Geoscience Reference
In-Depth Information
couple more decades. What it means is that you can develop and test new GC analysis tools using
HPC and be fairly confident in the knowledge that soon it will run on far less powerful and far more
available machines.
Another important development is to broaden the basis of the exploratory pattern search process
to include all aspects of spatial data (e.g. location in space, location in time and attributes of the
space-time event) and to make the search intelligent rather than systematic. Indeed, the added com-
plexity of additional data domains precludes a simple parallel brute force approach and emphasises
the importance of devising smarter search methods that can explore the full complexity of databases
without being too restricted. What we now need are geographical data mining tools. Only the most
primitive of methods have so far been developed due to a seemingly widespread distaste for induc-
tive analysis. That is a pity because this is exactly what the current era of massive data warehouses
and unbelievable spatial data riches require.
Openshaw (1994d, 1995b) describes the development of space-time attribute creatures, a form of
artificial life that can roam around what he terms the geocyberspace in an endless hunt for patterns.
The claim to intelligence results from the genetic algorithm used to control the search process and
the use of computational statistics to reduce the dangers of spurious results. It is strongly dependent
on having sufficient parallel computational power to drive the entire process. Openshaw and Perree
(1996) show how the addition of animation can help users envisage and understand the geographi-
cal analysis. This type of highly exploratory search technology is only just becoming feasible with
recent developments in HPC, and considerable research is still needed to perfect the technology.
More powerful computing is still needed but mainly in design and development of these methods
where they can dramatically speed up testing and be used to resolve design decisions via large-scale
simulation and animation of the behaviour of alternative algorithms.
1.7.8 B
uilding
g
eograPhical
k
nowledge
S
ySteMS
A final illustration describes HPC applications that are highly relevant to many areas of geography
but which are probably not yet feasible but soon will be. All the components needed probably exist
(a fairly common occurrence in GC research), usually in other contexts, and the trick is to find them,
understand them sufficiently so as to be able to safely use them, be bold enough to try them out and
have access to a sufficiently fast HPC platform to permit experimentation. Creativity is the name of
this game.
Consider the following view of the modern data landscape. Modern GISs have provided a
microspatial data-rich world, but there are no non-cartographic tools to help identify in any scientific
manner the more abstract recurrent patterns that may exist at higher levels of generalisation if only
we could
see
them. Geography is full of concepts and theories about space that can be expressed
as idealised 2D and 3D patterns that are supposedly recurrent. Traditionally, these concepts and
theories have been tested using aspatial statistical methods that require the geography to be removed
prior to analysis. For example, if you ask the question does the spatial social structure of Leeds as
shown by the 1991 census conform to a broadly concentric ring type of pattern?, then this hypothesis
can be tested by first defining a central point, then a series of three rings of fixed width, and then
a statistic of some kind is applied to census data to test the
a priori
hypothesised trends in social
class. However, this clearly requires considerable precision and is not really an adequate test of the
original hypothesis that specified no ring widths nor defined a central point nor defined at what level
of geographic scale the pattern exists. A possible solution is to use pattern recognition and robotic
vision technologies to see whether any evidence of a general concentric geographic structure exists
in the census data for Leeds, after allowing for the distorting effects of scale, site and topography.
If no idealised concentric patterns exist, then which of a library of different pattern types might be
more appropriate?
The old pre-quantitative geographical literature of the 1940s and 1950s contains spatial patterns
of various sorts that could never really be tested using conventional statistical methods. Moreover,
