Fiber Uncertainty Visualization for Bivariate Data With Parametric and Nonparametric Noise Models.

Area of Science:

• Data Visualization
• Scientific Computing
• Statistical Modeling

Background:

• Multivariate data visualization and uncertainty analysis are significant research challenges.
• Fiber surfaces offer a method for visualizing multivariate data, extending univariate level-set visualization.
• Existing methods often rely on Gaussian models for uncertainty in bivariate data.

Purpose of the Study:

• To develop a statistical framework for quantifying the positional probabilities of fibers derived from uncertain bivariate data.
• To extend existing uncertainty models beyond Gaussian distributions to include various parametric and nonparametric distributions.
• To enable more robust uncertainty analysis in multivariate data visualization.

Main Methods:

• Extended Gaussian models to other parametric (uniform, Epanechnikov) and nonparametric (histograms, kernel density estimation) distributions.
• Leveraged Green's theorem for closed-form computation of fiber probabilities with independent noise.
• Employed a nonparametric approach with numerical integration for correlated noise scenarios.

Main Results:

• Derived spatial probabilities for fibers across diverse bivariate data uncertainty models.
• Successfully computed fiber probabilities for both independent and correlated noise.
• Visualized probability volumes using volume rendering and probability thresholding for uncertainty analysis.

Conclusions:

• The proposed statistical framework effectively quantifies fiber positional probabilities in uncertain bivariate fields.
• The methods enhance the analysis of multivariate data uncertainty, applicable to synthetic and simulation datasets.
• This work provides a generalized approach to uncertainty quantification in fiber surface visualization.