AI in archaeology. Remote sensing and machine learning for Andosols landscapes analysis in Central-Southern Italy (2024-2025)

This paper investigates Andosols landscapes in Central-Southern Italy to evaluate the potential of high-resolution multispectral satellite imagery for archaeological prospection. The physical and chemical properties of Andosols play a crucial role in determining the visibility and readability of potential subsurface archaeological features within the landscape. In order to make the most of these soil characteristics, the research provides a rigorous comparative analysis of three complementary software platforms - QGIS, ENVI, and eCognition - evaluating traditional, semi-automated, and automated workflows, for the processing, analysis, and interpretation of multispectral satellite imagery. The research focuses on comparing the implementation and performance of Object-Based Image Analysis (OBIA) and Machine Learning (ML) techniques with conventional Pixel-Based Image Analysis (PBIA). Preliminary results demonstrate the capability of the proposed integrated methodology in successfully validating known archaeological features while revealing previously unidentified subsurface anomalies, including ancient road networks, settlements, land divisions, and funerary areas. In summary, the integration of Remote Sensing analysis (RS) and ML has been demonstrated to facilitate the identification of archaeological features in Andosols landscapes. However, the results indicate that automated, machine-driven interpretations must be integrated with traditional photo-interpretation, archaeological expertise, and rigorous ground-truthing to ensure reliable historical landscape reconstructions and effective cultural heritage management.