Details-First, Show Context, Overview Last: Supporting Exploration of Viscous Fingers in Large-Scale Ensemble Simulations

• Published in: IEEE transactions on visualization and computer graphics Vol. 25; no. 1; pp. 1225 - 1235
• Main Authors: Luciani, Timothy, Burks, Andrew, Sugiyama, Cassiano, Komperda, Jonathan, Marai, G. Elisabeta
• Format: Journal Article
• Language: English
• Published: United States IEEE 01.01.2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Computational fluid dynamics; Computational modeling; Computer simulation; Context modeling; Data points; Data visualization; details-first model; discourse paper; Domains; Exploration; Feature extraction; Fingers; Mathematical models; Spatiotemporal phenomena; Subject specialists; theory; Tracking; Visualization; visualization design; Workflow
• ISSN: 1077-2626; 1941-0506
• DOI: 10.1109/TVCG.2018.2864849

Visualization research often seeks designs that first establish an overview of the data, in accordance to the information seeking mantra: "Overview first, zoom and filter, then details on demand". However, in computational fluid dynamics (CFD), as well as in other domains, there are many situations where such a spatial overview is not relevant or practical for users, for example when the experts already have a good mental overview of the data, or when an analysis of a large overall structure may not be related to the specific, information-driven tasks of users. Using scientific workflow theory and, as a vehicle, the problem of viscous finger evolution, we advocate an alternative model that allows domain experts to explore features of interest first, then explore the context around those features, and finally move to a potentially unfamiliar summarization overview. In a model instantiation, we show how a computational back-end can identify and track over time low-level, small features, then be used to filter the context of those features while controlling the complexity of the visualization, and finally to summarize and compare simulations. We demonstrate the effectiveness of this approach with an online web-based exploration of a total volume of data approaching half a billion seven-dimensional data points, and report supportive feedback provided by domain experts with respect to both the instantiation and the theoretical model.