Dunn and Newton (1992) examine the performance of two popular approaches to network analysis, Dijkstra’s and out-of-kilter algorithms, in the context of population evacuation. At the time of publication, it’s clear that the majority of network analysis research has been conducted by computer scientists and mathematicians. It’s interesting how historical conceptualizations of networks, which appear to be explicitly non-spatial in the way that distortion or transformation are handled and the lack of integrated geospatial information, are transferable to GIS applications. What the authors describe as an “unnecessarily flexible” definition of a network for geographical purposes appears to be an insurmountable limitation of previous network conceptualizations for GIScience. However, I’ll admit that the ubiquity of Dijkstra’s algorithm in GIS software is a convincing argument for the usefulness of previous network concepts in GIS against my limited knowledge of network analysis.
The out-of-kilter algorithm provides a means to address the lack of integrated geospatial information in other network analysis methods. The authors demonstrate how one might incorporate geospatial concepts such as traffic congestion, one-way streets, and obstructions to enable geographic application more broadly. It’s striking that the processing time associated with network analysis is ultimately dependent on the complexity of the network. In the context of pathfinding, increased urban development and data availability will necessarily increase network complexity, and it was demonstrated in the paper how incorporating geographic information into a network can increase processing time. While it was unsurprisingly left out of a paper published in 1997, I would be curious to learn more about how heuristics might be applied to address computational concerns in the geoweb.