Novel visualization methods are presented for spatial probability density function data. These are spatial datasets, where each pixel is a random variable, and has multiple samples which are the results of experiments on that random variable. We use clustering as a means to reduce the information contained in these datasets; and present two different ways of interpreting and clustering the data. The clustering methods are used on two datasets, and the results are discussed with the help of visualization techniques designed for the spatial probability data.
This paper looks at spatial probability density function data where each pixel across a 2D space represents a random variable described by a PDF. This data can be examined either as a spatial scalar data set where a single instance is a 2D surface with a sample for each pixel, or as a collection of PDFS. They implement clustering techniques to reduce the complexity of having to show all of this data at once and group similar regions of the data together. Two types of clustering are implemented to address the two ways of thinking of the data. The first technique clusters on the individual PDFs, so each cluster will contain pixels with similar PDFs. The second type of clustering will cluster on realizations, thus reducing the number of realizations and allowing for a histogram across the number of realizations to be formed where we can determine a most occurring realization or outlying outcomes.
Data: 2D Uncertainty: 1D Visualization: 2D Technique: Clustering
@Article{ bordoloi:2003:VTPD,
author = {Udeepta D. Bordoloi and David L. Kao and Han-Wei Shen},
title = {Visualization techniques for spatial probability density function data},
journal = {Data Science Journal},
year = {2004},
volume = {3},
pages = {153--162},
}