A Review of Temporal Data Visualizations Based on

Space-Time Cube Operations

Benjamin Bach, Pierre Dragicevic, Daniel Archambault, Christophe Hurter,
Sheelagh Carpendale

To cite this version:

Benjamin Bach, Pierre Dragicevic, Daniel Archambault, Christophe Hurter, Sheelagh Carpendale.
A Review of Temporal Data Visualizations Based on Space-Time Cube Operations. Eurographics
Conference on Visualization, Jun 2014, Swansea, Wales, United Kingdom. 2014. <hal-01006140>

HAL Id: hal-01006140

https://hal.inria.fr/hal-01006140
Submitted on 25 Jun 2014

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
Eurographics Conference on Visualization (EuroVis) (2014), pp. 1–19
R. Borgo, R. Maciejewski, and I. Viola (Editors)

A Review of Temporal Data Visualizations

Based on Space-Time Cube Operations

B. Bach1 , P. Dragicevic1 , D. Archambault2 , C. Hurter3 and S. Carpendale4

1 INRIA, France
2 Swansea University, UK
3 ENAC, France
4 University of Calgary, Canada

Abstract
We review a range of temporal data visualization techniques through a new lens, by describing them as series of op-
erations performed on a conceptual space-time cube. These operations include extracting subparts of a space-time
cube, flattening it across space or time, or transforming the cube’s geometry or content. We introduce a taxonomy
of elementary space-time cube operations, and explain how they can be combined to turn a three-dimensional
space-time cube into an easily-readable two-dimensional visualization. Our model captures most visualizations
showing two or more data dimensions in addition to time, such as geotemporal visualizations, dynamic networks,
time-evolving scatterplots, or videos. We finally review interactive systems that support a range of operations. By
introducing this conceptual framework we hope to facilitate the description, criticism and comparison of existing
temporal data visualizations, as well as encourage the exploration of new techniques and systems.
Categories and Subject Descriptors (according to ACM CCS): H.5.0 [Information Systems]: Information Interfaces
and Presentation—General

1. Introduction isting techniques by their name, both for general visualiza-

Temporal datasets are ubiquitous but notoriously hard to vi- tions [Har99] and for temporal visualizations [AMST11].
sualize, especially rich datasets that involve more than one Although names are essential for indexing, retrieval and
dimension in addition to time. communication purposes, they are a poor thinking tool. Be-
Previous work on novel visualizations for temporal data cause there is no convention for naming techniques, names
has dramatically advanced the field of information visual- rarely reflect the essential concepts behind a technique. For
ization (infovis). However, there are so many different tech- example, names such as Value Flow Maps [AA04b] and
niques today that it has become hard for both researchers Planning Polygons [SRdJ05] say little about the possible
and designers to get a clear picture of what has been done, conceptual similarities between the two techniques (see Fig-
and how much of the design space of temporal data visual- ure 1). Names can also be ambiguous. For example, the term
izations remains to be explored. For similar reasons, teach- small multiples is commonly used to refer to a specific type
ing this research topic to students is challenging. Therefore, of temporal data visualization [Tuf86]. But Figure 2 shows
there is a clear need to structure and organize previous work that two visualizations can be based on small multiples de-
in the area of temporal data visualization. spite being very different conceptually.
Part of the problem is that information visualization re- There has been recent effort at proposing taxonomies,
searchers have mostly focused on nomenclature. Most fa- conceptual models and design spaces for temporal visual-
miliar charts have an agreed-upon name, e.g., bar charts or izations, mainly focusing on analytical tasks and data types
scatter plots, and this tradition has been continued in info- (e.g., object movement data [AAH11, AAB∗ 11, AA12],
vis, where each newly published visualization technique is video data [BCD∗ 12], or datasets with different temporal
given a different name. Many textbooks and surveys list ex- and spatial structures [AMM∗ 07]).

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

2 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

(a) Value flow diagram [AA04b]

Figure 3: A space-time cube based on an illustration by

Hägerstrand [Hä70] in 1970, showing social interactions
across space and time.

The term space-time cube originates from cartography,

(b) Planning Polygons [SRdJ05] where it refers to a geographical representation where time
is treated as a third dimension [AA03]. One of the earli-
Figure 1: Two conceptually similar temporal visualization est uses was by geographer Hägerstand in 1970, who de-
techniques showing: (a) the evolution of crime statistics in scribed a "space-time model which could help us to develop
every US state; (b) the evolution of high school population a kind of socio economic web model" [Hä70, p. 10]. His in-
in several districts across 3 years. tention was to analyze people’s behaviour and interactions
across space and time. For example, a moving person on a
2D map becomes a static 3D trajectory once visualized as
USA Japan 1990 1992 1994
8
a space-time cube (Figure 3). Since then, space-time cubes
Inflation (%)

8
Inflation (%)

8
Inflation (%)

8 8
Inflation (%)

Inflation (%)

6 6 6 6 6

00
90 4
90
4 4 4 have been employed in a number of interactive visualization
92
systems (e.g., [CCT∗ 99, FLM00, Kra03]), as well as for en-
2 2 92 2 2 2
98 96 94 94
0 0 98 0 0 0
96 00
-2
0 10 20
Unemployment rate (%)
-2
0 10 20
Unemployment rate (%)
-2
0 10 20
Unemployment rate (%)
-2
0 10 20
Unemployment rate (%)
-2
0 10 20
Unemployment rate (%)
tertainment purposes [CI05] (see Figure 4). However, they
France Spain 1996 1998 2000 have never been used as a conceptual model for reflecting
8
Inflation (%)

on temporal visualization techniques.

Inflation (%)

8 8 8
Inflation (%)

Inflation (%)

Inflation (%)

90 92
6 6 6 6 6

4
90 4 94 4 4 4
00
92 96
2 96 2 98 2 2 2
00 94
0 98 0 0 0 0

-2 -2 -2 -2 -2
0 10 20 0 10 20 0 10 20 0 10 20 0 10 20
Unemployment rate (%) Unemployment rate (%) Unemployment rate (%) Unemployment rate (%) Unemployment rate (%)

(a) By country (b) By year

Figure 2: Two visualizations using small multiples to show

the same indicator data for 4 countries over 6 years, but
which are conceptually very different.

We propose a simple way of describing temporal vi-

sualizations based on operations on conceptual space-time
cubes. Our work is specific in that it focuses on how to char- Figure 4: Khronos projector [CI05] lets users dig into video
acterize existing techniques, independently from the data cubes: here, a scene transitioning from day to night.
and the tasks, and without considering which technique is
the most effective. Hence our framework is unique in that it In this article we use the term space-time cube in a similar
is purely descriptive. fashion as in previous work, but with two major differences:
The merit of a clear and detailed descriptive framework 1. A space-time cube is a conceptual representation that
is that it helps i) connect techniques that are similar and ii) helps to think about temporal data visualization techniques
distinguish techniques that are dissimilar. For example, the in general, not only 3D visualizations. The space-time cube
two techniques from Figure 1 are the result of a similar op- does not necessarily have to appear explicitly in the final vi-
eration on a space-time cube and which we call sampling. sualization nor does it need to be implemented in the system
Figure 2(a) involves operations such as filtering, time flat- used to generate this visualization. For example, the visual-
tening and space shifting, while Figure 2(b) is the result of a izations in Figure 1 do not show a space-time cube. For most
compound operation we call time juxtaposing. observers, they are purely 2D visualizations.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 3

2. A space-time cube does not need to involve spa- 2.1. Time Cutting
tial data. Many visualizations (e.g., scatterplots, bar charts
or node-link diagrams) convey abstract, non-spatial data 1 2
[Mun08]. Nevertheless, they all occupy a 2D space. When
data changes over time, such as in GapMinder’s animated
2D scatterplots [Ros06], each animation frame can be con-
ceptually thought of as a slice of a space-time cube. In the
term space-time cube, space therefore refers to an abstract Time
2D substrate that is used to visualize data at a specific time.
Figure 5: The time cutting operation.
Thus it is important to stress that this article is not about
space-time cube visualizations, and that 3D space-time cube
A time cutting operation consists in extracting a particu-
representations like the one in Figure 3 only represent a very
lar temporal snapshot from the cube to be presented to the
small subset of the techniques we aim to cover.
viewer. Figure 5 illustrates this operation: the left part (1)
In addition, our conceptual framework does not consider shows the initial space-time cube and the temporal snapshot
how space-time cubes are built, e.g., whether or not 2D scat- that is being extracted, while the right part (2) shows the re-
terplots should be used to represent the value of country in- sulting image that is presented to the viewer.
dicators at any given time. Instead, it assumes that a con- For example, consider a photographer who captures a par-
ceptual 3D space-time cube is already given, and focuses ticular instant of a moving scene. If the scene being viewed
on how this cube can be transformed to accommodate 2D is represented as a space-time cube (i.e., all possible pictures
media like computer displays and paper while remaining are piled up to form a cube), then taking a photograph is
legible. We show how such transformations are enough to equivalent to applying a time cutting operation on this cube.
capture most known techniques for visualizing rich tempo-
ral datasets. We mostly focus on datasets that involve two In information visualization, an image produced by time
dimensions plus time (e.g., spatio-temporal data, dynamic cutting is typically called a time slice. But a temporal visual-
graphs, scatterplots, videos, or any two-dimensional numer- ization rarely consists in a single time slice. As we will see
ical data varying over time), although we later discuss how in Section 3, time cutting is typically either performed mul-
our model can be extended to other dimensionalities. tiple times and used in combination with other operations, or
is used in combination with animation and interaction.
We first review common temporal data visualization tech-
niques, and explain how they can be all seen as operations on
a conceptual space-time cube. We then describe our frame- 2.2. Time Flattening
work in more detail by providing definitions of key concepts,
1 2
as well as a taxonomy of elementary operations and how
they can be combined. We then review temporal data explo-
ration systems that show how a range of space-time cube op-
erations can be supported on a single system through inter-
activity. Finally, we discuss the limitations of our framework Time
and suggest avenues for future work.
Figure 6: The time flattening operation.

