Comparaison de différentes méthodes d'estimation des niveaux extrêmes en site à fort et à faible marnage

The estimation of probability of occurrence for extreme sea levels is a major issue for studies on the risks of coastal flooding. Return periods of extreme water levels are estimated here for two sites: a high tidal range, Brest, the other with a low tidal range, Sète. The direct approach is tested (on water levels) and also the revised joint probabilities methods (RJPM) combining probabilities associated with tide and surges. We compare the results of statistical adjustments of extremes for two different methods for selecting events: for the annual maxima method (AM) and the peak over the threshold method (POT). We discuss the different ways to treat eustatism, the choice of the probability function, the choice of threshold for POT method, the influence of the measurement duration and the time dependence with the use of the extremal index. For data from Sète, we try to quantify the seasonal dependence tide/storm surge. RJPM and direct approaches tend to the same result for the low tidal site. POT and AM tend toward the same result for the long series of Brest, and POT is better for the short series. The selection of the threshold for POT is justified by a set of additional statistical tests. Using the extremal index does not change significantly the result of convolution. The model of seasonal dependence used has a first interesting result and could be tested on other sites. We derive a general methodology and recommendations that can be useful for further analysis.