AI flagged 1,309 possible Nazca figures. Six months of walking the desert confirmed 303 of them

A survey of the Nazca plateau in southern Peru confirmed 303 figurative geoglyphs that were not previously catalogued. The number people tend to repeat is that one. The number that explains how the work was done is 1,309.

That is how many candidate locations a machine-learning model flagged from aerial imagery. Yamagata University's Institute of Nasca, working with IBM Research, trained the system to recognise the faint surface contrasts a geoglyph leaves and ran it across the plateau. What came back was a list of places worth walking to.

Then people walked to them. Roughly a quarter of the candidates were investigated during a six-month field survey, and from those investigations 303 figures were confirmed and documented. The remainder — something close to a thousand locations — have not been checked.

The order matters, and it is routinely reported backwards. The AI did not discover 303 geoglyphs. It proposed 1,309 places to look, and fieldwork converted a fraction of that list into confirmed finds. The confirmed figures are the output of the survey, not of the model.

Against a previously known total of 430 figurative geoglyphs, 303 additions is a substantial expansion of the catalogue, though not a doubling. It is also, on the evidence of the hit rate so far, an incomplete one: the unchecked candidates are not a rounding error.

The confirmed figures divide into two kinds, and the distinction carries the study's interpretive weight. Line-type geoglyphs are the larger ones, arranged along the networks of straight lines and trapezoids that the plateau is known for, and they mostly depict wild animals. The proposal is that these were made for community-level ceremony.

Relief-type geoglyphs are smaller, sit close to walking trails, and show different subjects: human figures, domesticated animals, decapitated heads — imagery of people and of things people had altered. The proposal there is quite different. These appear positioned to be seen by an individual or a small group passing on foot, functioning less like monuments and more like markers addressed to a traveller.

That is a more specific claim than the familiar one about the lines being meant for the sky, and it is more testable. It says the two categories had different audiences, different scales of use, and different placement logic — and it makes a prediction about where further relief-type figures should turn up, which the unchecked candidates could confirm or undercut.

The research was published in the Proceedings of the National Academy of Sciences and announced by the university on 24 September 2024. Open News could not retrieve the paper itself, which the publisher blocks to automated requests, so this account draws on the university's own announcement of its findings rather than on the paper text.