Hydrological models are widely used by water managers as a decision support tool for both real-time and long-term applications. Some examples of real-time management issues are the optimal management of reservoir releases, flood forecasting or water allocation in drought conditions. Long term-applications include the impact analysis of planned hydraulic structures or land use changes and the predicted impact of climate change on water availability. One of the obstacles hydrologists face in setting up river basin models is data availability, whether because the datasets needed do not exist or because of political unwillingness to share data which is a common problem in particular in transboundary settings. In this context, remote sensing (RS) datasets provide an appealing alternative to traditional in-situ data and much research effort has gone into the use of these datasets for hydrological applications. Many types of RS are now routinely used to set up and drive river basin models. One of the key hydrological state variables is river discharge. It is typically the output of interest for water allocation applications and is also widely used as a source of calibration data as it presents the integrated response of a catchment to meteorological forcing. While river discharge cannot be directly measured from space, radar altimetry (RA) can measure water level variations in rivers at the locations where the satellite ground track and river network intersect called virtual stations or VS. In this PhD study, the potential for the use of RA over rivers for hydrological applications in data sparse environments is investigated. The research focused on discharge estimation from RA as well as the use of RA for data assimilation to routing models with the objective of improving river discharge forecasts. In the first paper included in this PhD study, the potential for using altimetry for level and discharge monitoring in the Zambezi River basin was assessed. Altimetric levels were extracted using a detailed river mask at 31 VS located on rivers down to 80 m wide. Root mean square errors relative to in-situ levels were found to be between 0.32 and 0.72 m. Discharge was estimated from the altimetric levels for three different data availability scenarios: availability of an in-situ rating curve at the VS, availability of one pair of simultaneous measurement of cross-section and discharge and availability of historical discharge data. For the few VS where in-situ data was available for comparison, the discharge estimates were found to be within 4.1 to 13.8% of mean annual gauged amplitude. One of the main obstacles to the use of RA in hydrological applications is the low temporal resolution of the data which has been between 10 and 35 days for altimetry missions until now. The location of the VS is also not necessarily the point at which measurements are needed. On the other hand, one of the main strengths of the dataset is its availability in near-real time. These characteristics make radar altimetry ideally suited for use in data assimilation frameworks which combine the information content from models and current observations to produce improved forecasts and reduce prediction uncertainty. The focus of the second and third papers of this thesis was therefore the use of radar altimetry as update data in a data assimilation framework. The approach chosen was to simulate reach storages using a simple Muskingum routing scheme driven by the output of a rainfall-runoff model and to carry out state updates using the Extended Kalman Filter. The data assimilation approach developed was applied in two case studies: the Brahmaputra and Zambezi River basins. In the Brahmaputra, data from 6 Envisat VS located along the main reach was assimilated. The assimilation improved model performance with Nash-Sutcliffe model efficiency increasing from 0.78 to 0.84 at the outlet of the basin. In the Zambezi River basin, data from 9 Envisat VS located within 2 distinct watersheds was assimilated. Because of the presence of the large Barotse floodplain in the area, the routing scheme was coupled to a simple floodplain model. Overall model performance was improved through assimilation with Nash-Sutcliffe model efficiencies increasing from 0.21 to 0.65 and 0.82 to 0.88 at the outlets of the 2 watersheds. The results from both the Zambezi and the Brahmaputra showed that the low temporal resolution of the data could be compensated in part by the use of multiple VS which will acquire data on different days over the 35-day repeat period. This highlights the benefits which could be obtained from radar altimeter missions with denser spatial resolution allowing for more, narrower rivers to be monitored. In both case studies, the simple error model specification used was found to be one of the weak points of our approach and further research is suggested in this direction.