Item

Abstract

The robust estimation of extreme flood magnitude in poorly observed or ungauged basins is of critical importance for designing mitigation measures, particularly in the presence of anthropogenic environmental change and accelerating climatic changes. Traditional methods for estimating flood extremes are strongly limited by the availability of sufficiently long timeseries as these are typically designed to use annual maxima or a few values above a high threshold. In the present work we use a recent statistical model, the Metastatistical Extreme Values (MEV) distribution, in combination with a conceptual model of flood generation processes, the Phisically-based Extreme Values (PhEV) distribution, to explore the possible estimation of high quantiles where few or no observations exist. The main novelty of the approach is the ability of extracting extreme streamflow values from "ordinary" streamflow peaks and to provide a characterization based on a limited and physically meaningful set of hydrological parameters. The proposed methodology aims to overcome limitations in data availability by exploiting the relatively large number of daily observations available even in short time series (as opposed to the low number of yearly maxima) and a few hydrological attributes of the catchment that may be "guessed" on the basis of limited information. A large-scale application on 178 catchments in Germany allows us to formulate a reliable calibration technique and to show its controllable estimation uncertainty: the median relative error computed on predicted extreme streamflow values is globally contained between -25% and +50%:.