Geoscience Reference
In-Depth Information
It makes sense of course to choose an encoding strategy that retains some of the essential character
of the underlying data, such as its ordering from smallest to largest. Brewer and Harrower provide
an excellent tool (http://colorbrewer2.org/), to preserve some of this essential character during the
assignment process when choosing colours for a choropleth map.
When the recoloured regions are displayed, these changed visual variables in turn produce
new patterns and trends in the display, which the user's cognitive system then tries to make sense
of. Assuming some pattern of interest emerges in the map, we might say that the data attributes
selected, when passed through the classifier, produce a stimulus that shows a geographical pattern.
Or in other words, perhaps that the data attribute has an interesting spatial distribution that could
be further investigated? Before assuming this, however, the user might also check that the pattern
is not a spurious artefact of the classifier used. No matter how careful we may be, it is a statistical
certainty that data exploration by fitting models always produces false positives.
The classifiers used to bin data are typically simple 1D methods that utilise the frequency distri-
bution of the data. They might divide up the range of data attributes so that the same number of cases
is assigned to each category, or so that the range of values in each category is consistent, or perhaps
so that the largest apparent gaps in the distribution are used to define the classes. A full account of
choropleth mapping and associated classification tools is provided by Slocum et al. (2008).
As an additional twist on this theme, many GeoViz systems support bivariate choropleth map-
ping, where two data attributes are used to determine the colour used for a region or symbol on a
map, by using two colour ramps comprised of opposite colours (such as blue and yellow or purple
and green) to each encode one data attribute, but then these two colours are combined together into
a single colour used on the map. The map legend then becomes a matrix, as shown at the upper right
corner of the map displays in Figures 5.2 and 5.5.
5.5.3 g eneric e xPloratory V iSualiSation w orkflow
If we abstract the preceding process beyond the choropleth example and express it as a workflow,
then we get the following:
1. Select data attribute(s) for visualisation.
2. Project just the fields of interest, normalise or scale the range of the data if needed and
tidy up or avoid any errors or nulls. Just as with regular statistical analysis, missing values
are likely to cause problems, since they can dramatically change the appearance of the
d isplay.
3. Visually encode the chosen data using some combination of visual variables and graphi-
cal devices that carry the visual variable (such as a symbol or a layer). A first step in this
process is often to cluster or classify the data, followed by visually encoding the outcome,
such as in the earlier choropleth example. Sometimes it may be helpful to include the same
data attribute in several graphs or doubly encode it for emphasis by using more than one
visual variable. A common example of the latter is to use both size and colour of a symbol
to encode a single data attribute (such as disease incidence) on a map.
4. Fine-tune this visual encoding function so that it preserves useful or interesting emergent
properties of the data.
5. Interact with the display to explore the resulting visualisation.
6. Validate any emergent patterns by mentally translating each pattern back into the data
domain to see if it makes sense and if it tells you something new. Also test the robustness of
any emergent patterns by subtly changing the visual encoding strategy used in #3, to check
that the pattern is not an artefact of visual encoding. The human visual system is designed
to find patterns and structure in a given visual stimulus, even when there is none present.
So caution is needed to reflect carefully on what you think you may see.
7. Repeat a s ne cessa r y.
Search WWH ::




Custom Search