Geospatial Visualisation

6.
Geographic Visualization
Martin Nöllenburg
Geographic visualizations always played an important role in human history, es-

pecially in the earth sciences, long before computer visualizations became pop-
ular. The earliest examples of geographic visualization even date back to the
stone age with map-like wall paintings depicting the surroundings of our ances-
tors. Since then cartography, the art and science of map-making, has evolved
continuously until today. This is why computer-based geographic visualization
can build upon a large base of established cartographic knowledge. Well-known
examples of static visualizations beyond geographic maps are thematic maps
that display the spatial pattern of a theme such as climate characteristics or
population density. Moreover, the use of modern visualization technology offers
many new possibilities for geographical visualization tasks. These visualizations
may help to explore, understand, and communicate spatial phenomena.
Many readers will have a vague idea of what geographic visualization is about.
Nonetheless, to avoid misconceptions, the most common definitions of the term
geovisualization (short for geographic visualization) will be given. The following
notion according to the 2001 research agenda of the International Cartographic
Association (ICA) Commission on Visualization and Virtual Environments is
most widely accepted today: “Geovisualization integrates approaches from visu-
alization in scientific computing (ViSC), cartography, image analysis, informa-
tion visualization, exploratory data analysis (EDA), and geographic information
systems (GISystems) to provide theory, methods and tools for visual exploration,
analysis, synthesis, and presentation of geospatial data” [530]. Others take a more
human-centered view and describe geovisualization as “the creation and use of
visual representations to facilitate thinking, understanding, and knowledge con-
struction about geospatial data” [515] or as “the use of visual geospatial displays
to explore data and through that exploration to generate hypotheses, develop
problem solutions and construct knowledge” [474]. There are a few immediate
observations from these definitions. It is clear that geovisualization research is a
multidisciplinary task. Since it is the human who uses visualizations to explore
data and construct knowledge, effective geovisualization techniques must above
all take the user needs into account.
The chapter is structured as follows. First, the range of possible goals of
geovisualization and its driving forces are described in Section 6.1 and 6.2, re-
spectively. Then, Section 6.3 looks at some perceptual issues and theoretical
results in geovisualization. The main part of the survey, Section 6.4, covers a va-
254 M. Nöllenburg
riety of suitable visualization methods from map-based and abstract techniques

to animation. It also deals with interaction techniques and combined visual and
computational mining of geospatial data. A number of existing geovisualization
environments and tools are described in Section 6.5. Usability issues in the con-
text of geovisualization, namely user-centered design, results from user studies,
and geovisualization that supports group work, are reported in Section 6.6. The
final part of this chapter looks at current and future challenges in geovisualiza-
tion.
6.1 Goals of Geovisualization

The goals of geovisualization are manifold. The map use cube (Figure 6.1) by
MacEachren and Kraak [530] models the space of visualization goals with respect
to three dimensions:
– the task can range from revealing unknowns and constructing new knowledge
to sharing existing knowledge;
– the interaction with the visualization interface can range from a rather passive
low level to a high level where users actively influence what they see;
– finally, the visualization use ranges from a single, private user to a large, public
audience.
The four visualization goals exploration, analysis, synthesis, and presentation, as
identified in the International Cartographic Association’s research agenda [530],
are placed on a diagonal in this map-use space. On the one extreme, exploration
can be found as a private, highly interactive task to prompt thinking and to
generate hypotheses and ultimately new scientific insight. The other extreme
is formed by presenting knowledge in low-interaction visualizations to a wide
audience, e.g., on a professional conference or in a publication. DiBiase et al. [195]
described these two extremes as visual thinking which creates and interprets
graphic representations, and visual communication which aims at distributing
knowledge in an easy-to-read graphic form. The former task is exploratory, while
the latter one is explanatory.
In the beginning of the 1990s, geovisualization research focused on ex-
ploratory methods and tools. Communication was the realm of traditional car-
tography. Every map communicates a message by stressing certain aspects of the
underlying data. Cartographers, due to the lack of interaction in paper maps,
had the goal of finding an optimal map for the intended message. In exploration,
where the ‘message’ is yet to be discovered, there is no optimal map in the begin-
ning. The early focus on exploration has expanded recently to include the whole
range of visualization tasks as Dykes [220] observed. The reason is that sophisti-
cated interactive geovisualization methods are now recognized as useful not only
for exploration but also for presentation of knowledge through guided discovery.
The visualization experience [220] offers great benefits for understanding and
learning as it enables both experienced scientists and students to (re)discover
knowledge through interaction. The map is now frequently seen as an interactive
6. Geographic Visualization 255
–
sent public
– pre
size
nthe
s y
– Users
ly ze
na
e –a
lor
xp
–e private
k
con now
str led
uct ge
ion Ta low
sk ion
ract
inf
o Inte
sha rmat high
rin ion
g
Fig. 6.1. The map use cube after MacEachren and Kraak [530] characterizing geovi-
sualization objectives in a three-dimensional space by their level of interaction, their
audience, and the addressed tasks.
interface to access and explore geospatial data while it still retains its traditional
role as a presentational device [474]. Dykes argued that interaction appears to
be the key defining characteristic of geovisualization today and MacEachren and
Kraak [530] stated that geovisualization is characterized by interaction and dy-
namics. Concerning previously private tasks such as exploration, a shift from
individual use towards support of group work has been demanded in the ICA
agenda [530]. So recently, in terms of the map use cube, more research efforts
have been attracted by the high-interaction and group-use (or public) parts of
the geovisualization space.
6.2 Driving Forces of Geovisualization
So what is the reason for the increasing interest in geovisualization over the last
15 years? There are three driving forces for geovisualization.
The first is the rapid advances that have been made in graphics and display
technology. The availability of both low-cost 3D graphics hardware in personal
computers and the development of highly immersive 3D virtual environments re-
sulted in investigating the potential that these technologies have for visualizing
geospatial data. However, this emphasis on realism contrasts with the history of
cartography that points to centuries of successful abstraction making the world
easier to understand according to MacEachren [522]. Indeed, maps filter out
unnecessary details of the environment in order to highlight interesting infor-
mation. For example, a road map based on satellite images would be extremely
256 M. Nöllenburg
hard to use. The challenge is to study the relative advantages and disadvantages
of realism and abstraction in geovisualization and then, depending on the prob-
lem context, potentially integrate both abstract and realistic displays in a single
geovisualization environment.
The second driving force for geovisualization is the need to analyze and ex-
plore a dramatically increasing amount of geospatial data that are routinely
collected these days by a multitude of scientific and governmental institutions,
private companies, and individuals. This is due to an increasing availability and
decreasing cost of technology to acquire, store, and process these data. For ex-
ample, the location of a credit card purchase or a mobile phone call is recorded
by computers. A majority of the data, MacEachren and Kraak [530] estimated
up to 80 percent, contain geospatial references, e.g., coordinates of environmen-
tal measurements, census data, positions of vehicles, ships, planes, and parcels,
addresses of customers, etc. These data, often characterized by a high dimen-
sionality, are a vast source of potentially valuable information for research and
decision making, e.g., in studying disease incidence patterns, traffic flows, credit
card fraud, or climate change. Privacy issues with these data are an important
concern but they are out of the scope of this chapter. The large volume of many
data sets poses challenging problems for their exploration. While computers are
well suited for processing large amounts of data or finding well-known patterns,
they perform rather poorly in detecting and interpreting unknown patterns in
noisy data—at least in comparison to the human brain [399]. On the other hand,
with increasing data volume and complexity humans quickly reach the limit of
their capacities in analyzing raw numeric and textual data. The goal of geovi-
sualization is to combine the strengths of human vision, creativity, and general
knowledge with the storage capacity and the computational power of modern
computers in order to explore large geospatial data sets. One way of doing this
is by presenting a multitude of graphic representations of the data to the user,
which allow him or her to interact with the data and change the views in order
to gain insight and to draw conclusions, see Keim et al. [439].
Finally, the third driving force for geovisualization is the rise of the Internet
and its development into the prominent medium to disseminate geospatial data
and maps [474]. On the one hand, the Internet facilitates collaboration of expert
users at different places, which is one of the ICA Commission’s research chal-
lenges [530], and, on the other hand, it enables geovisualization applications to
address the public. Reaching the public is an important aspect both for govern-
mental agencies and for business companies who provide and sell services based
on geospatial information, see for example Steiner et al. [796] who developed
web-based tools to publish census data.
In general, there has been a shift away from technology-driven visualization
towards more human-centered approaches that base on usability engineering
principles and apply theoretical results from cognitive research as demanded by
Slocum et al. [776]. To exploit the full potential of geospatial data, geovisual-
ization tools need to adapt to their users. The question of what is a suitable
map or visualization method depends not only on the visualization task at hand
but also on the user’s background. According to Griffin [305], there are different
types of map readers who use geovisualization systems differently and who bring
different knowledge to the map reading process. Hence, this chapter will not only
present different methods and techniques applied in geovisualization, but also
focus on usability testing and user-centered design of geovisualization systems.
6.3 Cognitive Aspects

This short section describes a cognitive framework for visualization in general
and geovisualization in specific. The first part discusses visual thinking, i.e., how
human vision perceives maps and images and how it finds patterns. The second
part describes Bertin’s concept of graphic variables for traditional paper maps
and extensions for dynamic visualization techniques.
6.3.1 Visual Thinking

