Modeling the spatial variations in anthropogenic factors of soil heavy metal accumulation by geographically weighted logistic regression

Abstract

Quantifying the impacts of associated factors on soil heavy metal (HM) accumulation and mapping the accumulation risks can provide valuable information for the soil remediation and protection. This study investigates the risk of soil HM accumulation and its relationships with human activities, which were expressed by weighted industry distance (WID), weighted road density (WRD), and population density (LnPD). Three models, namely logistic regression (LR), geographically weighted logistic regression (GWLR), and kriging with external drift (KED), were used and compared. The results, which were based on the soil Pb contents in Wuhan city as an example, showed that the coefficients of LnPD, WRD and WID in LR model were all positive, meaning the increase of LnPD, WRD and WID will generally elevate the risk of Pb accumulation. Whereas in GWLR model, the coefficients were spatially varying, thus distinct dominant factors can be identified at every location by comparing the GWLR coefficients. Furthermore, GWLR gave significant higher model accuracy than LR, and had approximate but more straightforward explanatory power compared to KED. The results suggest that GWLR is a promising method in analyzing and mapping the spatial nonstationary relationships between the risk of soil HM accumulation and human activities.