摘要
The contamination of urban lakes by anthropogenic pollutants such as perfluorooctane sulfonate (PFOS) is a worldwide environmental problem. Large-scale, long-term monitoring of urban lakes requires careful prioritization of available resources, focusing efforts on potentially impaired lakes. Herein, a database of PFOS concentrations in 304 fish caught from 28 urban lakes was used for development of an urban-lake prioritization framework by means of exploratory data analysis (EDA) with the aid of a geographical information system. The prioritization scheme consists of three main tiers: preliminary classification, carried out by hierarchical cluster analysis; predictor screening, fulfilled by a regression tree method; and model development by means of a neural network. The predictive performance of the newly developed model was assessed using a training/validation splitting method and determined by an external validation set. The application of the model in the U.S. state of Minnesota identified 40 urban lakes that may contain elevated levels of PFOS; these lakes were not previously considered in PFOS monitoring programs. The model results also highlight ongoing industrial/commercial activities as a principal determinant of PFOS pollution in urban lakes, and suggest vehicular traffic as an important source and surface runoff as a primary pollution carrier. In addition, the EDA approach was further compared to a spatial interpolation method (kriging), and their advantages and disadvantages were discussed.