Visual thinking describes mental information processing through images instead
of words. DiBiase [194] saw the origins of the potential power of visual thinking
in the biological evolution where individuals with a quick reaction to visual cues
survived. While we communicate mostly through words, we are connected to our
environment primarily through vision. Hence, our visual perception has evolved
into a powerful system that actively seeks meaningful patterns in what we see,
sometimes even imposing them where they do not exist. Visual thinking often
does not follow a logical train of thought and hence has not been appreciated for
a long time in science. However, in 1990, MacEachren and Ganter [529] reported
a renewal of interest in human vision as a tool of advancing science.
Prominent examples of successful visual thinking are Wegener’s theory of
continental drift prompted by the similar shapes of the facing coasts of Africa
and South America, Kekulé’s ring-shaped model of the benzene structure, or the
discovery of the double helix structure of DNA by Watson and Crick stimulated
by an x-ray photograph of DNA.
MacEachren and Ganter [529] developed a cognitive approach to geovisual-
ization. They concentrated on the explorative side of visualization and thus on
gaining new scientific insight through visual thinking. In their article visualiza-
tion is seen in the first instance as a mental process rather than generating images
using computer graphics. Nonetheless, computer graphics are a valuable tool to
stimulate this mental process by creating graphics that facilitate visual identifi-
cation of patterns or anomalies in the data. Visualization should not primarily
focus on generating images but on using images to generate new ideas [194].
This also means that elaborate and highly realistic images are not necessarily
required to generate valid hypotheses. Instead it is often abstraction, in the past
achieved with pencil and paper, that helps to distinguish pattern from noise and
thus makes a map or some other graphic useful. One key aspect in visual data
exploration is to view a data set in a number of alternative ways to prompt both
hypotheses and their critical reflection.
258 M. Nöllenburg
Fig. 6.2. John Snow’s map of cholera deaths in London 1854. Deaths are marked by
c 1958 Blackwell Publishing.
dots and water pumps by crosses. Version of Gilbert [294]. °
Reprinted with permission.
An early example of how a cartographic picture was used to gain new insight
comes from medicine. In 1854 the London physician John Snow mapped cholera
cases to a district map and made the link between cholera and a specific water
pump that was used by the infected persons who ‘clustered’ around that pump
on the map, see Figure 6.2. In fact, it was reported [456] that an anomaly of
that pattern finally prompted his insight, namely the case of a workhouse with
very few infections in the center of the cholera outbreak: it had an independent
water source.
In order to successfully facilitate visual thinking it is necessary to understand
how the human mind processes visual information. MacEachren and Ganter [529]
and MacEachren [521] described visual information processing. Essentially, hu-
man vision produces abstractions from the complex input on the retina and these
abstractions are matched to the mind’s vast collection of patterns (or schemata)
from experience.
MacEachren and Ganter [529] proposed a two-stage model for interacting
with geovisualization tools in scientific exploration. At the first stage, called
seeing-that, the analyst searches for patterns in the visual input. They distin-
guished two types of pattern-matches: a pattern is recognized if it is expected in
the context; noticing, however, means detecting unexpected patterns that might
lead to new insight. Once a pattern is recognized or noticed the analyst enters
the second stage called reasoning-why, also known as the confirmatory stage of
scientific inquiry. At this stage, the judgment made before is carefully examined
to identify errors or to explain a pattern or anomaly. These two steps are iterated
to collect more evidence for or to modify a judgment.
Finally, when the scientist has confirmed a hypothesis, he or she will usually
want to share his insight with scientific peers through presentations and publi-
cations. Now, the goal is to lead fellow scientists to the same insight by invoking
their seeing-that and reasoning-why process through well-designed graphics. If
fellow scientists discover the patterns themselves the author’s arguments will be
much more convincing [529].
The model of MacEachren and Ganter implies that the success of a geovisu-
alization tool in scientific inquiry depends primarily on its ability of displaying
patterns that can be identified and analyzed by a human viewer [529]. How-
ever, individual users recognize and notice patterns differently based on their
individual experience. Hence, explorative geovisualization tools must be interac-
tive and permit a wide range of modifications to the visual display of the data.
In the reasoning-why process, a key to insight or error detection is examining
a judgment from different perspectives. Errors in pattern identification are di-
vided into two categories: Type I errors mean seeing-wrong, i.e., to see patterns
where they do not exist. Type II errors, on the contrary, denote not-seeing, i.e.,
missing patterns that are really there. Since human perception is adapted to
seeing patterns all the time humans are susceptible to Type I errors (e.g., seeing
shapes in clouds). Conversely, there is an effect known as scientific blindness, a
phenomenon describing the tendency to overlook what one is not actively search-
ing for. Consistency of patterns across multiple perspectives and modes is a cue
to a valid pattern while inconsistencies demand reconsidering or rejecting the
pattern [529].
6.3.2 Graphic Variables

On a map, information is usually represented by symbols, points, lines, and
areas with different properties such as color, shape, etc. Bertin’s concept of fun-
damental graphic variables for map and graphic design and rules for their use,
published as “Sémiologie graphique” [83] in 1967, has proposed a basic typology
for map design. This work was based on his experience as geographer and cartog-
rapher. For a discussion of Bertin’s graphic variables see also Section 4.1 in this
book. Since then his original set of variables has been modified and extended,
see MacEachren [521, 522]. Bertin’s fundamental graphic variables, namely lo-
cation, size, density/size of texture elements, color hue, color saturation, color
value, orientation, and shape, are means of communicating data to a map reader.
Especially the different variables of color have been studied with regard to their
efficiency in representing different kinds of data (categorical, sequential, binary,
etc.). See Brewer [111] for detailed guidelines on how to represent data by colors.
Variables can also be combined to represent the same information redundantly,
260 M. Nöllenburg
for example using both size and color value. This provides better selectivity and
facilitates the judgment of quantitative differences on a map in comparison to
using just one variable.
Originally, Bertin’s variables have been designed to describe information visu-
alization on paper maps. Today, advances in graphics display technology provide
a set of new graphic variables that can be utilized in geovisualization. Trans-
parency and crispness are regarded as static graphic variables, the latter for ex-
ample is suitable to represent uncertainty of some classification on a map [522].
However, geovisualization goes beyond static maps and therefore sets of tactile,
dynamic, and sonic variables have been proposed, e.g., loudness, pitch, dura-
tion, temporal position, rate-of-change, etc. Most of these variables are analogs
of graphic variables in another dimension, e.g., duration corresponds to size and
temporal position to spatial location. However, both dynamic and sonic variables
need to be observed over time and thus require more user attention than static
representations.
6.4 Visualization Methods and Techniques
A number of techniques and methods adapted from cartography and scientific

visualization are studied and applied in geovisualization. This section first in-
troduces geospatial data and their unique properties. Sections 6.4.2 and 6.4.3
discuss 2D and 3D map-based visualization techniques, respectively. A selection
of different techniques to display abstract multivariate attributes of geospatial
data is covered in Section 6.4.4. Animation as a dynamic visualization technique
that utilizes time as a visual dimension is the focus of Section 6.4.5. Spatial data
often contain temporal attributes and hence Section 6.4.6 discusses the display
of time, using animation as well as graphic variables like space or color. Interac-
tion is considered as one of the key characteristics of geovisualization. Interactive
methods and interfaces that control and link different visual representations are
introduced in Section 6.4.7. Finally, Section 6.4.8 deals with the combination of
visual and computational data mining techniques to explore very large data sets.
6.4.1 Geospatial Data
In contrast to information visualization displaying any abstract data, geovisual-

ization deals specifically with geospatial data, i.e., data that contain georeferenc-
ing. This is a unique feature and special methods and tools are needed to take full
advantage of it. Haining [323] decomposed geospatial data into an abstract at-
tribute space and a geographic space with two or three dimensions. According to
MacEachren and Kraak [530] geospatial data and information are fundamentally
different from other kinds of data since they are inherently spatially structured
in two or three dimensions (latitude, longitude, and possibly altitude). In case
of spatio-temporal data, time can be seen as a fourth dimension. Remaining di-
mensions are often unstructured and there are various visualization methods for
these abstract data, compare Section 4.3. Note also that distances and directions
have an immediate meaning in those dimensions in contrast to distances com-
puted on abstract data. While general visualization methods may be applied to
spatial data as well, they do not take into account the special characteristics of
the attributes. The georeference is usually either to a single point or to a whole
area. Whether geospatial data are defined as a point or as an area obviously
depends on the geographic scale at which they are examined. For example, a
village can be represented as an area on a large scale map, as a point on a map
of its province, and not at all on a country level.
6.4.2 2D Cartographic Visualization
The most common visualization method for geospatial data is a cartographic

display of some form, i.e., a map where the area under consideration is depicted,
and onto which the data of interest are plotted at their corresponding coor-
dinates. Space is used to depict space by mapping latitude and longitude to
the coordinate axes of the map drawing area. This might seem to be the most
natural way of using this graphic variable. However, there are good reasons of
linking for example population to space resulting in a cartogram1 where area
on the map is not proportional to a certain geographic area but in this case
to the number of people living in that very area. Still, cartograms usually try
to preserve the users’ mental map by keeping similar shapes and by preserving
the adjacencies between the depicted areas. An example of a world population
cartogram is shown in Figure 6.3. Tobler [832] gave an overview of algorithms
to automatically create contiguous value-by-area cartograms and van Kreveld
and Speckmann [853] studied drawing rectangular cartograms. Especially when
the focus of a map is on social, economic, or political issues, cartograms help
to draw the users’ attention to population as the map’s theme while avoiding
to emphasize large but sparsely inhabited regions. MacEachren [521] described
space as an indispensable graphic variable with a large influence on what the
user of a map sees. He argued that therefore space should represent the map
theme.
For both cartograms and geographic maps the interesting aspect is of course
how to depict abstract attributes of the data or at least a subset of them. Among
the most popular methods to represent categorical but also numerical data are
choropleth 2 maps. A choropleth map uses the graphic variables describing prop-
erties of color or texture to show properties of non-overlapping areas, such as
provinces, districts, or other units of territory division. A number of categories
is mapped to distinct colors or textures which are used to fill the areas ac-
cordingly. Examples are land cover/use with categories like forest, crop land,
housing, etc. or election results per district, e.g., displaying the percentage of
1
Among the most widely used cartograms are schematic maps of public transport
systems, e.g., the London Tube map [288] which emphasize the topology of the
transport network.
2
Greek from choros place and plethos magnitude.
262 M. Nöllenburg
Fig. 6.3. Geographical world map (top) and world population cartogram (bottom). In
the cartogram the size of a country is proportional to the size of its population. Image
courtesy of M. Newman.
votes for a certain party or the voter turnout as in Figure 6.4. For unordered
data well-distinguishable colors are needed while for ordered data it is impor-
tant to find a lightness or hue scale that represents the original range of numbers
efficiently, i.e., that the user can estimate values and differences from the colors.
Alternatively, a continuous range of attribute values is mapped to a continuous
color range without assigning values to a fixed number of classes. While choro-
pleth maps help to show general trends in the data there is certainly a loss of
information because the user cannot map a certain color to its exact numerical
value. Furthermore, a choropleth map can only express one or two attributes of
the data (by using a two-dimensional color scheme or by combining color and
texture). Andrienko and Andrienko [21] described a selection of methods to rep-
resent single and multiple attributes in a map. Depending on the type of the
attributes (logical, numeric, or nominal), they used bar and pie diagrams com-
mon in statistic visualization. Similarly, glyph-based techniques from visual data
mining can also be combined with map displays. These techniques, described in
more detail in Section 6.4.4, use compound glyphs to represent the values of
multiple abstract attributes. Using their geospatial reference, glyphs or statisti-
Fig. 6.4. A choropleth map showing turnout of voters in the 2005 federal elections in
Germany [127]. Image courtesy of Statistisches Bundesamt, Wiesbaden.
cal diagrams are placed on the map and thus both spatial and multidimensional
abstract attributes are represented on a single map. However, if the number of
symbols or attributes exceeds a certain limit the symbols become hard to com-
pare and other non-map based techniques from visual data mining should be
applied in addition to the display of a map or cartogram, see Section 6.4.4.
Other approaches for displaying high-dimensional data reduce the dimension-
ality of the data, e.g., by applying statistical techniques like principal component
analysis or by calculating compound indices representing, for example, the socio-
economic development of a region. The disadvantage, especially for explorative
visualization, is that through the loss of information potential patterns of some
attributes might get lost.
6.4.3 3D Cartographic Visualization
In contrast to traditional paper maps and two-dimensional visualization meth-

