The study of ancient irrigation plays a pivotal role in the archaeological investigation of arid landscapes. It covers various topics, including irrigation’s role in empires or complex societies, agricultural intensification, communal irrigation practices, and water harvesting techniques. Moreover, each irrigation landscape is unique and requires a partiularly adapted set of methods.

This paper outlines the results of UnderTheSands, a project aiming to investigate the application of remote sensing and machine learning methods to the analysis of ancient irrigated landscapes, in selected case studies in Morroco, Iraq, northern Iran, Turkmenistan and Afghanistan. The research integrates HEXAGON imagery, TanDEM-X digital elevation models, and LANDSAT, Sentinel-1, and Sentinel-2 multitemporal multispectral datasets in different parts of the workflow. Levees and paleochannels have been mapped by integrating the result of multitemporal vegetation indices (SMTVI) with Multi-Scale Relief Models (MSRM) calculated within the Google Earth Engine platform. Qanats, a type of underground water distribution system, have been mapped using the YOLOv8 qanat detection model we trained on the black-and-white HEXAGON high-resolution spy satellite images (1971-1986).

The applied methodology allows for a comprehensive understanding of irrigated landscapes. The research investigates the influence of environmental conditions and satellite imagery parameters on the visibility of features such as levees, paleochannels and qanats. Optimal results are achieved by combining several methods, adjusting seasonal parameters, and conducting a comparative analysis of visible features.