A hybrid data-driven machine learning method for mapping trace metals in urban soils: Integrating source...

Area of Science:

• Environmental Science
• Geochemistry
• Data Science

Background:

• Mapping regional soil pollution is crucial for environmental management, but urban heterogeneity complicates trace metal prediction.
• Conventional methods struggle with the spatial variability of soil trace metals in complex urban landscapes.

Purpose of the Study:

• To develop and validate a Hybrid Data-Driven Machine Learning Method (HDML) for high-precision soil trace metal distribution mapping.
• To assess the contribution of different data sources (environmental factors, spatial trends, source apportionment) to prediction accuracy.

Main Methods:

• The study proposes a Hybrid Data-Driven Machine Learning Method (HDML) integrating environmental factors, spatial distribution trends, and source apportionment.
• A case study in the Chang-Zhu-Tan urban agglomeration, China, was used to test the HDML model, specifically HDML-XGB, against Kriging and XGBoost.

Main Results:

• The HDML method achieved high predictive performance for Cd, As, Pb, and Cr (R²: 0.796–0.883) and moderate performance for Mn (R² = 0.669).
• HDML-XGB significantly reduced mean RMSE by 77.89% compared to Kriging and 51.55% compared to XGBoost.
• Spatial distribution trends and source apportionment information were the most important features, contributing over 68% to the model's explanatory power.

Conclusions:

• The HDML method provides a reliable tool for accurate soil trace metal mapping in heterogeneous urban environments.
• Integrating multi-source information, particularly spatial trends and source apportionment, enhances prediction accuracy and captures spatial details effectively.
• This approach supports targeted soil pollution remediation and effective environmental risk management strategies.