Mosses ML: Machine-Learning-Enhanced Biomonitoring of Emerging Contaminants Using Hylocomium splendens: An Integrated...

Area of Science:

• Environmental Science
• Ecotoxicology
• Computational Science

Background:

• Atmospheric deposition of trace elements poses environmental and health risks.
• Moss biomonitoring is a cost-effective tool for assessing air pollutants.
• Traditional moss analysis lacks predictive power for contaminant risk.

Purpose of the Study:

• To introduce Mosses ML, a machine learning framework integrating moss biomonitoring with deposition data.
• To enhance the detection, interpretation, and risk assessment of atmospheric contaminants.
• To improve the identification of high-risk pollution sites.

Main Methods:

• Used *Hylocomium splendens* transplants across industrial, urban, and rural sites.
• Combined trace element data, accumulation factors, PCA, and metadata with Random Forest and Gradient Boosting models.
• Applied SHAP analysis to determine feature importance.

Main Results:

• Machine learning models achieved high predictive accuracy (R² up to 0.91) for moss metal concentrations.
• Dry deposition load and co-occurring metal signals were key predictors of contamination.
• The ML approach improved high-risk site identification by 24-38% compared to traditional methods.

Conclusions:

• Mosses ML strengthens mosses as early-warning systems for atmospheric pollution.
• The framework accurately estimates contaminant levels and identifies pollution sources.
• Mosses ML is broadly applicable to biomonitoring and supports regulatory decisions for emerging contaminants.