DeepAndesArch: Assessing Performance of an AI Model for Satellite Imagery Survey of the Andes

Summary

This is an abstract from the "SAA 2025: Individual Abstracts" session, at the 90th annual meeting of the Society for American Archaeology.

Social and political networks in the Andes operated far beyond the scale that can be captured in any individual archaeological research project, while combining archaeological data from diverse projects presents challenges in data compatibility and unsystematic sampling. Satellite imagery and deep-learning computer vision models enable such trans-regional archaeological perspectives at scale. This poster presents the results of DeepAndesArch, a deep learning model for identifying archaeological structures in very high resolution multispectral satellite imagery of the Andes. This model utilized cutting-edge vision transformers (ViT) and a diverse dataset of manually identified archaeological features to map the locations of several classes of archaeological features across about 2 million square kilometers in Andean South America. We evaluate the model's performance by comparing the model's predictions to validation data collected by teams of archaeological professionals, showing both the quality of the model's results, and the strengths and weaknesses of the model when compared to manual imagery survey. We also discuss some of the many applications of this novel dataset and ongoing analyses.

This Resource is Part of the Following Collections

• Society for American Archaeology 90th Annual Meeting, Denver, CO (2025)   •
• SAA 2025: Individual Abstracts

Keywords

Record Identifiers

Abstract Id(s): 53814