Global marine plastic pollution has become a major concern for its negative effect on marine life and potential human health. However, research on the distribution and quantification of coastal sources remains limited, hindering the formulation of effective mitigation strategies. Here, we developed a machine learning-based framework using 25,892 data points from 3468 coastal sampling sites, incorporating 38 features including socioeconomic factors, plastic waste sectors, and types. The framework further integrates coastal geomorphological and tidal characteristics to account for the regulating effects of natural processes on plastic emissions. To systematically characterize pollution severity, we proposed the Coastal Plastic Pollution Index (CPPI), which classifies coastal input sites into four levels (I-IV) based on estimated emission. The model estimated global coastal plastic emission in 2019 between 15.32 and 59.18 Kt, with per capita values ranging from 2.17 to 8.38 g. Mismanaged plastic waste (MPW), plastic waste generation (PWG), consumer and institutional products (CIP), packaging (PAC) PET, and PP play dominant roles in coastal emissions. After clustering the sampling data using unsupervised learning, we found a high degree of consistency in CPPI distribution among countries within the same cluster. Our findings offer robust scientific support for both national and international policy actions.
Keywords: clustering analysis; coastal emissions; machine learning; marine plastics; policy mitigation.