Geostatistical data fusion to characterize contaminated sites for remediation.

Abstract

Environmental attributes are usually highly variable both in space and time leading to substantial uncertainty about their information. Furthermore, various types of additional information are usually available at environmental sites. Characterization of contaminated sites for remediation requires a sound geostatistical method to map soil contaminant concentrations or risk. Incorporation of different kinds of information (data fusion) to improve remediation decisions was one of the main objectives of this dissertation. Three types of soil contaminants were considered: dioxin, heavy metals, and unexploded ordnance (UXO). Each contaminant was supplemented by at least one kind of additional information. Measurement errors associated with dioxin concentrations were first incorporated into geostatistical simulation by considering a Gaussian distribution of errors. This approach resulted in more effective modeling of uncertainty about block average dioxin concentrations, based on which remediation decisions are commonly made. A variant of kriging techniques was also developed to account for source location and transport direction in trend modeling for the geostatistical prediction of heavy metal concentrations. This technique was shown to outperform traditional geostatistical methodologies if the correct transport direction was identified. Alternative technique that combines logistic regression and kriging was used to incorporate exhaustive information, such as analytic signal data, to map the risk of occurrence of UXO at any location. The technique improved the decision for further remedial action. The quality of secondary information was also assessed by accounting for uncertainty associated with prior information. Mapping the concentration or the risk is usually a preliminary step for further decision-making processes. A statistically rigorous multiphase strategy that accounted for uncertainty attached to the average contaminant concentration was developed to make effective remediation decisions. This technique reduced the amount of soils which may require remediation, while maintaining the risk of making wrong decisions below the regulatory threshold. These method developments resulted in the improvement of geostatistical approaches with respect to incorporating additional information of variable quality for risk mapping and remedial optimization. The approaches presented in this dissertation are general and not limited to the sites analyzed in this dissertation.