Spatial Modeling of Contaminated Sediments in Puget Sound

A data set of lead and copper concentrations in the aquatic sediments of Puget Sound is used to interpolate lead concentrations at unobserved locations in the Sound. The data are analyzed using random field models as well as spatial trend models. After the removal of trend, little anisotropy in the concentrations is evident, permitting theoretical isotropic variograms to be fit to the residual data. Kriging is performed on the residuals and the kriged residuals predictions are added to the trend model predictions to yield the final interpolated concentration contours in the Sound. It is demonstrated that the random field model has difficulty in accounting for hot spots in the data; the simple trend model works better than kriging in this case. However, in another region of the Sound where lead concentrations were more homogenous, the kriging model offers significant improvement over the predictions generated with the simple trend model.