Visualizing self-organizing maps with GIS

Self-organizing maps (SOM) are a powerful tool for detecting patterns in large, multi-dimensional data sets. Additional visualization techniques have been developed to support the user to gain insight into its structure. For complex data sets, even these techniques are not easily interpretable. Most of them consist of a grid where each cell contains a single value. Such a structure can be seen as an artificial landscape. This paper aims to explain the function of the SOM algorithm and to present a number of frequently used visualization techniques. We show a way to import traditionally created SOM into a GIS, so that operations created for spatial analysis can be applied to this originally non-spatial data. We present GIS operations that help the user to understand the structure of a visualization technique, its underlying SOM, and eventually the input data