Summary: | International audience Hazards need to be characterised both in temporal and spatial dimensions. When dealing with floods, the availability of data is limited in both extents. On the one hand, the observation periods are generally too short to fully appreciate the shape of the frequency-intensity curve. This problem is exacerbated by the non-linearity of the rainfall-runoff relationship which makes difficult to extrapolate this curve towards rare events. On the other hand, the number of river monitoring stations is limited and most catchments are actually ungauged, especially among the smallest ones. In France, the SHYREG method has been developed to address those issues. It couples a fully regionalised stochastic rainfall generator, and a simple event-based conceptual rainfall-runoff model. The rainfall generator is able to simulate realistic long time-series of rainfall events at any point of the French territory thanks to an extensive characterisation of the rainfall behaviour. The hydrological model solely necessitates one parameter to be calibrated against at-site discharge data to transform rainfall events into flood events. The regionalisation of this unique parameter would allow making flood quantile estimation throughout France for diverse return periods and time-steps. The at-site version of SHYREG appears to be able to deal with the temporal limit of flood data by providing robust and reliable flood quantiles [Arnaud et al., 2015]. Here we present a fully regionalised version of SHYREG which aims to overcome the spatial limit of flow data. The regionalisation is performed by transferring the parameter calibrated in gauged site to ungauged site relying on both physical characteristics and locations of catchments. The evaluation of the performance in ungauged sites is realised through cross-validation procedure. The results suggest that SHYREG extreme flood quantiles are less sensible to regionalisation than other regional flood frequency analysis techniques. Nevertheless, when providing a complete ...
|