Benjamin Ducke

(Institut für Ur- und Frühgeschichte Christian-Albrechts-Universität zu Kiel, Germany)

With roughly two decades of theoretical and practical work behind it, predictive modelling of archaeological sites (APM) has grown out of its infancy to provide a — sometimes the only — way of estimating a landscape’s hidden archaeological potential.
Criticism towards APM has traditionally been brought forward in one of two forms:

  1. philosophical (deterministic modelling does not reflect human nature of the subject under study)
  2. methodological (suitability of variables and techniques used; data quality; ignorance of uncertainty factors, etc.)

Aspects of methodology have been taken very seriously and resulted in at least two heavy-weight European projects during the last five years:

  1. In Germany, heritage management in the federal state of Brandenburg has conducted research into APM over a period of almost four years for an area encompassing ca. 30.000 sqkm and several thousand known archaeological sites.
  2. In the Netherlands, another three-year’s effort has been made to evaluate the Dutch APM expertise and practical experiences. It is hoped that this process will eventually produce methods and guidelines for predictive modelling to be used uniformly in the country’s heritage management.

This recent work has shed some light on several of the core concerns affecting APM:

  1. Handling uncertainties and biases in site and survey data
  2. Issues of scale and precision
  3. Inclusion of a priori expert knowledge in APM
  4. Predictive power of available landscape variables
  5. Reliability, ease-of-use and flexibility of different statistical and mathematical techniques
  6. Integration of predictive models into the decision schemes of heritage management

Using the open source Geographic Resources Analysis Support System (GRASS) GIS, a comprehensive predictive modelling environment has been developed that addresses most of these issues. The software is available completely under an open source license.