Predicting and regulating potential zone for heavy metal re-contaminated farmland: Case study in Taiwan

Abstract

Taoyuan City ranks second in Taiwan for land pollution area. Despite government remediation efforts, the dynamic nature of land characteristics continues to present risks of recontamination. This study utilizes the Random Forest (RF) algorithm to predict and manage potential heavy metal recontamination zones in farmland. Key model parameters were identified through a literature review, leveraging a dataset of 1,555 records from Taoyuan City's managed agricultural lands collected over 19 years (2004–2022). The proposed RF model demonstrates exceptional performance, achieving 75.76% accuracy in predicting soil monitoring concentration variations and an outstanding 99.95% accuracy for soil heavy metal concentration increments. It outperforms alternative models, including MLP, SVM, and DT, in establishing prediction thresholds for heavy metal recontamination, such as cadmium (Low: 0.0022, High: 0.0045), copper (Low: 0.1095, High: 0.1826), and zinc (Low: 0.2374, High: 0.5479), measured in mg/kg-day increments. This model provides critical insights into areas susceptible to heavy metal pollution, with a specific focus on cadmium, copper, and zinc contamination. By leveraging its predictive capabilities, authorities can proactively devise and implement strategies to mitigate risks, safeguarding both the environment and public health.