Advanced three-dimensional prediction model based on stable machine learning for soil pollution: A case study from a...

Area of Science:

• Environmental Science
• Geospatial Analysis
• Machine Learning

Background:

• Over five million global sites face soil contamination, necessitating accurate 3D characterization for risk assessment and remediation.
• Existing 3D interpolation methods struggle to incorporate spatial correlation and heterogeneity, crucial for understanding subsurface contamination patterns.

Purpose of the Study:

• To develop a refined 3D interpolation model for soil contamination that accounts for site characteristics, spatial position, correlation, and heterogeneity.
• To quantify prediction uncertainty in soil contamination modeling.
• To enhance model generalizability and reliability through a novel selection strategy.

Main Methods:

• Developed a machine learning (ML) model integrating site characteristics, spatial position, spatial correlation, and spatial heterogeneity.
• Implemented a stability analysis framework with random dataset partitioning and covariate ordering for model selection.
• Conducted 1000 random simulations for model screening.

Main Results:

• The ML model demonstrated high predictive performance, achieving R² values above 0.73 for four heavy metals (HMs).
• The stability analysis and model selection strategy enhanced model generalizability.
• The developed method provides a reliable basis for screening and selecting accurate 3D interpolation models.

Conclusions:

• Introduced a precise and versatile 3D spatial interpolation method for soil contamination.
• The model's accessible covariates facilitate widespread application in site assessment and remediation.
• This study significantly contributes to improving the accuracy and efficiency of managing contaminated sites worldwide.