Handling the stochastic uncertainty of flood statistics in regionalization approaches

Regional flood frequency analyses are highly affected by the number of gauged catchments and lengths of observation periods at the individual gauges. Therefore, information is unevenly distributed in the region of interest. In particular, the occurrence of single extreme floods that are observed or not observed at some gauges, depending on the observation period, may have a large impact on the regionalization. We evaluated the impact of the sample length on the regionalization of a type-specific statistical mixture-model. In a case study it is shown how the regionalization error can be reduced by more than 50% if the sample sizes increase. We compare three approaches to handling this stochastic uncertainty in regionalization. The alignment of the statistical distribution parameters to consider the impact of extreme floods proved to be most beneficial when aiming to obtain homogeneous regionalized flood quantiles for hydrologically similar regions in a region with heterogeneous observation periods.