Quantitative source apportionment and driver identification of soil heavy metals using advanced machine learning...

Area of Science:

• Environmental Science
• Geochemistry
• Data Science

Background:

• Accurate identification of soil heavy metal (SHM) pollution sources is crucial for effective environmental management.
• Conventional methods like positive matrix factorization (PMF) suffer from subjectivity in source interpretation due to reliance on existing research and expert experience.

Purpose of the Study:

• To develop a comprehensive source apportionment framework utilizing advanced machine learning techniques to overcome the limitations of traditional methods.
• To accurately identify and quantify the sources of SHM pollution in topsoils.

Main Methods:

• A novel framework combining self-organizing mapping (SOM) and PMF with a gradient boosting decision tree (GBDT) model was developed.
• Analysis of six heavy metals (Cd, Pb, Zn, Cu, Cr, Ni) in 272 topsoil samples.

Main Results:

• Average contents of six heavy metals exceeded background values, with Cadmium (Cd) pollution being particularly severe (66.91% of sites above risk screening values).
• PMF results indicated Pb primarily from vehicle emissions/atmospheric deposition (79.43%), Cd/Zn from sewage irrigation (79.32%/38.84%), and Cr/Ni from natural sources (85.97%/85.50%).
• GBDT identified industrial network density, water network density, and Fe2O3 content as key drivers influencing pollution sources.

Conclusions:

• The developed machine learning-based framework provides a more objective and accurate approach to SHM source apportionment.
• This improved methodology offers enhanced support for environmental agencies in formulating targeted soil pollution control strategies.