2. Static Visualizations as Space-Time Cube Operations

Time flattening collapses the space-time cube along its
In this section we illustrate how space-time cube operations time axis, by merging all time slices into a single 2D im-
can be used to describe a range of common static visual- age (Figure 6). An analogy is long exposure photography,
ization techniques for temporal data, all meant for screen which collapses several seconds, minutes or even hours of a
or paper media. We focus on a small but representative se- natural scene into a single image.
lection of examples from the literature, and describe oper-
One of the earliest uses of time flattening is Minard’s
ations informally, often using analogies from photography
illustration of Napoleon’s march towards Moscow (Figure
techniques and art.
7). The illustration shows on a single image the state of
The conceptual space-time cube we use to describe all Napoleon’s army (position, size, key events) at different
techniques has three major axes: a time axis, and two or- points in time during the Russian campaign in 1812 [Tuf86].
thogonal axes we call data axes. The 2D plane formed by Another early example is Dr. John Snow’s map showing
the two data axes is referred to as the data plane. While in where deaths from cholera occurred in London in 1854 (Fig-
Haegerstrand’s original illustration the time axis is vertical, ure 8(a)). The map shows events from several days aggre-
in our illustrations time goes from left to right. gated over time.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

4 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

An analogy for discrete time flatting is multiple expo-

sure photography, where several photos are taken at different
times and blended into a single image. Etienne-Jules Marey
pioneered this technique in 1882 with an instrument (the
chronophotographic gun) that records 12 photos per second
on the same film, and used it to visualize human and ani-
mal motion [Mar78]. Modern art has also employed a simi-
Figure 7: A famous example of time flattening: Napoleon’s
lar technique to convey movement, e.g., Marcel Duchamp’s
march to Moscow by Joseph Minard [Tuf86].
“Nude Descending a Staircase, No. 2”.

Spain
8
90 92
6 91 93
Inflation (%)

96 94
00 97
4
95
2 99 98 Figure 10: An example of discrete time flattening. For a bet-
0 ter infographic by Megan Jaegerman, see [Tuf].

-2
0 10 20
Unemployment rate (%) Tufte [Tuf86] comments on several examples of info-
(a) (b) graphics that employ discrete time flattening. He calls them
sequences. One of his famous examples is the life cycle of
Figure 8: Other examples of time flattening: (a) Detail of the
the Japanese beetle [Tuf86]. Figure 10 is a sequence show-
map of the cholera outbreak in London 1854, by Dr. John
ing a dancer’s move. Discrete time flattening has also been
Snow. Piled bars mark the number of death per house. (b)
used for summarizing videos [BDH04].
Connected scatterplot showing the relationship between in-
flation rate and unemployment in Spain from 1990 to 2000.
2.4. Colored Time Flattening
Many maps that show temporal data can be seen as time-
flattened space-time cubes. But the time flattening technique 1 2 3
is not limited to geographical data, and has been employed in
a large variety of information visualization systems as well
as in static data graphics. Figure 8(b) for example, shows the
evolution of inflation rate and unemployment in Spain from
Time
1990 to 2000. This diagram can be seen as time-flattened Time

version of a space-time cube representing a 2D scatter plot

with a single data point evolving over time. Figure 11: The colored time flattening operation.

2.3. Discrete Time Flattening The colored time flattening operation is similar to the
time flattening operation, but time slices are assigned a color
1 2 3
before being combined (Figure 11). Although this opera-
tion does not map to any photography technique we know
of, similar results could in principle be obtained by rapidly
switching color filters during a long-exposure photography.

Time Two examples of visualizations obtained by colored time

flattening are shown in Figure 12: (a) a dynamic graph where
Figure 9: The discrete time flattening operation. old links (in red) are distinguished from new links (in blue)
[CKN∗ 03]; (b) Chinese characters where first strokes (in
black) are distinguished from later strokes (in red) [Wik13].
Discrete time flattening is similar to time flattening, but Minard’s map (Figure 7) also makes use of a simplified form
instead of merging all time slices into an image, a selection of colored time flattening, since the army’s forward march
of time slices is made before combining them (Figure 9). and return are distinguished using two different colors.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 5

(a) (b)

Figure 12: Two visualizations using colored time flattening.

(a) Illustration of a dynamic graph visualization as used in
G EVOL [CKN∗ 03]. (b) Stroke order in Chinese characters
[Wik13]; the color legends have been added.
Figure 14: Time juxtaposing showing approved forest har-
Time
vest applications across 10 years [Gre11].
1 2 3

1 2

Time

Figure 13: The time juxtaposing operation.

Time

Figure 15: The space cutting operation.

2.5. Time Juxtaposing
Time juxtaposing consists in extracting multiple time slices
then moved while the film is being exposed [TGF08]. Slit-
then placing them side-by-side or on a grid (Figure 13).
scan photography has been used to create special effects in
An analogy is Eadweard Muybridge’s multiple camera movies, artwork and photo finishes in sports.
chronophotography [Muy87]. In contrast with Marey, Muy-
bridge used multiple cameras that recorded snapshots on dif-
ferent locations on the film. He used it for the scientific study
of for example horse gaits, and his pictures famously settled
the question as to whether horses have all four feet off the
ground while trotting. Time juxtaposing is also the base for
many forms of sequential art, from ancient Egyptian murals
and Greek vase paintings to today’s comics [McC94].
Time juxtaposing is often used in information visualiza-
tion to show temporal data such as time-evolving maps, tra-
jectories in space [TBC13] and dynamic graphs [LNS11,
BBL12, RM13, BPF14a]. Figure 2.5 shows forest harvest
data over 11 years. In information visualization time jux-
taposing is usually referred to as small multiples [CKN∗ 03],
although small multiples are not necessarily built from time Figure 16: Example of space cutting: horizontal lines indi-
slices (see Figure 2(a)). Time juxtaposing has been also cate train stops, vertical lines indicate times, and diagonal
widely used for video summarization [TV07]. lines indicate moving trains [Mar78].

Space cutting has also been employed for visualizing tem-

2.6. Space Cutting
poral data. In the 19th century, Marey created a visualiza-
Space cutting consists in extracting a planar cut in a direc- tion using space cutting to visualize train connections be-
tion perpendicular to the data plane (Figure 15). An anal- tween major French cities (Figure 16). Space is cut along the
ogy is slit-scan photography, a process where a plate into rails connecting cities and diagonal lines indicate positions
which a slit has been cut is inserted in front of a camera and of trains at any time [Tuf86, Mar78].

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

6 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

Figure 17: Space cutting used to show road traffic [TGF08].

More recently, space cutting was shown to be useful for

analyzing video logs [TGF08]: Figure 17 shows a space cut Figure 20: An example of space flattening for showing arti-
(called tear in the original work) extracted from a video cle citations over time [SA06, AS].
scene, and revealing traffic activity (car count, speed and di-
rection) on a road. The time slice at t1 is shown to the left, Space flattening has also been used for visualizing dy-
together with the position of the segment extracted. The sys- namic networks [FBS06,SA06,BVB∗ 11]. For example, Fig-
tem is also able to show multiple longitudinal slices on top ure 20 shows a screenshot from Semantic Substrates [SA06]
of each other (i.e. space juxtaposing). where the y-axis is a 1D graph layout, and the x-axis shows
when connections are established.
2.7. Space Flattening
2.8. Sampling
1 2
1 2 3

Time

Time
Figure 18: The space flattening operation.
Figure 21: The sampling operation.
Space flattening is similar to space cutting, but involves
flattening the cube along a particular direction on the data Sampling is a more complex operation that consists in ex-
plane instead of extracting a cut (see Figure 18). tracting space cuts (samples) from a space time cube at sev-
eral locations on the data plane, then rotating those samples
An example of use of space flattening in infovis is the in-place so they face the viewer (Figure 21).
History Flow technique for visualizing document histo-
ries [VWD04], illustrated in Figure 19: the right panel shows Two examples of sampling are mentioned in this article’s
the last revision of a Wikipedia article, each color corre- introduction (Figure 1). The top one shows the evolution of
sponding to a specific contributor. The left visualization crime statistics in every US state [AA04b], while the bot-
shows the history of the article, built by collapsing each ar- tom one shows the evolution of high school population in
ticle revision into a one-pixel column, and then displaying several districts across three years [SRdJ05]. Although addi-
all columns side-by-side. These operations are equivalent to tional operations are involved (e.g., using silhouette graphs
flattening the article’s space-time cube along the x data axis. to encode values), both examples are conceptually based on
a sampling operation. Sampling has also been used in dy-
namic network visualization, for conveying changes in edge
weight [BN11] and in attribute values [HSCW13].

2.9. 3D Rendering

Time

Figure 22: The 3D rendering operation.

3D rendering consists in showing a space-time cube the

Figure 19: An example of space flattening showing the edit way three-dimensional objects are typically displayed on 2D
history of a Wikipedia article [VWD04]. media, i.e. by projecting it onto a 2D plane (Figure 22).

