Visual data science to master complex simulation ensembles

Simulations and experiments in porous media research and other scientific domains yield large ensembles with information-rich members (spatial, temporal, multi-variate). In PN 6-8, we combine machine learning techniques for the visual exploration of ensemble member similarities (based on prior work in PN6-6) with a human-computer interaction perspective for intuitive data interfaces. We develop novel, machine learning-based approaches to yield expressive feature representations for the semantic clustering of members without a formal feature definition. This provides the basis for new approaches to visual sensitivity analysis, allowing analysts to better understand the parameter spaces that might trigger (sensitive) changes between clusters of different model behaviour or are responsible for the occurrence of different types of anomalies. We further leverage immersive technologies such as VR/AR displays to interact with, explore, and analyse ensembles (potentially in remote collaboration), and evaluate their benefits and drawbacks for visual ensemble analysis.