Abstract:This paper highlights the importance of taking into account the tail dependence in the context of bivariate frequency analysis based on copulas. Three nonparametric estimators of the tail-dependence coefficient are compared by simulations with seven families of copulas. We choose the two estimators most adapted to a bivariate frequency analysis of the annual maximum flows and the corresponding flow hydrograph volumes of the Loire River ͑France͒. In this example, the bivariate return period and the conditional … Show more
“…These works provide quite extensive simulation studies devised to assess the reliability of a set of parametric and nonparametric k U estimators as well as several conclusive warnings about their use in practical analyses. Poulin et al (2007) made an ad hoc Monte Carlo experiment to choose the most appropriate estimator for a specific case study, retaining the Coles-Heffernan-Tawn k CHT U estimator (Coles et al 1999) and Capéraà-Fougères-Genest k CFG U estimator (Capéraà et al 1997). Poulin et al (2007) also stressed the caveats previously reported by Schmidt (2003) and Frahm et al (2005).…”
Section: Introductionmentioning
confidence: 96%
“…Poulin et al (2007) made an ad hoc Monte Carlo experiment to choose the most appropriate estimator for a specific case study, retaining the Coles-Heffernan-Tawn k CHT U estimator (Coles et al 1999) and Capéraà-Fougères-Genest k CFG U estimator (Capéraà et al 1997). Poulin et al (2007) also stressed the caveats previously reported by Schmidt (2003) and Frahm et al (2005). Serinaldi (2008) exploited the relationship between k U and Kendall correlation coefficient s K to build a diagnostic plot useful for the model selection.…”
Section: Introductionmentioning
confidence: 99%
“…The use of k U in hydrological analyses can be dated back to the works of Poulin et al (2007) and Serinaldi (2008) concerning bivariate frequency analyses of the annual maximum flows and the corresponding flow hydrograph volumes, and the pairwise analysis of rainfall data at multiple locations, respectively. As for the introduction of copulas in hydrology, the theoretical apparatus used in those works was essentially borrowed from econometric literature, namely, Schmidt (2003), Frahm et al (2005) and Schmidt and Stadtmüller (2006).…”
Section: Introductionmentioning
confidence: 99%
“…In order to shed light on such a matter, we recall that the recommendations provided by Schmidt (2003), Schmidt and Stadtmüller (2006), Poulin et al (2007), and Frahm et al (2005) are based on simulation experiments in which a unique value of the overall correlation is used, namely s K ¼ 1=3 in Schmidt (2003) and Frahm et al (2005), Pearson correlation q P ¼ 0:25 in Schmidt and Stadtmüller (2006) and s K ¼ 0:51 in Poulin et al (2007). Based on these simulation settings, these studies conclude that:…”
The simultaneous occurrence of extreme events, such as simultaneous storms and floods at different locations, has a serious impact on risk assessment and mitigation strategies. The joint occurrence of extreme events can be measured by the so-called upper tail dependence (UTD) coefficient k U . In this study, we reconsider the properties of the most popular k U estimators and show that their strong bias and uncertainty make most of the empirical results reported in the hydrological literature questionable. In order to overcome the limits of k U analysis, we test several alternative tools such as a pool of formal statistical tests devised for recognizing upper tail independence and graphical diagnostics based on binary correlation and binary entropy. The reliability of all the methods is preliminarily checked by Monte Carlo experiments. Statistical tests and graphical diagnostics are therefore applied to three different rainfall data sets that allow us to explore the properties of the spatial dependence structure of rainfall extremes over a wide range of spatio-temporal scales ranging from 30 min and 1 km to 30 days and %3000 km. Results highlight that (1) classical estimators provide non zero tail dependence even for cases where it should be zero; (2) formal tests and binary correlation highlight that the pairwise spatial dependence structure can be weaker than Gaussian, thus excluding UTD calculated in a pairwise manner; (3) the binary entropy computed on triples of locations shows that the pairwise UTD is not enough to explain the spatial dependence structure of extreme rainfall, whose complexity becomes evident only after resorting to higher order correlation measures. The results concerning the bias and uncertainty of k U estimators are fully general and suggest avoiding their use especially for the short time series usually available in hydrology.