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 7

3D rendering is essentially a flattening operation but in The ultimate goal of space-time cube operations is to
contrast with time flattening and space flattening, it is (i) transform a space-time cube into a space-time object whose
typically done on a plane not orthogonal to the cube’s princi- shape is compatible with the shape of the media employed
pal axes; (ii) can involve a non-orthographic projection (e.g., to convey the information. By media we mean the visual-
perspective projection); (iii) can involve 3D shading, i.e. the ization’s physical presentation, which is the physical object
addition of light reflections and shadows. or apparatus that makes a visualization observable to the
viewer [JD13]. In the vast majority of cases (i.e. computer
displays and paper) the media has a planar shape.
For a given media, a space-time cube operation is com-
plete if it takes space-time volumes as input and produces
space-time objects whose shape match the media’s shape.
Otherwise the operation is incomplete: it cannot be used to
produce a valid visualization from a space-time cube. Sev-
eral elementary space-time cube operations can be chained,
in which case they form compound operations. A compound
operation is complete if the first operation takes space-time
(a) (b) volumes as input, and the last operation produces space-time
objects whose shape is compatible with the media.
Figure 23: Two examples of 3D rendering. (a) Occurrence of
earthquakes (authors’ illustration after [GAA04]), and (b) a
dynamic Network [DG04] 3.2. A Taxonomy of Elementary Space-time cube
operations
A taxonomy of elementary space-time cube operations is
In geography and geology, 3D rendering has been used shown in Figure 24 on the next page. The taxonomy breaks
to visualize events such as earthquakes (Figure 23(a)) or the down space-time cube operations into five main classes:
movement of objects [Kra03, GAA04]. 3D rendering is also
common in temporal information visualization. For exam- • Extraction consists in selecting a subset of a space-time
ple, in networks whose connectivity change over time, nodes object (e.g., extracting a line or cut from a volume),
can be represented as columns and links as bridges [DG04, • Flattening consists in aggregating a space-time object into
BC03] (Figure 23(b)). When the layout of the dynamic net- a lower-dimensional space-time object (e.g., projecting a
work also changes, nodes become worms [DE02, GHW09]. volume onto a surface),
• Filling consists in turning a set of disconnected space-
time objects into a fully connected space-time object,
3. The Design Space of Space-Time Cube Operations • Geometry transformation consists in transforming a
The previous section reviewed several common operations space-time object spatially without change of content,
that turn a conceptual time-space cube into a final two- • Content transformation consists in changing the content
dimensional visualization. Those were examples selected for of a space-time object without affecting its geometry.
illustration, and the list was not meant to be exhaustive. In The table in Figure 24 shows how general operations
addition, some operations were rather simple (e.g., time cut- break down into more specific operations. On each of the
ting), while others were more complex (e.g., sampling) and two columns, general operations are on the left while more
could be described as a composition of several lower-level specific operations are on the right. Operations that are the
operations. Therefore, we provide in this section a more sys- most specialized (i.e. leaves on the taxonomy tree) are shown
tematic description of the design space of space-time cube on a white background. Operations written in bold are those
operations. which produce planar surfaces, i.e. can be used as final op-
erations on screen-based and paper-based media.

3.1. Basic Terminology We quickly review the most specialized operations (white
background), going from top to bottom on the left column,
A space-time cube operation takes a space-time object and then on the right column. We also describe the parameters
produces another space-time object. A space-time object is necessary to specify each space-time cube operation. Most
a geometrical object within a space-time coordinate system of the operations have already been used in infovis, others
(i.e. two spatial dimensions and one temporal dimension). have been added for completeness.
Possible space-time objects include (i) space-time volumes
• Extraction:
(of which a complete space-time cube is an example), (ii)
space-time surfaces (planar and non-planar), (iii) space-time – Point extraction consists in selecting a specific point
curves, (iv) points, as well as (v) sets of disconnected vol- inside a space-time volume. This operation is defined
umes, surfaces, curves and points. by a 2D position on the data plane and a time value.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

8 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

Operations Time Space Operations Time Space

Point Point Extraction

Orthogonal
Interpolation

Filling
Time Space
Orthogonal
Interpolation Interpolation
Drilling Time Space
Planar Drilling Drilling
Volume Interpolation
Drilling
Curve Oblique Drilling

Planar Translation
Curvilinear Drilling Time Space
Shiting Shiting
Non-Planar Drilling Rigid
Transfor-

Geometry Transformation
Yaw
mation
Rotation
Orthogonal Pitch
Cutting Time
Planar Linear Space
Cutting Cutting Roll
Cutting
Extraction

Surface Oblique Cutting Scaling

Time Space
Scaling Scaling
Curvilinear
Non- Space Cutting Bending
Planar
Cutting
Other
Unfolding

Orthogonal
Chopping
Planar Time Linear Space
Chopping Chopping Chopping Time Space
Coloring Coloring
Recoloring

Volume Oblique Chopping

Diference
Curvilinear Coloring
Content Transformation

Non- Space Chopping

Planar
Chopping Labeling
Other Time Labeling

Stabilizing

Orthogonal Repositioning
Flattening
Flattening

Planar Bundling
Flattening Time Space
Flattening Flattening
Shading

Oblique Flattening
Filtering

Non-Planar Flattening Aggregation

Figure 24: Taxonomy of elementary space-time cube operations with schematic illustrations. Gray shading indicates non-
leaves, bold indicates complete operations. The Time column regroups operations that are applied according to the time axis,
while the Space column regroups operations that are applied according to the data plane.
submitted to Eurographics Conference on Visualization (EuroVis) (2014)
 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 9

– Time drilling consists in extracting a line parallel with This operation takes as parameters a time value, a
the time axis. It is uniquely specified by a 2D position projection function and an aggregation function. The
on the data plane. For example, sampling (Section 2.8) projection function maps 3D points to points on the
uses several drilling operations. plane. Examples include orthographic projection and
– Space drilling extracts a line perpendicular with the perspective projection. The aggregation function de-
time axis. It is specified by a 2D line and a time value. scribes how point values are combined. If values are
– Oblique drilling consists in extracting an arbitrarily defined in an RGBA color space, the function maps
oriented straight line from within a space-time volume. vectors of RGBA colors to a single RGBA color. Ex-
– Planar curvilinear drilling consists in extracting a amples of such functions include alpha-blending (e.g.,
planar 3D curve from a space-time volume. This op- averaging all colors) and overplotting (i.e. only keep-
eration, as well as all operations above, is complete ing the last color).
for 2D media. – Space flattening, oblique flattening and non-planar
– Non-planar curvilinear drilling consists in extract- flattening are similar operations, but the surface on
ing an arbitrary 3D curve from a space-time volume. which the volume is projected is different (see previ-
It is incomplete, and hence needs to be combined ous cutting operations as well as Sections 2.7 and 2.9
with other operations like flattening or unfolding. This for more details).
operation can be used to extract object trajectories
• Filling:
[KW04, RFF∗ 08].
– Time cutting consists in extracting a planar cut from – Time interpolation consists in filling “holes” in
a space-time volume in a direction orthogonal to the space-time objects (volumes, surfaces or curves) by
time axis (see Section 2.1). It takes as parameter a time interpolating between values along the time axis. It
value that defines the cut position on the time axis. It takes as parameter a monovariate interpolation func-
is a complete operation for 2D media. tion. For example, a piecewise linear time interpola-
– Linear space cutting consists in extracting a planar tion operation will transform a set of time slices into a
cut from a space-time volume in a direction orthogonal full space-time cube by linearly interpolating the val-
to the data plane (see Section 2.6). It is also complete, ues (e.g., RGBA colors) between pairs of successive
and takes as parameter a line or a segment parallel to time slices.
the data plane that once extruded over time defines the – Space interpolation consists in filling “holes” in
cutting surface. space-time objects by interpolating between values on
– Oblique cutting consists in extracting a planar cut each data plane. It takes as parameter a bivariate inter-
from a space-time volume that is neither orthogonal polation function. For example, a bilinear space inter-
to the time axis, nor orthogonal to the data plane (e.g. polation operation will transform a set of lines parallel
[FLM00]). It takes as parameter a 3D cutting plane. to the time axis into a full space-time cube.
– Curvilinear space cutting is similar to linear space – Volume interpolation consists in filling “holes” in
cutting except the cutting surface is produced by ex- space-time objects by interpolating across both space
truding a curve parallel to the data plane that is nei- and time. It takes as parameter a trivariate interpolation
ther a line nor a segment. This operation produces function. One example is interpolating video frames
non-planar space-time surfaces that further need to be using motion estimation techniques [CLK00].
flattened (e.g., using 3D rendering [TSAA12]) or un-
• Geometry Transformation:
folded (as in Figure 16).
– Time chopping is similar to time cutting but slices – Space shifting, time shifting, yaw, roll and pitch
have a thickness instead of being infinitely thin. Since consist in moving or rotating space-time objects. They
it produces volumes it is not complete for 2D media, can be used, e.g., for placing multiple cuts side-by-side
and thus needs to be complemented with additional op- or for rotating an entire space time cube rendered in
erations. It takes as parameter a time segment that de- 3D (e.g. [KW04,CCT∗ 99,BPF14b]). They each take a
fines the two cutting slabs (a slab is the infinite region single scalar value as parameter.
between two planes). – Time scaling and space scaling rescale space-time
– Linear space chopping, oblique chopping and objects along their principal axes. They take as param-
curvilinear space chopping are similar to the previ- eters one and two scalar values respectively, that define
ous cutting operations, with the difference that they the scaling factor.
produce volumes with a certain thickness instead of – Bending deforms space-time objects. For example, a
infinitely thin surfaces. space-time volume can be bent such that the time axis
follows an arc instead of a line [DC03]. This operation
• Flattening:
takes as parameter a deformation function that maps
– Time flattening aggregates a space-time volume into 3D locations to 3D locations.
a plane orthogonal to the time axis (see Section 2.2). – Unfolding transforms a non-planar space-time surface

