Machine learning-based prediction for settling velocity of microplastics with various shapes

Abstract

Microplastics can easily enter the aquatic environment and be transported between water bodies. The terminal settling velocity of microplastics, which affects their transport and distribution in the aquatic environment, is mainly influenced by their size, density, and shape. Due to the difficulty in accurately predicting the terminal settling velocity of microplastics with various shapes, this study focuses on establishing high-performance prediction models and understanding the importance and effect of each feature parameter using machine learning. Based on the number of principal dimensions, the shapes of microplastics are classified into fiber, film, and fragment, and their thresholds are identified. The microplastics of different shape categories have different optimal shape parameters for predicting the terminal settling velocity: Corey shape factor, flatness, elongation, and sphericity for the fragment, film, fiber, and mixed-shape MPs, respectively. By including the dimensionless diameter, relative density and optimal shape parameter in the input parameter combination, the machine learning models can well predict the terminal settling velocity for the microplastics of different shape categories and mixed-shape with R² > 0.867, achieving significantly higher performance than the existing theoretical and regression models. The interpretable analysis of machine learning reveals the highest importance of the microplastic size and its marginal effect when the dimensionless diameter D* = d_n(g/v²)^1/3 > 80, where d_n is the equivalent diameter, g is the gravitational acceleration, and ν is the fluid kinematic viscosity. The effect of shape is weak for small microplastics and becomes significant when D* exceeds 65.

Keywords: Machine learning; Microplastics; Optimal shape parameter; Shape classification; Terminal settling velocity.