Abstract. River basin management can greatly benefit from short-term river discharge predictions. In order to improve model produced discharge forecasts, data assimilation allows for the integration of current observations of the hydrological system to produce improved forecasts and reduce prediction uncertainty. Data assimilation is widely used in operational applications to update hydrological models with in situ discharge or level measurements. In areas where timely access to in situ data is not possible, remote sensing data products can be used in assimilation schemes. While river discharge itself cannot be measured from space, radar altimetry can track surface water level variations at crossing locations between the satellite ground track and the river system called virtual stations (VS). Use of radar altimetry versus traditional monitoring in operational settings is complicated by the low temporal resolution of the data (between 10 and 35 days revisit time at a VS depending on the satellite) as well as the fact that the location of the measurements is not necessarily at the point of interest. However, combining radar altimetry from multiple VS with hydrological models can help overcome these limitations. In this study, a rainfall runoff model of the Zambezi River basin is built using remote sensing data sets and used to drive a routing scheme coupled to a simple floodplain model. The extended Kalman filter is used to update the states in the routing model with data from 9 Envisat VS. Model fit was improved through assimilation with the Nash–Sutcliffe model efficiencies increasing from 0.19 to 0.62 and from 0.82 to 0.88 at the outlets of two distinct watersheds, the initial NSE (Nash–Sutcliffe efficiency) being low at one outlet due to large errors in the precipitation data set. However, model reliability was poor in one watershed with only 58 and 44% of observations falling in the 90% confidence bounds, for the open loop and assimilation runs respectively, pointing to problems with the simple approach used to represent model error.
Development and Evaluation of a Multi-Year Fractional Surface Water Data Set Derived from Active/Passive Microwave Remote Sensing Data