Reading the Memory of the Soil: Unveiling the Invisible with Artificial Intelligence

This article explores the potential of using open-source satellite data and fundamental artificial intelligence methods to identify prospective archaeological sites. In situations where traditional excavation methods are limited by cost and time constraints, remote sensing and machine learning techniques offer a supportive tool for the preliminary exploration process. Through spectral analyses and anomaly detection methods performed on freely available satellite data, such as Sentinel-2, the likelihood of subsurface anthropogenic structures can be determined. This study proposes a low-cost and accessible approach to digital archaeology.

Archaeology is a science dedicated to unearthing the traces of past civilizations from beneath the soil. However, traditional exploration methods are limited in their ability to survey vast areas and require significant fieldwork. With the advancement of satellite imaging technologies, structures not visible on the surface can now be analyzed through indirect indicators. Variations in vegetation, soil moisture distribution, and topographic irregularities can all be telltale signs of buried man-made structures.

This is where artificial intelligence comes into play. AI can analyze large volumes of satellite imagery to detect deviations from natural patterns. For example, symmetrical or linear forms that are rare in nature can be flagged as anomalies by algorithms. Vegetation indices calculated from multispectral bands can reveal how underground stone or wall remnants affect plant growth. The probability maps generated from these analyses guide archaeologists in prioritizing excavation sites.

This approach doesn't replace the archaeologist but rather provides a decision support mechanism. Model outputs are not definitive proof but statistical predictions. Final verification still requires fieldwork. However, in vast geographical areas, especially in projects with budget and time constraints, AI-assisted analysis can provide significant efficiency gains.

This system, which can be developed using open-source satellite data, makes digital archaeology more accessible. In the future, the accuracy of these methods will increase with higher resolution data and deep learning models. The past, hidden beneath the soil, can now be read not only with picks and shovels but also with algorithms.