Machine Learning and Gaussian Mixture Model for Delineating Soil Cadmium Risk Zones

Effective management of regional soil cadmium (Cd) contamination is limited by difficulties in accurately predicting Cd concentrations and delineating the risk zones. This study proposed an integrated approach, combining a Machine Learning Spatial Information Enhancement algorithm with a Gaussian Mixture Model (GMM) to address these challenges. The approach was applied to a case study in Jiande City, Zhejiang Province, eastern China. The results indicated that incorporating spatial regionalization covariates, which capture the spatial hierarchical heterogeneity of the study area, markedly improved the model’s predictive accuracy ( R 2 = 0.93). Based on prediction, the study area was divided into 4 risk zones using GMM. High-risk zones (1.65%) are concentrated in black shale regions, where a high geochemical background raises soil Cd levels to 6.30 mg/kg. Moderate-risk zones (2.78%), with an average concentration of 0.83 mg/kg, are primarily located along low-altitude valleys affected by intensive human activities. Low-risk zones (27.1%) exhibit higher Cd bioavailability and are dispersed across the region. Safety zones (68.4%) are situated in mountainous regions with lower temperatures, precipitation, and limited human activities. Of the total contaminated rice samples, 86.4% fell within identified risk zones, increasing with risk level, confirming the scientific robustness of the risk zones classification approach. This framework enables precise delineation of risk zones and supports the development of targeted, cost-effective soil management strategies, providing a practical tool for contamination control and ecosystem health protection.