Distributions of freak wave heights measured in the North Sea

We present a statistical analysis of some of the largest waves occurring during 793 h of surface elevation measurements collected during 14 severe storms in the North Sea. This data contains 104 freak waves. It is found that the probability of occurrence of freak waves is only weekly dependent on the significant wave height, significant wave steepness and spectral bandwidth. The probability does show a slightly stronger dependency on the skew and kurtosis of the surface elevation data, but on removing the contribution to these measures from the presence of the freakwaves themselves, this dependency largely disappears.Distributions of extreme waves are modelled by fitting Generalised Pareto distributions, and extreme value distributions and return periods are given for freak waves in terms of the empirical fitted parameters. It is shown by comparison with these fits that both the Rayleigh distribution and the fit of Nerzic and Prevosto severely under-predict the probability of occurrence of extreme waves. For the most extreme freak wave in our data, the Rayleigh distribution over-predicts the return period by about 300 times when compared to the fitted model.