Dependence of flood peaks and volumes in modeled discharge time series: Effect of different uncertainty sources

Abstract

Flood estimates needed for designing efficient and cost-effective flood protection structures are usually derived using observed peak discharges. This approach neglects, firstly, that floods are characterized not only by peak discharge but also by flood volume, and, secondly, that these characteristics are subject to modifications under climate and land use changes. Bivariate flood frequency analysis based on simulated discharge time series makes it possible to consider both flood peak and flood volume in design flood estimation. Further, this approach considers changes in discharge characteristics by using discharge series generated from climate time series used as an input for a hydrological model. Such series are usually not available at an hourly resolution but at a certain aggregation level (e.g. 24 h) and might not perfectly represent observed precipitation distributions. In this study, we therefore investigate how the aggregation and distribution of precipitation series and discharge distribution affect flood peaks and volumes and their dependence. We propose a framework for assessing the uncertainty in bivariate design flood estimates that is caused by different factors in the modeling chain, which consists of precipitation-discharge modeling, flood event sampling, and bivariate flood frequency analysis. The uncertainty sources addressed are precipitation aggregation and distribution, parameter and model uncertainty, and discharge resolution. Our results show that all of these uncertainty sources are relevant for design flood estimation and that the importance of the individual uncertainty sources is catchment dependent. Our results also demonstrate that substantial uncertainty is introduced already in the first step of the model chain because commonly used calibration procedures do not take into account the reproduction of flood volumes. Researchers should be aware of such deficiencies when performing bivariate flood frequency analysis on modeled discharge time series and should aim to tailor model calibration procedures to the problem at hand.