In this paper, we want to present a new perspective (or "way of thinking") for predictive modelling based on the analysis and comparison of the spatial context of archaeological settlements, using neighbourhood calculations and statistical classification and extrapolation methods. The basis for this method was originally developed by F.-P. Tourneux in the late 1990s in the Archaeomedes Project (Favory et al. 1999, van der Leeuw 1998; van der Leeuw et al. 2003; Nuninger et al. 2007) and was at the time used to enable the cross-regional comparison of the environmental context of rural settlements in the Rhône Valley in France dating from the Roman period. For this purpose, various environmental ‘descriptors’ were defined, like slope, aspect, solar radiation and soil type that characterize the environmental setting of a settlement within a specific search radius. This then resulted in the classification of settlements into groups with comparable environmental contexts. Since the definition of these contexts was done in the same way for each region, it enabled the interregional comparison of settlement location factors and the analysis of the development of settlement patterns for the whole Rhône Valley through time. For our current research, we have used this method not just for the characterisation of the settlements’ context, but also for predictive modelling purposes. In most published predictive modelling studies, comparison of settlement and environmental characteristics is done for the sites’ locations only, and the immediate surroundings of the settlements are not usually taken into account. The environmental contexts of settlements can however be defined with relative ease with the method described. It is then a small step to extrapolate the classification results to all locations in the landscape, and see how well these correspond to a context that is known to contain settlements. However, for predictive modelling we also need information on the occurrence and distribution of contexts that do not have settlements. For this reason, we have extended the analysis to characterize the context of each position (pixel) in the landscape. Furthermore, we have tried to add new, ‘socio-cultural’ descriptors to the ones already used, like visibility, accessibility, the hierarchical position of settlements and the effect of previous occupation (‘the memory of landscape’). In this way, it will become possible to move away from a purely environmentally based characterization and prediction of settlement patterns. We will report the results of applying this method using settlement data from the Roman period for three (very) different regions, the Argens-Maures region (Provence, France), the Vaunage region (Languedoc, France) and Zuid-Limburg (the Netherlands).
13 Apr 2011
Revive the Past. 39th Annual Conference of Computer Applications and Quantitative Methods in Archaeology.