How complex societies shaped their environment and managed resources, particularly water in arid environments, can tell us about their development and
decline. The main limitation of such studies is the creation of a uniform methodology, for instance, to detect irrigation features or study the landscape as a whole. As
each area is different, it requires an adapted set of research methods considering the topography, landscape character and diversity of irrigation features.
This paper aims to present the results of the Under the Sands project devoted to the study of irrigation using remote sensing and machine learning methods in
four selected case studies in Iraq, northern Iran and Turkmenistan. The research combines HEXAGON archival imagery, a 12 m resolution Tan DEM-X digital
elevation model, synthetic aperture radar (Sentinel-1), and multispectral (LANDSAT and Sentinel-2) multitemporal datasets available in Google Earth Engine. We
have prepared a methodology and parameters suitable to map diverse irrigation features in each case. Integrating multi-scale relief models (MSRM) computed
with multitemporal vegetation indices (SMTVI) within the Google Earth Engine platform, we could map paleochannels and levees that are the remains of canals.
The qanat, which is an underground water distribution system, was mapped using the YOLOv8 deep-leaning model.