Probabilistic Local Features in Uncertain Vector Fields with Spatial Correlation

Abstract

In this paper methods for extraction of local features in crisp vector fields are extended to uncertain fields. While in a crisp field local features are either present or absent at some location, in an uncertain field they are present with some probability. We model sampled uncertain vector fields by discrete Gaussian random fields with empirically estimated spatial correlations. The variability of the random fields in a spatial neighborhood is characterized by marginal distributions. Probabilities for the presence of local features are formulated in terms of low-dimensional integrals over such marginal distributions. Specifically, we define probabilistic equivalents for critical points and vortex cores. The probabilities are computed by Monte Carlo integration. For identification of critical points and cores of swirling motion we employ the Poincaré index and the criterion by Sujudi and Haimes. In contrast to previous global methods we take a local perspective and directly extract features in divergence-free fields as well. The method is able to detect saddle points in a straight forward way and works on various grid types. It is demonstrated by applying it to simulated unsteady flows of biofluid and climate dynamics.