Digital mapping and adaptive zoning of potentially toxic elements in selected floodplain soils of Khuzestan Province, Southwest Iran

Digital soil mapping, a cost-effective and data-driven method, is widely used for predicting soil properties. However, its targeted integration with pollution indices, particularly for spatial risk assessment in floodplains, remains limited in both research and practice. This study aimed to evaluate the performance of machine learning (ML) models, identify key environmental covariates for predicting potentially toxic elements (PTEs), and integrate model outputs with the Nemerow Pollution Index (NPI) to generate a cumulative pollution map for floodplain soils in Khuzestan Province, southwestern Iran. A total of 200 surface soil samples (0–10 cm) were collected using stratified random sampling. Three well-established ML models, Random Forest (RF), Support Vector Regression, and Partial Least Squares Regression, were evaluated for predicting concentrations of Zn, Ni, Cr, and Pb. Environmental covariates, derived from digital elevation model derivatives and remote sensing spectral indices, were reduced to 17 key predictors using Principal Component Analysis. Among the evaluated ML models, the RF model demonstrated comparatively better predictive performance for the PTEs studied [R 2 ranged from 0.38(Cr) to 0.52(Pb); RMSE values from2.59(Pb) to 16.53(Cr); and MAE values from 1.46(Pb) to 14.52 (Cr)]. NDVI and CNBL indices were identified as the most influential predictors, highlighting the complementary roles of remote sensing and topographic variables. The predicted NPI values ranging from 0.42 to 1.56, enabled adaptive zoning of the floodplain into three management classes. Uncertainty analysis via bootstrapping revealed a substantial underestimation of the mean NPI relative to field-observed values, particularly in high-risk hotspots. Overall, by integrating ML-based predictions with an integrated pollution index, this study provides a practical framework for data-informed soil pollution assessment and adaptive management in flood-prone areas.