Visualization of ensemble statistics (mean, variance, and skewness) for CFD using compressed particle data

Shimomura, Kazuya ; Kawamura, Takuma ; Idomura, Yasuhiro   

In data assimilation for computational fluid dynamics (CFD), ensemble-based methods are widely used to improve prediction accuracy, and statistical analysis of ensemble data is essential for validating assimilation schemes and quantifying uncertainty. In large-scale simulations such as nuclear safety analysis and weather prediction, it is necessary to maintain a large number of time-evolving three-dimensional fields, resulting in datasets that can reach several tens of terabytes. However, aggregating large ensemble fields consisting of dozens or hundreds of cases to compute statistics such as mean and variance suffers from bottlenecks due to intensive global communication and increased memory consumption, making interactive visualization of statistical quantities difficult. In this study, we develop an ensemble particle-based volume rendering (PBVR) method in which CFD solutions are compressed and stored as visualization particles, and statistical quantities are computed and visualized directly from the particle ensembles. We propose visualization algorithms for mean, variance, and skewness fields, and evaluate memory reduction and visualization quality using test datasets.