Generation of sets of synthetic events by multivariate stratified sampling

Joris Blanckaert, Gert Leysen, Fernando Pereira, Jan Swings, Tim Franken, Cedric Gullentops, Zhen Fang

Tuesday 30 june 2015

9:30 - 9:45h at North America (level 0)

Themes: (T) Extreme events, natural variability and climate change, (ST) Hydrological extremes: floods and droughts

Parallel session: 4I. Extreme events – Flood Drought

In order to develop optimized flood risk management plans for the Upper Scheldt and Lys/Leie basins, two major French/Belgian rivers with a confluence in the city of Ghent, a full probabilistic risk approach has been applied. Risk analysis requires considering the entire frequency domain of flood consequences. Hence statistical extrapolation is necessary. This was done by generating 2.500 extreme synthetic events for the river system. First, a continuous simulation of 60 years time series of recorded regional rainfall has been run through lumped hydrological models, simulating soil saturation conditions and yielding in joint occurrences of rainfall-runoff discharges at the different inflow locations. Next, for each catchment an extreme value distribution is fitted to the runoff discharges. Joint occurrences of extreme events in the Upper Scheldt and the Lys are taken into account by a bivariate extreme value copula. The bivariate discharge domain of the copula is then stratified in 100 classes. For each class synthetic hydrographs for both rivers are constructed, with a joint probability of occurrence that is calculated from the copula function. Synthetic hydrographs are derived from the combination of a standardized profile for baseflow and surface runoff, which is first shaped by a profile variation and then scaled with the discharge class. The standardized profiles are the mean expected baseflow and surface runoff hydrographs computed by normalizing all recorded extreme events. A probability distribution is fitted to the standardized events, yielding 5 profile classes through a second stratification. Next, the tidal water level variation at the downstream boundary, statistically modeled with a conditional distribution, was again stratified in 5 classes, eventually producing 100x5x5 = 2.500 sets of synthetic events. For each set the corresponding multivariate probability of occurrence is computed. All events are run through the hydrodynamic model of the river system. At this point, empirical frequency distributions can be drawn for flood depths in the floodplains. As a validation, these distributions are plotted against simulations of recorded extreme historical events, displaying good similarity. This methodology has the advantage that it determines a statistical distribution of consequences, rather than assigning frequencies to hydrodynamic boundary conditions.