ods, geovisualization can go one step further and use the potential of increas-
ingly experiential representation technologies. 3D visualization includes the full
range from regular 3D graphics hardware in desktop computers to immersive
3D displays, CAVEs (Cave Automatic Virtual Environments), and Power Walls
providing stereoscopic views. For an overview on display technology, the reader
264 M. Nöllenburg
is referred to Section 3.2. Since humans live in a three-dimensional environment

our perception and cognition is adapted to processing 3D visual stimuli. But
there is still little known about when 3D visualization is appropriate and how it
can effectively enhance visual thinking.
Cartography has a long and successful tradition using abstraction to depict
a wide range of data on maps. In contrast, the focus of computer graphics tech-
nology is on producing increasingly realistic images and virtual environments.
Virtual reality techniques are widespread, for example, in architecture and land-
scape planning where realism is very important. Depending on the geovisual-
ization task, realism can be a distraction and insight is more likely when using
abstract visual symbolism. But as MacEachren et al. [527] pointed out, there
had been only few efforts exploring abstract visualizations of geospatial data in
3D.
In terms of the 3D representation of the data, MacEachren et al. [527] dis-
tinguished between using the three dimensions of the representation to display
the three dimensions of physical space, using one or two dimensions for non-
spatial data, e.g., income or time, and using all three dimensions for abstract
data. Representing time as the third dimension is common and will be discussed
in Section 6.4.6 on spatio-temporal visualization.
Today, the most widespread use of 3D is at the level of visual representation
while the display is a 2D screen. It is important to be aware of the implications
that the projection of a 3D representation onto a 2D plane has. Depth cues such
as perspective and occlusion also cause problems because distances are harder
to estimate, and occlusion hides objects depending on the viewpoint. Ware and
Plumlee [886] observed that due to occlusion humans cannot perceive much in-
formation in the depth direction while the x- and y-directions, orthogonal to the
line of sight, can convey complex patterns. A set of interactive navigational con-
trols are necessary to move within the 3D representation, e.g., zooming or flying,
cf. Section 3.1 of this book or Bowman et al. [102]. As Wood et al. [912] pointed
out the effectiveness of the virtual environment metaphor relies to some extent
on navigational realism. While moving (e.g., walking or flying) slowly through
the visual space maintains a sense of orientation, faster modes of movement such
as teleporting lose the context and the user has to reorientate.
6.4.4 Visual Data Mining Tools

Visual data mining, also denoted as exploratory data analysis (EDA) in statis-
tics by Tukey [843], is a human-centered task that aims at visually analyzing
data and gaining new insights. This contrasts computational data mining tech-
niques which use algorithms to detect patterns in the data. Effective visual data
mining tools need to display multivariate data in a way that the human viewer
can easily perceive patterns and relationships in the data. Visual data mining
in general is not tailored specifically for geospatial data. See Section 4.3 for a
general discussion of visualizing multivariate data. Since geospatial data usually
have many abstract attributes (cf. Section 6.4.1) these general techniques can be
applied for displaying non-spatial attributes of the data. Visualization techniques
aj
...
ai a1 a2 a3 a4 ... an−1 an
(a) (b)
Fig. 6.5. Example of a 2D scatter plot in subfigure (a) and a parallel coordinate plot
in subfigure (b).
for multivariate data were broadly classified as geometric, glyph- or icon-based,

pixel-oriented, and hierarchical by Schroeder [745] and Keim and Kriegel [438].
In a geovisualization context, geometric and glyph-based techniques are most
common. Graph-drawing techniques that depict relationships between individ-
ual data items are also covered in this section.
Geometric Techniques Two geometric techniques commonly used in geovi-
sualization are scatter plots and parallel coordinate plots. Scatter plots in their
basic two-dimensional form depict objects as points in a coordinate system where
the axes correspond to two selected attributes, see Figure 6.5(a) for an exam-
ple. Elements in the data set with similar values in these attributes form visual
clusters in the scatter plot. The idea of a scatter plot can be extended to three
dimensions but then phenomena as occlusion and different perception of depth
as described in Section 6.4.3 may occur. The extension to more than three di-
mensions is often implemented by drawing a scatter-plot matrix containing one
scatter plot for each pair of attributes. This, however, makes the identification
of multidimensional patterns difficult because many plots in the matrix need to
be linked mentally.
Parallel coordinate plots (PCP) [388] are a means of displaying high-
dimensional data in a single plot. In a PCP, one dimension is used to place
multiple parallel axes, each of which represents one attribute of the data. Each
element of the data set is then characterized by the values of its attributes which
are connected along the axes and thus build a geometric profile of that element as
depicted in Figure 6.5(b). Since all elements are plotted in this way, the user can
identify similar objects by comparing the geometric shape of their profiles. How-
ever, depending on the number of profiles, overplotting occurs and may result in
poor legibility. Keim and Kriegel [438] estimated that about 1,000 items could
be displayed at the same time. Moreover it becomes difficult to compare profiles
based on an increasing number of attribute axes. Another important aspect of
PCPs is the order of the attributes plotted along the parallel axes since this
order has a strong influence on the shapes of the profiles. Hence, a user should
be able to rearrange the attributes manually or based on sorting algorithms.
266 M. Nöllenburg
Glyph-Based Techniques Glyph-based or icon-based techniques use a map-

ping of multiple attribute values to a set of different visual features of a glyph
which in turn represents one data object. Two examples of such techniques are
Chernoff faces [154] and star plots [253]. In a Chernoff face, different variables of
the data are related to facial features of an iconic face, such as size and shape of
mouth, eyes, ears, etc., see Figure 6.6(a). The motivation of using faces to depict
multidimensional data is that human mind is used to recognize and compare
faces. However, different features, e.g., shape of the eyes and area of the face,
are hard to compare and Chernoff faces are, in contrast to human faces, not
perceived pre-attentively [574] such that there is no advantage over other types
of glyphs.
Star plots depict the value of a set of attributes by the length of rays ema-
nating from the center of the glyph. The endpoints of the rays are connected to
create a closed shape as depicted in Figure 6.6(b). While the maximum number
of facial features in Chernoff faces is reached quickly, star plots can display data
with higher dimension by increasing the number of rays. Again, as for parallel
coordinate plots, the order of the attributes influences the shape of the star plots.
A nice property of glyph-based techniques is that they can be easily combined
with map displays described in Section 6.4.2 by placing each glyph according
to its geospatial coordinates on the map. However, with an increasing number
of symbols or attributes glyph-based techniques are of limited use due to the
difficulty of visually recognizing patterns and distinguishing features on a display
with too many glyphs or glyphs with too many features.
Graph-Drawing Techniques Geospatial data often contain links between re-
lated elements, e.g., routes, trade connections, etc. Exploring such data sets
includes the search for patterns in the link structure between items. Data con-
taining relationships between elements are mathematically modeled as a graph
consisting of a set of nodes, the data elements, and a set of (weighted) edges,
the links between elements. The research area of graph drawing provides a mul-
titude of algorithms for visualizing such graphs, see di Battista et al. [192].
Section 4.4 discusses visualization of graphs from a human-centered perspective
and Rodgers [717] gives an overview of graph drawing methods and their ap-
plication to geovisualization. In general, for graph drawing the emphasis is on
finding a layout, i.e., positions of nodes and edges of a given graph that satisfies
certain aesthetic criteria, e.g., few edge crossings. In geovisualization, there are
usually certain constraints on such a layout since nodes already have a spatial
location. In that case, finding a legible layout for the edges is of interest, for
example in schematizing road networks [137]. In other cases, such as drawing
metro maps [362, 618, 803], the network topology is more important and node
positions are only required to satisfy certain relative positions (e.g., left, right,
above, below) in order to preserve the user’s mental map. Finally, some data is
best analyzed by putting no restrictions to node positions and using a general
algorithm to find a graph layout in which link patterns can be identified visually.
Such methods are applied in visual social network analysis, see for example Bran-
des and Wagner [108]. In the latter cases, where node positions are modified, a
Latin America
375
100
(a) Former Soviet N. America
305 443 583
Asia
W. Europe 447
157
136 104
Africa Middle East
(b)
Fig. 6.6. Two examples of glyph- Fig. 6.7. A graph showing trade relationships.
based techniques: subfigure (a) Edges are weighted by trade volume and drawn
shows two Chernoff faces and sub- shorter and thicker with increasing weight.
figure (b) shows two star plots on Only edges with at least 100 billion dollars
six attributes. trade volume are shown.
map display of the true geography in combination with a graph layout focusing
on the link topology is helpful for identifying both spatial and link-based pat-
terns. An example by Rodgers [717] visualizing trade volume between regions of
the world as a graph is shown in Figure 6.7. The stronger a trade relationship
between two regions the more they attract each other in the graph layout.
6.4.5 Animation
The methods described so far are primarily static displays of geospatial data.
They can all be printed and studied on paper. In geovisualization, however,
dynamic and interactive displays are core concepts today. Since the 1930s car-
tographers are experimenting with map movies and animated maps. Leaving
interaction aside, which is covered in Section 6.4.7, animated maps are using
time to add another visual dimension to the display. It is intuitive to relate time
to time, just as space depicts space in most maps [333]. In this case, the time
period of the data is mapped to the animation time. Each scene or frame of the
animation shows the state of the data at one moment accordingly. Thus, the tem-
poral change of the attributes becomes visible. It may be necessary to smooth the
animation using interpolation to generate intermediate frames. Scenes can also
be reordered from chronological order to an order based on attribute values. This
may be helpful for studying at what points of time events with similar properties
took place. For example, earthquake data can be ordered by the number of hu-
man fatalities in such a way that the beginning of the animation shows the least
and the end the most catastrophic earthquakes [195]. Animation can also be used
to display spatial features, e.g., animations of flights over the terrain. In other
268 M. Nöllenburg
cases, the temporal dimension is used to display quantitative attributes by map-

