Enhanced Archaeological Predictive Modelling in Space Archaeology
Identifying and preserving archaeological sites before they are destroyed is a very important issue. In this paper, we develop a greatly improved archaeological predictive model $APM_{enhanced}$ that predicts where archaeological sites will be found. This approach is applied to remotely-sensed multispectral bands and a single topographical band obtained from advanced remote sensing technologies such as satellites and Airborne Laser Scanning (ALS). Our $APM_{enhanced}$ is composed of band transformation, image analysis, feature extraction and classification. We evaluate our methodology on the sensor bands over Ft.Irwin, CA, USA. A nested bi-loop cross-validation and receiver operating characteristics curves are used to assess the performance of our algorithm. We first validate our method on the east swath of Ft.Irwin and then test on a separate dataset from the west swath of Ft.Irwin. A convex combination of two methodologies: $APM_{conventional}$, which has been used among archaeologists for many years, and our $APM_{enhanced}$, is demonstrated to yield superior classification performance compared to either alone at low false negative rates. We compare the performance of our methodology on different band combinations, chosen based on the archaeological importance for these sensor bands. We also compare the two types of $APM$s in the aspects of input data, output values, practicality and transferability.