Search

National audience; Cette communication présente les résultats d’un exercice d’inter-comparaison des méthodes d’estimation des crues en France. Les méthodes comparées incluent l’estimation d’une loi de Gumbel ou d’une loi GEV, ainsi que la méthode de simulation SHYREG. Chacune de ces méthodes est déclinée en version locale, régionale ou mixte locale + régionale.

National audience; Présentation de la méthodologie de comparaison utilisée dans ExtraFlo

National audience; Résultats du projet ExtraFlo sur la comparaison des méthodes d'estimation des crues extrêmes en France.

International audience; An abundance of methods have been developed over the years to perform the frequency analysis (FA) of extreme environmental variable. Although numerous comparisons between these methods have been implemented, no general comparison framework has been agreed upon so far. The objectives of this presentation are: (i) to describe the foundation of a general framework devoted to the data-based comparison of FA implementations; (ii) to illustrate its application based on an extensive dataset of French gauging stations. The comparison framework is based on the following general principles: (i) emphasis is put on the predictive ability of competing FA implementations, rather than their sole descriptive ability measured by some goodness-of-fit criterion; (ii) predictive ability is quantified by means of reliability indices, describing the consistency between validation data (not used for calibration) and FA predictions; (iii) stability is also quantified, i.e. the ability of a FA implementation to yield similar estimates when calibration data change; (iv) the necessity to subject uncertainty estimates to the same scrutiny as point-estimates is recognized, and a practical approach based on the use of the predictive distribution is proposed for this purpose. This framework is then applied to a case study involving more than one thousand gauging stations in France, where several FA implementations are compared. These implementations correspond to the local, regional and local-regional estimation of Gumbel and Generalized Extreme Value distributions, completed with a “derived distribution” approach based on a rainfall simulator coupled with a rainfall-runoff model. Results show that reliability and stability indices are able to reveal marked difference between FA implementations. Moreover, the case study also confirms that using the predictive distribution to indirectly scrutinize uncertainty estimates is a viable approach, with distinct FA implementations showing marked differences in the reliability of their uncertainty estimates. The proposed comparison framework therefore constitutes a valuable tool to compare the predictive reliability of competing FA implementations, along with the reliability of their uncertainty estimates.

The French research project ExtraFlo (EXTreme Rainfall and FLOod estimation) aims to carry out an intercomparison of the main methods used in France for estimating extreme values of rainfall and floods, to obtain a better grasp of their respective fields of application. One work package is devoted to at-site flood frequency analysis, based either on extreme value distributions (Gumbel, GEV, Pearson III, Log Pearson III) or a mixed model Shypre (coupled rainfall generator and rainfall-runoff model). A general statistical framework has been developed to quantify model performances in term of reliability and stability (robustness). It has been applied on a set of 519 daily discharge series with at least 37 years of record. A first conclusion is that the extreme value distribution with three parameters remain difficult to fit at a local scale. The limited amount of at-site information induces a large uncertainty of the shape parameter. One suggestion is to apply a predictive computation, which improves the reliability of the estimates. A second result is that the Shypre model gives more robust estimates, with a similar reliability than the central estimate of extreme value distributions. The choice of a method of parameter estimation (moment, likelihood, L-moments, Bayesian) is of second order compared to the choice of a central or predictive distribution, or a simulation model., Le projet de recherche ExtraFlo a pour objectif de comparer les principales méthodes utilisées en France pour estimer des valeurs extrêmes de pluie et de crues. Une des actions de comparaison a porté sur les méthodes exploitant une série locale de données, à partir d'une distribution (Gumbel, GEV, Pearson III, Log Pearson III) ou d'un simulateur d'averses couplé à un modèle pluie-débit (SHYPRE). Un cadre d'analyse a permis de caractériser chaque méthode en terme de justesse et de robustesse. Les résultats obtenus sur un jeu de 519 stations avec au moins 37 ans de données montrent que la quantité limitée d'information disponible à une échelle locale ne permet pas de caler correctement le paramètre de forme d'une distribution. Une des alternative consiste à exploiter la distribution prédictive. L'approche par simulation SHYPRE donne de meilleurs résultats, à la fois en terme justesse et de robustesse. Le choix d'une méthode d'estimation des paramètres d'une distribution est du second ordre par rapport au choix entre distribution centrale ou prédictive, ou l'utilisation d'une méthode de simulation.

