Construction of multi-metal interspecies correlation estimation models based on typical soil scenarios.

The ecological risk assessment of metals in soils is essential for soil pollution management. However, regional soil heterogeneity and species diversity need to be considered when making these assessments. Therefore, an interspecies correlation estimation (ICE) model was constructed based on typical soil scenarios that could predict metal toxicity across species. A dataset comprising 1017 toxicity data points for 12 metals (including Cu, Zn, and Ni) across eight species and two microbial processes was analyzed. An information gain analysis revealed that soil properties contributed 0.687 to metal toxicity, which was significantly higher than that for metal structural characteristics (0.313). After clustering the soils into three typical scenarios (acidic low-clay, neutral high-clay, and alkaline medium-clay), the influence of soil properties on toxicity prediction decreased to 0.529 (neutral high-clay) and 0.496 (alkaline medium-clay). Hierarchical clustering was used to screen six metal elements with lower toxicity variabilities (inter quartile range: 0.270-169.895) for modeling and 32 optimized ICE models were established (R = 90.648-0.895, MSE = 0.183-0.614). Brassica napus was found to be the best surrogate species for predicting metal toxicity in Brassica chinensis L. under alkaline medium-clay soil conditions (R = 0.895, MSE = 0.303). This study is the first to systematically integrate soil scenario clustering, metal toxicity variability screening, and machine learning-enhanced ICE modeling and provides a more robust and adaptable framework for ecological risk assessment in heterogeneous soil environments.