Interpretable machine learning models reveal the partnership of microplastics and perfluoroalkyl substances in sediments at a century scale.

It is challenging to explore the complex interactions between perfluoroalkyl substances (PFASs) and microplastics in lake sediments. The partnership of perfluoroalkyl substances (PFASs) and microplastics in lake sediments are difficult to determine experimentally. This study utilized sediment cores from Taihu Lake to reconstruct the coexistence history and innovatively reveal the collaboration between PFASs and microplastics by using post-hoc interpretable machine learning methods. Microplastics and PFASs emerged in the 1960s and have significantly increased since the 1990s. PFASs and microplastics had the highest growth rate in the 0-10 cm range, with average growth rates of 35.96 pg/g/year and 4.40 items/year per 100 g, respectively. Extreme gradient boosting demonstrated the best simulation of PFASs and microplastics in machine learning models. Feature importance and Shapley additive explanations semi-quantitatively clarified the importance of transparent and pellet microplastics on PFASs concentrations, as well as the importance of perfluorooctane sulfonate (PFOS) and ΣPFASs on microplastics. Moisture content, redox potential, χfd, and χ were the key influencing factors on contaminants. Partial dependence plots showed the influencing thresholds were 0.30 ng/g for ΣPFASs and 0.15 ng/g for PFOS on microplastics, and 10 items per 100 g for pellets and 12 items per 100 g for transparent plastics on PFASs. This study elucidated the interactions between two typical emerging contaminants on a century-scale through the intersection of environmental geochemistry and interpretable machine learning.