A Bayesian approach for probabilistic streamline computation in uncertain flows

Abstract

Streamline-based techniques play an important role in visualizing and analyzing uncertain steady vector fields. It is a challenging problem to generate accurate streamlines in uncertain vector fields due to the global uncertainty transportation. In this work, we present a novel probabilistic method for streamline computation on uncertain steady vector fields using a Bayesian framework. In our framework, a streamline is modeled as a state space model which captures the spatial coherence of integration steps and uncertainty in local distributions using the conditional prior density and the likelihood function. To approximate the posterior distribution for all the possible traces originating from a given seed position, a set of weighted samples are iteratively updated from which streamlines with higher likelihood can be derived. We qualitatively and quantitatively compare our method with alternative methods on different types of flow field data sets. Our method can generate possible streamlines with higher certainty and hence more accurate flow traces.