Geoscience Reference
In-Depth Information
The process can be a bit daunting at first to those unfamiliar with the tools and theory. To make
good choices, the user must (1) have an idea as to what kinds of patterns they hope to explore and
which data attributes to use in order to see them; (2) understand the types of displays available to
them, the visual variables that they each provide and the kinds of discoveries and comparisons that
they can facilitate; and (3) choose sensible visual encoding and clustering strategies. As an example,
in a parallel coordinate plot, one can most readily compare values between neighbouring axes - it
is more difficult to compare values that are not immediate neighbours, but in a scatterplot matrix,
every variable can be compared with every other variable. Parallel coordinate plots work well for
comparing data attributes with a defined order, such as in a series or temporal sequence. In fact, the
stimulus produced is highly dependent on the ordering of the axes. Scatterplot matrices are a better
alternative if the aim is to search for pairwise correlations between attributes.
Turk (1994) provides a very useful taxonomy of approaches for the visual encoding of data,
reviewing a number of different ways by which the assignment of data to visual attributes can be
achieved. Some researchers have endeavoured to embed the kind of guidelines described previously
into intelligent visualisation systems that can make suitable choices for visual assignment based
on metrics and metadata (e.g. Beshers and Feiner, 1993; Gahegan and O'Brien, 1997; Senay and
Ignatius, 1998; Liu et al., 2005). A definitive set of rules for visualisation design that work in most
circumstances is still some way off, but in the meantime, the aforementioned guidelines can be
adopted as a starting point from which to construct a visualisation.
5.6 VISUALISATION AND THE PROCESS OF DISCOVERY SCIENCE
Recall from the previous section that the discovery process via GeoViz is one of repeatedly
posing hypotheses about the possible emergence of a pattern or structure, by changing the way
data are visualised and by exploring connections between different data patterns (possibly dis-
played in different graphs). Effectively, we discover by adding in some structure to the data and
then assessing if this structure helps to amplify themes or patterns of potential interest in the
scene.* If we begin with some kind of classification task to add structure, as described earlier
in the choropleth mapping example, then more formally, we recognise this as an application of
inductive reasoning . In inductive reasoning, we generalise the properties of individuals to create
categories that are useful devices for grouping the data and removing the confusion of dealing
with many data attributes with different values. We use the resulting classes as the basis for creat-
ing a new visual display in which the user searches for patterns of interest and then attempts to
translate these back into the domain of interest, that is, reasoning from a pattern observed (e.g. a
cluster of regions with a high unemployment rate) to an explanation that might support the pat-
tern (their proximity to factories that have shut down perhaps?). This latter form of inference is
called abduction - reasoning from an observation to an explanation. So, in GeoViz, the typical
inferential approach taken is to encourage abduction by first engaging in induction . Gahegan
(2005, 2009) provides more details of the reasoning processes and associated workflows used in
visualisation for discovery.
Where to begin visualising a new dataset and how to navigate through the seemingly endless set
of possibilities are major challenges. If we simply graph one variable after another - in the hope of
finding something of interest - we may well be disappointed or become bored, in which case we
may not notice a potentially interesting artefact. We are also likely to discover relationships that are
actually well known, even by ourselves! As an exploration activity, GeoViz must contend with a lack
of clarity around the discovery process in general: discovery in computational systems is still for the
* 'Sir Francis Bacon, in book II of his Novum Organum ( The New Organon ; 1620) states that “ Truth will sooner come
out from error than from confusion .” This famous epithet describes the idea that we understand the world by imposing
conceptual structure upon the confusion of data we receive. Our mistakes in doing so eventually lead us to a deeper
understanding.' (Gahegan, 2009).
Search WWH ::




Custom Search