This dissertation advanced the traditional hydrological prediction via multi-sensor satellite remote sensing products, numerical weather forecasts and advanced data assimilation approach in sparsely gauged or even ungauged regions and then extend this approach to global scale with enhanced efficiency for prototyping a flood early warning system on a global basis. This dissertation consists of six chapters: the first chapter is the introductive chapter which describes the problem and raises the hypotheses, Chapters 2 to 5 are the four main Chapters followed by Chapter 6 which is an overall summary of this dissertation. For regional hydrological prediction in Chapter 2 and 3, two rainfall – runoff hydrological models: the HyMOD (Hydrological MODel) and the simplified version of CREST (Coupled Routing and Excess Storage) Model were set up and tested in Cubango River basin, Africa. In Chapter 2, first, the AMSR-E (Advanced Microwave Scanning Radiometer for Earth observing system) signal/TMI (TRMM Microwave Imager) passive microwave streamflow signals are converted into actual streamflow domain with the unit of m3/s by adopting the algorithm from Brakenridge et al. (2007); then the HyMOD was coupled with Ensemble Square Root Filter (EnSRF) to account for uncertainty in both forcing data and model initial conditions and thus improve the flood prediction accuracy by assimilating the signal converted streamflow, in comparison to the benchmark assimilation of in-situ streamflow observations in actual streamflow domain with the unit of m3/s. In Chapter 3, the remote-sensing streamflow signals, without conventional in-situ hydrological measurements, was applied to force, calibrate and update the hydrologic model coupled with EnSRF data assimilation approach in the same research region, but resulting in exceedance probability-based flood prediction. For global hydrological predictions in Chapter 4 and 5, a physical based distributed hydrological model CREST is set up at 1/8 degree from 50°N to 50°S and forms the Real Time Hydrological Prediction System (http://eos.ou.edu) which was co-developed by HyDROS (Hydrometeorology and Remote Sensing Laboratory) lab at the University of Oklahoma and NASA Goddard center. In Chapter 4, the CREST model is described with details and then the Real Time Global Hydrological Monitoring System will be comprehensively evaluated on basis of gauge based streamflow observation and gridded global runoff data from GRDC (Global Runoff Data Center, http://www.bafg.de/GRDC/EN/Home/homepage_node.html). In order to extend the hydrological forecast horizon for the Real Time Global Hydrological Prediction System, the deterministic precipitation forecast fields from a numerical meteorological model GFS (Global Forecasting System) as well as the ensemble precipitation forecast fields are introduced as the forcing data to be coupled into the global CREST model in order to generate the global hydrological forecasting up to around 7 days lead time in Chapter 5. The July 21, 2012 Beijing extreme flooding event is selected to evaluate the hydrological prediction skills for extremes of both the deterministic and the ensemble GFS products.