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

10 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

into a planar space-time surface. An analogy is a map tive video fast-forward [PJH05]. Similarly, an unfolding op-
projection function that transforms a sphere or portion eration that works on any surface (as opposed to, e.g., only
of sphere into a plane. An example of space-time un- spheres), would be an adaptive unfolding operation.
folding is Maray’s train schedule (Figure 16), which
Semantic operations take into account the data seman-
can be seen as an unfolded curvilinear space cut per-
tics of the space-time objects they operate on. One exam-
formed on a time-evolving 2D map.
ple would be a semantic volume interpolation operation that
• Content Transformation: connects discrete sets of moving objects with lines or tubes
(see Figures 8(b) and 23(b), as well as [Ros06, BPF14a]).
– Time coloring consists in altering the colors of each
This type of operation is semantic because it needs to know
time slice according to time. Examples include col-
the identity of the objects to be able to match them on suc-
oring each time slice uniformly according to a linear
cessive time slices. Time labeling operations such as the one
color scale (Figure 12), changing the hue of each time
used in Figure 8(b) are also semantic, because they need to
slice, or dividing the time axis in different regions and
know the location of datapoints of interest to place the labels
applying a discrete color scale (Figure 7).
appropriately. Filtering operations can also be semantic, as
– Space coloring alters the color of points in a space-
well as recoloring operations [VWD04, RFF∗ 08, BPF14b].
time volume depending on their 2D position on the
Finally, semantic operations can also be used to cut cubes
data plane.
according to specific temporal cycles (days, weeks, etc.).
– Difference coloring consists in altering the colors
of each time slice according to the difference be-
tween time slices. One example is highlighting ap-
3.4. Compound Operations
pearing nodes and disappearing nodes in a dynamic
graph [RM13, BPF14a]. We previously defined compound operations as several oper-
– Time labeling consists in adding time labels to each ations applied in sequence. According to our taxonomy from
time slice or to objects inside a space-time volume Figure 24, some of the operations we introduced in Section
(Figure 8(b)). 2 are elementary, namely time cutting, time flattening, space
– Stabilizing consists in repositioning objects on each flattening. Others are compound and can be broken down as
data plane so that their trajectories are as parallel as indicated in Table 1. In our notation, the symbol + refers to a
possible to the time axis. Examples include comput- composition, the symbol ∗ refers to operations applied mul-
ing stable layouts for dynamic networks [AP12a] and tiple times and the symbols [ ] refer to optional operations.
stabilizing videos [BGPS07].
– Bundling consists in repositioning objects on each
Compound Operation Elementary Operations
data plane in order to bring their trajectories closer to
Discrete time flattening time cutting* + time flattening
each other. One example is bundling air plane routes
Colored time flattening time coloring + time flattening
[HEF∗ 13].
Time juxtaposing (time cutting + space scaling + space
– Shading consists in altering the color of a space- shifting)* + time flattening
time volume’s content by simulating light propagation Marey’s schedule curvilinear space cutting + yaw + un-
mechanisms (e.g., diffusion, specular reflection, drop folding
shadows). Slit tears (linear/curvilinear space cutting +
– Filtering consists in removing parts of a space-time yaw + [unfolding] + space shifting)*
volume’s content. One example is removing all points Sampling (time drilling + time scaling + yaw)*
of a certain color or value [CCT∗ 99, DC03, BPF14b]. 3D rendering [shading] + oblique flattening
– Aggregation replaces multiple space-time objects by
Table 1: Compound operations decomposed.
a single, larger space-time object. Different methods
exist. For example, 3D kernel density estimation trans-
forms a set of space-time points or space-time curves
A compound operation inherits the parameters of its sub-
into 3D volumes or 2D (iso) surfaces [DV10].
operations. For example, a discrete time flattening operation
is specified by a sequence of time values, as well as a pro-
3.3. Adaptive and Semantic Operations jection function and an aggregation function. But in practice,
most compound operations enforce constraints between their
So far we mostly described operations that are agnostic to the
parameters. For example, all space scalings from a time jux-
data and the content of the cube. Adaptive operations take
taposing operation are typically the same.
into account the shape or content of the particular space-time
objects they operate on. For example, an adaptive time cut- Many elaborate temporal data visualization techniques
ting operation can be used to cut cubes according to regions can be described as compound operations. For example, the
with large changes instead of cutting them into regularly- Visits technique (Figure 25) employs (time chopping + time
spaced slices. This technique is used, for example, in adap- flattening + space shifting)*.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 11

erations exist. For example, animated 3D rendering can ex-

plain a transition between two space-time cube operations
to a user by smoothly rotating a space-time cube representa-
tion [BPF14b]. This technique will be discussed in Section
4, where we review space-time cube visualization systems.

3.5.2. Interaction
Figure 25: Compound operation in Visits [TBC13]. Interaction is similar to animation except the changes in the
space-time cube operations are under the user’s control.

3.5. Dynamic Operations Consider animated time cutting: if the position of the cut-
ting plane is controlled by the user (e.g., by dragging a
So far we only considered operations (both elementary and slider) instead of being automatically moved, then the op-
compound) that transform a space-time cube into a static vi- eration becomes interactive time cutting. A common imple-
sual representation. On computer displays, operations can mentation of interactive time cutting is the seeker bar on a
also be applied in a dynamic manner. Dynamic operations video player. As with animations, any operation can be made
can involve either animation or interaction. interactive. Examples of interactive operations abound, and
we will review some of them in the next section.
3.5.1. Animation
We refer to animation as the process of applying different 4. Space-Time Cube Systems
operations on a space-time cube over time, or similarly, vary- Choosing an appropriate space-time cube operation depends
ing the parameters of an operation over time. on many factors and almost always involves tradeoffs. In this
The most common form of animation consists in changing section we review a representative sample of visualization
the position of a cutting operation over time, i.e. animated systems that address this issue by supporting multiple space-
time cutting. This results in the space-time cube content be- time cube operations. Such systems almost invariably use
ing “played back”. For example, if the space-time cube rep- 3D rendering as an explicit representation of the space time
resents a visual scene like video surveillance data, synchro- cube, both for showing an overview and for explaining how
nizing the motion of the slice with a clock will result in a different operations relate. We call these systems space-time
real-time playback of the original scene. When significant cube systems. Because they work by letting people switch
data is skipped during playback, the animation is closer to a between different operations and tune their parameters, in-
discrete time juxtaposing operation, except slices are shown teraction is a key feature.
in sequence instead of being laid out side-by-side.
An animated time cutting operation can be preceded by a 4.1. CommonGIS
filling operation in order to produce smooth animated tran-
sitions. Many examples exist in the literature, for exam-
ple when animating dynamic networks [ATMS∗ 11, RM13,
BPF14a] or scatterplots [Ros06, RFF∗ 08]. Most of these ex-
amples can be described as semantic volume interpolation
+ animated time cutting operations. Animated time cutting
can also be combined with other space-time cube operations
such as time flattening. For example, Gapminder can com-
bine scatterplot animations with static trails for points of in-
terest (a filtering + time flattening operation) [Ros06].
While many animation techniques can be described as an-
Figure 26: CommonGIS [AA99] (picture from [AA04a])
imated time cutting on static space-time cubes, more elabo-
rate techniques require operations to be applied in real-time.
CommonGIS [AA99] is a feature-rich analytical system
For example, Hurter et al.’s system [HEF∗ 13] uses animated
for spatio-temporal data. It supports several space-time cube
time chopping to animate a network over time while preserv-
operations, including time flattening and 3D rendering (Fig-
ing temporal context information. At every animation frame,
ure 26). The 3D rendering view is combined with a semantic
a time flattening is applied that produces colored trails and
filtering operation to make the space-time cube transparent:
a dynamic bundling operation is applied that guarantees a
geographical context is only shown on a single time slice, as
continuous animation without jumping bundles [HET12].
a reference plane. Two widgets provide control of the projec-
Although animated time cutting and its many variants are tion function (arrow in Figure 26). One controls the camera
the most common forms of animation, other animated op- position around the cube, the other one controls its height.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

12 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

4.2. GeoTime can, e.g., select a particular type of vegetation or a range of

vegetation ages. In addition, Tardis supports interactive or-
22:00
N
22:00
thogonal cutting, but in contrast with our previous examples,
cutting is always used in combination with 3D rendering.
20:00
20:00
18:00
18:00
16:00
Users can define and manipulate multiple orthogonal cut-
16:00
14:00 ting planes (Figure 28(a)). Further operations include open-
ing the cube like a book (interactive (volume extraction +
14:00
12:00
12:00

10:00
10:00
rotation)*) or apply a 3D fisheye effect (interactive (volume
extraction + translation)*). This fisheye effect, called “Vi-
08:00

08:00 N

sual Access Distortion”, pushes away voxels from the cursor

(Figure 28(b)).

4.4. VISUAL-TimePAcTS
(a) 3D rendering (b) Space flattening (on top)

Figure 27: Illustration after GeoTime [geo]

GeoTime is a carefully-designed commercial system for

analyzing spatio-temporal data [geo, KW04]. Events are
shown as spheres on a 3D rendering view that can be freely
rotated (Figure 27(a)). This view also uses a reference plane,
and a semantic volume interpolation operation is applied to (a) (b) (c)
indicate event ordering. Users can perform time chopping
operations by dragging on a timeline widget. GeoTime also
supports time flattening and space flattening. Figure 27(a)
shows a space flattening view where time runs from top to
bottom, and a reference plane is provided that can be rotated.
Thin gray lines connect the two views. Finally, pan & zoom
is supported through space chopping + space scaling.
(d) Shearing explained (e) The result of shearing

4.3. Tardis Figure 29: VISUAL TimePAcTS [VFC10]. (a) Space Flat-
tening on activities, (b) Oblique flattening, (c) Space flatten-
ing on individuals.

