How can we estimate extreme flood risk ?

Comparison of methods will allow to obtain a better grasp of their respective fields of application

The assessment of an extreme rainfall or discharge value is very sensitive to the statistical method used. It is therefore aimed to carry out an intercomparison of the main methods used in France to define which approach is better to use, according to the available data sets, the hydrological characteristics of the studied site and the accepted level of uncertainty. A large set of estimation methods have been considered on a large benchmark data set.

A common strategy of intercomparison and validation

The main statistical methods (theory of extreme values on a local series, or at a regional scale or with historical data; use of simulation methods) have been tested on a large data set on several thousands of rainfall and discharge series. A cross-validation procedure, based on calibration and validation sub data sets, is able to detect the methods which have some statistical drawbacks such as a bias of estimation or a large sensitivity to outliers. A second set of methods have been compared, based on hydro-geomorphological or paleoflood studies. The former is able to locate the spatial extent of past flood prone areas, as the latter is based on the dating of flood deposits located in the flood plain or in some caves above the main river.

Main results

The ExtraFlo project showed that design estimate of extreme values based on the fitting of a distribution on a limited sample of maximum values is very inaccurate when only a few decades of observation are available. Some interesting alternatives are possible, using a regional approach, or historical data when available, or simulation methods, or sedimentological or geochemical signatures of flood deposits.

Scientific outputs

At the end of the project, seven paper have been published or are in press:

·        one reference paper by Renard et al. (2013), with the presentation of a general framework for a data-based comparison of frequency analysis methods ;

·        two papers by Garavaglia et al. (2010, 2011) on the validation of a rainfall model based on weather pattern sub-sampling ;

·        one paper by Soubeyroux et al. (2011) on the digitalization of old rainfall record on paper ;

·        three papers by Tramblay et al. (2011, 2012, 2013) on a non stationary frequency model for extreme rainfall.

The ExtraFlo project is an applied research project coordinated by Irstea (ex-Cemagref), with the collaboration of Meteo-France (Toulouse), CNRS (HydroSciences and GeoSciences Montpellier) and EDF/DTG (Grenoble), and some exchanges with water authorities (Cete Mediterranee and Dreal Midi-Pyrenees) and consultancy firms (Artelia, Electrabel). The 4 year project started in February 2009 and ended in January 2013. It was funded by the French national agency of research (ANR) with a financial support of 650 k€ from the RiskNat program (global cost of 2 millions of €).