Author(s)

Y. HAMDI, C-M. DULUC, L. BARDET & V. REBOUR

Abstract

Nuclear power plants in France are designed for low probabilities of failure. Nevertheless, some exceptional surges, considered as outliers, are not properly addressed by statistical models. Regional information may be used to mitigate the paucity of data and the influence of outliers. A regional frequency analysis (RFA) model assumes a homogenous behavior of the variable of interest at a regional scale. The first stage in an RFA is the delineation of homogeneous regions. We propose herein a new approach to form a homogenous region centered on a target site and using the spatial extremal dependence between observations (the spatial extremogram) to measure the neighborhood between sites. Skew surge data set collated at a total of 19 sites located on the French coast (Atlantic and English Channel) were used as the case study for this contribution (with La Rochelle as a target site). Once a physically plausible region of interest is defined, a related issue regards the statistical homogeneity of the region of interest. The L-moment-based homogeneity tests of Hosking and Wallis, widely used in hydrology, are used in this paper. The principle of the extremogram allows us to form a physically and statistically homogenous group of sites centered on a target site. It also allows to overcome the problem of the so-called “border effect” and these are two key original points of the developed concept.

Keywords

extreme surges, homogenous region, regional frequency analysis, spatial extremogram, target site