Field-based hyperspectral imaging models for predicting multiple heavy metal pollution in soils based on soil component analysis and spectral indicators

This study developed a field-based SWIR hyperspectral imaging framework for detecting multi-element heavy metal pollution using regulatory-based pollution indices (PI). A controlled testbed incorporating mine tailings and native soils enabled investigation of spectral-geochemical mechanisms linking contamination levels to hydroxyl-bearing secondary minerals (vermiculite, montmorillonite, illite) through their cation-sorption capacity. By acquiring hyperspectral data under natural outdoor conditions, atmospheric absorption zones were identified and excluded, ensuring operational field applicability. Spectral variations in Al-OH, Fe-OH, and Mg-OH absorption regions (2160–2380 nm) showed strong correlations with PI values across both soil types despite different geological origins, reflecting the ubiquitous nature of hydroxyl-bearing minerals in most natural soils and supporting broad model applicability. Random Forest Regression using first-derivative reflectance achieved superior performance (R2v = 0.94, NRMSE = 9.88%), while Elastic Net Regression provided interpretable equations suitable for regulatory mapping. Spatial mapping revealed 81% of testbed soils exceeded national contamination thresholds, demonstrating the framework’s practical utility for rapid, non-destructive contamination assessment and remediation prioritization across diverse soil environments.