Accurately predicting the settling velocity of microplastics in aquatic environments is a prerequisite for reliably modeling their transport processes. An increasing number of settling models have been proposed for microplastics with fragmented, filmed, and fibrous morphologies, respectively. However, none of the existing models demonstrates universal applicability across all three morphologies. Scientists now have to rely on the predominate microplastic morphology extracted from filed samples to determine the appropriate settling model used for transport modeling. Given the spatiotemporal variability in morphologies and the coexistence of diverse morphologies of microplastics in natural aquatic environments, the extracted morphological information poses significant challenges in reliably determining the appropriate model. Evidently, to reliably model the transport of microplastics in aquatic environments, a universal settling model for microplastics with diverse shapes is warranted. To develop such a universal model, a unique shape factor, which can explicitly distinguish between the distinct morphologies of microplastics, was first proposed in this study by using a specifically-modified machine learning method. Using this newly-proposed shape factor, a universal model for predicting the settling velocity of microplastics with distinct morphologies was developed by using a physics-informed machine learning algorithm and then systematically evaluated against independent data sets. The newly-developed model enables reasonable predictions of the settling velocity of microplastic fragments, films, and fibers. In contrast to purely data-driven models, the newly-developed model is characterized by its transparent formulaic structure and physical interpretability, which is conducive to further expansion and improvement. This study can serve as a paradigm for future studies, inspiring the adoption of machine learning techniques in the development of physically-based models to investigate the transport of microplastics in aquatic environments.

中文翻译：

---

迈向不同形状微塑料的通用沉降模型：机器学习打破形态障碍


准确预测微塑料在水生环境中的沉降速度是对其运输过程进行可靠建模的先决条件。已经提出了越来越多的沉降模型，分别用于具有碎片、薄膜和纤维形态的微塑料。然而，现有的模型都没有证明在所有三种形态中具有普遍适用性。科学家们现在必须依靠从现场样品中提取的主要塑料微粒形态来确定用于运输建模的适当沉降模型。鉴于形态的时空变化以及自然水生环境中微塑料不同形态的共存，提取的形态信息对可靠地确定合适的模型构成了重大挑战。显然，为了可靠地模拟微塑料在水生环境中的迁移，需要为具有不同形状的微塑料建立一个通用的沉降模型。为了开发这样一个通用模型，本研究首先通过使用专门修改的机器学习方法提出了一种独特的形状因子，它可以明确区分微塑料的不同形态。使用这种新提出的形状因子，通过使用物理学机器学习算法开发了一个用于预测具有不同形态的微塑料沉降速度的通用模型，然后根据独立的数据集进行了系统评估。新开发的模型能够合理预测微塑料碎片、薄膜和纤维的沉降速度。 与纯粹的数据驱动模型相比，新开发的模型具有透明的公式化结构和物理可解释性，有利于进一步的扩展和改进。这项研究可以作为未来研究的范例，激发采用机器学习技术开发基于物理的模型来研究微塑料在水生环境中的运输。