André Skupin* - San Diego State University
Aude Esperbé -
Visualization has been recognized as a powerful strategy for understanding complex phenomena that are reflected in the multifaceted databases collected in all areas of contemporary society. The role of geographic visualization has typically been restricted to presenting geographic phenomena in terms of their geographic location, with geographic space acting as the dominant integrator of disparate data sources from the physical and human domains. One of the main reasons for the conceptual and visual richness of such depictions is the relatively high resolution of the geographic reference base, as compared to the relatively low resolution of the non-spatial attributes. This allows making inferences about low-dimensional attribute relationships in geographic space, but one learns relatively little about complex high-dimensional relationships and structures existing in attribute space.
In this paper, we present an alternative approach aimed at creating a high-resolution self-organizing map (SOM) whose geometry is constructed from the attributes of a large number of geographic objects. Specifically, we spatialize 200,000+ U.S. census block groups using a SOM consisting of 250,000 neurons. In addition, the attributes included represent a more holistic representation of geographic reality than in previous studies. Included are 69 attributes regarding population statistics, land use / land cover, climate, geology, and soils. The diversity of this set of attributes is informed by our desire to build a comprehensive two-dimensional base map of n-dimensional geographic space. The paper will discuss how standard GIS methods and neural network processing are combined towards the creation of an alternative map of the United States.