The recent survey doubles the count of known geoglyphs, bringing the total count to over 700, while showcasing how AI can be used in cultural heritage contexts.
It took a century for researchers to discover 430 geoglyphs, some massive, spread across the ground in the Nazca region of Peru. Now, a team of scientists has nearly doubled the count of known geoglyphs in just six months, using an AI-driven approach to reveal some of the harder-to-spot figures.
The Nazca Lines refer to a collection of these geoglyphs in the Nazca Pampa; between roughly 200 BCE and 500 CE, local groups removed the darker rocks on the arid landscapes surface, revealing the white sandy soil below. They did this on smaller scales but also on glyphs so large that they can only be properly seen from the air.
The lines are ancient but were rediscovered en masse in the 20th century; nevertheless, some went under the radar. In 2020, a feline geoglyph -- as in, it depicts a cat -- was found on a hillside in the region. Last year, two members of the recent team identified four new geoglyphs using an object detection algorithm, but they didn't pursue a more comprehensive survey. Now, the larger team has done just that and been rewarded for their AI-assisted efforts.
The recent model "focuses on relief-type geoglyphs that are small and difficult to identify since the distribution of the large line-type figurative geoglyphs is known from previous (manual) aerial studies," as the team wrote in the resulting paper, published earlier this week in the Proceedings of the National Academy of Sciences (PNAS).
While convolutional neural networks are generally trained on tens of thousands of images, the recent team's model only trained on images of the few hundred known geoglyphs. IBM's geospatial platform PAIRS first cleaned up the data before feeding it into the model; then, the model split up the Nazca Pampa into a grid and highlighted sections that appeared to contain a glyph. The ID'd candidate glyphs were then fed to human archaeologists who separated the wheat from the chaff.
Even with not much data to train its models, the team found the AI approach effectively identified geoglyphs. It proffered 1,309 potential glyphs, and over the next two months' worth of labor, the team identified 303 figurative geoglyphs and 42 newly identified geometric glyphs.
The team also identified diverging patterns in what the larger and smaller glyphs depicted. The monumental glyphs generally depicted animals and plants, while the the relief-type geoglyphs depicted humans, decapitated heads, and domesticated camelids. The smaller geoglyphs are located, on average, roughly 141 feet (43 meters) from ancient trails that crisscross the Nazca Pampa region, which were likely used to view the glyphs.
Of the newly discovered smaller geoglyphs, the 47 that depicted wildlife included figures of birds, cats, snakes, monkeys, foxes, killer whales, and fish.
Large geospatial data technologies and data mining have hastened the pace of discovery in aerial archaeology, the team wrote in its paper. "Thus, AI may be at the brink of ushering in a revolution in archaeological discoveries like the revolution aerial imaging has had on the field."