Visualization and exploration of spatial probability density functions: a clustering-based approach

We present an interactive visualization technique for spatial probability density function data. These are datasets that represent a spatial collection of random variables, and contain a number of possible outcomes for each random variable. It is impractical to visualize all the information at each spatial location as it will quickly lead to a cluttered image. We advocate the use of hierarchical clustering as a means of summarizing the information, and also as a tool to bring out meaningful spatial structures in the datasets. For clustering, we discuss a distance function which preserves the spatial correlation present in these datasets. To create an informative visualization of the clusters, we introduce a scheme of colors and patterns to represent various statistical properties of the clusters.