The contributions of “human activities (such as increasing human settlements) and climate change to an increase in the likelihood and adverse impacts of flood events” are clearly stated in the European Flood Directive (EU, 2007). These changes must be taken into account when assessing the flood hazard, from events with a high probability (return period 10 1000-year). For each range of period, the flood extent, the water depths and the flow velocities, must be estimated in the areas with potential significant flood risks and shown on flood hazard maps. Also, as stated in report n°13 of the French Nuclear Safety Authority (ASN) on the protection of nuclear installations against flooding, periods T up to 10000-year should be considered. However, while flood simulations for low return period events have become a common task, the prediction of the very high flows is not an easy task especially because of a lack of field data, but also because the flow processes are mostly controlled by the floodplains land occupation, which is increasingly inhomogeneous and complex with an increasing period T. The inhomogeneity can be observed in both lateral and longitudinal directions and the complexity is in large part due to the large variation of the confinement of the roughness elements with flow depth. As a result, the “hydraulic signature” of the roughness elements, i.e. the interaction flow/element, strongly varies with T. When simulating floods of high or medium probability, the classical numerical approaches (1D, 2D models) are calibrated and validated against observed field data. Field practitioners usually calibrate the models for the highest observed events for which data is available (typically T ~ 100-year) and use the same calibration parameters for the simulation of events with higher return periods T, as no additional data is available (see e.g., the ASN report). In the case of extreme events with an increasing inundated land area, of interspersed families of roughness elements and of a spatially varied confinement of these different types of elements, this practice is highly questionable and is very likely to lead to inaccurate flood hazard assessment. As a result, when dealing with extreme flood flows, engineers and scientists face (i) complex and still largely unexplored physical processes, and (ii) a lack of information regarding simulation validity as numerical modelling cannot be validated against field data. The project aims at improving the flood hazard assessment in floodplains during extreme events in: 1) investigating in laboratory the hydrodynamic structures specifically associated with extreme flood flows for various land occupations and flow discharge magnitudes; 2) assessing if the existing numerical modelling practices commonly used for T up to 100-year are still valid for T > 1000-year. The first task will particularly focus on the effects of lateral and longitudinal roughness transitions, of the confinement degree and the spatial distribution of the roughness elements. The experiments will be carried out in five flumes, under uniform or non-uniform flow conditions, relying on the state-of-the-art measurements on both large and small scales (i.e. the river reach scale or the roughness elements scale). For the second task, the previous experimental database will be compared to simulations performed with industrial and research codes (1D to 3D modelling). The classical methods to model flow resistance with an increasing complexity will be assessed and improved to capture the physics for the entire span of studied flow rates. The codes and methods will then be applied to the floods at Besançon, France. Events with T~100, 1000 and 10000-year will be simulated with both classical and improved methods, and the discrepancies will be calculated. This project will permit to quantify uncertainties on water levels and velocities computed for extreme events.