M. Lacayo, A. Skupin

San Diego State University



Advances in technology allow generating and storing unprecedented volumes of information, which has raised issues of how to convey such information in an efficient manner to diverse audiences. One approach is to rely on the human vision system to function as a conduit to which visualizations of large, high-dimensional data sets can be directed in a manner consistent with human cognitive facilities. One approach to visualization is spatialization, where data are preprocessed using specialized techniques so that they can be represented using standard methods and technologies used for geographic data, such as GIS. These techniques involve reducing high-dimensional data to lower dimensions irrespective of volume or content.

This paper presents an implementation of one of these techniques called self-organizing map (SOM) or Kohonen map in the context of GIS. The SOM method is an artificial neural network approach to interrelating elements of a high-dimensional data set and organizing them spatially to convey these relationships. The main goal of this GIS-based visualization module is to fill the need for generalized visualization tools that facilitate the display of a trained SOM and the mapping of other high-dimensional data onto the SOM within a standard GIS environment. The module imports a SOM from its standard codebook format into an object-relational database.  This database uses a schema specifically designed to accommodate the storage of SOMs in support of map-like visualizations. Application features include geometric transformation and dynamic re-expression among others.  The application of this module is a practical way to summarize large data sets and visualize their interrelatedness in meaningful ways. The paper discusses the methodology and implementation of this module in detail, the use and significance of specific features, and enumerates some areas requiring further development.