Classification of multispectral ASTER imagery in archaeological settlement survey in the near east

In alluvial areas of the Near East, the former locations of settlements are often represented by “tells”, small artificial mounds resulting from millennia of human settlement activity, especially the continual construction and decay of mud brick architecture. To identify such tells and other, smaller settlement sites in the modern landscape, we develop a classifier which screens wide areas for tell-specific soil-changes, based a characteristic spectral signature in ASTER imagery. Using data from sites identified from CORONA imagery and field survey on a north Syrian plain, a Random Forest classifier was trained, using the raw reflectances, vegetational features, correlation with prototype-spectra of the JPL ASTER SpecLib, and time flags as input features. A spatio-temporal sampling strategy allowed us to classify and fuse results from any ASTER images available for a certain region. The classifier was tested in an independent test area, centered around Tell Hamoukar, with close ground control from an archaeological survey. In this test area it was possible to identify 32 out of the 49 site bigger than 2ha. Overall we found that multi-spectral ASTER imagery can be used to provide highly specific information on character and composition of the ground, a tool which can be used in survey planning or the screening of wide regions for conservational issues or studies in landscape archaeology.