Extreme wave height estimation using a hybrid distribution approach

Accurate estimation of design wave heights is paramount for the safety and resilience of coastal infrastructure. However, extreme value analysis with limited samples faces two critical challenges: spurious identification of bounded EV-III distributions due to statistical fluctuations, and severe underestimation risks stemming from uncertain finite upper limits. To address this, we propose a novel hybrid framework: the Generalized Extreme Value–Generalized Gumbel Distribution (GEV–GGD) model for Annual Maximum data, and the Generalized Pareto Distribution–Weibull (GPD–WBL) model for Peaks-Over-Threshold data. If initial estimates indicate a bounded EV-III distribution, the model is adaptively substituted with an inherently unbounded alternative (GGD or WBL). Monte Carlo simulations show that, under the evaluated conditions, this methodology mitigates lower-tail underestimation and yields narrower empirical sampling intervals than the alternatives examined. Applications to an observed buoy record and a hindcast dataset demonstrate that, for these datasets, the approach avoids saturation at fitted upper endpoints while maintaining comparable goodness-of-fit within the sample range. The framework therefore offers a practical means of mitigating underestimation risk, while its design applicability depends on the target reliability and the consequences of conservative overestimation, particularly when both the positive shape parameter and return period are large.