Key points are not available for this paper at this time.
Artisanal and Small-Scale Mining (ASM) is a critical driver of soil contamination in West Africa, yet quantifying the relative contributions of natural and anthropogenic sources remains a major challenge. This study presents an integrated framework combining contamination indices, receptor modelling, and Machine Learning (ML) to evaluate Potentially Toxic Element (PTE) dynamics in soils from the Nangodi area, northern Ghana. A total of 552 grid-based soil samples were analysed for Cr, Co, Cu, Pb, V, and Zn using ED-XRF. Descriptive statistics revealed highly skewed distributions and elevated coefficients of variation for Cr, Co, and Cu, reflecting heterogeneous contamination patterns typical of ASM terrains. Pollution indices indicated divergent outcomes: the ecological Risk Index (RI) classified all samples as low risk (RI < 150), whereas the Nemerow Integrated Pollution Index (NIPI) identified 35.5% of samples as heavily polluted (NIPI ≥ 3). Positive Matrix Factorisation (PMF) resolved three distinct sources: a Cr–V lithogenic factor (51.7%), a Cu–Co–V assemblage reflecting mixed geogenic–agricultural influences (40.0%), and a Pb–Zn anthropogenic factor linked to ASM activities (8.3%). ML models (Artificial Neural Network and XGBoost) achieved high predictive accuracy (R² = 0.82–0.99), validating PMF factors and identifying Pb, Cu, and Cr as dominant predictors across source profiles. The integration of receptor and machine learning models provided both explanatory and predictive capacity, advancing current approaches to soil contamination assessment.
Sagoe et al. (Wed,) studied this question.