The integration of various data sets into desktop based 3D virtual environments, such as the virtual globe Google Earth, is quickly achieved with today's technological options. Nevertheless, we know little about the appropriateness of such representations. A number of research studies have looked at different aspects of 3D virtual environments, in particular interaction and navigation, but rarely at the use of virtual environments for data analysis. The visual combination of quantitative data with the three-dimensional virtual equivalent of the natural environment where a data set was collected may help the analysis of such data sets in regard to altitude and landform. Data sets demonstrating an interesting relationship between data and landscape may become increasingly available with the further development and application of sensor networks. The research summarised here aims to increase the understanding of the use of desktop based 3D virtual environments with a focus on the graphical representation of quantitative data through abstract symbols or graphics.
A mixed methods research approach is employed. Four different stages with different methodologies are combined to gain a holistic view regarding the goals of the study. The research stages are positioned along a 'bridge' from experimental 'in vitro' research to applied settings or 'in vivo' case studies driven by increasing context, data and task complexity. In the first stage, the effectiveness and efficiency of 2D bars in 3D virtual environments as compared to 2D displays was tested. Experiment participants identified the larger of pairs of bars and compared their lengths. The research stages IIa and IIb tested 2D bars in virtual environments with more complex data and tasks. In stage IIa participants answered complex tasks, such as pattern identification, in regard to several single value bars while in stage IIb a more open insight reporting approach was employed to let participants explore bar charts representing more complex data aggregations. The reported insights were analysed regarding their complexity, plausibility and the participants' confidence in them. In stage III a descriptive and explorative case study approach with three diverse cases including real world data sets and data experts was implemented to test and enhance the findings of the previous stages.
The results show that typical users are able to separate depth cues and distortions introduced by perspective viewing from absolute value changes in the representations of quantitative data in virtual environments when represented as 2D bars on billboards. While the users are able to relate multivariate data represented in virtual environments to altitude and landform, the 3D environment does not especially support this. Only insignificant variation between 2D representations and 3D visualisations are found. However, the different data sets and tasks influence the results. The participants' answers are strongly guided by the tasks and some data sets are more successfully analysed in 3D, others in 2D. Generally, analysis of data in relation to altitude and landform is successful in either visualisation but participants do it less habitually than data analysis in relation to location and distribution. The data experts of stage III comment positively about the possibilities of the quantitative data visualisations in virtual environments. But the usefulness is dependent on visualisation completeness and on the data expert's previous usage of visualisations for either communication and/or data exploration purposes. Displaying up to four variables at once is identified as maximum of acceptable data graphics complexity. Additionally, more interaction, such as switching on and off the reference frames of the bar charts, is requested. Navigation is imperative for data analysis in virtual environments.
Methodologically bridging between experimental 'in vitro' and case study based 'in vivo' research methods is appropriate as the results of each stage can inform the design of the following stages. Additionally, the outcomes of later stages lead to re-evaluations or different interpretation of earlier results as for the aspects of bar chart complexity, occlusion and use of reference frames. Thus, in combining different methods, particular strengths such as exclusion or inclusion of context can be added together and potential weaknesses, such as small numbers of data experts, overcome. A holistic understanding of the visualisation technique is gained but it is nevertheless possible to attend to details. The case studies indicate that it is difficult to capture the use of visualisation in real world settings as the kinds of data sets made available are likely to be well known, as they were in this study. Nevertheless, the results of stage III allow evaluating earlier findings in a more applied setting and explore further issues. For example, the data experts commented on improvements and further applications for the visualisations. This may serve as input to the design process of future visualisation prototypes.
2D screenshot of abstract data visualised in a 3D virtual environment
(Data © SNP; Virtual environment Google Earth © 2011 Google, © 2011 Cnes/Spot Image)
[Link]
Process files (documentation: see comments in the files, German only)
