Multi-dimensional parameter-space partitioning of spatio-temporal simulation ensembles. (May 2022)

• Record Type: Journal Article
• Authors: Evers, Marina; Linsen, Lars
• Is Part Of: Computers & graphics. Volume 104(2022)
• Page Start: 140
• Page End: 151
• Publication Date: 2022-05
• Subjects: Ensemble visualization -- Parameter-space analysis -- Spatio-temporal simulation -- Similarity; Computer graphics -- Periodicals; 006.6
• DOI: 10.1016/j.cag.2022.04.005
• Languages: English
• ISSNs: 0097-8493

Abstract

Abstract: Numerical simulations are commonly used to understand the parameter dependence of given spatio-temporal phenomena. Sampling a multi-dimensional parameter space and running the respective simulations leads to an ensemble of a large number of spatio-temporal simulation runs. A main objective for analyzing the ensemble is to partition (or segment) the multi-dimensional parameter space into connected regions of simulation runs with similar behavior. To facilitate such an analysis, we propose a novel visualization method for multi-dimensional parameter-space partitions. Our visualization is based on the concept of a hyper-slicer, which allows for undistorted views of the parameter-space segments' extent and transitions. For navigation within the parameter space, interactions with a 2D embedding of the parameter-space samples, including their segment memberships, are supported. Parameter-space partitions are generated in a semi-automatic fashion by analyzing the similarity space of the ensemble's simulation runs. Clusters of similar simulation runs induce the segments of the parameter-space partition. We link the parameter-space partitioning visualizations to similarity-space visualizations of the ensemble's simulation runs and embed them into an interactive visual analysis tool that supports the analysis of all facets of the spatio-temporal simulation ensemble targeted at the overarching goal of analyzing the parameter-space partitioning. The partitioning can then be visually analyzed and interactively refined. We compared our approach to alternative methods and evaluated it with experts within case studies from three different domains. Graphical abstract: Highlights: We visualize multi-dimensional parameter-space partitions of simulation ensembles. Hyper-slicer visualization of partitions to assess location and extent of segments. Partitions are imposed by semi-automatically generated clusters of simulation runs. Linked parameter- and similarity-space visualizations to analyze all facets. We evaluate our approach for three real-world datasets with domain experts.