Spatial Population Partitioning Using Voronoi Diagrams for Environmental Data Analysis

Kriging is the most commonly used geostatistical method for site characterization of a contaminated site if there is adequate spatial coverage by the samples collected from the site. Contaminant concentration data from the site are first used to estimate the variogram model, which in turn is used to compute the best linear unbiased estimates (BLUEs) of the contaminant concentrations at unsampled locations of the site. There are, however, situations when the data do not seem to be spatially correlated resulting in poor variogram model(s). There are also cases in which the presence of a few hospots (outliers) gives the appearance of good spatial correlations in the data. Just like all other classical statistical results, the variogram model, and the corresponding kriged estimates also get distorted by outliers. The purpose of this paper is to provide an alternative geostatistical method to handle such situations. The proposed method uses Q-Q plots for testing normality to partition the data into several normally distributed subsets. The Voronoi diagrams from computational geometry are then used for spatial visualization of the data set. This spatial partitioning of the data can be used for site remediation.