Integrated PMF-CFM source apportionment reveals contrasting microplastic signatures between rivers and drinking water sources on tropical islands.

Microplastics (MPs) are increasingly detected in freshwater systems, yet their sources and transformation pathways in tropical island environments remain insufficiently understood. This study investigates MPs in rivers and drinking water sources on a representative tropical island using a multidimensional framework that integrated polymer composition, environmental drivers, and dual-source identification models. A total of 285 MPs were detected, with higher mean abundance in drinking water sources (1.52 ± 0.91 n·L-1) than in rivers (0.69 ± 0.35 n·L-1). Although abundance differences were not statistically significant (P = 0.053), the two systems exhibited clear compositional divergence, with Rayon dominating rivers and cellophane and polyethylene terephthalate (PET) characterizing drinking water sources. To resolve the mechanisms controlling these differences, we assembled a multi-dimensional source-tracing framework that optimizes and integrates several existing analytical approaches, including redundancy analysis (RDA), positive matrix factorization (PMF), a conditional fragmentation model (CFM), and principal coordinate analysis (PCoA). PMF identified three major MPs sources, including packaging plastics, textile fibers, and mixed polymers. Meanwhile, CFM indicated stronger fragmentation signatures in rivers, consistent with continuous textile-related inputs. PCoA and further PMF confirmed significant compositional separation between the two systems (P < 0.01). Collectively, the results suggest that rivers are likely more influenced by textile-related inputs, whereas drinking water sources may preferentially accumulate buoyant, slowly fragmenting packaging-derived MPs. This multi-method framework integrates two-dimensional source tracing (composition and size) along with multi-index risk assessment, providing multi-dimensional complementary information for MPs source attribution.