Global Potential Risk of Thallium in Topsoil: A Cropland-Focused Quantification Framework.

Area of Science:

• Environmental Science
• Geochemistry
• Toxicology

Background:

• Thallium (Tl) pollution from natural and anthropogenic sources presents growing environmental and health concerns.
• Localized Tl pollution hotspots pose significant risks, despite generally low global background levels.

Purpose of the Study:

• To predict global topsoil Thallium concentrations exceeding 1 mg/kg using machine learning.
• To identify known and previously unreported Tl pollution hotspots.
• To assess the risk of Tl contamination in global croplands.

Main Methods:

• Utilized over 20,000 topsoil Tl measurements and 21 environmental variables.
• Applied CatBoost classification and regression models to predict Tl hotspots and concentrations.
• Integrated cropland cover maps with model predictions.

Main Results:

• CatBoost classification model achieved an AUC of 0.89, identifying Tl hotspots.
• CatBoost regression model achieved an R-squared of 0.62, predicting Tl concentration distributions.
• High-risk areas for Tl pollution are associated with mining, land cover, and geological factors.
• Approximately 9.9% of global croplands have a >47% probability of exceeding 1 mg/kg Tl.

Conclusions:

• Significant portions of croplands in South America, Asia, and Africa are at high risk for Tl contamination.
• Heightened attention to soil Tl testing in high-risk croplands is crucial for agricultural safety.
• The study provides a predictive framework for managing Tl pollution risks in agricultural soils.