Ecological risk zoning of soil heavy metal pollution using biomarker-enhanced indexing and machine learning.

Traditional soil heavy metal risk assessments suffer from fragmented indices and poor logical integration. Compared to single-chemical indicators, biomarkers more sensitively reflect biological effects and early ecological risks. To address these issues, a method for constructing a comprehensive index integrating biological and non-biological multi-indexes was proposed in this study. First, the Criteria Importance Through Intercriteria Correlation (CRITIC)-Fuzzy Biomarker Response Index (CFBRI) is developed by integrating the CRITIC method and fuzzy comprehensive evaluation into the Biomarker Response Index (BRI). It dynamically weights multi-timepoint biomarker data to resolve non-monotonic responses, improving the goodness of fit (R2) from 0.35 to 0.52 for the conventional BRI across different time points to 0.62 for CFBRI. Next, land use-specific comprehensive indexes (CI) were constructed by integrating abiotic and biotic indicators (Pollution Load Index, Nemerow Index, Potential Ecological Risk Index, Improved Geo-accumulation Index, and CFBRI) via Principal Component Analysis. The resulting Agricultural Land CI (AL-CI, R2 = 0.9388) and Construction Land CI (CL-CI, R2 = 0.9438) outperformed any single index. Using these CIs as reliable risk labels, multi-classification predictive models were built with five machine learning algorithms. The Random Forest models achieved the best cross-validation accuracies (RF-AL-CI: 0.888; RF-CL-CI: 0.905) and performed well in a regional case study, showing risk zoning highly consistent with land use functions and pollution logic. Overall, the proposed framework offers a transferable decision-support tool for soil ecological risk zoning, which can bridge the gap between biomarker responses and regional-scale land management decisions.