Search

• Two-component mixture distributions (TCMD) are used to model annual maximum floods. • Usage of flood timescale (FT) to classify distinct flood generating mechanisms. • Snowmelt-induced floods and rainfall-induced floods are identified in Norway. • The FT -based TCMD reduces uncertainty of design flood by 40% than traditional TCMD. The conventional flood frequency analysis typically assumes the annual maximum flood series (AMFS) result from a homogeneous flood population. However, actually AMFS are frequently generated by distinct flood generation mechanisms (FGMs), which are controlled by the interaction between different meteorological triggers (e.g., thunderstorms, typhoon, snowmelt) and properties of underlying surface (e.g., antecedent soil moisture and land-cover types). To consider the possibility of two FGMs in flood frequency analysis, researchers often use the two-component mixture distributions (TCMD) without explicitly linking each component distribution to a particular FGM. To improve the mixture distribution modeling in seasonally snow covered regions, an index called flood timescale (FT), defined as the ratio of the flood volume to peak value and chosen to reflect the relevant FGM, is employed to classify each flood into one of two types, i.e., the snowmelt-induced long-duration floods and the rainfall-induced short-duration floods, thus identifying the weighting coefficient of each component distribution beforehand. In applying the FT -based TCMD to model the AMFS of 34 watersheds in Norway, ten types of mixture distributions are considered. The design floods and associated confidence intervals are calculated using parametric bootstrap method. The results indicate that the FT -based TCMD model reduces the uncertainty in the estimation of design floods for high return periods by up to 40% with respect to the traditional TCMD. The improved predictive ability of the FT -based TCMD model is attributed to its explicit recognition of distinct generation mechanisms of floods, thereby being able to identify the weighting coefficient and FGM of each component distribution without optimization. [ABSTRACT FROM AUTHOR]

Design flood values give estimates of flood magnitude within a given return period and are essential to making adaptive decisions around land use planning, infrastructure design, and disaster mitigation. Many hydrologic applications where flood retention is important, e.g. floodplain management and reservoir design, need design flood values for different durations. Flood–Duration–Frequency (QDF) models extend the standard statistical flood frequency analysis framework to multiple flood durations and are analogous to intensity–duration–frequency models for precipitation. Implementations of QDF models commonly assume simple scaling, where only the magnitude of the index flood is assumed to change with duration, despite empirical analyses showing a more complex dependence structure. We propose a multiscaling extension to existing QDF models where the magnitude of the index flood and the slope of the growth curve may scale independently with duration. In an application to 12 locations in Norway, we assess how three different QDF models capture relationships between floods of different duration. Incorporating duration dependency independently in both the index flood and the growth curve (extended QDF model) improves modeling of both short-duration events and events with long return periods. This model extension further expands the models' ability to simultaneously model a wide range of durations. As measured by the integrated quadratic distance, the extended QDF model performs better than the original QDF model in 83% of the out of sample subdaily durations studied. Additionally, we find that the choice of durations used to fit QDF models is a highly influential aspect of the modeling process. • Flood–Duration–Frequency (QDF) models produce flood statistics at multiple durations. • We extend QDF models so that different quantiles can have different slopes. • This improves estimates and allows for a wider range of durations to be modeled. • Additionally, we introduce a Bayesian framework for QDF estimation. [ABSTRACT FROM AUTHOR]