Improving simulation accuracy of spread phenomena in a raster-based Geographic Information System

Abstract

Abstract Dijkstra's shortest path algorithm is widely used in spread simulation. While Dijkstra's algorithm is robust in solving classical network problems, its implementation in raster-based geographical information systems (GIS) has not been satisfying. A major reason for the inadequate simulation is the restricted number of raster cells (generally the 4 to 8 adjacent cells) included as links in the algorithm. We propose several alternative methods that incorporate non-adjacent cells as additional links, thus overcoming the limitation of a raster data structure. The simulation accuracy is increased from less than 60 per cent to above 95 per cent for a spatially homogeneous elliptical spread and appears to be increased significantly for spatially heterogeneous conditions. The computational cost of the new methods are higher but still reasonable in practical use.