Geostatistical assessment and validation of uncertainty for three-dimensional dioxin data from sediments in an estuarine river

Contaminated sediment management is an urgent environmental and regulatory issue worldwide. Because remediation is expensive, sound quantitative assessments of uncertainty about the spatial distribution of contaminants are critical, but they are hampered by the physical complexity of sediment environments. This paper describes the use of geostatistical modeling approaches to quantify uncertainty of 2,3,7,8-tetrachlorodibenzo-p-dioxin concentrations in Passaic River (New Jersey) sediments and to incorporate this information in decision-making processes, such as delineation of contaminated areas and additional sampling needs. First, coordinate transformation and analysis of three-dimensional semivariograms were used to describe and model the directional variability accounting for the meandering course of the river. Then, indicator kriging was employed to provide models of local uncertainty at unsampled locations without requiring a prior transform (e.g. lognormal) of concentrations. Cross-validation results show that the use of probability thresholds leads to more efficient delineation of contaminated areas than a classification based on the exceedence of regulatory thresholds by concentration estimates. Depending on whether additional sampling aims at reducing prediction errors or misclassification rates, the variance of local probability distributions or a measure of the expected closeness to the regulatory threshold can be used to locate candidate locations.