Visual Analysis of Spatial Variability and Global Correlations in Ensembles of Iso-Contours

• Published in: Computer graphics forum Vol. 35; no. 3; pp. 221 - 230
• Main Authors: Ferstl, F., Kanzler, M., Rautenhaus, M., Westermann, R.
• Format: Journal Article
• Language: English
• Published: Oxford Blackwell Publishing Ltd 01.06.2016
• Subjects: Analysis; Categories and Subject Descriptors (according to ACM CCS); Color coding; Computation; Computer graphics; Correlation; Datasets; I.3.3 [Computer Graphics]: Picture/Image Generation-Line and curve generation; Mathematical analysis; Mathematical models; Reduced order; Shape; Studies; Three dimensional; Visualization
• ISSN: 0167-7055; 1467-8659
• DOI: 10.1111/cgf.12898

For an ensemble of iso‐contours in multi‐dimensional scalar fields, we present new methods to a) visualize their dominant spatial patterns of variability, and b) to compute the conditional probability of the occurrence of a contour at one location given the occurrence at some other location. We first show how to derive a statistical model describing the contour variability, by representing the contours implicitly via signed distance functions and clustering similar functions in a reduced order space. We show that the spatial patterns of the ensemble can then be derived by analytically transforming the boundaries of a confidence interval computed from each cluster into the spatial domain. Furthermore, we introduce a mathematical basis for computing correlations between the occurrences of iso‐contours at different locations. We show that the computation of these correlations can be posed in the reduced order space as an integration problem over a region bounded by four hyper‐planes. To visualize the derived statistical properties we employ a variant of variability plots for streamlines, now including the color coding of probabilities of joint contour occurrences. We demonstrate the use of the proposed techniques for ensemble exploration in a number of 2D and 3D examples, using artificial and meteorological data sets.