مشخصات پژوهش

Accurate prediction of soil potentially toxic metals (PTMs) and identifying their relationships with environmental drivers is essential for environmental risk management in dust-affected areas of central Iran. Aluminum, copper, nickel, and manganese concentrations were measured in 107 surface and 32 subsurface samples. Environmental variables were selected using variance inflation factor analysis and the Boruta algorithm. Modeling was performed using Multivariate Adaptive Regression Splines (MARS), Random Forest (RF), and Cubist under based on readily-accessible drivers (Scenario i), sampling-dependent drivers (Scenario ii), and a combination of both (Scenario iii). Surface concentrations of all four metals were higher than subsurface levels. Calibrations were carried out with 80% data using 10-fold spatial cross-validation, and the results were validated on an independent test dataset (20%). RF-Scenario (iii) gave the best results, with test R2 values of 0.66 for Al, 0.81 for Cu, 0.61 for Ni, and 0.76 for Mn. It is worth mentioning that the first Scenario, using only the low-cost variables, also gave acceptable results for three metals (test R2 = 0.54-0.72), thus showing its potential for use in data-limited regions. SHapley Additive exPlanations (SHAP) analysis showed that the distance to industrial areas was the most important predictor of PTM concentrations, explaining about 25-41% of the total model importance. This was followed by the distance to roads, urban areas, landfills, and rivers. Partial Dependence Plots (PDPs) further confirmed the existence of strong nonlinear relationships and strong threshold effects. Specifically, PTM concentrations showed marked changes within about 10 km of industrial zones, 2–3 km of major roads, and approximately 1.8 km from rivers. The RF model showed high capability in mapping PTMs and identifying key environmental factors affecting their distribution in the study area under both Scenarios (i) and (iii). These capabilities provide a scientific basis for evidence-based land use planning and can help reduce land degradation in arid and dust-prone areas.