Regional frequency analysis of extreme significant wave heights with long return periods based on complete distribution characteristics

Abstract

Reliable assessment of extreme significant wave heights is crucial to the design and operation of ocean structures. However, due to the limited duration of available wave data, the uncertainty of return significant wave heights based on at-site extreme wave analysis may be large, especially for long return periods. In this study, a clustering method for identifying homogeneous regions is proposed, which directly diagnoses the frequency distribution of extreme samples to fully detect the distribution characteristics, rather than using statistical parameters of extreme samples to partially characterize this distribution. By standardizing extreme samples, measuring distribution differences, and iteratively calculating clustering centers, three homogeneous regions are identified in the study region. The diagnostic results of goodness-of-fit test show that the generalized extreme value distribution generally performs well in these regions. In each homogeneous region, the regional quantile is constructed based on all extreme samples within this region, and the site-dependent scale factor is used to extrapolate the quantile for each site. Compared with the traditional clustering method, the fitting performance of the model quantile based on the proposed method is generally improved, especially near the boundary of the homogeneous region. Compared with the at-site extreme wave analysis, the uncertainty of the return significant wave height extrapolated by the regional frequency analysis is generally reduced, especially for long return periods. For example, the width of the 95 % confidence interval for the 200-year return level is reduced by approximately 2 times at all study sites. In the homogeneous region, the distribution characteristics of extreme samples are similar due to the influence of some factors, such as driving weather. Sample information from all sites in this region can be used to describe the common distribution characteristics to construct a stable regional quantile, which is essential for extrapolation with long return periods.