“…These works provide quite extensive simulation studies devised to assess the reliability of a set of parametric and nonparametric k U estimators as well as several conclusive warnings about their use in practical analyses. Poulin et al (2007) made an ad hoc Monte Carlo experiment to choose the most appropriate estimator for a specific case study, retaining the Coles-Heffernan-Tawn k CHT U estimator (Coles et al 1999) and Capéraà-Fougères-Genest k CFG U estimator (Capéraà et al 1997). Poulin et al (2007) also stressed the caveats previously reported by Schmidt (2003) and Frahm et al (2005).…”
Section: Introductionmentioning
confidence: 96%
“…Poulin et al (2007) made an ad hoc Monte Carlo experiment to choose the most appropriate estimator for a specific case study, retaining the Coles-Heffernan-Tawn k CHT U estimator (Coles et al 1999) and Capéraà-Fougères-Genest k CFG U estimator (Capéraà et al 1997). Poulin et al (2007) also stressed the caveats previously reported by Schmidt (2003) and Frahm et al (2005). Serinaldi (2008) exploited the relationship between k U and Kendall correlation coefficient s K to build a diagnostic plot useful for the model selection.…”
Section: Introductionmentioning
confidence: 99%
“…The use of k U in hydrological analyses can be dated back to the works of Poulin et al (2007) and Serinaldi (2008) concerning bivariate frequency analyses of the annual maximum flows and the corresponding flow hydrograph volumes, and the pairwise analysis of rainfall data at multiple locations, respectively. As for the introduction of copulas in hydrology, the theoretical apparatus used in those works was essentially borrowed from econometric literature, namely, Schmidt (2003), Frahm et al (2005) and Schmidt and Stadtmüller (2006).…”
Section: Introductionmentioning
confidence: 99%
“…In order to shed light on such a matter, we recall that the recommendations provided by Schmidt (2003), Schmidt and Stadtmüller (2006), Poulin et al (2007), and Frahm et al (2005) are based on simulation experiments in which a unique value of the overall correlation is used, namely s K ¼ 1=3 in Schmidt (2003) and Frahm et al (2005), Pearson correlation q P ¼ 0:25 in Schmidt and Stadtmüller (2006) and s K ¼ 0:51 in Poulin et al (2007). Based on these simulation settings, these studies conclude that:…”
The simultaneous occurrence of extreme events, such as simultaneous storms and floods at different locations, has a serious impact on risk assessment and mitigation strategies. The joint occurrence of extreme events can be measured by the so-called upper tail dependence (UTD) coefficient k U . In this study, we reconsider the properties of the most popular k U estimators and show that their strong bias and uncertainty make most of the empirical results reported in the hydrological literature questionable. In order to overcome the limits of k U analysis, we test several alternative tools such as a pool of formal statistical tests devised for recognizing upper tail independence and graphical diagnostics based on binary correlation and binary entropy. The reliability of all the methods is preliminarily checked by Monte Carlo experiments. Statistical tests and graphical diagnostics are therefore applied to three different rainfall data sets that allow us to explore the properties of the spatial dependence structure of rainfall extremes over a wide range of spatio-temporal scales ranging from 30 min and 1 km to 30 days and %3000 km. Results highlight that (1) classical estimators provide non zero tail dependence even for cases where it should be zero; (2) formal tests and binary correlation highlight that the pairwise spatial dependence structure can be weaker than Gaussian, thus excluding UTD calculated in a pairwise manner; (3) the binary entropy computed on triples of locations shows that the pairwise UTD is not enough to explain the spatial dependence structure of extreme rainfall, whose complexity becomes evident only after resorting to higher order correlation measures. The results concerning the bias and uncertainty of k U estimators are fully general and suggest avoiding their use especially for the short time series usually available in hydrology.
“…In Malevergne and Sornette [2003] the authors showed that the usual Gaussian type of dependence is not appropriate to estimate financial risks, since it underestimates the dependence of extremes. There are several hydrological applications, most of them related to the analysis of extremes [Salvadori et al, 2007;Favre et al, 2004;Poulin et al, 2007].…”
[1] In many applications of geostatistical methods, the dependence structure of the investigated parameter is described solely with the variogram or covariance functions, which are susceptible to measurement anomalies and implies the assumption of Gaussian dependence. Moreover the kriging variance respects only observation density, data geometry and the variogram model. To address these problems, we borrow the idea from copulas, to depict the dependence structure without the influence of the marginal distribution. The methodology and basic hypotheses for application of copulas as geostatistical methods are discussed and the Gaussian copula as well as a non-Gaussian copula are used in this paper. Copula parameters are estimated using a division of the observations into multipoint subsets and a subsequent maximization of the corresponding likelihood function. The interpolation is carried out with two different copulas, where the expected and median values are calculated from the copulas conditioned with the nearby observations. The full conditional copulas provide the estimation distributions for the unobserved locations and can be used to define confidence intervals which depend on both the observation geometry and values. Observations of a large scale groundwater quality measurement network in Baden-Württemberg are used to demonstrate the methodology. Five groundwater quality parameters: chloride, nitrate, pH, sulfate and dissolved oxygen are investigated. All five parameters show non-Gaussian dependence. The copula-based interpolation results of the five parameters are compared to the results of conventional ordinary and indicator kriging. Different statistical measures including mean squared error, relative differences and probability scores are used to compare cross validation and split sampling results of the interpolation methods. The non-Gaussian copulas give better results than the geostatistical interpolations. Validation of the confidence intervals shows that they are more realistic than the estimation variances obtained by ordinary kriging.
During the recent years, there has been an increasing interest in multivariate frequency analysis of hydrological variables, e.g., those describing extreme events like rainfall, floods, or droughts. The multivariate analysis provides a better understanding of the phenomena under investigation and an additional insight about the interrelationships between the different variables (e.g., peak, volume, and duration of the flood), exploiting the complete structure of the problem and making a full use of the available data. However, while the developments on multivariate analysis of hydrological data have produced a large body of literature, a clear assessment of the use of these methods in the design and risk assessment of hydraulic structures is still a matter of debate. In the present work, we illustrate a general, structure-based framework for the design and/or risk assessment of hydraulic structures in a bivariate environment; we also compare it to recently proposed methods which are based on the assumption of hydrological design events (as is customary in the univariate context). For illustration purposes, both the structure-based and the design event-based approaches are applied to the design of an idealized structure, thus exploring the differences among the methods as function of the parameters involved. Our work highlights that the return period of structure failure in a multivariate environment strictly depends on the particular structure under design, and in most cases, the design of a hydraulic structure cannot be based on a single, hydrological multivariate design event. This acts as a warning for practitioners against the use of design methods based on single hydrological events, as usually done in the context of univariate hydrology, thus neglecting the interplay between the structure and the hydrological loads acting on it.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.