ping their values to the blinking frequency of symbols or to highlight classes in a
choropleth map by blinking. The presence of the temporal dimension in dynamic
visualizations also introduces the potential to use acoustic variables. Although
visualization is mostly concerned with visual aspects, using sound to comple-
ment dynamic graphics expands the possibilities of visualization. Krygier [480]
summarized the role of sound in geovisualization from attracting attention over
narrative voice and representing quantitative attributes (e.g., using the variable
pitch) to sound maps for visually impaired.
Bertin’s notion of visual variables [83], as introduced in Section 6.3.2, has
been extended to dynamic animated displays, see MacEachren [522] and Chap-
ter 6 of MacEachren [521]. Six dynamic variables are suggested: (1) temporal
position, i.e., when something is displayed, (2) duration, i.e., how long something
is displayed, (3) order, i.e., the temporal sequence of events, (4) rate of change,
e.g., the magnitude of change per time unit, (5) frequency, i.e., the speed of the
animation, and (6) synchronization, e.g., the temporal correspondence of two
events.
Animation and its set of additional visual variables represents a powerful
tool for map designers. However, animation should not be used carelessly and
it is always worth asking “why do I need to animate these data?” [334]. From
a user’s perspective, things that change on a map attract more attention than
the static background and moving objects attract more attention than objects
that appear and disappear. The fact that animated displays change over time
has an important disadvantage, especially if there are no interactive controls:
there is always the risk that the user will miss important information. While a
static display can be analyzed at an individual speed, an animation has to be
followed at the predefined pace. Therefore, it is hard to compare data displayed
at different points of time as the human brain usually forgets most details of
previous frames of an animation. A study by Rensink et al. [705] showed how
difficult perceiving changes in an animation can be.
Harrower [334] gave some guidelines for designing effective animated maps.
Because dynamic variables attract more attention than static variables the in-
formation conveyed by static variables should be kept simple, e.g., by using a
choropleth map with only few data classes (high, medium, and low ) and a rather
low level of detail. Details are often more effectively displayed using static vi-
sualization techniques. Temporal exaggeration, i.e., displaying durations not to
scale, is often necessary since otherwise a short event, e.g., an earthquake, in
an animation spanning several decades will be missed by the viewer. Directing
the user’s attention to critical events can be done for example by initially flash-
ing new symbols on the map. In general, animated maps are better suited to
depicting geographic patterns, e.g., growth or shrinkage of an area, rather than
specific rates of change according to Harrower [334]. Finally, he observed that
people are less confident with animated maps than with static maps, due to less
experience and training.
Many of the above problems can be avoided by giving the control of the
animation to the user. In interactive animations, where the user can control the
displayed level of detail and the speed of the animation, information is less likely
to be missed and users feel more confident with the animation. Still, the study
of dynamic displays with regard to their geospatial expressiveness is identified
as one of the challenges in the ICA research agenda [530] and further usability
studies are required, see Section 6.6.
6.4.6 Spatio-Temporal Visualization

Spatio-temporal data are very common in the earth sciences and related disci-
plines. The goal of many studies is to reveal, analyze, and understand patterns
of temporal change of phenomena, such as global warming, population devel-
opment, or spread of diseases. The previous section has presented animation as
a means of displaying temporal data. Animation works well in displaying pat-
terns if they are based on the same temporal sequence as the animation itself,
e.g., showing trends like urban growth over time. However, Andrienko et al. [26]
criticized that for less evident patterns it is necessary to compare the data at
different points in time which involves memorizing a large number of states in
an animated display, even if interactive controls allow to pause, jump, and step
through specific points in time. Thus it might be more effective to statically
display selected moments in time simultaneously using small multiples. Then,
an analyst can directly compare attribute properties of different points in time
at his or her own speed. However, the number of simultaneous images on the
screen is limited and long time series have to be evaluated piecewise. Andrienko
et al. argue that, for all these reasons, spatio-temporal data exploration must be
supported by a variety of techniques, possibly in combination with an animated
display [26].
Andrienko et al. [25,26] classified spatio-temporal data according to the type
of temporal changes: (1) existential changes, i.e., appearance and disappearance
of features, (2) changes of spatial properties, i.e., change of location, shape, size,
etc., and (3) changes of thematic properties, i.e., qualitative and quantitative
changes of attributes. Following this classification, they presented corresponding
visualization techniques. All techniques involved a map display to visualize the
spatial attributes of the data.
Data of existential changes usually consist of events or observations at spe-
cific moments or time periods during which a certain property holds, e.g., road
congestion data. Hence a map showing these data always considers a selected
time interval. If data items are represented by glyphs, one way to display the
time associated with them is by using textual labels. Another possibility is using
a color scheme to represent the age of the data. A 3D representation of space and
time is a third and common method [521]. In such a space-time cube, the third di-
mension corresponds to time while two dimensions represent geographical space.
The reference map is usually displayed in the coordinate plane corresponding to
time 0 and data items are positioned above the map depending on their spatial
locations and their times of appearance. An example of a space-time cube is
270 M. Nöllenburg
Fig. 6.8. A space-time cube visualization of Napoleon’s march in Russia. Image cour-
tesy of M.-J. Kraak.
shown in Figure 6.8. It shows the trajectory of Napoleon’s troops during the
French campaign against Russia in 1812.
For data that contain moving objects, comparing object trajectories is of
interest. Static 2D maps are able to show the trajectories of a small number of
objects but in this simple form it is not possible to evaluate aspects like speed
or whether two objects met at a crossing or just visited at different points in
time. Andrienko et al. [25] suggested animating object movements, either as a
sequence of snapshots in time in which at each moment objects are shown at their
current positions or using the movement history and showing the trajectories up
to the current point in time. Movement history can optionally be limited to a
specified time interval. It was found that the snapshot technique was suited for
a single object while several objects were better observed displaying also the
movement history. MacEachren [521] suggested using the space-time cube to
display trajectories which avoids the disadvantages of 2D trajectories mentioned
above as it shows when and not just if an object visited a point.
There are several methods of displaying thematic attributes on a map. A
very effective and common method is the choropleth map, see Section 6.4.2.
Animating a choropleth map is able to give a good overview of the values in
a selected attribute. However, it is difficult to estimate trends in a particular
Fig. 6.9. Cartographic representation of the spatial distribution of the burglary rates
in the USA. Image courtesy of G. Andrienko.
area on the map or to compare trends between different areas [26]. Change
maps [25], adapted from conventional cartography, use the choropleth map to
show the differences of an attribute between two selected points in time. Mapping
increase and decrease in attribute value to shades of two different colors allows
to evaluate regional changes for two moments. Such a map is restricted to two
points in time and the map can be misleading because information on the actual
attribute values is lost, e.g., concerning crime data two areas can have very
different burglary rates but still be colored the same if the rates both decrease by
the same value. Andrienko and Andrienko [23] combined time-series graphs with
maps to avoid these disadvantages. A time-series graph is a two-dimensional plot
of the temporal variation of attribute values, where time is represented on the x-
axis and attribute values on the y-axis. Plotting all data in the same time graph
gives an overview of the dynamics of the whole data set. To assess local behaviors,
Andrienko and Andrienko plotted the time-series data individually for each area
on the map and used the closed shape of the plot as a symbol superimposed on
each area similarly to the glyph-based techniques in Section 6.4.4, see Figure 6.9
for an example. This technique allows to evaluate changes and actual values of
an attribute for the whole time period under consideration. The user can explore
both spatial patterns and patterns in the attribute space in the same view.
Shanbhag et al. [757] presented three techniques that modify choropleth maps
in order to display temporal attribute data. They did not color each district area
in the map uniformly but partitioned it into several regions, each representing
one point of time in the data. Their first technique builds on a cyclical, clock-
like metaphor for perceiving time and partitions the area polygon into wedges.
The second technique draws from the metaphor of annual rings of a tree trunk
and assigns time points to ‘rings’ of the polygon. Finally, they suggested time
slices for a linear perception of time, i.e., polygons were partitioned into vertical
slices. Using any of the three techniques, temporal trends in each area could be
detected by observing the variation (e.g., brightness) of the different regions of
the district area. However, for an effective visualization the number of simulta-
neously displayed time points must be limited with respect to the size of the
polygons in order to avoid clutter [757].
272 M. Nöllenburg
For detecting periodic temporal patterns, Hewagamage et al. [348] suggested

using spirals to depict time in a 3D representation similar to MacEachren’s
space-time cube [521]. They used this technique to display events, i.e., data with
existential changes like the sight of a bird at a specific time and place. Depending
on the semantics of the data, events often show some periodic appearance pat-
terns, e.g., bird migration depends on the season and observed birds may rest at
a certain place every year. Hewagamage et al. took a linear time line and coiled
it such that one loop of the resulting three-dimensional spiral corresponded to
a user-specified time interval, e.g., a year or month. At each location of inter-
est such a spiral was positioned, and the events at that position were placed as
small icons along the spiral. Thus, points in time whose temporal distance was a
multiple of the selected time period were vertically aligned on the spirals. Since
parts of the spirals were occluded the display needed to have interactive controls
for zooming and panning, as well as for changing the period of the spirals.
6.4.7 Interactive User Interfaces

Interaction is paramount in geovisualization, especially for visual exploration,
recall the map-use cube in Figure 6.1. The communication aspect of geovisu-
alization is also shifting towards higher levels of interaction. Dykes introduced
guided discovery as a communication task [220]. For example, consider a student
who is learning by interactively (re-)discovering known relationships in a data
set. Dykes saw interaction as the key defining characteristic of visualization early
in the 21st century and according to MacEachren [522] interaction is a key factor
distinguishing geovisualization from traditional cartography. He continued that
geovisualization “is an active process in which an individual engages in sorting,
highlighting, filtering, and otherwise transforming data in a search for patterns
and relationships” [522]. In this sense, the present section focuses on common
interaction principles and interactive user interfaces in current geovisualization
systems.
Map use in traditional cartography, too, has been benefiting from interaction
albeit to a far lesser degree than current geovisualization. For example, drawing
with a pencil or using colored pins to mark spots on a paper map is a way of
interactively changing the map. For centuries using these and similar techniques
were the only way of exploring geospatial data. However, such manual interaction
techniques are time consuming and often give only very limited insight into
the data. Generating individual maps on demand had to be done manually by
cartographers and at a prohibitive cost for most explorative purposes. This has
changed drastically in computer-based geovisualization.
Buja et al. [126] introduced a taxonomy for general interactive multivari-
ate data visualization. In the following, interactive visualization techniques are
grouped using this taxonomy. Their two main classes are focusing individual
views and linking multiple views. For a detailed description of interaction meth-
ods see Chapter 3 of this book.
Focusing By focusing Buja et al. [126] mean any interactive modification that
selects what to see in a single display and how it is seen. They compared focusing
Fig. 6.10. Screenshot of Google Earth [300].
to operating a camera: choosing a perspective, deciding about magnification

