Sribas Chowdhury, Adamas University, Kolkata
Scientists at the University of Pennsylvania recently used machine learning and deep learning methods to study Native American shell rings that date back to the mid-Holocene era (approximately 3000-5000 years ago). The existence of shell rings was reported way back in the early 19th century. However, not much had been researched about them and hence, their function remained a subject of debate. Recently, researchers used the LiDAR technique to study the extent of archaic shell rings in Southeastern parts of North America. It was revealed that these rings were much more extensive and way more in number than originally speculated.
Archaic shell rings
Shell rings are circular middens (heaps) composed of faunal and floral remains that contain a central plaza devoid of midden material. They have been reported in several countries, including Colombia, Puerto, Japan and southeastern parts of the United States. In this particular case, the study was concerned around shell rings found in the Southeastern US, particularly, Beaufort, Charleston and Georgetown counties in South Carolina. Archaic shell rings are particularly found around coastal regions, with diameters ranging from 30-250 m and reaching a height of 1-6m.
Despite undergoing elaborate investigations, the knowledge about archaic shell rings remained sparse. This was mainly due to poor documentation, inaccessibility and conclusion being drawn based upon observable sites. To date, only ~50 shell rings have been studied properly. The function of these shell rings remains a topic of debate. Some experts are of the opinion that these rings were annual living structures due to the presence of flora and fauna throughout the year. While others argue that these were not used for living, and suggestions have been made that these rings were just built coincidentally and were not in regular use.
Deep learning
Deep learning is an AI function that mimics the working human brain to process data and creates patterns to use in decision making. It has been rapidly gaining popularity amongst archeologists for its computational capabilities. In particular, CNN (Convoluted Neural Networks) give highly accurate results. This is because CNN takes input from multidimensional matrices and hence, can make multidimensional patterns, which eases the process of deciding whether an outcome is true or false. However, due to limitations like a shortage of data and expertise, this arena remained quite unexplored. In this study, the researchers showed how to study archaic rings using augmentation (creation of synthetic data) and transfer learning techniques. It also focuses on showcasing how deep learning can help learn about the geographical extent of human behavioural patterns like demographic shifting, sociopolitical domains etc.
Using LiDAR to study shell rings
To start with, the researchers obtained LiDAR (Light Detection and Ranging) point data of the targeted area from the National Oceanic and Atmospheric Administration (NOAA). Then this data was converted into DEM (Digital Evaluation Model) points. Next, two additional visualizations: a hillside map and a slope map, were created from the LiDAR point data. Succeeding it was the making of a Mask R-CNN using ArGIS pro to study the shell ring structures. The model was then tested and finally applied to the target area and the outcome was noted. To further supplement the verification of data obtained, they also obtained Sentinel 1 and 2 data and evaluated the training data. The data model was then validated and the results were cross-checked with the CNN data. The cross-checked outcomes were then sent for final evaluation.
Observations in the Study
- Using deep learning and LiDAR data, it was eventually found that the practice of ring building was more widespread than anticipated earlier.
- The researchers were able to identify 120 potential ring structures in the total area of 6712 sq. km in South Carolina.
- The density of ring structures gradually decreased as they moved further north, justifying that these structures don’t extend beyond South Carolina.
With the discovery of these new shell rings, it can be concluded that these structures were not rare or coincidental structures, but rather, a common practice amongst the coastal dwelling communities in the southeastern USA.
Significance of deep learning and LiDAR
The study clearly showed how deep learning and LiDAR can help in studying the geographical aspects of behavioural patterns in Native American tribes. However, these archaeological sites are at risk of damage due to climate change and development activities. More advances have to be made in data processing capabilities to ensure the protection and study of undiscovered sites. The researchers are hopeful that with further progress in machine learning, they will be able to get more insights into the extent and structure of these archaic shell rings. This can lead to clarification of the role of ring building in Archaic Native American communities and socioeconomic networks of the coastal groups dwelling in South Carolina.
Also read: Male Pregnancy: A Miracle of Nature?
Reference:
Davis, D. S., Caspari, G., Lipo, C. P., & Sanger, M. C. (2021). Deep learning reveals extent of Archaic Native American shell-ring building practices. Journal of Archaeological Science, 132, 105433. https://doi.org/10.1016/j.jas.2021.105433
One thought on “Archaic shell rings: New info revealed by deep learning methods”