WP20 Contaminated Harbor and River Sediment
Wednesday, 16 November 2005: 8:00 AM - 6:30 PM in Exhibit Hall

(THO-1117-836405) Comparative evaluation of geostatistical methods for delineating PCB remediation areas and volumes, Lower Fox River.

Thornburg, T.¹, Wolfe, J.², Barabas, N.², ¹ Anchor Environmental, L.L.C., Portland, OR² Limno-Tech, Inc., Ann Arbor, MI

ABSTRACT- The performance of three different interpolation methods and a range of statistical significance levels were evaluated for delineating sediment remediation areas and volumes for the remedial design of the Lower Fox River, Wisconsin. The three methods evaluated included Thiessen polygons, a hybrid of ordinary kriging and indicator kriging, and full indicator kriging. The kriging methods are generally preferred over Thiessen polygons because they provide valuable information about uncertainty and spatial correlation structure, and can be cross-validated. However, Thiessen polygons are useful for delineating sediments above 50 ppm PCBs which require special disposal; these areas constitute the upper 10 percent of the PCB data and kriging of these extreme values is difficult. For areas where dredging is the selected remedy, dredge prism designs at a range of significance levels (defined as the probability of exceeding the cleanup level at a given location) were evaluated using cross-validation metrics to inform risk management decisions. These decisions involve balancing the risk of leaving contaminated sediment behind (false negative error) versus the risk of unnecessarily dredging clean material (false positive error). Additional protection against false negative errors will be provided by post-dredge confirmation sampling. A significance level of 0.5 provided the optimum balance of sensitivity and specificity, the highest accuracy, and the least bias in the dredge cut depth. In practice, more sediment will be removed than the selected significance level would indicate because additional deepening of the dredge prism occurs during dredge plan design and to account for contractor overdredging allowance. The hybrid ordinary/indicator kriging method and the full indicator kriging method showed excellent agreement and very comparable performance metrics, confirming the robustness of the geostatistically-based sampling grid. However, full indicator kriging showed less attenuation of extreme values and is the preferred method for use in remedial design.

Key words: geostatistics, kriging, dredge prism, remedial design