Decoding the transport thresholds of emerging contaminants in watersheds using explainable machine learning.

Area of Science:

• Environmental Science
• Water Resource Management
• Machine Learning Applications

Background:

• Understanding watershed emerging contaminants (ECs) transport is crucial for pollution control but hindered by complex land-climate interactions and limited predictive models.
• Microplastics (MPs) and antibiotics are significant ECs posing threats to water security, necessitating detailed investigation into their transport mechanisms.
• Existing models often fail to capture the intricate relationships between land use, climate variables, and ECs transport in river systems.

Purpose of the Study:

• To develop and validate a novel machine learning framework (ML-SHAP) for modeling ECs transport in the Huangshui River.
• To quantify the levels of MPs, antibiotics, heavy metals, and water quality indicators in seasonal water samples.
• To identify the key land use and climate drivers influencing the transport of MPs and antibiotics within the watershed.

Main Methods:

• Collected 517 seasonal water samples from the Huangshui River between 2020-2024 for contaminant analysis.
• Developed a machine learning (ML-SHAP) framework integrating multiscale land use data, landscape metrics (PD, LPI, CONTIG-MN), and 11 climate variables.
• Modeled ECs transport using the ML-SHAP framework, achieving high accuracy on training data (R² = 0.94) and good performance on test data (R² = 0.65).

Main Results:

• Microplastic levels reached 1831 items/L and antibiotic concentrations 55.33 ng/L, indicating significant water security risks.
• MPs transport was linked to fragmented urban land and connected cropland, while antibiotic transport intensified in less connected cropland.
• Forest and grassland cover, particularly with enhanced connectivity, effectively mitigated ECs transport; specific climate conditions influenced MP and antibiotic transport differently.

Conclusions:

• The ML-SHAP framework provides a robust tool for understanding and predicting ECs transport, integrating complex environmental factors.
• Riparian forest and grassland enhancement, alongside reduced urban fragmentation, are key strategies for mitigating ECs.
• Climate change scenarios (SSP585) are projected to increase MP and antibiotic loads, underscoring the need for adaptive watershed management strategies, including reforestation (SSP245).