Towards High-resolution Self-organizing Maps of Geographic Features

This chapter introduces the use of high-resolution self-organizing maps (SOM) to represent a large number of geographic features on the basis of their attributes. Until now, the SOM method has been applied to geographic data for both clustering and visualization purposes. However, the granularity of the resulting attribute space representations has been far below the resolution at which geographic space is typically represented. We propose to construct SOMs consisting of several hundred thousand neurons, trained with attributes of an equally large number of geographic features, and finally visualized in standard GIS software. This is demonstrated for a data set consisting of climate attributes attached to 200,000+ U.S. census block groups. Further, overlays of point, line, and area features onto such a high-resolution SOM are shown. INTRODUCTION This volume demonstrates the range of approaches currently pursued in the field of geographic visualization. Geographic visualization has clearly captured the public’s imagination. Evolutionary changes in creation, distribution, and interaction with cartographic depictions have powerfully converged in early realizations of the digital earth concept (see chapter by Goodchild in this volume). Further convergence of various technologies and methodologies is likely, including trends towards high-resolution imagery (see preceding chapter by Orford) This is a pre-publication draft only. For the final, published version, please refer to: Skupin, A. and Esperbé, A. (2008) Towards High-Resolution Self-Organizing Maps of Geographic Features. In: Dodge, M., Turner, M., and Derby, M. (Eds.) Geographic Visualization: Concepts, Tools and Applications. Wiley.