Articles by Giuseppe Arbia

The aim of this paper is to introduce a class of testable statistical models aimed at modelling archaeological sites locations (ASL) or a continuous space and at producing probability maps of ASL. These models are based on collected statistical and auxiliary information (such as information about the slope or exposure of the land, the topography, the hydrology, the topology, etc.). More explicitly the model for the ASL is based on the assumption that the location of the existing archaeological sites is fixed and that the location of new archaeological sites not yet found is a realization of a non-stationary point process depending on the spatial interaction with other existing sites, on the distance from important location or communication networks, and on a set of auxiliary variables. It is also possible to introduce into the discussion information we have about the archaeological site dimensions. In fact, it is plausible that the interaction between existing archaeological sites and candidate locations for new archaeological sites is dependent on their dimension. These models can prove valuable in that they formalize some “common sense” knowledge and can help in producing automated probability maps on a continuous space to support excavation decisions.

Forecasting statistical models are becoming increasingly important in archaeological research. One of the reasons of this popularity is that archaeological sites tend to present themselves in particular environments so that forecasting models can help in identifying areas where the probability is higher based on previously collected statistical information. In the present paper we consider a class of statistical models designed to produce maps of the probability for archaeological site location (ASL) which incorporate both deductive and inductive considerations. In the discussion we criticise the use of the logistic regression for the production of ASL probability maps, a popular approach known in archaeological literature as the “integrated strategy”. The application of the method is statistically incorrect since in archaeological studies the hypothesis of independence between sites, which is at the basis of the logistic regression model, is violated. To overcome such limitations we propose two alternative models. The first one is an autopredictive model in which the probability of ASL is modelled as a function of the observations coming from field surveys in neighbouring zones. This approach accounts for the problem of non-independency of observations, but neglects a priori auxiliary information on the archaeological area. The second approach is a more comprehensive one which overcomes the problems of logistic regressions while preserving the role of a priori information.