small multiples and linked views require high cognitive load
and have been found to be hard to compare [25].
Furthermore, the use of animation has been exploited for
better understanding of relationships between different
properties of a data set and various visual pieces in a representation.
Animated maps are a result of employing animation
to show changes of attribute data over time and on a
map [26], [27]. Griffin et al. [25] have researched the effectiveness
of animated maps and compared it to static smallmultiple
maps. The results of their controlled experiment
shows that map readers can identify more moving clusters
more quickly using animated maps. The type and speed of
animation can also play an important part in whether or not
animation is effective or not. In [28], authors compare various
types of animations for showing the relationships
between different structures. Smooth transitions were
shown to help users maintain the visual relationships
between the different views. Also animation speeds that
complete a viewpoint change in one second are sufficient
for maintaining perceptual constancy. Effects of smooth
transitions and have further been investigated in [29] finding
dramatic benefits on user performance and guidelines
on how to avoid some of the costs associated with animated
transitions.
An alternative to animated playback is to provide the
user with an interactive time slider (e.g., [22], [23]). Theoretically,
this could allow the user to more quickly navigate
through the data and find valuable information, by sliding
more quickly through time spans of lower interest, and
slowing down when there is more temporal detail. Previous
evaluations of time-varying visualizations have sometimes
only allowed animated playback, foregoing the evaluation
of a time slider [30]. In our study, users had access to both
animated playback and a time slider, making the evaluation
more realistic and giving users more flexibility.
2.3 Abstract Space Representation
The 2D approaches mentioned above all keep the original
spatial structure intact hence incorporation of extra attributes
and time in the visualization makes them cluttered.
There are, however, another group of 2D visualization
methods proposed in the literature, which exploits abstract
space representations. For instance, authors in [31] use the
line graph metaphor to represent time on an abstract space.
The result is a proximity-based visualization of movement
trace data in which the spatial relationships (e.g., distance
among objects) are preserved.
2.4 3D Space-Time Representation
Adding the third axis to represent time takes us to the
alternative 3D group of visualizations that combine space
and time in a single display. Originally proposed by
H€agerstraand [32] and known as space-time cube, in
this form of representation, space and time are thought of as
being inseparable and movement is depicted as trajectories
in 3D with time being one of the coordinates. This idea has
been expanded by other researchers in the field [33], [34]. A
potential problem with the STC approach is occlusion in
case many trajectories are involved. To facilitate manipulation
and perception of information, STC has been extended
with interactive techniques [35]. A more advanced version
of STC enhanced with timeline as the main interaction
device, time zooming or focusing, and linking of maps with
corresponding symbols is presented in [36]. The enhanced
version of STC that supports many of these features has
been turned into a commercial software application called
GeoTime [3], [37].
Recent work by Tominski et al. [38] presents a solution
based on the STC with the focus on trajectory attribute
data, i.e., movement data which includes other attributes.
By stacking 3D color-coded bands on a 2D map and
ordering the bands based on the temporal information,
the trajectories and their attributes are visualized while
temporal information is directly perceivable. Extra visual
cues are also added to the bands to depict direction and
other properties.
Another drawback of the STC approach, besides occlusion,
is distortion of both space and time due to projection
which makes it hard to perceive depth. Even though 3D
representation of movement data has been introduced,
much research is being devoted to finding suitable forms of
representing this complex data set.
2.5 Evaluation of 2D versus 3D Visualization
of Movement Data
We next survey evaluations of 2D and 3D visualizations of
movement data. The most closely related work to our study
is by Kristensson et al. [2]. The authors compared 2D and
3D visualizations of movement data, asking users to answer
four types of questions:
Category1: simple “when” & simple “whatþwhere”
Category2: simple “when” & general “whatþwhere”
Category3: general “when” & simple “whatþwhere”
Category4: general “when”& general “whatþwhere”
The 2D visualization was found to be significantly less
error prone for category 2 questions, and 3D was found to
be significantly faster with category 4 questions. One explanation
for these results is the design choices made for both
2D and 3D visualizations.
In their [2] experiment, the 2D view allowed for interactive
pan and zoom, but did not have any time slider or animation.
Instead, text labels showing the times of “critical
time points” could be displayed or hidden by hitting a keyboard
key. The 3D view allowed for pan, zoom, and rotation,
and a “measurement plane” could be moved up or
down the time axis, with the current time of the plane displayed.
We intentionally made different choices in the
design of our own experiment, to test a more consistent
implementation in both 2D and 3D visualizations. In particular,
we note that Kristensson et al. [2] argue against using
time sliders and animations “because users cannot get an
overview of the data set at a glance with such representations”.
This is true, but in their 2D condition, users had
to read text labels to understand timing and sequencing,
which we suspect is slower than using a time slider or animation.
Furthermore, the ability to move a “measurement
plane” in their 3D condition is very similar to having a time
slider, whereas an analogous feature was not available in
their 2D condition. Temporal zooming or focusing in the
form of a timeline has been shown to be a necessity with