Physically-Based Detection Algorithm of Buried Archaeological Remains Using Spectral Signatures

Detection of buried archaeological remains based on identification of archaeological proxies, such as cropmarks, has been widely used. Nevertheless, physically-based models for such archaeological prospection surveys are still missing from the literature. In this work we present a spectral classification criterion procedure for the detection of buried archaeological remains (cropmarks) using remote sensing techniques, in particular top-of-canopy hyperspectral data. The criterion is built by using (1) the radiative transfer model PROSAIL in inverse and forward mode to produce physically-based simulations of spectral signatures of an observed cropmark dataset captured from an artificial test-field, and (2) machine-learning methods (decision trees) to identify the highest importance wavelengths and the associated classification thresholds. This is done by statistically analyzing different simulated dataset size (synthetic hyperspectral image size) and different contents in signatures affected by buried ‘remains’ relatively to healthy crop signatures. The analysis of the results does indeed allow the formulation of a well-performing criterion, with above 70% detection rate in test synthetic datasets. Our findings show that the physical reduction of the degrees of freedom forming cropmarks plays a significant role in their modelling and successful detection. The underlying hypotheses and issues, as well as the generalizability potential of the method in different conditions are discussed.