Pietro Bernardara

Application of regional frequency analysis to the estimation of extreme storm surges

Traditionally, extreme value theory is applied to single-site series of surge observations in order to estimate the probability of occurrence of extreme events at that particular site. However, single-site analyses give uncertain estimation of extreme quantiles, mainly because of the limited duration of observation periods. In order to reduce this uncertainty, regional frequency analysis (RFA) approaches suggest collecting information not only from a single-site series but also from all (statistically) similar available series of observation. The use of RFA is widely increasing in geosciences, but few applications have been attempted yet for surge estimation. The aim of this study is to examine the applicability of RFA to extreme storm surges. The surge data observed at 18 French harbors, located on the Atlantic coast from the Spanish to Belgian borders, were collected. The series span a period of 30 years, on average, with the longest series going back to the 19th century. Stationary and independent samples of extreme surges (peaks over a given threshold) are extracted and their (statistical) homogeneity has been tested via heterogeneity and discordancy measures based on L moments. Homogeneous regions have been identified and, in order to merge information on frequency of occurrence of surges from all the sites, a surge index pooling method is defined. Finally, a regional frequency distribution has been estimated. The hypothesis and the applicability of RFA application are discussed, with some ideas for future developments in the research direction.

Dependence of model-based extreme flood estimation on the calibration period: case study of the Kamp River (Austria)

Abstract The Kamp River is a particularly interesting case study for testing flood frequency estimation methods, since it experienced a major flood in August 2002. Here, the Kamp catchment is studied in order to quantify the influence of such a remarkable flood event on the calibration of a rainfall–runoff model, in particular when it is used in a stochastic simulation method for flood estimation, by performing numerous rainfall–runoff model calibrations (based on split-sample and bootstrap tests). The results confirmed the usefulness of the multi-period and bootstrap testing schemes for identifying the dependence of model performance and flood estimates on the information contained in the calibration period. The August 2002 event appears to play a dominating role for the Kamp River, since the presence or absence of the event within the calibration sub-periods strongly influences the rainfall–runoff model calibration and the extreme flood estimations that are based on the calibrated model.