The ability to estimate terrestrial water storage (TWS) realistically is essential for understanding past hydrological events and predicting future changes in the hydrological cycle. Inadequacies in model physics, uncertainty in model land parameters, and uncertainties in meteorological data commonly limit the accuracy of hydrological models in simulating TWS. In an effort to improve model performance, this study investigated the benefits of assimilating TWS estimates derived from the Gravity Recovery and Climate Experiment (GRACE) data into the OpenStreams wflow_hbv model using an ensemble Kalman filter (EnKF) approach. The study area chosen was the Rhine River basin, which has both well-calibrated model parameters and high-quality forcing data that were used for experimentation and comparison. Four different case studies were examined which were designed to evaluate different levels of forcing data quality and resolution including those typical of other less well-monitored river basins. The results were validated using in situ groundwater (GW) and stream gauge data. The analysis showed a noticeable improvement in GW estimates when GRACE data were assimilated, with a best-case improvement of correlation coefficient from 0.31 to 0.53 and root mean square error (RMSE) from 8.4 to 5.4 cm compared to the reference (ensemble open-loop) case. For the data-sparse case, the best-case GW estimates increased the correlation coefficient from 0.46 to 0.61 and decreased the RMSE by 35%. For the average improvement of GW estimates (for all four cases), the correlation coefficient increases from 0.6 to 0.7 and the RMSE was reduced by 15%. Only a slight overall improvement was observed in streamflow estimates when GRACE data were assimilated. Further analysis suggested that this is likely due to sporadic short-term, but sizeable, errors in the forcing data and the lack of sufficient constraints on the soil moisture component. Overall, the results highlight the benefit of assimilating GRACE data into hydrological models, particularly in data-sparse regions, while also providing insight on future refinements of the methodology.