Statistical modelling of extreme storms using copulas: A comparison study

Coastal risk assessment and structure design heavily rely on the statistical analysis of the extreme wave climate, such as the wave height, storm duration, wave period and surge height. Due to the dependence among these variables, the multivariate dependence structure should be taken into account for statistical simulation of the storm events. Here, three modelling framework are described and compared for probabilistic modelling of extreme storms, i.e. the Gaussian copula, conditional mixture method and entropy copula. This paper demonstrates the functionality of the approaches in simulating the sea storm while maintaining their statistical characteristics. None of the three methods can fit all the bivariate cases best, and the Gaussian copula gives the best overall fitting quality for the four dimensional data in the case study, while the conditional mixture method gives the lowest fitting quality. The entropy copula gives a comparable simulation results and shows its great potential to be universally applicable for modelling the multivariable joint distribution by avoiding the procedure of assigning any copula family before fitting data.