In Journal of hazardous materials
From the perspective of the mechanism of soil pollution, it is difficult to explain the process of predicting the spatial distributions of soil heavy metal pollution using traditional geostatistical methods at a regional scale. Furthermore, few methods are available to proactively identify potential risk areas for preventing soil contamination. In this study, we selected 13 environmental factors related to the accumulation of soil heavy metals based on the source-sink theory. Then, the fuzzy k-means method in combination with the random forest (RF) method was used to classify potential risk areas. The concentrations and spatial distributions of the heavy metals were well predicted by RF, and the average values of the root mean square error of the prediction and R2 were 4.84 mg kg-1 and 0.57, respectively. The results indicated that the soil pH, fine particulate matter, and proximity to polluting enterprises significantly influenced the heavy metal pollution in soils, and the environmental variables varied significantly across the identified subregions. This study provides a theoretical basis for the sustainable management and control of soil pollution at the regional scale.
Jia Xiaolin, Fu Tingting, Hu Bifeng, Shi Zhou, Zhou Lianqing, Zhu Youwei
Cluster analysis, Heavy metal pollution, Machine learning methods, Potential risk areas, Source-Sink theory