and detail, etc. The individual visualization methods introduced in the previous
sections all profit from one or the other form of interaction—as static displays
most of them are of very limited use. Two- and three-dimensional maps (see
Section 6.4.2 and 6.4.3) usually come along with a set of navigational controls to
move within the map space either by scrolling, shifting, or rotating a map or by
walking or flying through a virtual 3D environment. Further controls allow users
to zoom in or out of the map and thus to increase or decrease the level of detail
and the amount of generalization of the underlying geographic features (rivers,
cities, lakes, etc.). Once a perspective on the map is chosen, its appearance can
be modified further, depending on the type of map. The display of geographic
background features such as rivers, mountains, roads, etc. can be switched on
or off. Depending on the user’s task such surface details may be informative
or distracting. In Figure 6.10, a screenshot of the satellite image viewer Google
Earth3 [300] is shown. With the navigational controls in the bottom of the figure
the user can move around and change the scale. The check boxes on the left can
be used to hide or show information like roads or borders.
The next mode of interaction concerns the way that the actual data items are
displayed. For the general case of multivariate data the user must be able to select
the attributes of interest and the type of their visualization, such as choropleth
map or glyph-based map. Deselected attributes as well as the actual numeric
values of selected attributes should be accessible, e.g., as tool tip information
when moving the mouse over a symbol or area on the map. Dynamic isolation
was used by Andrienko and Andrienko [21] to highlight a subset of data items
in maps that use different symbols to depict different classes of data items, e.g.,
to select only coal mines in a map of natural resources. Spatial patterns may be
3
http://earth.google.com
274 M. Nöllenburg
cluttered in a map of all resources while they become evident once dynamically
isolated.
Modification of the color scheme in a choropleth map is another important
aspect. Assigning colors to classes or color ranges to numerical ranges is a very
simple way of interacting with choropleth maps. While traditional paper maps
use well-established color schemes that are suitable for the majority of map
users these schemes may still not be optimal for an individual analyst and hence
should be adaptable. In an unclassed map, i.e., numbers are mapped to color
shades directly, Andrienko and Andrienko [21] used dynamic visual comparison
to display attribute values with respect to a number N in the value range. To
this end they applied a diverging color scheme which maps N to a neutral color,
e.g., white, values lower than N to shades of red, and values higher than N to
shades of green. The greater the difference the darker the color. This is a way
to visually split the areas on a choropleth map. The reference value N can be
selected by clicking on an object in the map to select its value or by using a
slider unit representing the value and color range. Statistic values like mean,
median, or quantiles are typically used as reference value. In classed maps the
value range of an attribute is partitioned into a set of intervals and a distinct
color is assigned to each interval. Here dynamic classification was applied, i.e.,
interactive modification of the underlying classes by shifting interval boundaries
or changing the number of classes.
Finally, the mapping of a numeric value range to color shades in a choropleth
map or to height in bar maps can be modified. This mapping is usually a linear
function and hence outliers in the value distribution may cause a poor color reso-
lution for the bulk of values. To reduce this effect Andrienko and Andrienko [21]
used dynamic focusing to map the full scale of colors to a subrange of values
with the outliers removed. Values to the left or right of the subrange are simply
colored by the minimum or maximum color value. Using non-linear functions to
map values to colors, as it is common for statistical graphics, is undesirable for
choropleth maps as visual analysis greatly depends on the immediate expres-
sion [21]. Similar colors would no longer necessarily reflect numeric closeness of
values. For scatter plots, however, non-linear transformations (e.g., logarithmic
scale) are acceptable if the goal is to find a functional relationship between two
attributes.
As mentioned in Section 6.4.4, visual tools like the parallel coordinate plot
(PCP) or glyph-based techniques depend on an order of the attributes. Ordering
the axes in a PCP or star plot and assigning facial features to attributes in
Chernoff faces has an important effect on the geometric shape of PCP profiles
and glyphs. Giving users the opportunity to change these properties allows them
to see the same data from multiple perspectives. Concerning outliers and visual
resolution, similar arguments and solutions as for mapping a value range to a
color range in choropleth maps apply for the scale of PCP axes.
For animated maps there is a need to give control of the animation to the user.
Even simple animations have VCR-like interfaces to control the animation speed,
to pause the animation, or to select individual scenes. Harrower [334] described
these abilities to navigate in time as equivalent to navigating in space, i.e.,

zooming and panning on static maps. Tools to navigate in time vary from linear
time lines with sliders, which the user can move to select points or intervals in
time, to more sophisticated devices like the time wheel4 . The time wheel supports
a cyclic view of time and consists of three concentric circles, the innermost
representing hours, the next one days, and the outer circle represents months.
In each circle, the respective time units can be marked and thus it enables to
select recurring periods in time, e.g., the hours between 7 and 9 p.m. on the first
five days of January, April, and May.
Linking and Brushing The full potential of interaction in geovisualization lies
in linking multiple views of the same data on the screen which is the defining
criterion of the second class of Buja et al.’s taxonomy of interactive multivariate
visualization [126]. See also Section 4.5 which discusses multiple and integrated
views from a more general perspective. Linking basically means simultaneous
highlighting of data items in multiple views. It is usually combined with brush-
ing, i.e., selecting display objects by pointing on them or encircling them on
the screen. Brushing was originally used for scatter-plot matrices, where points
highlighted in one scatter plot are simultaneously highlighted in the other plots
of the matrix to evaluate for example whether a relationship in two attributes
also holds for other pairs of attributes. Monmonier [568] extended this idea as
geographic brushing and links a map with scatter plots or other visual data min-
ing tools. An example of linking and brushing is shown in Figure 6.11, where
an outlier point in the scatter plot of per-capita income and percentage of poor
is selected. The map shows the geographic location of the corresponding county
New York, which has the highest average income in the USA. The PCP in the
bottom of the screenshot allows to evaluate the remaining statistical attributes
of New York.
Different views should be linked in a geovisualization system. Highlighting
a point cluster in a scatter plot or a cluster of PCP profiles thus shows the
spatial pattern of the corresponding objects in the associated map view. If there
is, for example, a set of objects that are visually similar in a PCP the analyst
might ask whether these object are located in the same region on the map or
whether they are spread over the whole country but only in rural areas etc.
Consequently, outliers that deviate from a general pattern can be located and
examined more thoroughly. Similarly, one can select spatial object clusters on
the map and subsequently analyze their behavior in attribute space using the
remaining views. If the map is a choropleth map then linking and brushing can be
done automatically using the same color scheme for the representations of objects
in all views. Then, the analyst can mentally connect multiple views because the
classes of the choropleth map are marked identically in all views. Changing the
class assignment in the choropleth map immediately updates the linked displays.
Conversely a scatter plot can be used to define new class boundaries for the map.
A very simple example is the cross map [521]. It divides the value ranges of both
4
Demonstration applet by Rob Edsall available at http://www.geovista.psu.edu/
products/demos/edsall/Tclets072799/cyclicaltime.htm.
276 M. Nöllenburg
Fig. 6.11. Screenshot of GeoVISTA Studio showing linking and brushing between
a scatter plot, a map view, and a PCP. The red item marked in the scatter plot is
simultaneously highlighted in the other views.
attributes in a scatter plot into a lower and upper range thus defining four
classes, each of which is represented with a distinct color on the map. Changing
the class assignment is as simple as moving the class-break point, i.e., the center
of the cross separating the four classes in the scatter plot.
The possibilities of interacting with multiple linked views through highlight-
ing and brushing are numerous. The number, type, and arrangement of the views
depend on the specific geovisualization task, the individual user, and the avail-
able space on the screen. In any case, it is linking and brushing that make the
use of multiple views more than simply the sum of its parts. The interactive
principles introduced in this section all concern a core aspect of geovisualiza-
tion: stimulate visual thinking by presenting the data in different ways and from
a large number of perspectives. This is a key aspect of avoiding both Type I
(seeing-wrong) and Type II (not-seeing) errors as false patterns are unlikely to
be visible from many perspectives and patterns hard to see in a single view are
more likely to be discovered in other views.
6.4.8 Combining Visual and Computational Exploration
The visualization and interaction techniques described so far seem most success-
ful for data exploration tasks in small and medium-sized data sets, i.e., up to
a few hundred items and a few tens of attributes. Geospatial data sets in prac-
tice, however, are continuously growing in size and are often characterized by a
high number of dimensions as a report by the US National Research Council ob-

served [601]. In such large data volumes with high-dimensional attributes human
vision cannot be successful in isolating patterns [530]. Visualizing large data sets
result in maps and other displays that are cluttered with overlapping items and
small symbols such that properties of data items are hardly visible. Zooming is
no remedy to these problems: it does help to avoid information overload and
magnifies small symbols but at the cost of losing the overall view of the data
which is just as important.
On the other hand, computational methods have been developed in areas
like machine learning and data mining that can analyze large data volumes and
automatically extract knowledge. But data mining methods have limited pattern
interpretation abilities. They are susceptible to missing patterns with unusual,
non-linear shapes. Interpreting potential patterns is also extremely difficult for
computational methods which do not have the domain knowledge of a human
expert.
With their strengths and weaknesses, computational methods and visual ap-
proaches complement each other. The integration of both approaches to combine
their advantages promises further advances in the exploration of geospatial data.
MacEachren and Kraak [530] reported “integrating advantages of computational
and visual approaches for knowledge discovery” as one of the four primary themes
on the ICA research agenda. Data exploration with tools that integrate both ap-
proaches is an iterative process. Results of the initial computational analysis are
displayed graphically for further examination by an analyst. Using visual tools
the patterns detected automatically need to be explored and interpreted. Ques-
tioning the results of a single run is very important. The user must be able to
verify patterns and their stability by interactively changing the parameters of
the data mining methods.
Examples of this integrative approach comprise the use of self-organizing
maps (SOM) as a form of neural networks by Guo et al. [315] and Koua and
Kraak [472] as well as applying k-means clustering by Andrienko and An-
drienko [22]. Both methods are used for detecting clusters in a data set. A
cluster denotes a subset of data items that are similar to each other and differ-
ent from items in other clusters. This implies the need for a similarity measure
on the attribute space, e.g., based on the Euclidean distance. In an explorative
environment, where the goal is to discover unknown relationships, patterns have
no predefined shape. Therefore, it is important to apply computational methods
that do not impose a-priori hypotheses about the shape of patterns and instead
let the data speak for themselves [315]. Kohonen’s SOM is an example of such
a method. The basic idea is to project high-dimensional data to a given grid of
nodes on a two-dimensional surface such that (potentially non-linear) similari-
ties are preserved and transformed into spatial proximity: data items within the
same node are most similar, and the similarity between an item in one node and
items in other nodes decreases with increasing grid distances to these nodes. A
detailed description of SOMs can be found in Hastie et al. [340].
278 M. Nöllenburg
Fig. 6.12. Visualization of a cancer data set with a choropleth map, a parallel co-
ordinate plot, and a SOM view. Screenshot of an application built with GeoVISTA
Studio [816].
A SOM for geospatial data by itself is missing some important information.

