The peak over threshold (POT) method and the generalized Pareto distribution (GPD) model are studied based on 22-year hindcasted wave data. The initial data are divided into many groups with the fixed time interval to ensure the independence of the maximal value within every group. To realize this function, 3-day and 5-day are selected as the fixed time intervals, which are used to obtain the sample to extrapolate extreme significant wave heights with return periods of 100-year, 200-year and 500-year at 4 locations in the Yellow Sea (YS).
Adequate extrapolation of extreme significant wave heights is highly important for the design, installation and operation of the coastal and ocean structures (Soukissian and Kalantzi, 2006). A reasonable design wave height can guarantee the security of structures and reduce the engineering expense, which are two key points for the engineering practices.
To reasonably determine the design wave height, a reliable sampling method and an appropriate distribution method are needed. The sampling methods can be grouped into the annual maximum (AM) method (Carter and Challenor, 1981; Soares and Scotto, 2001), the peak over threshold (POT) method (Petruaskas and Aagaard,1971), the total sample method (Ochi,1992), the triple annual maxima method (Sobey and Orloff, 1995), and the annual N-Largest method (Weissman,1978). Because of the utilization efficiency of initial data, the POT method has widely been employed to select the samples for extrapolation (Caires and Sterl, 2005; Soukissian et al., 2006; Mazas et al., 2014; Shao et al., 2017). According to the samples of the POT method, the generalized Pareto distribution (GPD) model is commonly accepted to extrapolate the extreme values for different return periods with reasonable uncertainties (Tancredi et al., 2006; Izadparast and Niedzwecki, 2010; Os et al., 2011).
The POT method extracts a series of independent peak significant wave heights above a selected value (the threshold) from the initial data. To ensure the independence of the sample, many methods have been proposed, such as the meteorological phenomena method (Shao et al., 2017), the time window method (Thompson et al., 2009) and the double-threshold method (Mazas and Hamm, 2011; Bernardara et al., 2014). In this study, the initial data come from the hindcasted wave data in the Yellow Sea (YS), a winter storm-dominated area. The time window method is more applicable for selection of independent peak significant wave heights, due to the characteristic of winter storms. After preparing the initial sample from the hindcasted wave data, the threshold needs to be determined. The graphical diagnostic referred to as the GPD parameter plot (Coles, 2001) is usually used as the threshold selection method, due to the graphical inspection and comprehension of data and the assessment of the candidate thresholds.
