Eléments d'analyse sur les crues éclair

Overview & Synopsis

Flash floods (i.e.floods produced by severe thunderstorms of limited areal extent) are certainly the most destructive natural hazard in France. They also give rise to numerous questions but no systematic studies have been conducted in the past on such events. The relatively poor hydrological knowledge about this type of flood is not in accordance with the risk that they represent or with the ambitious prevention policy objectives of the stake holders in France. In the first part of this document, a post flood investigation methodology is proposed and tested on five case studies. Besides the analysis of the available rainfall data and the estimations of peak discharges on the basis of high water marks, elements concerning the timing of the floods given by eyewitnesses were gathered. The first results obtained are hopeful and reveal some original aspects of the rainfall-runoff relationship during flash floods : 1) the watershed response to the intense storm bursts is late and relatively sudden, 2) a large amount of rainwater (150 to 200 millimetres in the present case studies) is retained on the catchments and does not contribute to the flood flow, 3) no significant effects of the land use type could be identified. The second part of the document, is devoted to the theoretical analysis of the flood peak distributions. The potential of the so called derived distribution methods consisting in coupling a stochastic rainfall simulator and a mathematical rainfall-runoff model is assessed. It appears that, due to the properties of the rainfall-runoff process, flood peak distributions (FPD) belong most probably to none of the three extreme value distribution types. The asymptotic behaviour of the FPD is controlled by the maximum rainfall intensities measured over a duration characteristic of the studied watershed. The shape of the FPD depends on the rainfall-runoff model used for the intermediate range of return periods - typically 10 to 106 years. Some realistic hypothesis, according to the results of the post flood investigations, lead to multi modal distributions, which local gradient on a Gumbel paper can be much larger than the Gradex of the rainfall intensities over a limited range of return periods.