First, the geographic locations of clustered items cannot be extracted from the
SOM view and second, the attribute values representative for a node cannot be
displayed. Guo et al. [315] solved this problem by linking a view of the SOM with
a map and a PCP view which provided the missing information. A screenshot
of their tool is shown in Figure 6.12. The map depicts 156 counties in the US
states Kentucky, Pennsylvania, and West Virginia. They used an appropriate
(user-adjustable) 2D color scheme to color SOM nodes such that nearby nodes
have similar colors. This can be seen in the upper right window in Figure 6.12.
The hexagons in this windows colored with shades of grey depict the distance
in attribute space between adjacent SOM nodes. In the choropleth map view,
each county is colored according to the SOM node it belongs to. This enables
the user to compare proximity in attribute space with proximity in geographic
space in a single view. A PCP is used to display the summarized attribute values
of all SOM nodes. Summarizing data items in the same node avoids overplotting
in the PCP since the number of SOM nodes (e.g., 10 × 10) is usually much
smaller than the number of data items. The profile of each node is again colored
identically and the line thickness is adapted to the number of items in that
node. Brushing and linking of the three views are supported and the user can
select either counties on the map, profiles in the PCP, or nodes in the SOM. The
corresponding objects in the other displays are immediately highlighted.
The example of Guo et al. [315] shows that the integration of a computational
data-mining technique into a geovisualization system allows the exploration of
large geospatial data sets. The visual information load is reduced by automati-
cally clustering the data and only displaying summary information while details
are still available on demand. Users can explore the data and generate hypotheses
by interacting with the system and by bringing in their expertise.
6.5 Geovisualization Tools

In the previous section, a wide range of methods and techniques used in geovi-
sualization have been presented. However, the challenge for designers of geovisu-
alization tools is to put these methods together effectively in order to help users
solving their respective tasks. It has become clear that both users and tasks
vary considerably. Hence, there cannot be a universal tool that fits all users and
tasks. Instead, tools need to be flexible and easily configurable by users in order
to be applicable to more than just a tightly defined problem. Users should be
able to select freely from the methods discussed in the previous section and to
link multiple views. In the following, five examples of geovisualization systems
are briefly described to show how multiple views are combined and linked in
practice.
ArcView. ArcView5 is a commercial tool for visualizing and analyzing geo-
graphic data. It is a component of ArcGIS, one of the world’s leading geographic
information systems. ArcView offers a range of methods for creating customized
thematic maps and analyzing spatial data. It is primarily used in administrative
and industrial settings, e.g., for emergency planning, site planning, or marketing.
XGobi. XGobi [814], initially released in 1996, is a general data visualization
system supporting linked scatter-plot matrices and parallel coordinate plots.
XGobi is a freely available tool, and it can also be linked to ArcView.
Cartographic Data Visualizer. The Cartographic Data Visualizer (CDV) [219]
integrates the mapping and abstract data visualization components into a sin-
gle application. Its latest release is from 2000, and it is freely available on the
Internet. CDV offers dynamic choropleth maps, population cartograms, and sta-
tistical graphics like PCPs. Graphic symbols serve as interface elements to access
detailed information or to select subsets of items. The views are linked such that
highlighting items in one view is passed on to the other views.
CommonGIS. CommonGIS [24] is another integrated geovisualization environ-
ment developed at Fraunhofer AIS which is in practical use in academic, ad-
ministrative, and commercial settings. It combines a multitude of geovisualiza-
tion techniques from (multiple) dynamic choropleth maps, optionally combined
5
More information at http://www.esri.com/software/arcgis/arcview.
280 M. Nöllenburg
Fig. 6.13. Screenshot of CommonGIS. Image courtesy of G. Andrienko.
with bar plots and pie plots, over animated maps, time-series diagrams, and
space-time cubes to multivariate visualizations like PCPs or scatter plots, see
Figure 6.13. The user interface of CommonGIS supports interaction through
focusing, brushing, linking of multiple views, and dynamic range selection for
attributes. It is also possible to complement the visual data analysis by compu-
tational data-mining techniques. One focus in the development of CommonGIS
was that it could be used even by users with no expertise in cartography and
geosciences. The tool is commercially distributed, but it is free of charge for
academic users.
GeoVISTA Studio. The approach taken by GeoVISTA Studio [816] is differ-
ent. It is an open source software development environment for geovisualization
rather than a static application. It is a component-oriented system with the goal
to integrate a wide range of both computational and visual analysis activities
and ultimately to improve geoscience research. Creating custom applications is
done via a visual programming user interface that allows to connect different
geovisualization components, provided as Java Beans, according to the desired
data and control flow. Visual programming is a key aspect according to Takat-
suka and Gahegan [816] because it allows geoscientists with little computational
background to rapidly create prototypes when they are searching for useful in-
sight. GeoVISTA Studio comes with a range of standard components, such as
choropleth maps, 3D renderers, PCPs, scatter plots, color maps, spreadsheet

views, and computational tools as k-means clustering and self-organizing maps.
Additionally, Java Beans created by third-party developers can easily be plugged
in. Once a specific application or applet has been designed, its deployment over
the Internet, for example to students for educational purposes, is supported.
Sample applications can be downloaded from the project web site6 . One exam-
ple is the SOM-based tool by Guo et al. [315] described in Section 6.4.8 and
shown in Figure 6.12 which is built using GeoVISTA Studio.
6.6 Usability of Geovisualization Systems

In the early days of geovisualization, the development of new tools was mainly
technology-driven. New features of computer hardware had to be used exten-
sively just because they were available. The tools were often seen as proofs of
concept or designed for personal use and not really intended to be utilized in
practice by geoscientists and other professionals. Currently, this scope is shifting
away from innovators and early adopters towards a broader audience of prag-
matic and conservative users [274]. Hence, it becomes increasingly important to
provide geovisualization systems that are useful and usable for their target users.
As Slocum et al. [776] put it, “the most sophisticated technology will be of little
use if people cannot utilize it effectively”. In other words, considering cognitive
and usability issues throughout the whole development process of a geovisual-
ization tool is highly important for its success with the users. Cartography has
a long tradition of applying perceptual and cognitive principles to map design,
e.g., devising color schemes [111]. However, the cognitive theory for static maps
does not easily generalize to current geovisualization systems. Slocum et al. [776]
argued that cognitive and usability issues had to be considered in the context
of (1) dynamic interactive maps, (2) immersive geovisualization environments,
(3) interface design, (4) individual and group differences, (5) collaborative visu-
alization, and (6) the effectiveness of geovisualization methods.
Usability issues and evaluation of visualizations are covered in depth in Chap-
ter 2 of this book. Here, only the most important aspects with respect to geo-
visualization are discussed. Usability and usefulness are terms originating from
the field of human-computer interaction (HCI) which are increasingly influencing
geovisualization research after being put on the 2001 ICA research agenda [530].
Usability in the HCI community refers to the effectiveness, efficiency, and sat-
isfaction with which specified users can achieve specified goals in a software
system. This is not to be confused with usefulness which denotes the system’s
ability to achieve the specified goal in principle. A tool can be useful for some
spatial analysis task (it provides all necessary information) but still not be usable
e.g., because of poorly designed interfaces or because users don’t understand its
intended usage. Usability engineering in HCI refers to a human-centered design
process in which the usability of the system at different stages is evaluated, see
6
http://www.geovistastudio.psu.edu
282 M. Nöllenburg
Nielsen [609] for a comprehensive introduction. Quantitative usability evalua-

tion, as introduced in Section 2.4 of this book, is successfully applied for tasks
that are well defined. However, in geovisualization well-defined tasks are hard
to identify when it comes to data exploration and knowledge discovery. These
scenarios are inherently poorly defined and goal achievement becomes difficult
to assess. Fuhrmann et al. [274] gave an overview on how principles of usabil-
ity engineering could be modified to take the characteristics of geovisualization
into account. Common quantitative measures like the required time to com-
plete a task or the number of correctly answered questions are often not directly
applicable. Rather are system developers interested in whether their tool sup-
ports the user’s abilities to understand the data and to uncover hidden patterns.
Therefore, qualitative assessment methods using various interview and discus-
sion techniques, participant observations, or thinking-aloud protocols proved to
be suitable for evaluating geovisualization tools, see the results reported in Sec-
tion 6.6.2.
One difficulty of usability testing in geovisualization, reported by Andrienko
et al. [27], is that often the goal is to evaluate a concept, e.g., a parallel coor-
dinate plot, but the actual test must be performed with an implementation of
that concept. A positive result provides evidence in favor of the concept (and the
implementation), but a negative result does not imply the failure of the concept.
It could simply mean that this implementation of the concept is not appropri-
ate. It is therefore important to evaluate preliminary prototypes qualitatively to
reduce the risk of poor implementations of a concept.
A user group that may not be forgotten—and one of the most challenging
according to Plaisant [654]—is the general public. Problems that need to be
addressed in this context mainly deal with the diversity of users. But reaching
every citizen is becoming important for governmental agencies and companies
offering geospatial services for end users. Such services need to be distributed
to a wide audience, most likely over the Internet. However, users range from
children to senior citizens, have different levels of education, speak different lan-
guages, and—not to forget—may have visual disabilities like color blindness.
Additionally, they use different technology in terms of network bandwidth, pro-
cessor speed, screen size, etc. Scientific visualization methods such as scatter
plots or PCPs are hard to understand for the average user. Simple interfaces
and displays are paramount to keep the required training as low as possible. In
general it is important that the visualizations are adjustable for the needs of
the specific user group. All these challenges are subsumed in the term universal
usability by Shneiderman [764].
6.6.1 Involving Users in the Design of Geovisualizations