VISUAL-TimePAcTS is a system for analyzing activity

diaries [VFC10]. It uses non-geographical space-time cubes.
The cube’s two data axes can be mapped to data dimensions
such as individuals, locations, or activities. We focus on the
case where one axis maps to individuals while the other axis
maps to activities. Activities are also encoded using color.
VISUAL-TimePAcTS supports linear space flattening on
both data axes. Figure 29(a) shows 6 individuals (horizontal
(a) 3D rendering with multiple (b) 3D rendering with multiple
time and space cutting volume extraction+translation axis) and their activities (colors) across time (vertical axis).
Figure 29(c) shows the evolution of activities aggregated
Figure 28: Tardis [CCT∗ 99] and visual access [CFC∗ 96] across all people over time. VISUAL-TimePAcTS supports
a seamless transition between the two operations through in-
teractive 3D rendering (Figure 29(c)). Since 3D rendering
Tardis [CCT∗ 99, CFC∗ 96] is a system for visualizing en- employs orthographic projection and no shading, it is essen-
vironmental data using 3D rendering in combination with tially an oblique flattening operation.
advanced space-time cube operations. The voxels in the cube
VISUAL-TimePAcTS supports a more elaborate space-
are color-coded depending on the type of vegetation, its age,
time cube operation that prevents visual marks from overlap-
soil characteristics or the presence of bush fires (Figure 28).
ping due to flattening. In Figure 29(c), for example, individ-
Tardis implements interactive semantic filtering: users uals are horizontally offset when several of them do the same

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 13

activity at the same time. This technique is called shearing

by the authors, and is further explained in Figures 29(d),
29(e). This technique is essentially a (linear space cutting +
space offset)* + space flattening operation, and is a hybrid
between space juxtaposing and space flattening.

4.5. Cubix (a) (b)

Figure 31: (a) Video Cubism [FLM00]; (b) V 3 [DC03].

with an interactive volume extraction operation that is de-

fined by manipulating a planar cutting plane (Figure 31(a)).
Similary, Khronos projector [CI05] supports manipulation
of a non-planar cutting plane using touch or mid-air gestures
(Figure 4). V 3 [DC03] (Volume Visualization for Videos)
supports different operations, including time juxtaposing and
a 3D rendering view that can be combined with a bending
operation (Figure 31(b)). V 3 also supports filtering opera-
tions that allow removal of pixels of a certain color, or pixels
that do not change across a given time period.
Besides the space-time cube systems reviewed in this sec-
tion, there is a wealth of general 3D visualization systems.
Commercial and research tools exist in domains such as geo-
visualization (e.g., Voxler [Vox], ArcGIS [arc]), scientific vi-
Figure 30: Different operations applied to a time-evolving sualization (e.g., VTK [SAH00], Matlab [mat] and R [r]),
adjacency matrix in Cubix [BPF14b] and medical visualization [MTB03]. Although these tools do
not treat time as a specific dimension, they can be used to in-
spire the design of interactive space-time cube systems.
Cubix is a system for analyzing dynamic weighted net-
works through adjacency matrix representations [BPF14b].
A 3D rendering provides an overview of the data (Figure 5. General Discussion
30(a)). Time goes from left to right. Each cell of the cube We now discuss the limitations of our descriptive framework
represents a connection between two nodes at a given time, and consider areas for future research, including: unify-
with size depending on connection weight. Cells can be ing our framework with the infovis pipeline model, extend-
color-coded according to time, weight, or direction. ing it to other dimensionalities, considering non-planar me-
Cubix supports a range of space-time cube operations, in- dia such as physical visualizations, characterizing the inner
cluding time juxtaposing (Figure 30(b)), space juxtaposing structure of time space cubes, and considering the strengths
(detail in Figure 30(d)), animated time cutting , animated and weaknesses of different space-time cube operations.
space cutting, time flattening and space flattening. For flat-
tening operations, cells can be made translucent to visually 5.1. Comparison with the Infovis Pipeline Model
aggregate the history of connections. Cubix also supports se-
mantic filtering on connections based on their weight. Since our framework builds on the notion of composition
of operations, it shares similarities with another common
Cubix provides a control widget in the form of a stylized model: the infovis reference model, also called the info-
cube, and whose different parts can be clicked or dragged vis pipeline [CMS99, Chi00, JD13]. The infovis pipeline
to switch between operations. All operation switches are ex- sees visualization as a data-flow process, i.e., a sequence of
plained using animated transitions through rotations of the stages and transformations that turn raw data into a final im-
3D rendering representation, or through staged animations age. These transformations commonly include data trans-
of extraction and rigid transformation operations. formation, visual mapping, presentation mapping and ren-
dering [JD13]. Interactivity is implemented by having data
4.6. Video Cube Systems analysts alter these transformations at different stages.
Several space-time cube systems have been proposed to sup- There is clearly an analogy between transformations and
port video analysis [MB98,FLM00,DC03,CI05]. Video Cu- operations. However, the infovis pipeline and our framework
bism [FLM00] uses a 3D rendering representation together differ in several important respects. The infovis pipeline is a

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

14 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

general model for visualizations, where the sequence of op-

erations is fixed, but the operations themselves are rather ab-
stract. For example, the pipeline model provides no specific
details about what happens in the visual mapping transfor-
mation. In contrast, our model only captures a specific fam-
ily of visualizations (temporal visualizations), its sequence
of operations is not fixed, and the operations are more con-
crete. The infovis pipeline is more general but too high-level
to capture the similarities and differences between the vi-
sualizations we presented. On the other hand, our model is
incomplete in that it does not define how the space-time cube
is built. The two models are therefore complementary.
Since the infovis pipeline has inspired the software archi- Figure 32: Two physical implementations of the Matrix
tecture of several infovis tools [Fek04], it is worth consid- Cube visualization [BPF14b] for dynamic networks, made
ering to what extent the two models can be unified. Sev- by this article’s first author.
eral space-time cube operations could in principle be imple-
mented at different stages of the infovis pipeline. For exam- since the richness of time-series visualizations and timelines
ple, time flattening can be performed at the data transforma- mostly stem from the visual encodings used, the usefulness
tion stage, by aggregating raw data over time. Alternatively, of our framework would be less clear in this case.
time flattening could be emulated by explicitly rendering a
3D space-time cube on the screen and using a proper cam- Other temporal visualization techniques, although less
era placement and projection transformation. In that case, it common, show three spatial dimensions plus time. We be-
would be implemented at the rendering level. lieve most of them can be captured with operations on 4-
dimensional space-time hypercubes. For example, Tufte ex-
However, these approaches would only provide a very plains how small multiples can be used to show the evolu-
partial support for space-time cube operations. For a full sup- tion of a three-dimensional storm [BB95]. This approach
port, our space-time cube needs to be reified as a first-class amounts to applying a time juxtaposition operation on a
object. Since it is both abstract and visual, our space-time space-time hypercube, where each time cutting operation
cube best aligns with the abstract visual form stage of the yields a 3D image. Similarly, FromDady [HTC09] uses 3D
pipeline [JD13]. Thus space-time cube operations are best trails to show the trajectories of airplanes in space. This tech-
seen as presentation mapping transformations, i.e., trans- nique amounts to performing a time flattening on a space-
formations that turn the abstract visual form into a fully- time hypercube. Extending our framework to higher data di-
specified 2D image or 3D model [JD13]. In other terms, our mensionalities is an exciting topic for future research. How-
space-time cube operations can be used to decompose and ever, it is less easy to imagine a hypercube than a cube, so
refine the presentation mapping transformation of the info- the merits of such a conceptual model still remain to be seen.
vis pipeline. We believe that implementing our framework
in this way could dramatically facilitate the exploration of a
wide range of temporal visualization techniques. 5.3. Non-Planar Media
Throughout this review we assumed the presentation
medium to be planar. Although these are by far the most
5.2. Other Dimensionalities
common, other display shapes are being explored in HCI,
This review focused on temporal visualizations that involve some of which are even deformable [RPPH12, HV08]. In
two spatial dimensions plus time. These two dimensions can these cases, the conditions for an operation to be complete
be inherently spatial or can result from 2D spatial encodings are not the same. This opens up a wide range of possibili-
of abstract data. However, temporal visualizations with other ties for new visualization designs. For example, one imple-
dimensionalities are possible. mentation of the Khronos projector (Figure 4) employs back
projection on a freely deformable cloth, allowing the use
Most notably, a rich variety of temporal visualizations ex-
of non-planar cutting operations that are complete. In ad-
ist that involve a single spatial dimension plus time, e.g.,
dition, physical visualizations make it possible to faithfully
timelines and time-series visualizations [AMST11]. In prin-
display 3D space-time cubes without any additional oper-
ciple, our framework still applies if the 3D space-time cube
ations [JDF13, JD13]. Many such visualizations have been
is turned into a 2D space-time plane. Operations analogous
already crafted by scientists, artists and designers [DJ13].
to our geometry transformation operations would capture
techniques such as spiral visualizations, calendar visualiza- Physical temporal visualizations can even be made mod-
tions or cycle plots [AMST11]. However, since a 2D space- ular to support interactive space-time cube operations. Fig-
time plane already naturally maps to a 2D planar display, and ure 32 shows two physical representations of a dynamic net-

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 15

work [BPF14b] made of laser-cut and laser-engraved acrylic.

The left version supports interactive time cutting while the
right version supports interactive space cutting. Cuts can be
taken apart and manipulated freely, allowing for time juxta-
posing and space juxtaposing as well as time flattening and
space flattening, if viewed from a proper orthogonal angle
and distance. For another example see [STB].

