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. 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 1INRIA, France 2Swansea University, UK 3ENAC, France 4University 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 8 USA Japan 1990 1992 1994 8 8 8 8 6 a space-time cube (Figure 3). Since then, space-time cubes 6 6 6 6 Ination (%) Ination(%) Ination(%) 4 90 Ination(%) Ination(%) 4 4 4 4 90 have been employed in a number of interactive visualization 2 00 92 2 94 92 2 2 2 ∗ 98 96 94 0 0 98 0 0 0 systems (e.g., [CCT 99, FLM00, Kra03]), as well as for en- 96 00 -2 -2 -2 -2 -2 0 10 20 0 10 20 0 10 20 Unemployment rate (%) 0 10 20 0 10 20 Unemployment rate (%) Unemployment rate (%) Unemployment rate (%) 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 8 8 8 8 90 92 6 6 6 6 6 on temporal visualization techniques. Ination(%) Ination(%) 4 4 94 Ination(%) Ination(%) Ination(%) 90 00 4 4 4 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.