AI-assisted survey discovers hundreds of new Nazca geoglyphs
Using AI-supported screening, researchers were able to discover 303 previously unknown geoglyphs in the Nazca region. The scientists hope to find more.
Twelve of the 303 newly discovered, relief-like geoglyphs from the study. The researchers confirmed the existence of the geoglyphs with drone images from the field research and added contours to the images for clarification.
(Image: PNAS)
Researchers have discovered 303 previously unknown geoglyphs in the Nazca region of Peru with the help of artificial intelligence. This is the result of a new study published in the Proceedings of the National Academy of Sciences (PNAS). The Nazca Lines are a large number of huge line and relief drawings in the Nazca Desert in Peru. They were created by the Nazca culture between 500 BC and 500 AD and have repeatedly given rise to speculation – among others, the pseudo-scientist Erich von Däniken interpreted them as landing strips and welcome symbols for extraterrestrials. Even if von Däniken's theories can be ruled out, the exact function of the Nazca geoglyphs has not yet been conclusively clarified. Experts assume that the glyphs were created for ritual and religious purposes.
The international team of researchers led by Masato Sakai from Yamagata University in Japan has now used a neural network to study the entire Nazca UNESCO World Heritage Site. In just six months of field research, they were able to identify 303 new geoglyphs, almost doubling the previous number. It took almost a century to discover the first 430 geoglyphs in Nazca, the authors explain in their study. With the new finds, there are now 733.
(Image: PNAS)
Few training data as a challenge
The greatest challenge was the recognition of rare geoglyphs with low contrast and few training examples. While conventional neural networks are trained with thousands of images per class, only a few hundred examples were available for the Nazca geoglyphs. To overcome this hurdle, the researchers say they used a model pre-trained on nature photographs that was only fine-tuned to the geoglyphs. This reduced the need for training data. They also developed a classification model that creates a continuous probability map for geoglyphs instead of rectangular bounding boxes, thus breaking down each training example into smaller, reusable fragments.
(Image: PNAS)
New insights into purpose and meaning
The improved inventory of the geoglyphs enabled the researchers to analyze motifs and distribution across the Nazca Plain. This revealed a clear difference between the large linear geoglyphs and the smaller relief-like geoglyphs: Linear geoglyphs mainly depict wild animals (64 percent). Relief geoglyphs mainly depict human motifs or things altered by humans (81.6 percent), such as people, farm animals and severed heads.
According to the study, the results suggest that the two types of geoglyphs served different purposes. The linear geoglyphs were likely used at the community level for ritual activities, while the relief geoglyphs were more likely built and viewed at the individual or small group level. Based on their analysis, the researchers estimate that at least 248 more figurative geoglyphs suggested by the neural network could be discovered in future fieldwork.
(vza)
