Improved ecological risk assessment of heavy metal contamination using machine learning-corrected portable XRF measurements in a high-sulfur mining landscape

High-sulfur mining generates severe acid mine drainage (AMD) and heavy metal contamination, posing substantial risks to environmental and human health. Accurate and efficient risk assessment is critical for land management. In a high-sulfur mining region of East China, this study investigated soil contamination by heavy metals (Mn, Pb, Zn, As, Cu, Ni), measured key soil properties (pH, organic matter, available phosphorus, available iron, sulfate), and evaluated human health and ecological risks using the USEPA model and Hakanson potential ecological risk index (RI), respectively. We developed an integrated Random Forest (RF) machine learning framework to enhance assessment accuracy. The framework successfully corrects portable X-ray fluorescence (pXRF) data using environmental covariates, significantly improving reliability (R² > 0.85 for key metals). The RF models also effectively classified ecological risk categories (accuracy=0.74), identifying sulfate and pH as the most critical influencing factors. Building on this, we proposed an Adapted Ecological Index (AEI) that incorporates RF-derived environmental weights for a refined, site-specific assessment. Results revealed severe contamination dominated by As, posing significant non-carcinogenic and carcinogenic human health risks, primarily via oral ingestion. The AEI confirmed arsenic's dominance in ecological risk, classifying 24.3% of sites as very high risk. This study successfully developed an integrated, machine learning-enhanced framework to improve pXRF accuracy and refine ecological risk assessments in high-sulfur mining landscapes. This framework provides a robust and rapid tool for environmental risk assessment, offering valuable insights for policymakers and environmental managers in formulating targeted remediation strategies and land-use policies for mining-affected areas globally. • Machine learning effectively corrects pXRF data for accurate soil metal assessment. • Novel Adapted Ecological Index (AEI) provides machine learning-weighted, site-specific risk assessment. • Severe arsenic contamination dominates ecological and human health risks. • Sulfate and pH critically influence ecological risk classification accuracy.