A new method for spatial three-dimensional prediction of soil heavy metals contamination

Abstract

Soil heavy metals contamination are highly correlated to human health. It is crucial to employ three-dimensional heavy metals modeling and mapping for site assessment and remediation. However, current methods are limited due to the poor consideration of both spatial auto-correlation and stratified heterogeneity concurrently. The present study established a novel methodology 3D-MSN to model soil metals, encompassing autocorrelation and heterogeneity. In addition to considering in-strata correlation and between-strata heterogeneity like traditional methods, 3D-MSN accounted for the between-strata correlation to enhance the accuracy of prediction. A former agrochemical plant was used as a case to validate the superiority of the 3D-MSN method over traditional approaches. The accuracy of different methods was evaluated using mean absolute error (MAE) and root-mean-square error (RMSE), through leave-one-out cross-validation. Results demonstrated significant spatial autocorrelation and stratified heterogeneity for the presence of As and Cu in soil. 3D-MSN exhibited the lowest MAEs (2.424 mg/kg for As, 4.863 mg/kg for Cu) and RMSEs (3.439 mg/kg, 7.279 mg/kg) compared to 3D-ordinary kriging (MAEs (2.949 mg/kg, 6.482 mg/kg) and RMSEs (3.890 mg/kg, 8.364 mg/kg)) and 3D-stratified kriging (MAEs (2.571 mg/kg,5.184 mg/kg) and RMSEs (3.570 mg/kg, 7.412 mg/kg)). 3D-MSN also accounted for estimation uncertainties. Considering autocorrelation and stratified heterogeneity, 3D-MSN presented superior performance. This research contributes to advancing the field of three-dimensional heavy metal modeling and provides valuable insights for site assessment and remediation efforts.