Key points are not available for this paper at this time.
The soil environment is being continually contaminated with heavy metals from emissions that are introduced from both the atmosphere and water under the condition of rapid urbanization and industrialization, which cause land use regression (LUR) models could not easily capture complex relationship between soil heavy metal and potential indicator. Random forest is a non-parametric statistical method that can manage non-linear relationships. This study aims to explore the application of random forest (RF) models in predicting the soil concentration and spatial distribution of six heavy metal(loid)s (Pb, Cd, Cr, As, Hg and Zn) comparing with land use regression (LUR) models. Finally, R2 values for the RF models were approximately 0.90 and presented a larger cross-validation R2 and lower root mean square error (RMSE) than LUR models. The comparison between the RF and LUR models demonstrates that the RF model performed better and RF can accurately predict the concentration and spatial distribution of heavy metal(loid)s in soils. Moreover, in the study area, human activities and transportation are the main sources of soil heavy metals Pb, sewage irrigation is the main source of Cd, Cr and Zn, and atmospheric deposition from thermal power stations is an important source of soil heavy metals Hg. Parent materials is the most likely source of As. Given the above, application of random forest in soil heavy metal(loid)s may assist soil environmental management departments to focus on controlling the diffusion of heavy metal(loid)s pollution sources in a practical way, and providing targets for pollution control and prevention.
Wang et al. (Fri,) studied this question.