Relevance of an at-site flood frequency analysis method for extreme events based on stochastic simulation of hourly rainfall

Abstract

Extreme events are rarely observed, so their analysis is generally based on observations of more frequent values. The relevance of flood frequency analysis (FFA) method depends on its capability to estimate the frequency of extreme values with reasonable accuracy using extrapolation. A FFA method based on stochastic simulation of flood event is assessed based on its reliability and stability. For such an assessment, different training/testing decompositions are performed for a set of data from more than 1,000 gauging stations. We showed that the method enables relevant "predictive" estimates, e.g. by assigning correct return periods to the record values that are systematically absent in calibration data sets. The model is also highly stable visa -vis the sampling. This characteristic is linked to the use of regional statistical rainfall data and a simple rainfall-runoff model that requires calibrating only one parameter. Résumé: Les événements extrêmes sont par nature rarement observés, c'est pourquoi leur estimation est généralement basée sur l'observation de valeurs plus courantes. La pertinence d'une méthode de prédétermination des événements extrêmes dépend donc de sa capacité à raisonnablement extrapoler les distributions de fréquences vers les valeurs extrêmes. Dans cette étude, une méthode de prédétermination de crues basée sur la simulation de scénarios de pluies horaires, est évaluée sur sa capacité à produire des estimations justes et stables. Cette évaluation s'appuie sur différents tests d'échantillonnage sur les périodes de calage et de validation, appliqués sur un jeu de données conséquent (plus de 1000 stations). Nous montrons que la méthode est capable de fournir une estimation pertinente sur les événements extrêmes bien que ceux-ci soit systématiquement ôtés de la période de calage. La méthode montre aussi une grande stabilité face à l'échantillonnage. Cette caractéristique est liée à l'utilisation d'une information statistique régionale sur la pluie et à la simplicité de la modélisation hydrologique paramétrée par un seul paramètre.

Comparison of Flood Frequency Analysis Methods for Ungauged Catchments in France

“…In the present case, for a long return period, the low dependence between the regionalised SHYREG extreme quantiles and calibration discharge data provide quantile estimates relatively close to that obtained in a gauged configuration, i.e., quantiles that can be and have already been validated [18,45].…”

“…Most of the time, operational FFAs are implemented by fitting a statistical distribution to local data in gauged sites or by estimating flood quantiles through empirical relationships called "rational" approaches [44]. Recent studies implemented a data-based comparison of diverse FFA methods used in France [9] and concluded that methods using regional data or exploiting the rainfall information should be preferred to purely at-site statistical data [18,45]. Nevertheless, the transfer of this kind of FFA method to ungauged catchments has remained limited with only one regionalisation method tested and the performance was insufficient to draw general conclusions [45].…”

“…Different FFA methods were applied and associated with diverse regionalisation techniques. The FFA methods were an at-site Gumbel distribution (also called GEV type I), an at-site GEV distribution fitted with a Bayesian process using a regional a priori, a regional GEV distribution fitted on pooled data (the index flood method) and a process-based simulation approach (the SHYREG method [18,46,47]). To be as exhaustive as possible, different regionalisation schemes were associated with each of the FFA approaches.…”

“…In this way the non-linearity of rainfall-runoff transformation can be explicitly taken into account. For instance, a rainfall generator can be coupled to a rainfall-runoff model in order to simulate long flow time-series from which flood quantiles can be extracted [18,19].…”

“…The SHYREG flood quantiles (in gauged sites) were evaluated in previous studies [18,45], which demonstrated the high stability of the approach regarding calibration data and better accuracy than at-site distribution fitting, especially regarding high return period quantiles.…”