The cuurent study was performed to predict spatial distribution of some heavy metals (Ni, Fe, Cu, Mn) in western Iran, using environmental covariates and applying two machine learning methods comprised Random forest (RF), and Cubist. In this respect, a combination of different input environmental variables (remote sensing data, topographic attributes, thematic maps and soil properties) were used in modeling under four scenarios (I: remote sensing data (RS); II: RS + topographic attributes resulted from digital elevation model (DEM); III: RS + topographic attributes + thematic maps; IV: RS + topographic attributes + thematic maps +soil properties). The maps of Euclidean distance from mines and roads as well as the geology map have been used as thematic maps. A total of 346 soil samples were taken using stratified random sampling from the surface layers (0–20 cm depth) of the studied area and selected heavy metals (Ni, Fe, Cu, Mn), and soil properties were measured in the laboratory. RF and Cubist models were used to predict soil heavy metals in four scenarios. The results indicated that the best prediction accuracy was achieved for the fourth scenario (IV) when all input variables were combined to predict selected heavy metals. Moreover, two models showed different capability for various metals. According to our results, the random forest model had a high accuracy in predicting Ni (R2 = 0.67) and Cu (R2 = 0.60), In contrast, the Cubist model had a higher accuracy in predicting Mn (R2 = 0.55). For predicting Fe, both models provided a similar accuracy (R2 = 0.73). This study proved the high capability of machine learning methods to use easily available environmental data to predict studied heavy metals in the large scale that are essential for decision making in sustainable management in agricultural and environmental concerns.
