Site-scale groundwater pollution risk assessment using surrogate models and statistical analysis.

Area of Science:

• Environmental Science
• Geochemistry
• Machine Learning

Background:

• Petroleum pollution poses a global environmental threat, necessitating effective remediation strategies.
• Monitored Natural Attenuation (MNA) is a promising in-situ technique for petrochemical-contaminated sites.
• Traditional deterministic models struggle with site heterogeneity and parameter uncertainties in MNA feasibility assessments.

Purpose of the Study:

• To develop a reliable method for assessing MNA feasibility at petrochemical sites, considering parameter uncertainties.
• To simulate phenol leakage and natural attenuation using physical models and time-series data.
• To employ machine learning for constructing stochastic parameter models to enhance risk assessment.

Main Methods:

• Established a physical model using GMS software for a petrochemical-contaminated site in northern China.
• Utilized time-series phenol concentration data (2018-2020) to simulate contaminant transport and attenuation.
• Applied Random Forest Regression (RFR) to build stochastic parameter models, validated against numerical outputs (NSE >0.96).

Main Results:

• Adsorption coefficient and maximum adsorption capacity were identified as key factors influencing model outcomes.
• The stochastic RFR model predicted a significant reduction in phenol concentration exceedance probability from 64.0% to 15.7%.
• Porosity was identified as the most influential parameter in the stochastic model for mitigating phenol pollution risk.

Conclusions:

• Machine learning-based stochastic models offer a reliable approach for MNA feasibility assessment, overcoming limitations of deterministic models.
• The study presents a novel method for rapid pollution risk assessment at petrochemical sites with heterogeneous conditions or data limitations.
• The findings provide valuable insights for risk assessment and management of petrochemical-contaminated sites, particularly those with complex hydrogeological characteristics.