摘要：Modern applications in computational science are increasingly focusing on understanding uncertainty in models and parameters in simulations. In this paper, we describe visitation graphs, a novel approximation technique for the well-established visualization of steady 2D vector field ensembles using visitation maps. Our method allows the efficient and robust computation of arbitrary visitation maps for vector field ensembles. A pre-processing step that can be parallelized to a high degree eschews the needs to store every ensemble member and to re-calculate every time the start position of the visitation map is changed. Tradeoffs between accuracy of generated visitation maps on one side and pre-processing time and storage requirements on the other side can be made. Instead of downsampling ensemble members to a storable size, coarse visitation graphs can be stored, giving more accurate visitation maps while still reducing the amount of data. Thus accurate visitation map creation is possible for ensembles where the traditional visitation map creation is prohibitive. We describe our approach in detail and demonstrate its effectiveness and utility on examples from Computational Fluid Dynamics.
