Geoscience Reference
In-Depth Information
a highly
under-constrained
problem: meaning that there are a vast number of visual assignment
and graphing possibilities, with too few firm rules or constraints to narrow down the solution
to one or a few
good
visualisations. Current psychometric guidelines only get us so far, as do
more holistic approaches to design. Comprehensive, long-term, evaluations of effectiveness are
urgently needed. Two useful examples are provided by Andrienko et al. (2006) and Lloyd and
Dykes (2011), but more are required; in fact, evaluation is far more necessary at this stage than
the development of new visualisation tools. At this point, too little is known about the relative
merits of the different graphing methods, even the popular ones, and their relative utility for
specific visual tasks. The envisioned target is to have enough verified knowledge of perception
and utility for a specific task to be able to support the different stages of the spatial analysis with
the most appropriate visualisation tools, configured together and loaded with data in a visually
effective way.
The following four problems seem particularly pressing in this regard:
1. It is difficult to define the psychometric principles that a
good
visualisation should fol-
low. In very simple scenarios, existing perceptual and cartographic knowledge can be
applied with relative ease. However, the richness of the geographic domain (includ-
ing the spatial and temporal dimensions) and the large number of visual variables that
might be required concurrently point to the need for this existing knowledge to expand
(e.g. Mackinlay, 1986; Rheingans and Landreth, 1995; Ware, 2000; McCandless, 2009;
MacEachren et al., 2012).
2. There is a related problem of using acquired perceptual and cartographic knowledge to
good effect (Andrienko et al., 2006), that is, a system that can actually apply these rules
and guidelines on behalf of the user in the building of a scene or can warn a user when their
course of action will likely cause perceptual problems. Experimental systems that do so
have been suggested by several researchers including Duclos and Grave (1993), Senay and
Ignatius (1994) and Gahegan and O'Brien (1997). The previously mentioned ColorBrewer
tool (http://colorbrewer2.org/) is a good example of a successful recommender system that
will adapt to a given set of task-specific parameters to help the researcher make perceptu-
ally grounded choices on their use of colour. We need to extend ideas like this over more
visual variables and more display types.
3. A second issue is the need to embed or otherwise connect GeoViz systems with more
traditional analysis and modelling tools in a more seamless manner (e.g. Takatsuka and
Gahegan, 2002; Johansson et al., 2004; Andrienko et al., 2011). Most tools to this point
are either stand-alone or not closely integrated with existing GIS or spatial analysis tool-
kits. Closer integration would encourage greater uptake. A service-oriented approach,
based on open standards, would seem to offer a good way forward (Hildebrandt and
Döllner, 2010).
4. The task of evaluating the effectiveness of a scene is problematic (Carpendale, 2008).
Ideally, some quantification of the
utility
of a visualisation should be computable, as feed-
back to the preceding first and second problems. Effectiveness evaluation is complicated
because the results are ultimately evaluated by humans, whose judgements are subjec-
tive and can vary between individuals and over time. In order to make this much-needed
progress as a community, we require an agreed means of discerning the
good
from the
bad
. For example, how do we establish that one visualisation paradigm or graph type
works better than another for a given task (Kosara et al., 2003; Koua et al., 2007; Isenberg
et al., 2008)? Ideally, we would measure effectiveness as experienced by human operators,
requiring that we perform experiments, use control groups, questionnaires and so forth,
in the tradition of psychometric research. Even simple feedback on combinations of tasks,
graphs and visual encoding strategies that worked (or did not) would contain a wealth of
currently unrecognised information. For a user, providing a coarse evaluation could be as