The French research project ExtraFlo (EXTreme Rainfall and FLOod estimation) aims to carry out an intercomparison of the main methods used in France for estimating extreme values of rainfall and floods, to obtain a better grasp of their respective fields of application. One work package is devoted to at-site flood frequency analysis, based either on extreme value distributions (Gumbel, GEV, Pearson III, Log Pearson III) or a mixed model Shypre (coupled rainfall generator and rainfall-runoff model). A general statistical framework has been developed to quantify model performances in term of reliability and stability (robustness). It has been applied on a set of 519 daily discharge series with at least 37 years of record. A first conclusion is that the extreme value distribution with three parameters remain difficult to fit at a local scale. The limited amount of at-site information induces a large uncertainty of the shape parameter. One suggestion is to apply a predictive computation, which improves the reliability of the estimates. A second result is that the Shypre model gives more robust estimates, with a similar reliability than the central estimate of extreme value distributions. The choice of a method of parameter estimation (moment, likelihood, L-moments, Bayesian) is of second order compared to the choice of a central or predictive distribution, or a simulation model., Le projet de recherche ExtraFlo a pour objectif de comparer les principales méthodes utilisées en France pour estimer des valeurs extrêmes de pluie et de crues. Une des actions de comparaison a porté sur les méthodes exploitant une série locale de données, à partir d'une distribution (Gumbel, GEV, Pearson III, Log Pearson III) ou d'un simulateur d'averses couplé à un modèle pluie-débit (SHYPRE). Un cadre d'analyse a permis de caractériser chaque méthode en terme de justesse et de robustesse. Les résultats obtenus sur un jeu de 519 stations avec au moins 37 ans de données montrent que la quantité limitée d'information disponible à une échelle locale ne permet pas de caler correctement le paramètre de forme d'une distribution. Une des alternative consiste à exploiter la distribution prédictive. L'approche par simulation SHYPRE donne de meilleurs résultats, à la fois en terme justesse et de robustesse. Le choix d'une méthode d'estimation des paramètres d'une distribution est du second ordre par rapport au choix entre distribution centrale ou prédictive, ou l'utilisation d'une méthode de simulation.

International audience; Frequency analysis (FA) is one of the cornerstones of hazard quantification and risk assessment. Its basic objective is to estimate the distribution of some environmental variable X, e.g. annual maximum of the areal rainfall over some catchment, annual maximum flood, etc. This distribution can be used to estimate the exceedance probability of a given value of X (often expressed in terms of return period), or alternatively, to estimate the p-quantile of X, i.e. the value having an exceedance probability equal to 1-p. The estimation of quantiles is of primary importance since they are used to design civil engineering structures (e.g. dams, reservoirs, bridges) or to map hazard-prone areas where restrictions may be enforced (e.g. building restrictions in flood zones). FA has been the subject of extensive research, yielding an abundance of approaches. In practice, FA users and practitioners may feel lost facing such an abundance of methods. Consequently, several initiatives aimed at assisting users in realizing their analyses using best-practice methods. In addition to these end-user-oriented guideline documents, a large number of comparisons between competing methods have been reported in the research literature. The French National research project EXTRAFLO aims to perform a thorough comparison between FA approaches currently used in France, based on an extensive dataset of long series of rainfall and runoff. This poster provides a detailed description of the methodology used to perform the comparison, and presents preliminary results of its application to a large runoff dataset. More precisely, the following topics are presented: Difference between descriptive and predictive ability, and the need to decompose the dataset into calibration / validation sub-samples The issue of scrutinizing uncertainty estimates is discussed, and a method based on the concept of predictive distribution is proposed in order to compare the reliability of competing uncertainty estimates. Reliability indices are derived in order to compare the performances of competing methods on an objective basis. This methodological framework is first applied to synthetic data to verify the ability of the comparison framework to distinguish between good and bad FA approaches. Preliminary results based on a large runoff dataset are then discussed.