5.4. The Inner Structure of Space-Time Cubes

In this review we considered space-time cubes as monolithic
entities. Two space-time cubes can however look quite dif-
ferent, depending on what data is visualized: migration of
animals, earthquakes, changes in vegetation and ecosystems,
networks with changing connections, surveillance videos, or
scatterplots evolving over time. We can refer to this lower- Figure 33: Visualization of energy consumption over time
level decomposition as the cube’s inner structure. [vWvS99]. One horizontal axis is mapped to days while the
other one is mapped to hours.
Inner structure is likely to be an important factor when
choosing between space-time cube operations. For example, transformations they use (e.g., aggregation [BCH07] or fil-
videos produce maximally dense inner structures and there- tering [AABW12]), and in the interactions they implement.
fore are not well-suited to 3D rendering, unless comple- Overall, findings are very hard to compare for a lack of a
mentary operations such as filtering are used (Figure 31(b)). consistent descriptive terminology.
Structures that present large variations over time may not
be well-suited to time juxtaposing because objects could be A clear and descriptive framework would allow us to bet-
difficult to relate across time slices. We plan to extend our ter structure this body of evidence and tease apart the ef-
descriptive framework by characterizing different types of fects of different subtle design features and to better control
inner structures for space-time cubes. for confounds. We showed how many temporal visualiza-
tion techniques can be decomposed into elementary opera-
tions. These operations can be combined in many ways or
5.5. Which Operation to Choose? made dynamic at different levels (either through animation
or interaction). The characterization of the inner structure
Again, our contribution is a descriptive conceptual frame- of space-time cubes, may already provide many elements to
work, and we chose not consider the relative merits and lim- discuss the practical strengths and weaknesses of space-time
itations of different techniques. We believe that a detailed cube operations, mostly based on well-established knowl-
descriptive model is a necessary first step, before consider- edge on perception and HCI, and on common sense.
ing performance issues. The question however remains as of
which operation works best under which context. Running studies for answering specific research questions
will naturally remain important, and we hope our descriptive
Several studies comparing space-time cube operations framework will facilitate the design of such studies and the
have been reported in the past. Many of them com- discussion of findings in a more informative manner.
pared animation (usually animated time cutting) against
static visualizations (usually time juxtaposing) [TMB02,
5.6. Other Limitations
GMH∗ 06, RFF∗ 08, APP11, FQ11, AP12b], finding benefits
and drawbacks for both approaches. Other studies evalu- Our framework is designed as a thinking tool. Like any
ated space time-cube representations (i.e., 3D rendering) model, it is necessarily incomplete. First, our taxonomy of
against a range of baseline conditions, including time flat- elementary operations in Figure 24 is not meant to cover all
tening [KDA∗ 09, WvdWvW09, KK12, AABW12], time cut- possible operations. Second, our framework does not pro-
ting [WvdWvW09, BCH07], and space cutting [AABW12]. vide much guidance for interaction design: the design space
Although such empirical investigations are crucial for the for interactive operations has only been partially explored.
advancement of science, only a subset of all possible oper- Finally, not all techniques for visualizing temporal data can
ations have been covered, and many of the existing findings be captured with space-time cubes. For example, temporal
are hard to generalize beyond the domain, type of data, and data can be visualized using two time axes instead of a sin-
operation parameters used in the stimuli. For example, ani- gle data axis (see Figure 33). Our framework is not meant to
mation can mean either animated time cutting or interactive restrict creativity but rather to help visualization designers
time cutting, and space-time cube visualizations can be im- find new solutions, extend or generalize existing ones, and
plemented in a multitude of ways, including in the content think out of the box.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

16 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

Again, our framework assumes that the space-time cube References

already exists. It does not provide guidance for producing the [AA99] A NDRIENKO G., A NDRIENKO N.: Making a GIS in-
space-time cube itself. For abstract data, many visual map- telligent: CommonGIS project view. Proceedings of AGILE99
pings can be used to produce individual slices. For example, (1999), 19–24. 11
locations on a map can yield values for altitude, temperature, [AA03] A NDRIENKO N., A NDRIENKO G.: Visual Data Explo-
rain, vegetation and soil type. How to visualize all these at- ration using Space-Time Cube. In Proceedings of International
tributes at any particular point in time is a general problem Cartographic Conference (Durban, South Africa, 2003), Interna-
tional Cartographic Association, pp. 1981–1983. 2
of information visualization, but the choice may also affect
the efficiency of later space-time cube operations. [AA04a] A NDRIENKO G., A NDRIENKO N.: Research
on visual analysis of spatio-temporal data at Fraun-
hofer AIS: an overview of history and functionality of
CommonGIS. http://geoanalytics.net/and/
KDworkshopPaper2004/KDworkshop.html, 2004.
6. Conclusion [online, accessed:12-apr-2014]. 11
We reviewed various techniques to visualize temporal data, [AA04b] A NDRIENKO N., A NDRIENKO G.: Interactive visual
by describing them as sequences of parametric operations tools to explore spatio-temporal variation. In Proceedings of the
working conference on Advanced visual interfaces (New York,
applied to a conceptual space-time cube. Our operations are NY, USA, 2004), AVI ’04, ACM, pp. 417–420. 1, 2, 6
independent from the underlying data and can be applied
[AA12] A NDRIENKO N., A NDRIENKO G.: Visual analytics of
across a range of application domains, be they cartography, movement: An overview of methods, tools and procedures. In-
dynamic network analysis, geopolitics, or video analytics. formation Visualization 12, 1 (2012), 3–24. 1

By introducing domain-agnostic concepts and a clear ter- [AAB∗ 11] A NDRIENKO G., A NDRIENKO N., BAK P., K EIM D.,
K ISILEVICH S., W ROBEL S.: A Conceptual Framework and
minology, this article aims at facilitating the comparison of Taxonomy of Techniques for Analyzing Movement. Journal of
different approaches for visualizing temporal data. Exist- Visual Languages and Computing 22, 3 (June 2011), 213–232. 1
ing visualizations from one data domain can be analyzed in [AABW12] A NDRIENKO G., A NDRIENKO N., B URCH M.,
terms of elementary operations and then be adapted to other W EISKOPF D.: Visual analytics methodology for eye movement
domains and problems. studies. Visualization and Computer Graphics, IEEE Transac-
tions on 18, 12 (Dec 2012), 2889–2898. 15
By giving a better vision of the richness of the design [AAH11] A NDRIENKO G., A NDRIENKO N., H EURICH M.: An
space, we hope our model will also motivate the exploration Event-based Conceptual Model for Context-aware Movement
of new approaches. It also stresses the importance of devel- Analysis. International Journal on Geographic Information Sci-
oping fully-integrated interactive systems and toolkits that ence 25, 9 (Sept. 2011), 1347–1370. 1
can support a range of techniques in a consistent manner. [AMM∗ 07] A IGNER W., M IKSCH S., M ÜLLER W., S CHU -
MANN H., T OMINSKI C.: Visualizing Time-oriented data-A
Our model also aims at facilitating the design of studies Systematic View. Computers and Graphics 31, 3 (June 2007),
and discussing their results in a more informative manner. 401–409. 1
We hope the presented terminology and low-level concepts [AMST11] A IGNER W., M IKSCH S., S CHUMANN H., T OMIN -
SKI C.: Visualization of Time-Oriented Data, 1st ed. Human-
will help design better experiments that tease out important
Computer Interaction. Springer Verlag, 2011. 1, 14
factors in dynamic data visualization. Many more controlled
studies are needed to understand the trade-offs between dif- [AP12a] A RCHAMBAULT D., P URCHASE H. C.: Mental Map
Preservation Helps User Orientation in Dynamic Graphs. In Pro-
ferent space-time cube operations and how they perform de- ceedings of Graph Drawing (2012), GD ’12, Springer, pp. 475–
pending on the task, the data, and the people that use them. 486. 10

This work mostly arose out of the need to teach temporal [AP12b] A RCHAMBAULT D., P URCHASE H. C.: The Mental
Map and Memorability in Dynamic Graphs. In Proceedings of
information visualization to undergrad students. We there- Pacific Visualization Symposium (2012), PacificVis ’12, IEEE
fore hope that it will help other people teach this field effec- Computer Society, pp. 89–96. 15
tively, by providing a clear structure and a clear terminology [APP11] A RCHAMBAULT D., P URCHASE H., P INAUD B.: An-
on which to base higher-level discussions and analyses. imation, Small Multiples, and the Effect of Mental Map Preser-
vation in Dynamic Graphs. IEEE Transactions on Visualization
and Computer Graphics 17, 4 (2011), 539–552. 15
[arc] ArcGIS. http://www.esri.com/software/
Acknowledgements arcgis. [online, accessed:02-apr-2014]. 13
Collaboration on this work was initiated by Dagstuhl Semi- [AS] A RIS A., S HNEIDERMAN B.: NVSS: Network Visualiza-
tion by Semantic Substrates. http://www.cs.umd.edu/
nar “Putting Data on the Map 12261”, June 2012, and sup- hcil/nvss/. [online, accessed:02-apr-2014]. 6
ported in part by NSERC, SMART Technologies, AITF,
[ATMS∗ 11] A HN J.-W., TAIEB -M AIMON M., S OPAN A.,
Surfnet and GRAND NCE, as well as by Clique Strate- P LAISANT C., S HNEIDERMAN B.: Temporal visualization of
gic Research Cluster funded by Science Foundation Ireland social network dynamics: prototypes for nation of neighbors. In
(SFI) Grant No. 08/SRC/I1407. Proceedings of International conference on Social computing,

