Broadscale Machine Learning Model for Archaeological Feature Detection in the Maya Area

Summary

This is an abstract from the "2023 Fryxell Award Symposium: Papers in Honor of Timothy Beach Part II" session, at the 88th annual meeting of the Society for American Archaeology.

Comprehensive maps of ancient structures across the Maya area of Central America can help archaeologists to deepen knowledge of past settlement patterns and regional interactions, potentially leading to enhanced understanding of thousands of years of Maya civilization. However, most Maya archaeological sites are not comprehensively or systematically mapped because ruins, often hidden under dense subtropical forest canopy in rugged topographic settings, can take decades to map. Here we present the preliminary version of a broadscale multi-site-based machine learning model for archaeological feature detection across 1,500 km2 of the Maya area that will enable researchers to map sites in hours instead of decades. We find that a model trained on multiple sites across varied topographies produces better results than small, site-specific models, which are to date the only kind of models that exist for the region. Our model produced an F1 score of 0.80, and results also include many potential new structure detections. This approach suggests that a single machine learning model may be capable of broadscale mapping of Maya archaeological features across Central America. Such a model would be the first of its kind for the Maya area and demonstrates how big data can be integrated into traditional archaeological workflows.

This Resource is Part of the Following Collections

• 2023 Fryxell Award Symposium: Papers in Honor of Timothy Beach Part II   •
• Society for American Archaeology 88th Annual Meeting, Portland, OR (2023)

Keywords

Spatial Coverage

Record Identifiers

Abstract Id(s): 36018.0