A mutual information-based framework for identifying the controlling factors of potentially toxic element soil distribution and geochemical anomalies in an abandoned tungsten mining area (Peña del Seo, NW Spain)

Historical mining areas require robust methods to assess and manage soil contamination by potentially toxic elements (PTEs). We describe a methodology for identifying and characterizing arsenic (As), copper (Cu), iron (Fe), lead (Pb), and tungsten (W) spatial anomalies and the factors controlling them, applied to soils from the San Vicente River basin (NW Spain) contaminated by the abandoned Peña del Seo mine. Spatial distribution of PTE concentrations in soils was mapped via high-density sampling (140 points) combined with geostatistical interpolation. To disentangle complex element-environment relationships, we combined mutual information (MI), which captures non-linear relationships, with Spearman's rank correlation (ρ) for monotonic trends. A key contribution is a reproducible workflow integrating domain-based data curation and permutation-based significance testing. Results reveal significant soil anomalies for As, Cu, and W near historical workings, and identify the mineralized parent rock as the primary source of W. Topographic variables showed significant associations with As and Fe, highlighting geomorphological controls on redistribution. MI identified significant non-linear dependencies (e.g., land cover with As, Pb, and W) missed by the correlation analysis. The proposed workflow provides a transferable and robust framework for source identification and environmental risk assessment in complex mining landscapes.