Towards High-resolution Self-organizing Maps of Geographic Features - Geographic Visualization: Concepts, Tools and Applications - page 160

using GPS or cell phones (see Chapters 15 and 16). One example of such convergence is the
extension of space-time paths to n -dimensional attribute space (Skupin, 2007).
Geographic visualization does not merely represent the reemergence of cartography, to
which GIS was once thought to have delivered a deadly blow. Instead, the confluence of
various academic and market forces towards formation of such new disciplinary categories
as information visualization and visual data mining and the success of products and services
like MapQuest and Google Earth suggest the emergence of a cartographic imperative ,a need
to map , aimed at making sense of voluminous, multi-facetted data. This imperative delivers a
powerful impulse to create meaning-bearing visualizations, even of non-georeferenced data
and of the non-spatial elements of geographic data. In order to achieve this, one is, however,
required to shed notions of cartography as being essentially about attaching symbols to
geometry in order to communicate geographic reality. On a much more fundamental level,
such approaches as spatialization remind us that cartography is all about transformation
(Tobler, 1979) and that the impact of visualization often derives from novel combinations
of transformative processes.
As reflected in this volume, geographic visualization at its core grew out of the cartographic
tradition of representing geographic objects in a representational space derived through
projection of locations from a curved two-dimensional space (i.e. latitude and longitude )
into a planar map space (i.e. x and y ). New techniques of representation have emerged, such
as parallel coordinate plots (PCP) and self-organizing maps (Kohonen, 2001), and are now
being linked to form increasingly powerful means for discovering interesting patterns and
relationships in large, multidimensional geographic databases. However, the geographic map
tends to be the element binding it all together, the one with which all other representations
interact and are bound to, and to which users will ultimately refer. One major reason for
this is the well-deserved recognition given to the possible effects of spatial autocorrelation -
or the First Law of Geography (Tobler, 1970) - in any such geographic investigation. In fact,
the effects of spatial autocorrelation, such as the appearance of spatial clusters, are the very
subject of many investigations. However, apart from such effects, one might argue that an
additional impetus for referring back to the geographic map is the sheer richness provided
by it. Ultimately, this richness derives not simply from a choice of symbols for point, line,
area and text objects - because that would apply to many non-geospace representations as
well - but from the finely grained geometric base to which such symbols become attached.
The inherent geometric detail or resolution provided by a geographic map tends to remain
unmatched by alternative representations.
Out of these considerations the theme of this chapter then emerges, to advance the conver-
gence of intense computation with the cartographic tradition, to spatialize an n -dimensional
geographic attribute space with a resolution detailed enough to mimic geographic maps,
and to apply a range of transformations towards eventual visualization. This is demonstrated
through a visualization derived from climate attributes associated with more then 200 000
US census block groups.
8.2 Self-organizing maps
First introduced a quarter-century ago, the self-organizing map has become a popular
method for visual modelling of complex, n -dimensional data. A number of excellent
overviews of the method, edited volumes, as well as a comprehensive monograph on the