Evaluating a geovisualization tool according to usability principles right before
its release should only be one of the final stages of usability testing in a human-
centered design process. Fuhrmann et al. [274] reported that carrying out ‘last-
minute’ evaluations often revealed major flaws and thus caused timely and costly
repairs. In order to avoid this, user participation should take place right from
the early stages of tool development as discussed in Section 2.4 of this book.
Fuhrmann et al. identified setting an early focus on users and tasks involving
their participation and empirical testing iteratively during the whole design pro-
cess as common principles in the literature on user-centered design approaches.
Determining the characteristics of the target user group in a user analysis is one
of the first steps. Variables that are important range from cultural background
and sex over domain expertise and education to computer literacy and potential
disabilities. Learning about users and their needs is usually done in individual
interviews, user observations, or group discussions (called focus groups). Narra-
tive story telling is another method to gain insight into a user’s subjective view
on a problem by listening to users’ reports of critical incidents and personal
experiences. These methods usually require that users have already been using
a similar geovisualization system. For novel technologies like virtual environ-
ments or highly dynamic geovisualization systems this is, according to Slocum
et al. [776], often not the case. Hence, users cannot be interviewed about or
observed during analogous situations.
During later stages, such as concept development and prototyping, domain
experts’ feedback gives valuable hints about the usefulness of the system. In
the beginning, this comprises discussing drafts and paper and pencil prototypes.
Later on, prototypes with still limited functionality can, for example, be tested in
a situation where one of the designers acts as the interface to the tool for expert
users who perform their actions verbally. Once the desired level of usability for
all parts of the product can be ensured, the final implementation takes place. If
similar systems are available comparative evaluations are common in practice.
Otherwise the tool is evaluated on its own by users interacting directly with the
full product or with certain components individually. Depending on the specific
visualization goal the tool is built for, a set of representative user tasks may be
defined and subsequently be solved by the test participants. In this case, results
can be measured quantitatively. Alternatively, users are asked to explore a data
set and report on their findings using methods like interviews, focus groups, etc.
as described in Chapter 2 of this book.
The procedure described above is still vague and actual usability studies
in geovisualization fill in the details differently. The reason for this is that the
ideas of usability testing and user-centered design from HCI research are novel
in geovisualization. The need for a comprehensive user-centered design approach
and formal methods for usability assessment in geovisualization was expressed by
MacEachren and Kraak [530]. A theoretical cognitive framework providing the
basis for novel geovisualization methods and user interfaces is needed as well as
the development of usability engineering principles for designing geovisualization
tools [776]. Slocum et al. [776] also observed that authors of geovisualization tools
often claimed that their methods facilitated science and decision making without
providing empirical support for their claims. In the following section, some results
from user studies that were inspired by these challenges are reported.
284 M. Nöllenburg
6.6.2 Results from User Studies
Griffin [305] studied how experts use geovisualization to explore epidemiological

data with a tool using three types of displays: maps, scatter plots, and time-series
graphs. The thinking-aloud method was used to generate protocols of the system
usage of 18 experts from epidemiology and related domains. Participants were
asked to use the model for exploring the data set and generating hypotheses after
an introduction to the user interface. The main finding matches the intuition that
users attending a greater diversity of displays also generated more complex and
diverse hypotheses. Since facilitating hypotheses generation is an important goal
of explorative geovisualization, Griffin’s results provide evidence for including a
wide range of different display types in such tools.
A study by MacEachren et al. [524] reported on the use of a single manipu-
lable map by nine experts for the analysis of health data. The authors analyzed
mortality rates and risk factors for three types of cancer between 1979 and 1993.
The map display supported selection of up to two attributes represented in the
choropleth map. Interactive controls comprised dynamic classifiers and a cross
map to compare two attributes. Additionally VCR-like temporal controls were
offered to analyze changes over time. Verbal protocols from participants were
recorded as well as interaction logs during thinking-aloud sessions. Unlike in the
study described above [305], participants were given six explorative tasks in-
volving locating areas with certain properties, examining temporal trends, and
making space-time comparisons. Depending on how users interacted with the
tool they noticed or missed certain patterns. This study provides evidence that
users who animated time-series data noticed trends missed by others, who used
manual stepping to navigate through time.
Animations in comparison to static graphics, as examined in the previous
study, have been the focus of several other studies. Koussoulakou and Kraak [473]
compared the communicative expressiveness of animated maps versus static
maps. In a study with 39 geodesy students using quantitative evaluation, they
found that the correctness of answers was not influenced by the map type.
However, spatio-temporal phenomena were perceived more quickly on animated
maps. They also stated that using the full potential of animated maps requires
an interactive user interface, which was not available in their study.
Harrower et al. [335] studied two interactive controls for animated maps:
temporal brushing and focusing with temporal legends (linear and time wheel).
In their test setting, 34 undergraduate students were using the interactive learn-
ing environment EarthSystemVisualizer. The advanced interaction methods had
little impact on the students’ performance. The reason is that many students
were confused by the new interactive controls and hence did worse than students
who did not use brushing and focusing. However, those students who understood
the tools performed best in answering a series of objective questions. In contrast
to the results from the main evaluation, a preliminary focus group using Earth-
SystemVisualizer gave mainly positive feedback about the interactive tools. This
example of a useful but not very usable system clearly indicates the importance
of sufficient user training, especially for novel techniques. A visualization system

that is too complex for its users has rather negative than positive effects.
Perceptual salience7 of dynamic displays was studied by Fabrikant and Golds-
berry [242]. They applied a neural-net based and neurobiologically plausible
computational model by Itti et al. [395] to predict the most salient locations
on either static or animated maps. They found that the most salient locations
determined by the vision model differ between static small multiples and non-
interactive animations. Future plans include comparing these predicted results
with empirical eye-movement measurements in a user study and exploring the
perceptual differences of dynamic displays between expert and novice users.
Morrison et al. [581] compared several studies on the efficacy of animations
in general. They observed that animation was not superior to static displays in
studies where both displays provide equivalent information. Studies reporting
positive effects of animation were using non-equivalent static displays, e.g., with
the microsteps of an animation not being shown in the static graphics. So, these
microsteps might be the power of animations. Slocum et al. [776] pointed out
that the methodology of Morrison et al. [581] was limited by comparing only
non-interactive animations to static graphics. However, just as users can control
(in a sense interactively) the focus location in a static graphic, they should be
able to do the same with respect to the temporal dimension of an animation.
Andrienko et al. [27] reported on an experimental usability study of inter-
active techniques in CommonGIS assessing tool learnability, memorability, and
user satisfaction. Since they primarily wanted to test whether users are able to
handle the tools and not whether the tools facilitate knowledge construction,
users had to solve well-defined tasks and answer multiple-choice questions. In a
first round, the authors gave a tutorial lesson to nine participants before the test
which was repeated a month later with the same group to assess memorability
of the tool usage. The third web-based test was taken by 102 students who did
not receive the introduction to the tools. However, they could read illustrated
explanations for each task series. Overall, the performance in the web-based test
was much worse in comparison to the first two rounds. This confirms the find-
ings of Harrower et al. [335] that demonstrating and teaching the novel tools is
very important. Teaching users to interact with web-based tools is a challenging
task. Many participants did not read the instructions thoroughly and interactive
online tutorials were suggested instead. Positive results were reported for some
of the tested techniques, such as visual comparison of values in choropleth maps
and dynamic linking of map, scatter plot, and PCP. Negative results were ob-
tained for the dynamic query technique. This technique allows to specify value
ranges for attributes. Objects with values outside these ranges are removed from
the display.
Edsall et al. [231] assessed and compared the effectiveness of a map dynam-
ically linked to a PCP with a map linked to a scatter plot. At first, the 37
participants of the study had to use either of the tools to accomplish a series of
7
An object in a graphic is said to be perceptually salient if it can be detected quickly
with respect to its surrounding objects.
286 M. Nöllenburg
well-defined tasks. In the second step, participants were observed during unre-
stricted exploration of a new data set with all available tools. They did not find
a statistically significant advantage of one method over the other. However, scat-
ter plots were slightly ahead for studying bivariate patterns while PCPs seem
to be more suitable for multivariate data. In the second test, those participants
were most successful in finding patterns who made extensive use of a variety of
visualization tools and thus taking multiple perspectives on the data.
A software tool to visualize uncertainties of global water balance due to
climate change on a wall-size display was described by Slocum et al. [777]. Their
goal was to communicate scientific results to decision makers, e.g., to politicians
or executives. They followed usability engineering principles and first built a
prototype which was evaluated by domain experts. The prototype was followed
by two software revisions, the first one evaluated by usability experts and the
final version evaluated by decision makers. This approach clearly improved the
software, however Slocum et al. recognized that decision makers, the actual users
of the system, were involved too late in the design process. Their comments would
have been of greatest value in evaluation of the prototype [777].
Another example of a system designed according to usability engineering
principles is ESTAT (Exploratory Spatio-Temporal Analysis Toolkit) by Robin-
son et al. [716]. It was developed in a cooperation of geovisualizers and epidemi-
ologists to facilitate cancer epidemiology research. ESTAT is built in GeoVISTA
Studio, and it combines and dynamically links a bivariate choropleth map, a
scatter plot, a time series plot, and a PCP. At all stages of the development end-
users were involved to assess the usability of the system. A prototype of the PCP
module was evaluated in a thinking-aloud study with a group of students and
resulted in reorganizing the PCP interface. Then, a refined version of ESTAT
was evaluated with a group of epidemiologists in a tutorial and task session and
a focus group discussion. Participants expressed the need for more training with
the PCP and the time series plot and the desire to see descriptive statistics of the
data. it was decided to perform an in-depth case study with an epidemiologist to
understand how users would like to work with a geovisualization tool. This case
study emerged as a particular beneficial way of gaining a deeper understanding
of an expert’s analysis strategies and directed attention to previously neglected
issues like loading and sorting data and interface icons and controls. All parts of
usability evaluation resulted in redesigning significant parts of ESTAT, e.g., the
scatter plot was enhanced by a number of descriptive statistics and regression
methods. The study clearly demonstrates how important it is to adapt innovative
methods to the needs of real users doing real work.
6.6.3 Geovisualization to Support Group Work

