Automated detection of archaeological mounds using machine-learning classification of multisensor and multitemporal...

Area of Science:

• Archaeology
• Remote Sensing
• Machine Learning

Background:

• Cholistan Desert holds significant Indus Civilization sites (3300–1500 BC).
• The region is crucial for understanding water availability, civilization dynamics, and aridification.
• Previous studies have not fully captured the extent of archaeological features.

Purpose of the Study:

• To develop and implement a novel multisensor, multitemporal machine learning approach for detecting archaeological mounds.
• To map mound-like signatures across a large area of the Cholistan Desert.
• To reassess the archaeological significance and settlement patterns of the region.

Main Methods:

• Utilized a machine learning classifier algorithm.
• Employed a large collection of synthetic-aperture radar and multispectral satellite images.
• Implemented the approach on the Google Earth Engine platform for large-scale processing (36,000 km²).

Main Results:

• Accurately mapped numerous archaeological mounds, significantly more than previously recorded.
• Detected a wide range of mound sizes, from small (<5 ha) to large (>30 ha).
• Identified mounds extending further south and east into the desert.

Conclusions:

• The findings necessitate a revision of the archaeological significance of Cholistan.
• Observed settlement shifts indicate adaptation to a dynamic hydrological network and desertification.
• The results provide crucial insights into the long-term abandonment of settlements during the Late Harappan period.