Multi-Temporal Classification of Multi-Spectral Images for Settlement Survey in Northeastern Syria

Abstract

Nearly all Near Eastern surveys have employed some form of satellite remote sensing, although usually at coarse resolution or for very limited study areas only, and nearly always in a qualitative fashion. In this chapter we briefly review the multi- and hyper-spectral remote sensing approaches that have been used in archaeological survey in the Near East, and report on a novel satellite remote sensing approach we recently developed. This approach recognizes anthropogenic sediments via scenes from multi-spectral ASTER images using a multi-temporal classification strategy, guided from results of a visual inspection of CORONA images. We apply it to the Khabur Basin, in north-eastern Syria, returning a probabilistic map of anthrosols that is indicating the locations of some 10,000 settlement sites of all times – from the eighth millennium B.C. to modern – at a resolution of 15 m for an area of about 22,000 km² and with an accuracy that is comparable to modern ground survey. This makes it, to the best of our knowledge, the largest systematic satellite imagery based survey in archaeology. Our multi-temporal classification strategy can integrate information from any other multi- or hyper-spectral sensor, and it will easily generalize to other related detection tasks in archaeological remote sensing.