then / now / next
Satellite photos were insufficient for identifying ruins hidden under jungle canopy.
AI agents can process lidar data to identify structures, though they may miss existing academic records.
Developers should monitor improvements in agent literature review capabilities to prevent redundant discoveries.
Story timeline
- Then Satellite photos were insufficient for identifying ruins hidden under jungle canopy.
- 11 Nov 2026 My AI agents found an unmapped Maya site
- Now AI agents can process lidar data to identify structures, though they may miss existing academic records.
- Next Developers should monitor improvements in agent literature review capabilities to prevent redundant discoveries.
The strongest doubt
Counterpoint from the sources
The agents failed to identify that the site was already mapped in 2025, indicating significant limitations in their ability to conduct comprehensive literature reviews.
Try it
Analyze lidar data for structures
Good result: Agents can identify pyramids and plazas in lidar-derived images without custom models.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
lidar
A remote sensing method that uses laser pulses to map ground features through canopy gaps.
INAH
Mexico's National Institute of Anthropology and History, which maintains the official register of archaeological sites.
agents
AI systems that can perform multi-step tasks, such as reviewing images and filtering data.
Questions readers ask
Did the AI agents discover a new Maya site?
No, the site had been mapped in 2025 but was not yet in the official Mexican government register.
How many agents were used in the analysis?
The author used 48 agents to review the images generated from the lidar data.
What data source was used for the analysis?
The analysis used public NASA lidar data covering approximately 1,200 square kilometres of jungle.
Connected stories
Reading receipt
- The agents processed 1,200 square kilometres of jungle lidar data to find structures.
- The site was already mapped in a 2025 paper but absent from the official INAH register.
- The process involved 48 agents filtering images without a specialised classifier model.
Still open: How can agents be improved to better cross-reference existing academic literature?
Still open: What other public datasets could benefit from this multi-agent visual inspection workflow?