Introduce multivariate two-dimensional information to establish a data-driven volume estimation model for complex microplastic fibers

Microplastic (MP) fibers are pervasive in the environment, posing significant ecological risks due to their persistence and potential toxicity. Accurate quantification of fiber volume is crucial for assessing its environmental flux and ecological impacts. Traditional methods, reliant on geometric assumptions, often miscalculate the volume of complex-shaped fibers, particularly those with curvature, due to limitations in accurately extracting length and width. This study introduces a novel framework leveraging machine learning and image recognition to estimate MP fiber volume. Our volume estimation model incorporates shape descriptors and additional two-dimensional (2D) features, such as area, allowing it to learn volumetric patterns across diverse fiber and fragment shapes. Compared to geometric models, it demonstrated superior performance, achieving 89.43 % accuracy and a mean absolute percentage error (MAPE) of 10.58 % ± 4.30 % in external testing with real-world MP samples. Error analysis indicates that our machine learning approach overcomes the limitations of geometric assumptions. Interpretability analysis highlights area as the most significant feature for volume estimation, followed by aspect ratio, circularity, and solidity for fibrous MPs. This study underscores the potential of machine learning in MP volume estimation, offering a valuable tool for environmental scientists and policymakers to better understand and manage MP pollution.