Predicting the geographic distribution of ancient Amazonian archaeological sites with machine learning.

Abstract

Amazonia has as least two major centers of ancient human social complexity, but the full geographic extents of these centers remain uncertain. Across the southern rim of Amazonia, over 1,000 earthwork sites comprised of fortified settlements, mound villages, and ditched enclosures with geometric designs known as geoglyphs have been discovered. Qualitatively distinct and densely located along the lower stretches of major river systems and the Atlantic coast are Amazonian Dark Earth sites (ADEs) with deep anthropogenic soils enriched by long-term human habitation. Models predicting the geographic extents of earthworks and ADEs can assist in their discovery and preservation and help answer questions about the full degree of indigenous landscape modifications across Amazonia. We classify earthworks versus ADEs versus other non-earthwork/non-ADE archaeological sites with multi-class machine learning algorithms using soils, climate, and distances to rivers of different types and sizes as geospatial predictors. Model testing is done with spatial cross-validation, and the best model at the optimal spatial scale of 1 km has an Area Under the Curve of 0.91. Our predictive model has led to the discovery of 13 new geoglyphs, and it pinpoints specific areas with high probabilities of undiscovered archaeological sites that are currently hidden by rainforests. The limited, albeit impressive, predicted extents of earthworks and ADEs means that other non-ADE/non-earthwork sites are expected to predominate most of Western and Northern Amazonia.