A Bayesian Spatial Contamination Model

Abstract

In environmental research, it is often the case that to cluster observations into environmentally polluted and natural groups is an important issue. The Bayesian contamination model which adopts a multivariate mixture regression model has been developed in that it aims to cluster observations and estimate the average amount of pollution. However, because the Bayesian contamination model does not take spatial correlations between observations into consideration, a Bayesian spatial contamination model is proposed. A simulation study was conducted showing that the proposed model has an advantage over the Bayesian contamination model in terms of biases and RMSE of estimators of the logistic regression parameters. We applied the proposed model into environmental data and confirmed the improvement on the model fit. Also, the clustering was reasonably performed from the environmental perspective, which was coherent with the fact that the underground water flows from the southwest side to the northeast side. This model is expected to be utilized effectively to monitor the quality of a ground or groundwater and capture the heterogeneity in it which is suspected of environmental pollution especially when the interested site consists of areas with strong spatial dependency.