Marine litter, predominantly plastic, is a global environmental threat, which management is hindered by the absence of large-scale monitoring tools. This study extends a framework to detect Floating Marine Macro-Litter (FMML) in the complex waters of the Pearl River Estuary (PRE) and Hong Kong, using the MARIne Debris Archive benchmark dataset and Sentinel-2 imagery. A comparative analysis showed that a Random Forest (RF) classifier – leveraging spectral indices and textural features – significantly outperformed a U-Net model, achieving higher macro-F1 scores and precision. Applying the RF classifier to the PRE from 2017 to 2024, results revealed distinct spatial–temporal patterns, with FMML peaks in March and September linked to seasonal hydrology and complex circulation. Detection results deviated from summer clean-up ground truth data, suggesting that high hydrodynamic energy processes may cause rapid transport, submergence, or stranding, reducing satellite visibility. Although the model faced generalization challenges, it demonstrated strength in identifying large accumulations related to circulation and regional debris traits. This research underscores the potential of satellite monitoring while emphasizing the need for regionally calibrated models that integrate hydrodynamic data to align remote sensing with on-ground realities, ultimately supporting mitigation in dynamic coastal environments.