Disentangling natural and anthropogenic influences on the spatial variability of soil cadmium contamination

Soil cadmium (Cd) contamination poses serious environmental risks to cultivated land use, driven by a combination of natural processes and anthropogenic activities. Traditional single-factor analyses often struggle to capture its spatial heterogeneity and the interactions among influencing factors. To address this, we propose an integrated framework that combines geographical detectors (GD), random forest (RF), and structural equation modeling (SEM) to systematically identify the key drivers of Cd contamination and quantify their direct and indirect effects. This hybrid approach advances spatial modeling by integrating data-driven analysis with causal inference. The results reveal significant spatial variability in Cd concentrations, with a coefficient of variation reaching 103.70 %, underscoring the non-uniformity of the contamination landscape. Two pollution indices, the geoaccumulation index (I_geo) and exceeding standard rate (ES), consistently indicated increasing contamination from northwest to southeast with hotspots near mining areas. Results from both the GD and RF models indicate that major contributing factors include distance to the mining site, soil organic carbon (SOC), pH, topographic elevation (DEM), and surrounding residential zones. Interaction analysis showed that the combined effect of mining proximity and soil pH intensifies Cd accumulation, a pattern further confirmed by SEM. SEM also demonstrated that distance to the mining site is the most significant direct driver of Cd contamination, while DEM indirectly affect Cd levels by influencing mining-related spatial patterns. This integrated framework offers a comprehensive methodology for understanding the spatial dynamics of soil Cd contamination and provides new insights into the interplay between natural and human-induced drivers.