Monitoring coarse-grained sediment is essential for managing riverbed stability, flood capacity, and ecological resilience in the Yellow River, where high sediment loads originate from the erosion-prone Loess Plateau. Although large-scale ecological restoration has been implemented since the 1980s, its long-term impact on sediment grain-size dynamics remains unclear due to limited field observations. This study developed a duallayer random forest model that synergizes Landsat satellites reflectance (1986-2022) with multi-scale watershed attributes (hydrological information, vegetation coverage, erosion susceptibility) to remotely quantify particle size distribution (PSD) of suspended sediment. The model achieved high precision (root mean square error: 2.94-4.82 %; mean absolute percentage difference: 13.44-19.87 %), enabling the first basin-wide PSD reconstruction. Key findings reveal: (1) Medium-sized particles (0.01-0.05 mm) dominated the mainstream (63.96 %), while coarse (>0.05 mm, 67.80 %) and fine particles (<0.01 mm, 25.70 %) were concentrated in the Fen and Wei Rivers, respectively; (2) Median grain size decreased by 7.25 % during the 1980s-2020s, reflecting the cumulative effects of ecological restoration, though localized coarsening (1.27, 2.39 and 2.61 %) occurred in the Huangshui, Wei, and Jing Rivers; and (3) Vegetation expansion (8.50-51.23 %) and urbanization (impervious surfaces (12.90-17.04 %)) drove particle fining, while increased wind/water erosion increased the proportion of coarse particle. This study fills a critical gap in monitoring suspended particle size dynamics and provides a scalable framework for evaluating ecological restoration outcomes and informing suspended sediment management in large, sediment-rich watersheds.