Riverine dissolved organic carbon (DOC) fluxes to the ocean is an important component of the global carbon cycle, yet its high-resolution assessment in China’s coastal rivers remains limited, especially for small-scale rivers. Here, we developed a cascaded framework integrating multi-parameter conversion with high-resolution remote-sensing retrieval. Using 471 field samples from 186 coastal rivers, we established an XGBoost-based DOC conversion model and applied it to national automatic water-quality monitoring data to generate quasi-in situ DOC estimates. More than 13,000 quasi-in situ estimates were collocated with Sentinel-2 spectral features and environmental variables to train a 10 m RF-based DOC retrieval model, achieving a validation R² of 0.70 and an RMSE of 0.57 mg/L. Retrieved DOC concentrations were integrated with river discharge estimates to quantify DOC fluxes to the ocean. Retrieved DOC concentrations showed a north–south gradient and contrasting seasonality: northern basins had higher concentrations with spring–summer maxima and winter minima, whereas southern basins had lower concentrations with more frequent winter–spring maxima and summer–autumn minima. During 2019–2025, annual mean DOC concentrations declined significantly at −0.02 (mg/L·yr) (p = 0.01), mainly driven by decreases in the Songliao, Haihe, and Yellow River basins. The multi-year mean DOC flux was 4.75 × 10⁶ t/yr. Small-scale rivers (Strahler orders 1–5) contributed approximately 12.6% of the total DOC flux and showed an increasing relative contribution during 2019–2024. This study advances high-spatiotemporal-resolution monitoring of riverine DOC and supports more complete quantification of land-to-ocean carbon export.