Towards reconstructing cities with AI: a novel machine learning approach for automated archaeological surveying and preservation by learning canopy structures

Abstract

Archaeological landscapes face increasing threats from climate change and human activity, necessitating scalable methods for site detection. LiDAR has revolutionized archaeological surveys by revealing features beneath dense vegetation, but manual interpretation remains labor-intensive. This study introduces an Automated Archaeological Survey Method (AASM) that utilizes machine learning to analyze tree canopy structures as proxies for underlying archaeological features. Using a LiDAR-derived digital canopy model (DCM), the Bi-path Ensemble (BPE) model predicts the locations of the land-use typologies of "public" and "private" space at Angamuco. The model evaluation shows moderate to high agreement between predicted and true labels, particularly for dense public and un-terraced private spaces. These results suggest that vegetation patterns can serve as reliable indicators of past human activity, offering a scalable approach for prioritizing areas for archaeological surveys. By integrating creative computational methods with remote sensing data, this study advances the use of machine learning in archaeological landscape reconstruction.