Are stochastic point rainfall models able to preserve extreme flood statistics?

✍ Scribed by Niko E. C. Verhoest; S. Vandenberghe; P. Cabus; C. Onof; T. Meca-Figueras; S. Jameleddine

• Publisher: John Wiley and Sons
• Year: 2010
• Tongue: English
• Weight: 239 KB
• Volume: 24
• Category: Article
• ISSN: 0885-6087
• DOI: 10.1002/hyp.7867

✦ Synopsis

the European Directive on the Assessment and Management of Flood Risks entered into force requiring that all European Member States identify areas at risk of flooding. For these areas, flood risk maps need to be drawn by 2013, and flood risk management plans are to be developed by 2015, where flood risk refers to 'the combination of the probability of a flood event and of the potential adverse consequences of human health, the environment, cultural heritage and economic activity associated with a flood event ' (Directive 2007/60/EC, 2007

). An important aspect within the exercises to be carried out by each Member State will thus be the correct assessment of the return periods of extreme flood events.

A conventional approach, still frequently used, is to estimate flood discharge for individual precipitation events of particular interest (Institute of Hydrology, 1999), these being a significant historical rainfall event or a simulated design storm with given statistical properties (Wheater, 2002). The disadvantage of such an approach is that the antecedent wetness state of the catchment is generally not properly accounted for. This antecedent condition, however, determines the magnitude of Dunne runoff production and therefore the fluvial response to each precipitation event, which means that the return period of a flood event does not generally correspond to that of the causing storm event. This is also demonstrated in Figure 1, from which it can be seen that extreme discharge events often result from less extreme rainfall which must have fallen on a wet catchment. It follows that the conventional approach cannot be used to provide flood frequency maps needed as input for flood risk maps. Continuous discharge modelling using longterm rainfall records overcomes the problem mentioned earlier, and a frequency analysis of the obtained discharge time series enables the calculation of the requested probability of flood events. Unfortunately, this approach is limited by the length of the available rainfall time series which hampers the accurate modelling of very low frequency flood events. To circumvent this, one could consider using very long stochastically modelled rainfall time series for the continuous simulation of streamflow (Boughton and Droop, 2003).

Rainfall modelling can generally be classified into four categories (Onof et al., 2000): (1) meteorological models involving complex sets of differential equations representing the physical processes controlling precipitation and other weather variables; (2) stochastic multi-scale models describing the spatial evolution of the rainfall process independently of scale; (3) statistical models which can allow for the modelling of trends; and (4) stochastic process models which, although they make simple assumptions with respect to the physical processes, enable the description of the hierarchical structure of the rainfall process with a minimal set of model parameters. The latter stochastic process models, more specifically point-scale Bartlett-Lewis rectangular pulse models, are the focus of this commentary.