Another aspect of human-centered geovisualization is the design of systems that
support collaborative use. While many potential applications of geovisualization
require coordinated efforts by groups, most geovisualization methods are pri-
marily designed for single users at a workstation. Technological advances such as
large screen displays or high-bandwidth communication enable both same-place
and different-place multi-user systems but current geovisualization systems often

do not use these capabilities. MacEachren and Kraak [530] listed geovisualization
methods that support group work as a fundamental multidisciplinary research
challenge. Section 3.4 of this book discusses distributed collaborative visualiza-
tion in general. Here, both same- and different-place collaboration is covered in
the context of geovisualization.
Scenarios in which small groups collaborate can be found in scientific re-
search, environmental and urban planning, education, or economic decision mak-
ing. Commonly, group work is classified by same- or different-place and same-
or different-time collaboration as described by Applegate [32], see also the dis-
cussion in Section 3.4.1 of this book. Current geovisualization research on group
work focuses on both same-place/same-time and different-place/same-time col-
laboration. Same-place collaboration certainly provides the most natural mode
of interaction between humans but in certain situations groups of experts with
various backgrounds need to be consulted and it is often impossible to convene
at a single place, particularly for time-critical tasks such as emergency manage-
ment. Same-place collaboration often utilizes wall- or desk-based large screen
displays and multimodal user interfaces. In contrast, different-place collabora-
tion introduces the problem of synchronizing and linking the visualizations at
multiple places. Brodlie [116] described three models of distributed visualization
systems. A single, shared application runs on a single machine and the user in-
terface is replicated to different displays, i.e., all collaborators see exactly the
same image. Either a single user controls the application, e.g., for a demonstra-
tion, or control of the application is passed between users. A single, replicated
application is executed at each location and parameter settings are shared. This
is a more flexible approach, e.g., users can change their personal screen layout
or language settings. Finally, multiple, linked applications allow the highest de-
gree of flexibility. Users can work individually and share data and parameters
as they wish. However, flexibility is also the disadvantage of this approach as
it is hard for a user to get a feeling of what others are doing and that makes
coordinated work difficult. MacEachren [523] suggested a participant watcher as
a tool that shows each user’s screen layout and currently active windows as well
as activity displays that show for example the amount of time each user was
in control of the joint display. Such a tool arises concerns about privacy and
therefore its acceptance by users but it was not yet evaluated in a user study.
Video conferencing is a another common mode of human-human interaction in
such applications [116, 523].
Being the object of collaboration is not necessarily the role of geovisualiza-
tion. MacEachren [523] listed supporting dialogue and supporting coordinated
activity as alternative roles. In supporting dialogue, the display can be used,
for example, to depict an argument directly or to serve as a framework for ar-
guments. Geovisualization can support coordinated activity, for example, in an
emergency response situation, where the command center staff needs to com-
municate with individuals in the field. The visualization environment serves to
288 M. Nöllenburg
guide navigation, to input data and to provide an overall view for assessing the
situation.
Collaborative geovisualization systems make different demands on usability
and user-centered design in comparison to the single-user case discussed before
in Section 6.6.1. Users, such as experts from different fields, are no longer homo-
geneous and tasks that support and mediate dialogue have not been considered
before in single-user systems. MacEachren and Brewer [525] identified several re-
search challenges for geocollaboration. For example, working with a large screen
display requires more natural, multimodal user interfaces, such as speech and
gestures, to support group work. See Section 3.3 of this book for a survey on
multimodal interaction techniques. Furthermore, a theoretical understanding of
cognitive and social aspects of collaboration must be developed as a basis for us-
ability studies. In the following, two prototype systems designed for collaborative
geovisualization tasks are described.
Brewer et al. [112] developed a prototype system to support same
time/different place collaboration of a team of environmental scientists. The
system was designed to mediate and enhance knowledge construction in a sam-
ple climatic time-series data set. A focus of their study was on usability and
human-centered design aspects and they reported the results of a user-task anal-
ysis. They interviewed six domain experts and demonstrated their prototype to
prompt discussion. The prototype presented a 3D map-based view of precipi-
tation and temperature. The temporal dimension is controlled by a time wheel
interface and the map can be animated accordingly. The environment can be
displayed on regular displays as well as on an ImmersaDesk virtual reality sys-
tem. In their user-task analysis, Brewer et al. found that drawing the attention
of others is a crucial task [112]. Participants frequently used gestures to draw
attention to features on the display. Joint interface controls were endorsed but
users added that conflict avoidance is necessary, e.g., with a turn-based approach
or separate windows for each user. Alternatively they suggested a single user to
be in control. Concerning the display a workbench-like display was preferred
over a wall-mounted solution. In terms of different place collaboration, the ex-
perts favored voice and video for communication. They expressed the need for
a private workspace and for the ability to see what others are doing to the joint
workspace, not just the outcome. Again, drawing attention, e.g., with the mouse
cursor, was regarded as very important.
A second prototype system is DAVE G (Dialogue-Assisted Visual Environ-
ment for Geoinformation) by MacEachren et al. [526]. DAVE G was designed for
a map-mediated crisis management scenario where decision makers need to coor-
dinate the emergency response. Geoinformation systems (GIS) are used in prac-
tice today but they are operated by GIS technicians in response to requests from
decision makers. Valuable time is often lost due to this indirect GIS usage. Hence,
the goal of MacEachren et al. was to develop a multimodal, dialogue-assisted tool
that uses speech and gesture recognition to enhance human-human collabora-
tion around a large screen map display. In this approach, human-computer and
human-human interaction use the same modalities which is an important aspect
for user acceptance. An initial work-domain analysis with four emergency man-
agers identified a set of required map functionality, e.g., zoom, pan, buffer, select
data, and free-hand drawing. Challenges for the user interface include recogni-
tion of implicit commands (“we needed to view this in more detail”) and sloppy
speech. Speech and gestures are fused, i.e., a spoken command, e.g., “zoom in
to this area”, is associated with a pointing gesture. A dialogue manager re-
quests missing parameters for incomplete commands. Technical details of the
user interface can be found in MacEachren et al. [526]. Fuhrmann et al. [275]
evaluated the usability and performance of DAVE G in a user study with ten
graduate students. The system was unknown to the participants. In the begin-
ning, they set up individual voice profiles and completed a DAVE G learning
session of about ten minutes after which they felt comfortable with the system.
They solved two unguided tasks where they had to load data, zoom, pan, and
identify certain objects. About 70% of the requests were successful and 90% of
the gestures were correctly recognized. In a subsequent interview, participants
were positively describing their experience with DAVE G and emphasized quick
learnability. Fuhrmann et al. [275] reported that hand tracking sometimes lost
the cursor and that the dialogue capabilities of DAVE G needed to be improved.
6.7 Chapter Notes
This chapter aimed to give a comprehensive overview of the state of the art in
geovisualization research. However, the abundance of developments and results
in this lively research field cannot be covered completely within this book. A
good starting point for further reading is “Exploring Geovisualization” [222], a
book that contains the results of a cross-disciplinary workshop with the same
title held in 2002 at City University, London.
6.7.1 Future Trends
How will geovisualization develop over the next years? In the future, Dykes et
al. [221] still see the map as the primary tool to present, use, interpret, and
understand geospatial data. However, it has become apparent in the past that
the map can and will take a variety of forms, some of which quite astonishing.
The map has evolved from its traditional role as a presentational device to an
interactive and highly dynamic interface to access and explore geospatial data.
Another common feature of current geovisualization tools is that they consist
of multiple linked displays, each depicting an alternative representation of the
data under examination to stimulate visual thinking. This will certainly remain
a central aspect in geovisualization.
Concerning the tasks and users of geovisualization tools a transition from
explorative, individual use by experts towards the whole range of tasks from
exploration to presentation and heterogeneous groups of users is taking place.
While individual expert tools will continue to develop, there will be an increasing
290 M. Nöllenburg
number of applications designed for the public and disseminated over the web.
It is clear that the usability requirements differ significantly between research
tools for specialists and public applications for the mass of non-expert users.
Human-centered aspects need to play a key role in the design of future tools in
order to make them fit the needs of their audience. The goal of universal us-
ability, i.e., creating applications for wide ranges of users from children to senior
citizens in different languages and respecting visually disabled users, remains a
big challenge.
Advances in hardware technology will also have a strong influence on geovi-
sualization. On the one hand, large 2D and 3D displays and virtual environments
will be used for geovisualization tasks, especially for collaborative use by a group
of experts and decision makers. Usability studies for these new technologies will
have to investigate their advantages and disadvantages over traditional visual-
ization methods as well as possible cognitive and social impacts. In particular,
the right balance between abstraction and realism needs to be determined for 3D
displays. On the other hand, portable devices like PDAs or mobile phones will
provide location-based services, e.g., route finding in foreign places which build
on some sort of map display. This means additional challenges for application
developers in terms of efficient memory and bandwidth use as well as in terms
of visualization design for multi-platform, small-size displays.
A last point that is raised by Gahegan [283] is the need for interoperabil-
ity of geovisualization components. Following the efforts of the Open Geospatial
Consortium8 to define open standards for geospatial information systems, Gahe-
gan promotes a similar initiative for geovisualization. Visualization components
should define standard interfaces in order to be reused and integrated into a
variety of geovisualization tools. A lack of interoperability means that for devel-
oping a new tool too much effort is spent on re-implementing basic functionality
and too little effort can be invested in developing new methods.
The coming years will show what directions geovisualization research will
take and how it will influence both research in the earth sciences and related
disciplines and our everyday handling of geospatial information.
8
See http://www.opengeospatial.org.

Geospatial Visualisation

Uploaded by

Copyright:

Available Formats

Geospatial Visualisation

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Geospatial Visualisation

Uploaded by

Copyright:

Available Formats

6.

Geographic visualizations always played an important role in human history, es-

riety of suitable visualization methods from map-based and abstract techniques

6.1 Goals of Geovisualization

6.2 Driving Forces of Geovisualization

6.3 Cognitive Aspects

6.3.1 Visual Thinking

6.3.2 Graphic Variables

6.4 Visualization Methods and Techniques

A number of techniques and methods adapted from cartography and scientific

6.4.1 Geospatial Data

In contrast to information visualization displaying any abstract data, geovisual-

6.4.2 2D Cartographic Visualization

The most common visualization method for geospatial data is a cartographic

6.4.3 3D Cartographic Visualization

In contrast to traditional paper maps and two-dimensional visualization meth-

is referred to Section 3.2. Since humans live in a three-dimensional environment

6.4.4 Visual Data Mining Tools

for multivariate data were broadly classified as geometric, glyph- or icon-based,

Glyph-Based Techniques Glyph-based or icon-based techniques use a map-

(a) Former Soviet N. America

305 443 583

cases, the temporal dimension is used to display quantitative attributes by map-

6.4.6 Spatio-Temporal Visualization

For detecting periodic temporal patterns, Hewagamage et al. [348] suggested

6.4.7 Interactive User Interfaces

Fig. 6.10. Screenshot of Google Earth [300].

to operating a camera: choosing a perspective, deciding about magnification

these abilities to navigate in time as equivalent to navigating in space, i.e.,

6.4.8 Combining Visual and Computational Exploration

high number of dimensions as a report by the US National Research Council ob-

A SOM for geospatial data by itself is missing some important information.

6.5 Geovisualization Tools

Fig. 6.13. Screenshot of CommonGIS. Image courtesy of G. Andrienko.

choropleth maps, 3D renderers, PCPs, scatter plots, color maps, spreadsheet

6.6 Usability of Geovisualization Systems

Nielsen [609] for a comprehensive introduction. Quantitative usability evalua-

6.6.1 Involving Users in the Design of Geovisualizations

6.6.2 Results from User Studies

Griffin [305] studied how experts use geovisualization to explore epidemiological

of sufficient user training, especially for novel techniques. A visualization system

6.6.3 Geovisualization to Support Group Work

and different-place multi-user systems but current geovisualization systems often

6.7 Chapter Notes

6.7.1 Future Trends

You might also like