Comparative visual analysis of vector field ensembles

We present a new visual analysis approach to support the comparative exploration of 2D vector-valued ensemble fields. Our approach enables the user to quickly identify the most similar groups of ensemble members, as well as the locations where the variation among the members is high. We further provide means to visualize the main features of the potentially multimodal directional distributions at user-selected locations. For this purpose, directional data is modelled using mixtures of probability density functions (pdfs), which allows us to characterize and classify complex distributions with relatively few parameters. The resulting mixture models are used to determine the degree of similarity between ensemble members, and to construct glyphs showing the direction, spread, and strength of the principal modes of the directional distributions. We also propose several similarity measures, based on which we compute pairwise member similarities in the spatial domain and form clusters of similar members. The hierarchical clustering is shown using dendrograms and similarity matrices, which can be used to select particular members and visualize their variations. A user interface providing multiple linked views enables the simultaneous visualization of aggregated global and detailed local variations, as well as the selection of members for a detailed comparison.