New Approach to Identifying Critical Initial Conditions for Extreme Flood Simulations in a Semicontinuous Simulation Framework

Abstract

Extreme flood simulation with synthetic extreme precipitation events raises unavoidable questions about the choice of initial conditions. State-of-the-art extreme flood estimation frameworks propose to address these questions with the help of semicontinuous modeling and reanalysis of simulated state variables. In this context, the present work proposes a new method for the selection of initial conditions for extreme flood simulation. The method is based on generating sets of initial conditions from the matrix of state variables corresponding to a long simulation run of the selected hydrological model. Two sets of initial conditions are obtained: a deterministic set composed of selected state variable quantiles and a stochastic set composed of state variable vectors randomly drawn from the complete state variable matrix. The extreme flood simulations corresponding to both sets are compared in detail, and the stochastic simulations are used in a sensitivity analysis to identify the dominant state variables and possible interactions. The aim hereby is to provide a tool to analyze the role of initial conditions and the importance to account for state variable interactions in extreme flood estimation. The proposed method is applied to probable maximum flood estimation for the Swiss Mattmark Dam catchment with a semilumped hydrological model. The obtained results for this case study show that for high flood peak quantiles, the initial soil saturation is dominating other state variables, and deterministic initial conditions are sufficient to generate extreme floods.

Downsizing parameter ensembles for simulations of rare floods

“…The quantification of extreme floods and associated return periods remains a key issue for flood hazard management (Kochanek et al, 2014). Extreme-value analysis was largely developed in this field for the estimation of flood return periods (Katz et al, 2002); corresponding methods have been recently extended to bivariate approaches that assign probabilities jointly to flood peaks and flood volumes (Favre et al, 2004;De Michele et al, 2005;Brunner et al, 2016) and to trivariate approaches to assign probabilities jointly to flood peaks, volume and duration (Zhang and Singh Vijay, 2007); for a review of this field, see the work of Graler et al (2013).…”

“…Most modern applications, however, require the estimation of not only extreme peak flow, associated flood volumes and duration but also of hydrograph shapes, in particular in the context of reservoir design or for safety checks of hydraulic infrastructure (Kochanek et al, 2014;Gaál et al, 2015;Zeimetz et al, 2018). The key is thus the construction of design hydrographs with different shapes, peak flows and volumes, with a corresponding probability of occurrence.…”

“…Modern extensions of this approach, however, use continuous hydrological modelling for design flood estimation to generate either (i) a range of initial conditions for use in combination with design or randomly drawn storms (Paquet et al, 2013;Zeimetz et al, 2018) or (ii) long discharge time series from long observed-precipitation records or from synthetic precipitation time series obtained with a weather generator (Calver and Lamb, 1995;Cameron et al, 2000;Blazkova and Beven, 2004;Hoes and Nelen, 2005;Winter et al, 2019). The above approach (ii) is computationally intensive, especially if long time series are to be simulated using ensembles of hydrological parameter sets or if very high return periods (> 1000 years) have to be robustly estimated.…”

Clustering model responses in the frequency space for improved simulation‐based flood risk studies: The role of a cluster number

“…Within such a PFRA chain, continuous hydrologic models are often employed to simulate a possible hydrological response to a considered input and to estimate the extent of possible floods. In place of these hydrologic models, conceptual models are often employed as they are very fast in simulation and thus enable the use of long time series, many meteorological scenarios or different states of initial conditions (Blazkova & Beven, 2004; Cameron et al, 1999; Haberlandt & Radtke, 2014; Paquet et al, 2013; Winter et al, 2019; Zeimetz et al, 2018). They thus appear to be more suitable for PRFA than physically based models that, although do not need a direct calibration (Cunha et al, 2011), are slow in simulation.…”

Hydrological Risk Analysis of Dams: The Influence of Initial Reservoir Level Conditions

“…Moreover, human actions also provide additional sources of uncertainty to the analysis. By the use of probabilistic approaches, variability of the variables involved can be assessed [9][10][11][12][13]. One such variable related to human actions is the variability of the initial reservoir level, due to its connection to the operation of reservoirs.…”