A V ISUALIZATION AND E XPLORATION T OOL FOR G EO -SOM BASED C LUSTERING David Moena, Fernando Bacao 1 , Victor Lobo 1,2 1 Instituto Superior de Estatística e Gestão de Informação, Universidade Nova de Lisboa 2 Academia Naval Portuguesa Lisboa, Portugal bacao@isegi.unl.pt , vlobo@isegi.unl.pt Abstract – Clustering geographically referenced data is an important issue in Geographic Information Science. There are several algorithms that can be used in such tasks, the self- organizing map (SOM) is one of the most popular tools in this context. Although the standard SOM can be used geographically referenced data, it is useful to have a clustering tool that takes into account the special importance that geographic location has in these problems. The GEO-SOM was developed with that specific purpose in mind: to incorporate the unique perspective of the spatial analyst while analyzing geo-referenced data. In this paper we briefly present the differences between the training and mapping algorithms of the standard SOM and GEO-SOM, and give some simple examples of applications. This paper is focused on the issue of implementing a visualization software through the integration of the Geo-SOM with Geographic Information Systems (GIS). It is shown that Geo-SOM can easily be integrated in such systems, and examples of relevant visualization tools are presented. Key words – Geography, geo-referenced data, spatial data, SOM variants. 1 Introduction Clustering geographically referenced data, such as census data or remote sensing data, has been an important issue in Geographic Information Science (GIScience) for a long time. With the widespread use of GPS, mobile phones, and other location aware technologies, the amount of geo-referenced data has increased dramatically, and the need for new data reduction and analysis tools has became more urgent than ever. Self-Organizing Maps (SOMs) [1] have been used in GIScience both for clustering geo- referenced data [2-4] [5] and for the spatialization of various non-geographic datasets [6-10]. The original SOM proposed by Kohonen does not take into account the particular role that geographic location has in most problems involving the clustering of geo-referenced data. In the original SOM algorithm, all variables are treated equally. When clustering geo-referenced data, spatial location is particularly important, since objects that are geographically far away should not be clustered together, even if they are similar in all other aspects. This is neatly expressed in the 1 st Law of Geography [11] “everything is related to everything else, but near things are more related than distant things”.