Tracking the exposure of geoglyphs after Amazon deforestation bouts using deep learning of satellite imagery

Summary

This is an abstract from the "Practice, Theory, and Ethics of Machine Learning in Archaeology" session, at the 90th annual meeting of the Society for American Archaeology.

The Amazon rainforest contains an abundant record of human occupation, including evidence of extensive landscape modification, and construction of extensive earthworks and roads. However, reconstructing the extent and precise location of archaeological sites is impractical without analyzing either LiDAR imagery of forested areas or satellite imagery of deforested areas. Detecting and locating these archaeological sites is increasingly important as they are exposed, and often destroyed through deforestation and the subsequent effects of agropastoral practices. We discuss approaches to training machine learning models to detect geoglyphs in satellite imagery of deforested areas of the Western Amazon. Using a dataset of ~1,300 geoglyph sites, we trained a computer vision model using pre-trained deep learning architecture for object detection on satellite images of those known geoglyphs. That model was then applied to images of tiles extracted from Google Earth and Bing satellite imagery in Western Brazil both from deforested areas without evidence of geoglyphs, as well as areas with known evidence for geoglyphs. We discuss the reliability of this approach to detecting archaeological features across such a broad area, and its potential in aiding with the management of archaeological resources in a rapidly changing landscape.

This Resource is Part of the Following Collections

• Society for American Archaeology 90th Annual Meeting, Denver, CO (2025)   •
• Practice, Theory, and Ethics of Machine Learning in Archaeology

Keywords

Record Identifiers

Abstract Id(s): 51937