Investigation of techniques enabling a seamless analysis of data from multi-run simulations (also known as ensemble data) on multiple degrees of detail.
The key idea of this project is that a holistic analysis of simulation data requires to look at multiple levels-of-detail. At the lowest level, the data generated by single simulation runs can be analyzed in isolation. Examples of such data include state information in case of agent-based simulations (e.g., traffic simulations), event data for simulations of communication networks, and 4D volumetric data in case of climate simulations. An analysis at this level enables a detailed understanding and validation of the underlying application-specific simulation. However, each single simulation run typically involves massive amounts of data, and the contained information is much too detailed for tasks like comparison, sensitivity analysis, or decision making.
At the highest level, the result of a simulation run can be aggregated to a few quantitative performance indicators. This enables the comparison of a large number of simulation runs based on common techniques for visualizing multivariate data, and is much more useful for an analysis of the parameter space as well as ultimate decision making. However, the inevitable loss of information does neither allow for validating the simulation itself at this level, nor to understand details of the simulated system. In some cases, there might also be intermediate levels, e.g., reducing volumetric results from a voxel level to an object-level.
Previous work by VRVis as well as by the international scientific community has addressed the analysis of ensemble data at particular levels of detail or the integration for very particular cases (e.g., ensembles of 2D functions). In contrast, the focus of this project is to research a much more general approach towards a seamless transition between the various levels as well as towards a synchronized analysis at multiple levels simultaneously. In a bottom-up way, this may include defining key indicators for comparisons on higher-levels by interacting on low-level simulation data. In a top-down way, selections on high levels may specify detailed comparisons on lower levels.