Geoscience Reference
In-Depth Information
Without some means of avoiding these strong distractors, it will be more difficult to make
genuinely new discoveries. Targeting the search to regions of the hypothesis space that
might prove most fruitful is a challenging and unresolved problem in any discovery sys-
tem, no matter what technologies it uses.
5.
Test that any finding is not simply an artefact
of the mappings used between the data
and the visual variables. This is the visualisation equivalent of avoiding Type I statistical
errors. For example, visual classifiers must create somewhat arbitrary bins into which to
place the data. The methods used can themselves introduce a
false positive
into the spatial
distribution of some visual variable in the display that would not occur with a slightly dif-
ferent classifier or with a different number of classes used. Some measure of the
stability
of a discovery gauged against these kinds of perturbations may help to avoid errors of
commission.
6.
Create some evidence of the visual discovery
and translate that evidence into follow-up
analyses that can be carried out using more traditional toolsets, such as correlation analy-
sis or spatial statistical tests. It is important to remember that the visualisation software
does not evaluate relationships in the data, but instead aids the user in identifying such
relationships for themselves. It is also often difficult to reproduce findings in current sys-
tems or to record them in a convenient manner so that they can be shared. Screenshots
usually fail to show the settings used and hence may not convey a description of the
hypothesis that the visualisation represents (see #3 and #4 earlier). Yang et al. (2007)
and Groth (2007) provide some useful ideas for addressing these shortcomings based on
provenance tracking.
7.
Keep the user engaged and alert
, since the user is an essential part of the workflow. If
there is not enough novelty in the various displays to maintain a good level of interest
and curiosity, then the whole enterprise will fail. Likewise, if the graphical devices used
are at odds with the visual tastes of the user, then they will not want to look. Perhaps the
overall effectiveness of data visualisation (Tufte, 1990) is sometimes overlooked in our
adherence to lower-level psychometric principles (Jones et al., 2010)? But it is difficult to
present information in a compelling way if you don't yet know what you are looking for!
Samara (2007) provides a very interesting perspective on visual design rules and when to
break them.
5.7 CONCLUSIONS AND CHALLENGES IN GEOVISUALISATION
GeoViz has come a long way, even in the time since the irst version of this topic was produced in
2000. With a greater number of visualisation tools available, a better understanding of the roles they
play and the beginnings of a reconceptualisation of the discovery process that features visualisation
centrally (Gahegan, 2005; Thomas and Cook, 2005; Andrienko et al., 2011), GeoViz is in a strong
position, going forward.
As has been shown earlier, GeoViz in the context of discovery science has much in common
with other methods that attempt the same ends: the problem is essentially one of search and the
challenges are to search a potentially massive hypothesis space effectively and then to ensure that
any findings are validated, preserved and communicated effectively. Recognition of this degree of
structure and formality in the process of GeoViz is a necessary step on the way to becoming an
established and legitimate scientific method. As I hope I have shown earlier, this journey is not yet
complete, but it has begun in earnest.
Many of the outstanding research questions relate to visual effectiveness: specifically, how to
make the best use of the available functionality to engage more successfully with the user. This
is a familiar problem to cartographers, but the large increase in visualisation possibilities raises
some difficult and as-yet-unresolved problems around choice of display types and visual encod-
ing strategies. The development of a visualisation (specifically the building of a scene) is often
