It is well-recognized that floods represent one of the most important natural hazards in terms of frequency of occurrence, human impacts and casualties, as well as economic losses. As stated by the United Nations report «reducing disaster risk » in 2004, in the time period 1980-2000 around 196 million people in more than 90 countries were exposed on average every year to catastrophic flooding and 170,010 deaths were associated with floods. There is thus a need for an improved understanding and evaluation of flood hazard at global scale. Yet estimates of flood hazard remain of insufficient quality in many regions of the world and hamper the implementation of appropriate mitigation measures. The proposed CASCADE project (Combining eArth obServation with a large scale model Cascade for Assessing flood hazarD at high spatial rEsolution) aims at developing a Satellite Earth Observation (SEO)-based modelling framework that enables an assessment of flood hazard at large scale and high spatial resolution. The project intends to unlock the potential offered by recent developments in terms of high performance computing, parsimonious and efficient hydrological and hydraulic models, as well as the availability of globally coherent remote sensing data. By using SEO data and other globally and freely available data sets as default data for driving and parameterizing the model, the project aims at developing a modelling solution that is no longer relying entirely on the availability of long records of reliable in situ observations. Such developments are considered a pre-requisite for hydrology-related disaster risk reduction worldwide. The main novelty of the project relies on the optimal use of the complementarity between SEO-derived soil wetness and flood observations. We expect the former to help controlling model predictions mainly during low flows while the latter should allow keeping model predictions on track especially when rivers overflow their banks. The main working hypothesis of the project is that the joint use of these two complementary datasets enables sufficient constraining of model predictions in the absence of any in situ observations.The related innovative research and development activities will be focused on (i) the enhancement of automatic algorithms for the production of flood maps (flood extent and water levels) from long-term SEO archives, (ii) the development of a large scale flood prediction chain based on the loose coupling of a distributed conceptual hydrological model with a hydraulic model as well as (iii) the development of efficient and robust algorithms for the joint assimilation of long-term SEO archives of soil moisture and flood maps into the flood prediction chain. In the end the integration of the scientific algorithms shall lead to a significant reduction of predictive uncertainty, especially in poorly gauged regions.