Applying POT methods to the Revised Joint Probability Method for determining extreme sea levels

Abstract

Newly exposed concepts of POT declustering (Bernardara et al., 2014) within the GPD-Poisson model are applied to the joint probability of tide and surge for determining extreme sea levels, as a variation of the Revised Joint Probability Method (RJPM, Tawn and Vassie, 1989). A mixture model is proposed for the meteorological residual (surge) component with a non-parametric (empirical) density for the bulk values and parametric models for both the lower and upper tails. In particular, a distinction is made between values observed at regular time steps, called sequential values, and the clusters of extreme values, or events, on which the statistical extrapolations are performed. The sea level distribution is obtained by convolution of the tide and surge density functions. Confidence intervals are also proposed. This model is applied to the case study of Brest, France using both hourly and high water values. Two methods for handling tide–surge interaction are presented and discussed and a comparison with a direct approach is made.