Adaptive Spatial Sampling for Investigating and Remediating Contaminated Properties. L. A. Cox, Jr. and D. Orvosh, Cox Associates, 503 Franklin Street, Denver, Colorado, 80218
Consider a hazardous waste site surrounded by hundreds of residential properties with soil contamination concentrations ranging over two or more orders of magnitude. In many cases, it may be difficult or impossible to determine where site-related contamination ceases and «background» concentration begins. Sampling soils at residential properties helps to reveal contaminant levels but is expensive and imprecise. In such cases, deciding which properties to sample, how many samples to take, and where to pay to clean up based on sample evidence, presents a challenging problem in applied statistical decision theory and multi-criterion decision-making. We present an adaptive approach and decision support software for sequentially choosing where to sample next and which properties to clean, based on sampled concentrations, to minimize expected costs and errors. The method applies Monte-Carlo uncertainty analysis methods and bootstrapping principles to estimate soil contamination frequency distributions at different locations and to quantify trade-offs among remediation objectives (e.g., probabilities of falsely cleaning vs. falsely failing to clean a property). Results from a case study in Chicago are used to illustrate the method and to show how adaptive spatial sampling can reduce costs and error rates.