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 17

behavioral-cultural modeling and prediction (Berlin, Heidelberg, WAMPLER K.: A system for graph-based visualization of the
2011), SBP’11, Springer-Verlag, pp. 309–316. 11 evolution of software. In Proceedings of ACM Symposium on
Software Visualization (New York, NY, USA, 2003), SoftVis ’03,
[BB95] BAKER M. P., B USHELL C.: After the storm: Consid-
ACM, pp. 77–ff. 4, 5
erations for information visualization. Computer Graphics and
Applications, IEEE 15, 3 (1995), 12–15. 14 [CLK00] C HOI B.-T., L EE S.-H., KO S.-J.: New frame rate up-
conversion using bi-directional motion estimation. IEEE Trans-
[BBL12] B OYANDIN I., B ERTINI E., L ALANNE D.: A Qualita-
actions on Consumer Electronics 46, 3 (2000), 603–609. 9
tive Study on the Exploration of Temporal Changes in Flow Maps
with Animation and Small-Multiples. Computer Graphics Forum [CMS99] C ARD S. K., M ACKINLAY J. D., S HNEIDERMAN B.
31, 3pt2 (2012), 1005–1014. 5 (Eds.): Readings in Information Visualization: Using Vision to
Think. Morgan Kaufmann Publishers Inc., San Francisco, CA,
[BC03] B RANDES U., C ORMAN S. R.: Visual unrolling of net- USA, 1999. 13
work evolution and the analysis of dynamic discourse. Informa-
tion Visualization 2, 1 (Mar. 2003), 40–50. 7 [DC03] DANIEL G., C HEN M.: Video Visualization. In Proceed-
ings of IEEE Visualization (Washington, DC, USA, 2003), VIS
[BCD∗ 12] B ORGO R., C HEN M., DAUBNEY B., G RUNDY E., ’03, IEEE Computer Society, pp. 409–416. 9, 10, 13
H EIDEMANN G., H ÖFERLIN B., H ÖFERLIN M., L EITTE H.,
W EISKOPF D., X IE X.: State of the Art Report on Video-Based [DE02] DWYER T., E ADES P.: Visualising a Fund Manager Flow
Graphics and Video Visualization. Computer Graphics Forum Graph with Columns and Worms. In Proceedings of Interna-
31, 8 (Dec. 2012), 2450–2477. 1 tional Conference on Information Visualisation (2002), IV ’02,
pp. 147–152. 7
[BCH07] B RUNSDON C., C ORCORAN J., H IGGS G.: Visualis-
ing space and time in crime patterns: A comparison of methods. [DG04] DWYER T., G ALLAGHER D. R.: Visualising changes in
Computers, Environment and Urban Systems 31, 1 (2007), 52 – fund manager holdings in two and a half-dimensions. Informa-
75. Extracting Information from Spatial Datasets. 15 tion Visualization 3, 4 (Dec. 2004), 227–244. 7
[BDH04] BARTOLI A., DALAL N., H ORAUD R.: Motion [DJ13] D RAGICEVIC P., JANSEN Y.: List of Physical Visualiza-
Panoramas. Computer Animation and Virtual Worlds, 15 (2004), tions. http://www.tinyurl.com/physvis, 2013. [On-
501–517. 4 line; accessed 04-Sep-2013]. 14

[BGPS07] BATTIATO S., G ALLO G., P UGLISI G., S CELLATO [DV10] D EMŠAR U., V IRRANTAUS K.: Space–time density
S.: SIFT features tracking for video stabilization. In Proceedings of trajectories: exploring spatio-temporal patterns in movement
of Conference on Image Analysis and Processing (2007), ICIAP data. International Journal of Geographical Information Science
’07, IEEE, pp. 825–830. 10 24, 10 (2010), 1527–1542. 10
[FBS06] FALKOWSKI T., BARTELHEIMER B., S PILIOPOULOU
[BN11] B RANDES U., N ICK B.: Asymmetric Relations in Lon-
M.: Mining and Visualizing the Evolution of Subgroups in Social
gitudinal Social Networks. IEEE Transactions on Visualization
Networks. In Web Intelligence, 2006. WI 2006. IEEE/WIC/ACM
and Computer Graphics 17, 12 (Dec. 2011), 2283–2290. 6
International Conference on (2006), pp. 52–58. 6
[BPF14a] BACH B., P IETRIGA E., F EKETE J.-D.: GraphDi- [Fek04] F EKETE J.: The infovis toolkit. In Information Visual-
aries: Animated Transitions and Temporal Navigation for Dy- ization, 2004. INFOVIS 2004. IEEE Symposium on (2004), IEEE,
namic Networks. IEEE Transactions on Visualization and Com- pp. 167–174. 14
puter Graphics (2014). to appear. 5, 10, 11
[FLM00] F ELS S., L EE E., M ASE K.: Techniques for Interactive
[BPF14b] BACH B., P IETRIGA E., F EKETE J.-D.: Visualiz- Video Cubism (poster session). In Proceedings of International
ing Dynamic Networks with Matrix Cubes. In Proceedings of conference on Multimedia (New York, NY, USA, 2000), MUL-
SIGCHI Conference on Human Factors and Computing Systems TIMEDIA ’00, ACM, pp. 368–370. 2, 9, 13
(New York, NY, USA, 2014), CHI ’14, ACM. to appear. 9, 10,
11, 13, 14, 15 [FQ11] FARRUGIA M., Q UIGLEY A.: Effective temporal graph
layout: A comporative stydy of animations versus statuc display
[BVB∗ 11] B URCH M., V EHLOW C., B ECK F., D IEHL S., methods. Information Visualization (2011), 47–64. 15
W EISKOPF D.: Parallel Edge Splatting for Scalable Dynamic
Graph Visualization. Visualization and Computer Graphics, [GAA04] G ATALSKY P., A NDRIENKO N., A NDRIENKO G.: In-
IEEE Transactions on 17, 12 (2011), 2344–2353. 6 teractive Analysis of Event Data Using Space-Time Cube. In
Proceedings of the Information Visualisation, Eighth Interna-
[CCT∗ 99] C ARPENDALE S. T., C OWPERTHWAITE D. J., tional Conference (Washington, DC, USA, 2004), IV ’04, IEEE
T IGGES M., FALL A., F RACCHIA F. D.: The Tardis: A Vi- Computer Society, pp. 145–152. 7
sual Exploration Environment for Landscape Dynamics. In Pro-
ceedings of Conference on Visual Data Exploration and Analysis [geo] GeoTime. http://www.geotime.com. [Online; ac-
(1999), no. 3643, pp. 110–119. 2, 9, 10, 12 cessed 24-Jan-2014]. 12

[CFC∗ 96] C ARPENDALE S. T., FALL A., C OWPERTHWAITE [GHW09] G ROH G., H ANSTEIN H., W OERNDL W.: Interac-
D. J., FALL J., F RACCHIA F. D.: Case study: visual access for tively Visualizing Dynamic Social Networks with DySoN. In
landscape event based temporal data. In Proceedings of Visual- Workshop on Visual Interfaces to the Social and the Semantic
ization (1996), VIS ’06, IEEE Computer Society, pp. 425–428. Web (February 2009). 7
12 [GMH∗ 06] G RIFFEN A. L., M AC E ACHREN A. M., H ARDISTY
F., S TEINER E., L I B.: A Comparison of Animated Maps with
[Chi00] C HI E. H.- H .: A Taxonomy of Visualization Techniques
Static Small-Multiple Maps for Visually Identifying Space-Time
using the Data State Reference Model. In Proceedings of IEEE
Clusters. Annals of the Association of American Geographers 96,
Symposium on Information Visualization (2000), Infovis ’00,
4 (2006), 740–753. 15
IEEE, pp. 69–75. 13
[Gre11] G RETCHEN P.: A Cartographer’s Toolkit - Small Mul-
[CI05] C ASSINELLI A., I SHIKAWA M.: Khronos Projector. In
tiples. http://www.gretchenpeterson.com/blog/
Proceedings of SIGGRAPH 2005 (2005). 2, 13
small-multiples, 2011. [Online; accessed 24-Jan-2014].
[CKN∗ 03] C OLLBERG C., KOBOUROV S., NAGRA J., P ITTS J., 5

