A generative physics-informed machine learning model for soil microplastic accumulation dynamics.

Area of Science:

• Environmental Science
• Soil Science
• Data Science

Background:

• Microplastic pollution poses significant environmental challenges.
• Traditional methods for studying microplastic transport in soils are limited by heterogeneity and complex interactions.
• Developing robust models for soil microplastic dynamics is crucial for understanding environmental fate.

Purpose of the Study:

• To develop an integrated, mechanistically informed approach for modeling soil microplastic dynamics.
• To combine experimental data with advanced machine learning and statistical methods.
• To identify key parameters influencing microplastic transport in soils.

Main Methods:

• Utilized a machine learning framework (TabNet) integrated with first-principles PDEs for predictive modeling.
• Employed advanced statistical dependency analyses including Spearman's rho, Kendall's tau, distance correlation, HSIC, copula-based modeling, and Granger causality.
• Enhanced model interpretability using SHAP, partial dependence plots, symbolic metamodeling, Double ML, and TCAV.

Main Results:

• Density solution emerged as the most influential parameter, acting as a composite variable integrating multiple input interactions.
• Secondary influential factors include land use (≈0.9-0.93), size range (≈0.77-0.86), sampling depth (≈0.73-0.81), and soil organic matter (SOM) operations (≈0.64-0.72).
• Statistical analyses revealed significant nonlinear interactions and temporal/causal importance of density solution, land use, and size range.

Conclusions:

• The integrated approach provides a robust framework for understanding and predicting soil microplastic dynamics.
• Density solution is a critical, dominant factor in microplastic transport, simplifying complex interactions.
• Land use, size range, and sampling depth are significant context-dependent factors influencing microplastic movement in soils.