SAMPLING DESIGN FOR CLASSIFYING CONTAMINANT LEVEL USING ANNEALING SEARCH ALGORITHMS

A stochastic method for sampling spatially distributed contaminant lev el is presented. The purpose of sampling is to partition the contamina ted region into zones of high and low pollutant concentration levels. In particular, given an initial set of observations of a contaminant w ithin a site, it is desired to find a set of additional sampling locat ions in a way that takes into consideration the spatial variability ch aracteristics of the site and optimizes certain objective functions em erging from the physical, regulatory and monetary considerations of th e specific site cleanup process. Since the interest is in classifying the domain into zones above and below a pollutant threshold level, a n atural criterion is the cost of misclassification. The resulting objec tive function is the expected value of a spatial loss function associa ted with sampling. Stochastic expectation involves the joint probabili ty distribution of the pollutant level and its estimate, where the lat ter is calculated by means of spatial estimation techniques. Actual co mputation requires the discretization of the contaminated domain. As a consequence, any reasonably sized problem results in combinatorics pr ecluding an exhaustive search. The use of an annealing algorithm, alth ough suboptimal, can find a good set of future sampling locations quic kly and efficiently. In order to obtain insight about the parameters a nd the computational requirements of the method, an example is discuss ed in detail. The implementation of spatial sampling design in practic e will provide the model inputs necessary for waste site remediation, groundwater management, and environmental decision making.