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

18 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations

[Har99] H ARRIS R. L.: Information Graphics: A Comprehensive [McC94] M C C LOUD S.: Understanding Comics. HarperCollins,
Illustrated Reference. Oxford University Press, New York, 1999. 1994. 5
1 [MTB03] M C G UFFIN M. J., TANCAU L., BALAKRISHNAN R.:
[HEF∗ 13] H URTER C., E RSOY O., FABRIKANT S., K LEIN T., Using Deformations for Browsing Volumetric Data. In Proceed-
T ELEA A.: Bundled Visualization of Dynamic Graph and Trail ings of IEEE Visualization (October 2003), VSI ’03, pp. 401–
Data. IEEE Transactions on Visualization and Computer Graph- 408. 13
ics, 99 (2013). 10, 11 [Mun08] M UNZNER T.: Process and Pitfalls in Writing Informa-
[HET12] H URTER C., E RSOY O., T ELEA A.: Graph Bundling by tion Visualization Research Papers. In Information Visualization.
Kernel Density Estimation. Computer Graphics Forum 31 (June Springer-Verlag, 2008, p. 134–153. 3
2012), 865–874. 11 [Muy87] M UYBRIDGE E.: Animal Locomotion. Published under
[HSCW13] H ADLAK S., S CHUMANN H., C AP C. H., W OLLEN - the auspices of the University of Pennsylvania, 1887. 5
BERG T.: Supporting the Visual Analysis of Dynamic Networks [PJH05] P ETROVIC N., J OJIC N., H UANG T. S.: Adaptive Video
by Clustering associated Temporal Attributes. IEEE Transactions Fast Forward. Multimedia Tools and Applications 26, 3 (2005),
on Visualization and Computer Graphics 19, 12 (2013), 2267– 327–344. 10
2276. 6
[r] R. http://www.r-project.org/. [online,
[HTC09] H URTER C., T ISSOIRES B., C ONVERSY S.: From- accessed:02-apr-2014]. 13
DaDy: Spreading Aircraft Trajectories Across Views to Support [RFF∗ 08] ROBERTSON G., F ERNANDEZ R., F ISHER D., L EE
Iterative Queries. IEEE Transactions on Visualization and Com- B., S TASKO J.: Effectiveness of Animation in Trend Visualiza-
puter Graphics 15, 6 (2009), 1017–1024. 14 tion. IEEE Transactions on Visualization and Computer Graph-
[HV08] H OLMAN D., V ERTEGAAL R.: Organic user interfaces: ics 14, 6 (Nov. 2008), 1325–1332. 9, 10, 11, 15
designing computers in any way, shape, or form. Communica- [RM13] RUFIANGE S., M C G UFFIN M. J.: DiffAni: Visualizing
tions of the ACM 51, 6 (2008), 48–55. 14 Dynamic Graphs with a Hybrid of Difference Maps and Anima-
[Hä70] H ÄGERSTRAND T.: What about People in Regional Sci- tion. IEEE Transactions on Visualization and Computer Graph-
ence. In Papers of the Regional Science Association (1970). 2 ics 19, 12 (2013), 2556–2565. 5, 10, 11
[JD13] JANSEN Y., D RAGICEVIC P.: An interaction Model for [Ros06] ROSLING H.: Stats that reshape your world-
Visualizations Beyond The Desktop. IEEE Transactions on Vi- view. http://www.ted.com/talks/hans_rosling_
sualization and Computer Graphics (2013). In press. 7, 13, 14 the_good_news_of_the_decade.html, 2006. [online,
accessed:24-jan-2014]. 3, 10, 11
[JDF13] JANSEN Y., D RAGICEVIC P., F EKETE J.-D.: Evaluating
the Efficiency of Physical Visualizations. In Proceedings of the [RPPH12] R ASMUSSEN M. K., P EDERSEN E. W., P ETERSEN
SIGCHI Conference on Human Factors in Computing Systems M. G., H ORNBÆK K.: Shape-changing interfaces: a review of
(2013), CHI ’13, ACM, pp. 2593–2602. 14 the design space and open research questions. In Proceedings
of the SIGCHI Conference on Human Factors in Computing Sys-
[KDA∗ 09] K RISTENSSON P., DAHLBACK N., A NUNDI D., tems (2012), ACM, pp. 735–744. 14
B JORNSTAD M., G ILLBERG H., H ARALDSSON J., M ARTENS -
SON I., N ORDVALL M., S TAHL J.: An Evaluation of Space Time
[SA06] S HNEIDERMAN B., A RIS A.: Network Visualization by
Cube Representation of Spatiotemporal Patterns. IEEE Transac- Semantic Substrates. IEEE Transactions on Visualization and
tions on Visualization and Computer Graphics 15, 4 (July 2009), Computer Graphics 12, 5 (2006), 733–740. 6
696–702. 15 [SAH00] S CHROEDER W. J., AVILA L. S., H OFFMAN W.: Vi-
sualizing with VTK: a tutorial. Computer Graphics and Applica-
[KK12] K VELADZE I., K RAAK M.-J.: What do we know about
tions, IEEE 20, 5 (2000), 20–27. 13
the space-time cube from cartographic and usability perspective?
In Proceedings of the 2012 AutoCarto International Symposium [SRdJ05] S HANBHAG P., R HEINGANS P., DES JARDINS M.:
on Automated Cartography (2012), AuroCarto’12. 15 Temporal Visualization of Planning Polygons for Efficient Par-
titioning of Geo-Spatial Data. In Proceedings of the 2005
[Kra03] K RAAK M. J.: The Space-Time Cube revisited from a IEEE Symposium on Information Visualization (Washington, DC,
Geovisualization Perspective. Proceedings of the 21st Interna- USA, 2005), INFOVIS ’05, IEEE Computer Society, pp. 28–. 1,
tional Cartographic Conference (2003), 1988–1996. 2, 7 2, 6
[KW04] K APLER T., W RIGHT W.: GeoTime Information Vi- [STB] S TUSAK S., TABARD A., B UTZ A.: Can physical visu-
sualization. In Proceedings of IEEE Information Visualization alizations support analytical tasks? In Poster Abstracts of IEEE
(Washington, DC, USA, 2004), INFOVIS ’04, IEEE Computer VIS 2013. 15
Society, pp. 25–32. 9, 12
[TBC13] T HUDT A., BAUR D., C ARPENDALE S.: Visits: A
[LNS11] L IU Z., NAVATHE S., S TASKO J.: Network-based vi- Spatiotemporal Visualization of Location Histories. [online,
sual analysis of tabular data. In Proceedings of Conference accessed:24-jan-2014]. 5, 11
on Visual Analytics Science and Technology (2011), VAST ’11,
pp. 41–50. 5 [TGF08] TANG A., G REENBERG S., F ELS S.: Exploring Video
Streams using Slit-Tear Visualizations. In Proceedings of the
[Mar78] M AREY É.-J.: La méthode graphique dans les sci- working conference on Advanced visual interfaces (New York,
ences expérimentales et particulièrement en physiologie et en NY, USA, 2008), AVI ’08, ACM, pp. 191–198. 5, 6
médecine. G. Masson, 1878. 4, 5
[TMB02] T VERSKY B., M ORRISON J., B ETRANCOURT M.:
[mat] MatLab. http://www.mathworks.fr/products/ Animation: Can it facilitate? International Journal of Human-
matlab/. [online, accessed:02-apr-2014]. 13 Computer Studies 57, 4 (2002), 247–262. 15
[MB98] M ACKAY W., B EAUDOUIN -L AFON M.: DIVA: Ex- [TSAA12] T OMINSKI C., S CHUMANN H., A NDRIENKO G.,
ploratory Data Analysis with Multimedia Streams. In Proceed- A NDRIENKO N.: Stacking-Based Visualization of Trajectory At-
ings of SIGCHI Conference on Human Factors in Computing tribute Data. Visualization and Computer Graphics, IEEE Trans-
Systems (apr 1998), CHI ’98, ACM, pp. 416–423. 13 actions on 18, 12 (Dec 2012), 2565–2574. 9

submitted to Eurographics Conference on Visualization (EuroVis) (2014)

 B. Bach & P. Dragicevic / A Review of Temporal Data VisualizationsBased on Space-Time Cube Operations 19

[Tuf] T UFTE E.: Megan Jaegerman’s brilliant news graph- His research interests lie in the areas of information visual-
ics. http://www.edwardtufte.com/bboard/ ization and human-computer interaction, with a focus on vi-
q-and-a-fetch-msg?msg_id=0002w4. [online,
sualization of multivariate data in space and time, the design
accessed:12-apr-2014]. 4
of scalable visual interfaces and the development of pixel-
[Tuf86] T UFTE E. R.: The visual display of quantitative informa-
based algorithms such as bundling rendering techniques.
tion. Graphics Press, Cheshire, CT, USA, 1986. 1, 3, 4, 5
[TV07] T RUONG B. T., V ENKATESH S.: Video Abstraction: Sheelagh Carpendale is a professor at University of Cal-
A systematic Review and Classification. ACM Transactions gary, where she also holds a Canada Research Chair in In-
on Multimedia Computing, Communications, and Applications formation Visualization. She leads the Innovations in Visual-
(TOMCCAP) 3, 1 (2007), 3. 5
ization (InnoVis) Research Group. Her research on informa-
[VFC10] V ROTSOU K., F ORSELL C., C OOPER M.: 2D and 3D tion visualization, large interactive displays, and new media
representations for feature recognition in time geographical diary
data. Information Visualization 9, 4 (Dec. 2010), 263–276. 12 draws on her interdisciplinary background in art, design and
computer science. She has contributed a number of novel
[Vox] Voxler. http://www.goldensoftware.com/
products/voxler. [online, accessed:02-apr-2014]. 13 techniques for temporal data visualization.
[VWD04] V IÉGAS F. B., WATTENBERG M., DAVE K.: Study-
ing cooperation and conflict between authors with history flow
visualizations. In Proceedings of the SIGCHI Conference on Hu-
man Factors in Computing Systems (New York, NY, USA, 2004),
CHI ’04, ACM, pp. 575–582. 6, 10
[vWvS99] VAN W IJK J. J., VAN S ELOW E. R.: Cluster and Cal-
endar Based Visualization of Time Series Data. In Proceedings of
the IEEE Symposium on Information Visualization (Washington,
DC, USA, 1999), INFOVIS ’99, IEEE Computer Society, pp. 4–.
15
[Wik13] W IKIMEDIA: "stroke order project". http://www.
tinyurl.com/strokeOrderProject, 2013. [Online; ac-
cessed 24-Jan-2014]. 4, 5
[WvdWvW09] W ILLEMS N., VAN DE W ETERING H., VAN
W IJK J. J.: Visualization of Vessel Movements. In Proceedings
of the 11th Eurographics / IEEE- VGTC Conference on Visualiza-
tion (Aire-la-Ville, Switzerland, 2009), EuroVis’09, Eurograph-
ics Association, pp. 959–966. 15

Biography
Benjamin Bach is a PhD student in the Aviz group at IN-
RIA, France, since 2011. He received his Masters Degree
in 2010 from the Department of Computer Science at the
University of Technology, Dresden, Germany. His research
focuses on network visualization, visualization of temporal
data, interaction for information visualization.
Pierre Dragicevic is a permanent researcher in the Aviz
group at INRIA, France, since 2007. He is doing research on
human-computer interaction and information visualization,
with a focus on animation, direct manipulation, physical vi-
sualization, visual perception and decision making.
Daniel Archambault is a lecturer at Swansea University
(since 2013). His research interests include information vi-
sualization, graph drawing, and perceptual issues in visu-
alization. Daniel has worked in the area of dynamic graph
drawing and has conducted human-centered experiments in-
volving presentation methods for dynamic graphs such as
animation and small multiples as well as drawing stability
(or the “mental map”).
Christophe Hurter is an associate professor at the Inter-
active computing laboratory (LII) at the French Civil Avi-
ation University (ENAC) in Toulouse, France, since 2011.

submitted to Eurographics Conference on Visualization (EuroVis) (2014)