基于随机森林与地统计预测城市土壤PAHs分布

收集南京城区采样点位置信息和环境变量等数据，应用地统计和随机森林方法，以及两种方法相结合分别预测土壤多环芳烃（polycyclic aromatic hydrocarbons，PAHs）含量，并比较不同方法预测精度.结果表明：随机森林与地统计方法结合能大幅度提高城市土壤污染物制图精度，整合克里金与随机森林预测残差模型拟合优度R²相比克里金插值法提高74.8%.PAHs空间制图结果能够较好拟合污染物的变化范围，识别污染高值区与低值区的空间分布.随机森林输出特征重要性发现影响南京城区土壤PAHs分布的主控因子为土壤碳和土壤粒度以及工厂密度.本研究可为城市污染物高分辨率和高精度制图以及污染防控治理提供参考.

Distribution prediction of soil PAHs based on random forest and geostatistics methods in urban area

Based on the locational and environmental vairables collected at sampling points in Nanjing city, the Geostatistics and Random forest models were combined to predict the distribution of soil PAHs. Results showed that combination of these two could improve the prediction accuracy of PAHs in the research area. The model fitness achieved by the combined model was 74.8% higher than that from the traditional Kriging method. The generated map also characterized the spatial variation pattern better, and identified the high and low polluted areas. The importance of environmental variables in the output from the random forest model showed that soil carbon, soil texture and plant density were the main controlling factors for PAHS distribution. This study could provide a methodology framework for high-resolution and high-precision mapping of urban pollutants, such as PAHs.