Refinement of a Method for Identifying Probable Archaeological Sites from Remotely Sensed Data

Abstract

To discover and locate archaeological sites, we aim to develop scientific and efficient approaches to identify these sites with high accuracy. In this chapter, we present a statistical learning model consisting of imaging processing, feature extraction and classification. Our analysis uses the remotely sensed data composed of eight WorldView-2 imagery bands and one slope band. In the imaging processing step, we use a particular annuli technique; in the feature extraction step, principal component analysis is applied; in the classification step, linear discriminant analysis is carried out. We test this procedure on 33 lithic sites, 16 habitation sites and 100 non-sites from the western portion of Ft. Irwin, CA, USA. The receiver operating characteristic curve, used for assessing the performance of the algorithm, shows that our new approach generates higher classification power than the Archaeological Predictive Model (APM). When APM is convexly combined with our new model, the classification accuracy is even higher.

Publication
Mapping Archaeological Landscapes from Space