Spatially resolved distribution models of POP concentrations in soil: a stochastic approach using regression trees.

Background concentrations of selected persistent organic pollutants (polychlorinated biphenyls, hexachlorobenzene, p,p'-DDT including metabolites) and polyaromatic hydrocarbons in soils of the Czech Republic were predicted in this study, and the main factors affecting their geographical distribution were identified. A database containing POP concentrations in 534 soil samples and the set of specific environmental predictors were used for development of a model based on regression trees. Selected predictors addressed specific conditions affecting a behavior of the individual groups of pollutants: a presence of primary and secondary sources, density of human settlement, geographical characteristics and climatic conditions, land use, land cover, and soil properties. The model explained a high portion of variability in relationship between the soil concentrations of selected organic pollutants and available predictors. A tree for hexachlorobenzene was the most successful with 76.2% of explained variability, followed by trees for polyaromatic hydrocarbons (71%), polychlorinated biphenyls (68.6%), and p,p'-DDT and metabolites (65.4%). The validation results confirmed that the model is stable, general and useful for prediction. The stochastic model applied in this study seems to be a promising tool capable of predicting the environmental distribution of organic pollutants.