The differential responses of large river sink triggered by spatially and temporally differentiated factors in complex earth systems are still unclear. Based on the machine learning methods, a new data-driven method is proposed here for prediction of satellite missing value, generating the long- term continuous dataset with high resolution for temporal and spatial analysis. After evaluation of several machine learning algorithm including K-Nearest Neighbor, Support Vector Machine, Deep Neural Network and Random Forest, RF is ultimately selected for establishing model named SatelliteFixer with its best performance on typical case, identied by PCC and Taylor diagrams. Based on MODIS data, trained model is able to provide reliable continuous inversion value of daily suspended sediment concentration with constraint of cruise data. Here, surface suspended sediment concentration data are used to study the interdecadal cariation trend of the Yellow River sink area in the Bohai Sea-North Yellow Sea as an example. The results show that the proximal sink near the estuary shows a trend of becoming turbid over the decadal period, while the distal sink has the opposite trend. The inconsistency is mainly due to the reduction of riverine material and coarsening of the delta under the inuence of human activities, together with those caused by increased northerly wind speed and frequent storm events in winter.