REDUCING INTER-ANALYST VARIABILITY IN THE CLASSIFICATION OF IRON AGE STONE-WALLED STRUCTURES FROM SATELLITE IMAGERY OF THE SOUTHERN GAUTENG PROVINCE, SOUTH AFRICA

Previous remote sensing studies of South African Iron Age stone-walled ruins revealed considerable variation in how different analysts subjectively classify the ruins into one or another of several architectural types. In previous studies, such high disagreement in classifying individual ruins co-existed, curiously, with similarity in the regional site distribution maps produced by the different analysts. Here, with the assistance of my students and a private GIS company in India, I present and test two hypotheses to explain this anomaly and to help reduce inter-analyst variations in classifying the ruins. The results show support for both hypotheses and indicate that in proceeding with our settlement pattern studies, we should use only the sites identically classified by two experienced analysts. These sites should then be treated as a representative sample, and spatial statistical techniques used to approximate the whole population.