Towards A universal settling model for microplastics with diverse shapes: Machine learning breaking morphological...

Area of Science:

• Environmental Science
• Fluid Dynamics
• Machine Learning

Background:

• Accurate prediction of microplastic settling velocity is crucial for modeling their transport in aquatic environments.
• Existing models are morphology-specific (fragmented, filmed, fibrous), lacking universal applicability.
• Reliance on predominant morphology from samples complicates transport modeling due to spatiotemporal variability and mixed morphologies.

Purpose of the Study:

• To develop a universal settling model for microplastics with diverse shapes.
• To address the challenge of reliably determining appropriate settling models for complex microplastic mixtures.
• To create a physically interpretable and expandable model for microplastic transport.

Main Methods:

• Proposed a unique shape factor using a modified machine learning method to distinguish microplastic morphologies.
• Developed a universal settling velocity model using a physics-informed machine learning algorithm.
• Validated the model against independent datasets for microplastic fragments, films, and fibers.

Main Results:

• The newly developed universal model accurately predicts the settling velocity of microplastics across different morphologies.
• The model demonstrates reasonable predictive performance for microplastic fragments, films, and fibers.
• The model's transparent, formulaic structure enhances physical interpretability and potential for future improvements.

Conclusions:

• A universal model for microplastic settling velocity has been successfully developed, applicable to diverse shapes.
• The physics-informed machine learning approach with a novel shape factor overcomes limitations of existing models.
• This study provides a paradigm for integrating machine learning into physically-based models for environmental transport studies.