Estimation of Extreme Floods Using a Statistical and Conceptual Model of the Hydrological Response

The robust estimation of flood peak discharge values is critical for designing mitigation measures and increasing preparedness to natural hazards. Traditional flood estimation methods are, however, severely limited by data series shorter than the return period of interest, as they only use annual maxima or a few values above a high threshold. Here we couple two recent advances in flood estimation from short data samples, namely the Metastatistical Extreme Value Distribution (MEVD) and a conceptual model of flood generation processes, the Physically based Extreme Value (PhEV) distribution of river flows. The result is a methodology, defined through a few hydrologic attributes describing runoff generation and catchment response, to estimate extreme discharge in poorly gauged basins, which we test on data from 178 catchments in Germany. We find that extremes are best estimated when PhEV runoff-generation parameters are set using the long-term mean discharge and precipitation depth, while catchment response parameters are estimated by statistical fitting to observed peak streamflow values. This estimation method interestingly outperforms a methodology in which all parameters are tuned to optimize the reproduction of the statistics of observed peak streamflow values. Our results show that the median relative error associated with MEVD-PhEV, across the large data set explored, consistently remains between −50% and +50%. Hence, MEVD-PhEV yields useful estimates of extreme flows with limited observational information and with no need of preselecting a suitable distribution for ordinary peak discharge values, a step that is substituted by the inclusion of catchment hydrologic information.