Bayesian sampling design for contaminated land investigation

Abstract

The problem of sampling design for contaminated land investigation is approached using Bayesian methods. We develop a decision tool designed to aid site investigators and decision makers in the process of site investigation. Current legislation and guidance is considered, and used to drive the development of a spatial model to describe the contamination levels over a site. This model is updated using a full Bayes approach and combined with a detailed loss structure in order to calculate the expected losses associated with the possible decisions. A sampling search algorithm looks for good designs with which we can further update beliefs and improve decision making ability through reduced uncertainty and therefore increased confidence. We also offer an MCMC approach to learn about multiple contaminants which are believed to be related. The decision tool provided offers a flexible environment in which multiple decisions, outcomes and contaminants may be considered simultaneously in order to assist the site investigator in implementing a cost effective sampling strategy