KARTOGRAFISCH TIJDSCHRIFT
2003-XXIX-2
Kraak, M.J. (2000), About maps, cartography,
geovisualization and other graphics. GeoinformaticsJournal,
vol. 3, December 2000.
Kohonen, T. (1989), Self-Organizing Maps. Berlin: Springer-
Verlag.
MacEachren, A.M. et al. (1999), Constructing knowledge
from multivariate spatiotemporal data: integrating
geographical visualization with knowledge discovery in
databases methods. International Journal of Geographical
Information Sciencevol. 13, nr. 4, pp. 311-334.
MacEachren, A.M. M.J. Kraak (2001), Research
challenges in geovisualization. Cartography and
Geoinformation science, vol. 28, nr. 1.
Miller, H.J. J. Elan (2001), Geographie data miningand
knowledge discovery. London: Taylor and Francis.
Norman, D.A. S.W. Draper (1986), User centered System
design: New perspectives on Human-Computer Interaction.
New Jersy: Lawrence Erlban Associates, Publishers.
Openshaw, S. C. Openshaw (1997), ArtificialIntelligence
in geography. Chichester: John Wiley Sons, pp. 21-25.
Rubin, J. (1994), Handbook ofusability testing: How to plan,
design, and conduct ejfective tests. New York: John Wiley
Sons, Inc.
Ultsch, A. (1993), Self-organizing Neural Networks for
Visualization and Classification, in O. Opitz, B. Lausen
R. Klar (Eds.), Information and ClassificationBerlin:
Springer-Verlag, pp. 307-313.
Summary
Etien L. Koua Exploring Seif Organising Maps for representation
and visualisation of complex geospatial data sets
Key words: visualisation, spatial data exploration, exploratory
analysis, Self-Organising Maps
New techniques for spatial data acquisition generate large
amounts of data to process. Analysis of these data is difficult
because of the amount, complexity, scaling issues, and existing -
yet hidden - patterns. Traditional deduetive and statistically
oriented analyses may not be suitable, partly due to the complexity
of the attributes, partly due to the inejficiency of existing
mathematical methods. Data-rich environments require the
strueture and support offered by techniques for data exploration in
Order to reduce the complexity and to facilitate knowledge
building. Ideally, users would be able to visualise spatial data in
any combination of variables and at any scale to discover spatial
relations and patterns.
To explore and use data sets more efficiently and intuitively,
geoinformatics can learn from the techniques developed by other
diseiplines (e.g. Information visualisation) such as Statistical
analysis, artificial intelligence, and artificial neural networks fior
data analysis and pattern recognition. Especially, the Self-
Organising Map (SOM), is a neural network model marked as a
promising technique for exploring large spatial data sets. Here, the
SOM-algorithm is applied to explore the socio-economic
characteristics of municipalities in the Dutch province Overijssel.
The purpose of the SOM-algorithm is to bring to the fiore the
strueture and patterns in the data set. Spatial analysis,
exploration, and knowledge building are combined to create a
visual representation of the data that stimulates pattern
recognition and generation of hypotheses. Some examples of visual
representation (information Spaces) are explored. Spatial
metaphors such as distances, regions, and scale are applied to
simplify this.
Resume
E.L. Koua Representation et visualisation des series de donnees
geospatiale complexe: le cos des 'Self-Organising Maps(SOM)
Mots-cles: visualisation, exploration de donnees spatiales, Self-
Organising Maps
Les nouvelles techniques d'exploration des donnees spatiales
produisent des donnees innombrables dont l'analyse est dijficile en
raison de leur quantite, complexite, problemes d'echelle et leur
strueture, souvent cachee. Les methodes d'analyse traditionnelles
deduetives et statistiques semblent d'est pas etre appropriees pour
des series de donnees aussi larges, vu la complexite des attributs,
d'une part. Dans un tel environnement l'utilisateur doit pouvoir
faire appel, de plus en plus, l'organisation et l'appui
d'exploration des donnees par une simplification fiacilitant leur
prise en connaissance. Idealement, il devrait pouvoir explorer les
donnees spatiales dans n'importe quelle combinaison et
n Importe quelle echelle, en vue de decouvrir les relations et
struetures spatiales.
A cette fin la communaute des geo-informaticiens peut faire appel
d'autres diseiplines, teile que la visualisation d'information et
peut utiliser leur methodes d'une fagon plus efficace et intuitive,
aussi que Celles de la statistique et de l'intelligence artificielle en,
dans l'avenir, des reseaux artificiels neuvaux pour l'analyse de
donnees et de struetures. II s'agit surtout d'un modele de reseaux
appele SOM ('Self-Organising Maps'), une technique
d'exploration des grandes bases de donnees prometteuse.
L 'algorithme de SOM estpresente, dans l'article pour l'analyse des
donnees socio-economiques des communes de la province de
Overijssel pour en faire ressortir la struetures spatiales. Sont ainsi
combinees les techniques d'analyse spatiale, d'exploration et de
collecte de connaissance sous une forme visuelle, permettant de
decouvrir les struetures et de proposes des hypotheses. Les exemples
de visualisation utilisent des metaphores spatiales telles que
distances, regions et echelles, dans un but de simplification.
9
