Irmela Herzog

(Bonn, Germany)

Simulation studies help to understand the limitations and the power of methods used for data analysis. In this paper simulation experiments will be presented that focus on settlement patterns. In most GIS studies, relief is considered an important factor for settlement location. But the Rhine area is fairly flat, and for this reason other models may be more appropriate. Five different models for settlement distribution in a flat terrain will be presented. It is the aim of this paper to test various methods proposed in the GIS and spatial statistics literature for detecting and describing settlement patterns. When dealing with point patterns, quadrat and near neighbour analysis have been in the tool box for quite a while, and the advantages and disadvantages are well known. Since the mid-1990s, kernel density estimation (KDE) has been considered by most authors to be the method of choice for reconstructing the distribution function of the point pattern. But it is not easy to determine the appropriate bandwidth parameter for KDE. In addition, KDE procedures are not readily available in GIS programs. An alternative approach popular with German archaeologists is based on voronoi diagrams and the largest empty circle (LEC) which can be drawn touching any three points with a common vertex of the voronoi tessellation. Where contouring the point density is the aim, an interpolation method is needed in addition to the LEC calculation, whereas no interpolation is necessary for KDE. This paper will compare the results of the contouring processes of different methods with the models used for the simulated data sets.

Keywords: Simulation, settlement models, point density estimation, GIS