Machine learning reveals climate-urbanization interactions controlling soil PCBs in megacities

Area of Science:

• Environmental Science
• Urban Ecology
• Geochemistry

Background:

• Urbanization-climate interactions significantly influence pollutant dynamics in megacities.
• Understanding these complex interactions is crucial for effective environmental management.
• Soil polychlorinated biphenyls (PCBs) are persistent organic pollutants of concern in urban environments.

Purpose of the Study:

• To develop an integrated machine learning framework to analyze soil PCB contamination.
• To identify key urbanization and climate drivers of PCB dynamics along an urban-rural gradient in Shanghai.
• To quantify the impact of natural attenuation by vegetation on PCB contamination.

Main Methods:

• Integrated machine learning framework (Random Forest, XGBoost, SHAP analysis).
• Year-long soil monitoring across 100 soil columns along Shanghai's urban-rural gradient.
• Analysis of urbanization variables (population density, road density) and meteorological conditions.

Main Results:

• Three distinct soil PCB contamination zones identified, peaking in the peri-urban industrial belt.
• Population density, road density, vegetation, and winter meteorological conditions were key drivers.
• Vegetation indices counterbalanced 80-92% of urbanization-driven contamination during growing seasons.
• High-chlorinated PCBs linked to cold/wet conditions and intensive development; low-chlorinated PCBs to warm periods and road density.

Conclusions:

• Machine learning framework effectively reveals drivers of soil PCB contamination.
• Urban development creates stable baseline contamination, while climate modulates mobility.
• Findings support climate-adaptive contamination management and sustainable urban planning in megacities.