Interactive Visual Analysis of Scientific Data

Steffen Oeltze (University of Magdeburg, Germany)
Johannes Kehrer (Vienna University of Technology, Austria)
Helwig Hauser (University of Bergen, Norway)

In a growing number of application areas, a subject or phenomenon is investigated by means of multiple datasets being acquired over time (spatiotemporal), comprising several attributes per data point (multi-variate), stemming from different data sources (multi-modal) or multiple simulation runs (multi-run/ensemble) [KH13]. Interactive visual analysis (IVA) comprises concepts and techniques for a user-guided knowledge discovery in such complex data. Through a tight feedback loop of computation, visualization and user interaction, it provides new insight into the data and serves as a vehicle for hypotheses generation or validation. It is often implemented via a multiple coordinated view framework where each view is equipped with interactive drill-down operations for focusing on data features. Two classes of views are integrated: physical views, such as direct volume rendering, show information in the context of the spatiotemporal observation space while attribute views, such as scatter plots and parallel coordinates, show relationships between multiple data attributes. The user may drill-down the data by selecting interesting regions of the observation space or attribute ranges leading to a consistent highlighting of this selection in all other views (brushing-and-linking). Three patterns of explorative/analytical procedures may be accomplished by doing so. In a feature localization, the user searches for places in the 3D/4D observation space where certain attribute values are present. In a multi-variate analysis, relations between data attributes are investigated, e.g., by searching for correlations. In a local investigation, the user inspects the values of selected attributes with respect to certain spatiotemporal subsets of the observation space.

In this tutorial, we discuss examples for successful applications of IVA to scientific data from various fields: climate research, medicine, epidemiology, and flow simulation / computation, in particular for automotive engineering. We base our discussions on a theoretical foundation of IVA which helps the tutorial attendees in transferring the subject matter to their own data and application area. In the course of the tutorial, the attendees will become acquainted with techniques from statistics and knowledge discovery, which proved to be particularly useful for a specific IVA application. The tutorial further comprises an overview of off-the-shelf IVA solutions, which may be be particularly interesting for visualization practitioners. It is concluded by a summary of the gained knowledge and a discussion of open problems in IVA of scientific data.