A POSEIDON-R box model was applied to the Yellow and East China Seas to assess the role of sediment flux in the fate of Cs-137. Key parameters, including suspended sediment concentration, sediment flux, bioturbation intensity, top sediment layer thickness, and porosity, were updated with recent field and satellite data. Atmospheric deposition was reconstructed from Global Precipitation and in situ measurements and imposed as a time-dependent input. Although the model reproduced the overall observed trends, it consistently underestimated Cs-137 concentrations in both water and surface sediments when the observational sediment flux data were used. Three sensitivity tests were performed: (1) the re-estimation of sediment fluxes, (2) the application of differentiated sediment fluxes (settling velocities) between the water column and the sediment layer, and (3) the incorporation of observed distribution coefficients (Kd) into the model. The model responses were quantitatively assessed to determine the contributions of each parameter individually. The adjustment of sediment fluxes led to partial improvement in model performance within both the seawater and sediment layers; however, Cs-137 concentrations in both compartments remained underestimated compared with observations. In contrast, applying differentiated sediment fluxes to the seawater and sediment layers significantly enhanced model reproducibility. In addition, the application of observed Kd values improved the accuracy of Cs-137 predictions in the sediment layer, particularly in the Yellow Sea. These results highlight the importance of accurately characterizing both sediment flux and Kd, especially in high-turbidity regions such as the Yellow and East China Seas, to enhance model performance. However, the limited availability of reliable sediment flux and Kd data remains a limitation of this study. This study also demonstrates the importance of sensor-based observational data for improving the parameterization and predictive capability of marine radioactivity models.