A bidirectional feedback model for enhancing heavy metal pollution sensitivity Assessment: Case study of Queshan Longhu Wetland

Abstract

Soil heavy metal pollution poses serious threats to ecological security and public health. Traditional sensitivity assessments often rely on static weighting methods, which overlook dynamic interactions between pollutants and environmental factors. This study proposes a Bidirectional Feedback Model (BFM) that integrates Random Forest and SHAP to dynamically quantify these interactions. Taking the Queshan Longhu Wetland as a case study, we applied an AHP-based BFM and validated its stability through Monte Carlo simulations. The findings revealed that: (1) The Environmental Sensitivity Index (ESI) exhibited high values in the southwest and eastern regions, closely aligning with the spatial distribution of the Natural Sensitivity Index (NSI) and underscoring the dominant role of natural factors in environment sensitivity assessments. (2) The Human Sensitivity Index (HSI) Level V areas accounted for 11.83 %, concentrated in urban centers and transportation hubs. (3) Monte Carlo simulations confirmed that the BFM model was highly robust, with minimal deviations in the Soil Pollution Index (SPI) and ESI under weight perturbations. Remote sensing image interpretation and field sampling verification corroborate the reliability of the findings. Cross-regional validation using the Huangshui River Basin in Qinghai Province demonstrated the model's strong transferability and applicability in diverse geographical settings. This study provides a novel, interpretable framework for precision soil pollution assessment and offers practical insights for environmental management in ecologically sensitive areas.