Feasibility of using environmental covariates and machine learning to predict the spatial variability of selected heavy metals in soils