Geoscience Reference
In-Depth Information
The introduction describes the connections between GC and discovery, followed by a section that
recaps the history and development of various approaches and provides definitions for the key
terminology used in later sections. Section 5.3 dissects the anatomy of a GeoViz display and intro-
duces some of the more common visual devices and metaphors. In Section 5.4, the perceptual and
cognitive issues underlying GeoViz are summarised, followed in Section 5.5 by an account of how
a visualisation is constructed, in the light of these perceptual and cognitive issues. A workflow-like
description of the main stages of construction is provided. Section 5.6 describes how GeoViz can
be used in the discovery process and notes seven challenges that must be overcome in order for
visually led discovery to be effective. A summary and list of outstanding problems are presented
by way of conclusions.
5.1 INTRODUCTION: GEOCOMPUTATION AND
INFORMATION VISUALISATION
When the term
GeoComputation
(GC) was first proposed in the 1990s by Openshaw and colleagues
(Openshaw, 2000), the focus was clearly on the analytical methods and techniques needed to allow
geographical analysis to scale up to bigger, or more complex, or more detailed problems, extend-
ing what was possible using existing analytical methods. In this sense, it actually foreshadows the
current interest in big data (e.g. Hey et al., 2009; Sui, 2014) that we see taking hold right across
the sciences and also in business and government. These new analytical methods were typically
directed at improved scalability, or dealing with noisy data, or situations where existing parametric
assumptions to analysis were found wanting (e.g. Gould, 1970; Gahegan, 2003). Since its intro-
duction, the meaning of the term
GeoComputation
has become somewhat broadened, perhaps a
casualty of the constant battle for conference attendance. But in this original sense, geovisualisation
(GeoViz) aligns well with the ethos of GC because both aim to avoid a priori assumptions about data
distributions and patterns where they are not needed or likely to be misleading or just plain wrong.*
Another shared aim between GC and GeoViz is to scale to large or complex multivariate datasets.
And while no analysis technique can avoid making some assumptions or scale infinitely, neverthe-
less the role that GeoViz can play here is a critically important one, which geographical analysis
will need to draw more deeply upon in the future. Within this chapter, the focus is placed on the
process
of GeoViz, to show the tasks involved in creating visual displays, the complexities in some
of the steps that can still cause problems, the kinds of inferences and insights gained and the nature
of the results produced. In other words, if we treat GeoViz as just another analysis toolbox, how do
we use it and how does it compare to other emerging GC approaches, such as machine learning or
a r tiicial life?
In contrast to McNoleg's (2008) tongue-in-cheek claim that GeoViz is '… the recycling of
surplus numbers into abstract art', GeoViz has by now demonstrated that it is much more than
window dressing at the end of the research enterprise and that it has a great deal to offer as a
method for data exploration and discovery science. For example, many compelling visualisation
success stories are regularly reported in the results of the International Science and Engineering
Visualization Challenge hosted periodically by the US National Science Foundation. This chal-
lenge has the tag line 'Science and Engineering's most powerful statements are not made from
words alone'
†
and its findings are reported in an accompanying special issue of the premier jour-
nal
Science
(2010).
As a simple example of the power of exploratory visualisation to grant insight, consider the
distributions shown in Figure 5.1. The four datasets graphed are known as Anscombe's quartet
(Anscombe, 1973). Each of these datasets has identical values - to two decimal places or better - for
the following descriptive statistics:
mean in x
,
variance in x
,
mean in y
,
variance in y
,
correlation
*
The same can be said of the machine learning methods that emerged as part of the same GeoComputation movement.
†
http://www.nsf.gov/news/special_reports/scivis/winners_2009.jsp.
