Interpretable machine learning reveals transport of aged microplastics in porous media: Multiple factors co-effect

Abstract

Microplastics (MPs) easily migrate into deeper soil layers, posing potential risks to subterranean habitats and groundwater. However, the mechanisms governing the vertical migration of MPs in soil, particularly aged MPs, remain unclear. In this study, we investigate the transport of MPs under varying MPs properties, soil texture and hydrology conditions. Under nearly all controlled conditions, aged MPs demonstrated a higher vertical mobility compared to virgin MPs. By employing interpretable machine learning models (IML), we not only identified the dominant role of individual parameters in the vertical migration of MPs but also discovered that the increased contribution of carbonyl index and O/C ratio in aged MPs, along with the enhanced interaction with other feature parameters, collectively promotes the elevated vertical mobility of aged MPs. The varying contributions of different feature parameters under individual control variables revealed the mechanisms of MPs vertical migration under different gradients and highlighted the dual constraints of physical obstruction and chemical retention between MPs and soil particles. The established machine learning model was also utilized to predict the differences in vertical mobilities of MPs with varying degrees of aging. The nonlinear increasing relationship between MPs vertical mobility and simulated aging time suggests that MPs can migrate to deeper soil layers shortly after entering the soil environment. The integration of laboratory experiment with IML elucidates the key drivers of vertical MP migration. It also provides a theoretical basis for the timely removal of MPs from soil and the assessment of their potential risks.