5 May 2025 The Future is the Product of the Past

Scientists discover 4 new Nazca Geoglyphs using AI deep learning

Scientists from Japan used AI deep learning to discover new geoglyphs in the Arid Peruvian coastal plain, in the northern part of Peru’s Nazca Pampa.

The research has been ongoing since 2004 by a team from Yamagata University, led by Professor Makato Sakai. Yamagata University has been conducting geoglyph distribution surveys using satellite imagery, aerial photography, airborne scanning LiDAR, and drone photography to investigate the vast area of the Nazca Pampa covering more than 390 km2.

The Nazca Lines are thought to have been made over centuries, starting around 100 BC by the Nazca people of modern-day Peru. They were first studied in detail in the 1940s, and by the time they were designated a UNESCO World Heritage Site in 1994, around 30 were identified. They’re remarkably well-preserved considering their age, helped by the desert’s dry climate and winds that sweep away the sand but are being obscured by floods and human activity.

Archaeologists discovered 142 new designs in the desert over the course of ten years by manually identifying them using aerial photography and on-site surveying. Then, in collaboration with IBM Japan researchers, they used machine learning to search the data for designs that had been missed in previous studies.

Geoglyphs can be categorized into three main types: figurative, geometric, and lineal. (A) “Line-type figurative geoglyphs” were made by removing black stones in a linear pattern exposing the white sand underneath. (B to E) “Relief-type figurative geoglyphs” are often located on slopes and comprise a combination of black stone and white sand surfaces.
Geoglyphs can be categorized into three main types: figurative, geometric, and lineal. (A) “Line-type figurative geoglyphs” were made by removing black stones in a linear pattern exposing the white sand underneath. (B to E) “Relief-type figurative geoglyphs” are often located on slopes and comprise a combination of black stone and white sand surfaces.

In order to create a thorough survey of the area in 2016, the researchers used aerial photography with a ground resolution of 0.1 m per pixel. The team has identified numerous geoglyphs over time, but because the process takes a long time, they have turned to AI deep learning to analyze the images much more quickly.

A study published in the Journal of Archaeological Science revealed the discovery of four new Nazca geoglyphs using this new method by developing a labeling approach for training data that identifies a similar partial pattern between the known and new geoglyphs.

The four new geoglyphs depict a humanoid figure, a pair of legs, a fish, and a bird. The humanoid geoglyph is shown holding a club in his/her right hand and measures 5 metres in length. The fish geoglyph, shown with a wide-open mouth measures 19 metres, while the bird geoglyph measures 17 metres and the pair of legs 78 metres.

Four new Nasca geoglyphs identified by Deep Learning. (A) A humanoid, relief-type. (B) A pair of legs, line-type. (C) A fish, relief-type. (D) A bird, line-type. (B to D) are presented to the public for the first time in this paper.
Four new Nasca geoglyphs identified by Deep Learning. (A) A humanoid, relief-type. (B) A pair of legs, line-type. (C) A fish, relief-type. (D) A bird, line-type. (B to D) are presented to the public for the first time in this paper. Science Direct

“We have developed a DL pipeline that addresses the challenges that frequently arise in the task of archaeological image object detection,” the study’s authors write. Our method enables the discovery of previously unattainable targets by enabling DL to learn representations of images with better generalization and performance. Additionally, by speeding up the research process, our approach advances archaeology by introducing a novel paradigm that combines field research and AI, resulting in more effective and efficient investigations.

These results serve as yet another illustration of how machine learning can be useful to scientists, particularly when tackling tasks involving sizable datasets. Just like humans, algorithms can be taught to sift through specific types of data looking for patterns and anomalies. Although creating these tools can be challenging, once trained, such algorithms are tireless and consistent.

Cover Photo: Yamagata University / IBM Japan

https://doi.org/10.1016/j.jas.2023.105777

Related Articles

Mysterious and Life-size camel carvings have been found in Saudi Arabian desert

4 October 2023

4 October 2023

Archaeologists have found life-size camel carvings on a rock near the southern border of Saudi Arabia’s Nafud desert. The Neolithic...

Archaeologists Discovered a Fragmentary Inscription in Cypriot Syllabary Found Dating to the Cypro-Archaic Period

1 December 2024

1 December 2024

During excavations at Palaepaphos, located within the municipal boundaries of the modern village of Kouklia-Martsello on the southwest coast of...

Scientists Reconstruct Face Of 16th Century Italian ‘Vampire’ Buried With Brick In Mouth

23 March 2024

23 March 2024

A 16th-century ‘vampire‘ who was buried with a stone brick jammed in her mouth over fears she would feed on...

Carvings at Göbeklitepe could be World’s Oldest Calendar

6 August 2024

6 August 2024

Experts suggest that markings on a stone pillar at the 12,000-year-old Göbeklitepe archaeological site in Türkiye probably represent the oldest...

Detectorist Finds 2,500-Year-Old Unique Bronze Brooch

26 August 2024

26 August 2024

A metal detectorist, who chose to remain anonymous, uncovered three artifacts, including a massive fibula, i.e. a bronze brooch dating...

Drone photos reveal Venice of the Fertile Crescent

16 October 2022

16 October 2022

A drone survey of Lagash, a site located between the Tigris and Euphrates rivers, revealed that the 4,900-year-old settlement was...

1500-year-old Stunning Pendant Amulet Depicting the Prophet Solomon Spearing the Devil on Horseback Found in Türkiye

15 November 2024

15 November 2024

During the excavations in the ancient city of Hadrianopolis in Eskipazar district of Karabük, Türkiye, an amulet from the 5th...

Roman-era Pottery Workshop discovered in Alexandria

29 April 2022

29 April 2022

The Egyptian archaeological mission discovered a Roman-era pottery workshop at the site of Tibet Mutawah, west of Alexandria. The researchers...

Apocalypse Ship of the Vikings

26 April 2021

26 April 2021

Researchers discovered a stone boat made by Vikings and surprising gifts inside a cave in Iceland. Aside from the cave,...

Turkey to Present 12 Historic Artifacts to Istanbul Patriarch

10 August 2021

10 August 2021

The government said on Monday that Turkey will deliver stolen icons from ancient local churches to Istanbul’s Fener Greek Patriarch...

7,000-Year-Old Canoes Reveal Early Development of Nautical Technology in Mediterranean

21 March 2024

21 March 2024

The discovery of five “technologically sophisticated” canoes in Italy has revealed that  Neolithic people were navigating the Mediterranean more than...

Hidden past of Ani ruins in eastern Turkey to be uncovered by excavations

31 May 2021

31 May 2021

Archaeological excavations will reveal the historical mystery behind the ruins of Ani on the present-day Turkey-Armenia border. The Ani archaeological...

More than 1,300 prehistoric burial mounds in western Azerbaijan systematically surveyed for the first time

2 January 2025

2 January 2025

Over 1,300 archaeological sites in Azerbaijan were systematically surveyed and documented in two field campaigns in 2021 and 2023 by...

Ancient Mythical Castle “Sörby Borg” Discovered on Swedish Island Creates Archaeological Sensation

4 August 2021

4 August 2021

A text from the early 18th century mentions the castle, which has become a bit of a legend. It has...

Cosmic cataclysm 1,500 years ago may have caused downfall of the Hopewell Culture

3 February 2022

3 February 2022

Researchers at the University of Cincinnati find evidence of cosmic cataclysm 1,500 years ago at 11 ancient sites in three...