datasets cannot be represented by a normal distribution and
therefore the non-stationary trends have to be removed. By
removing a fit of monthly mean and diurnal variation from wind
speed dataset, the overall distribution approximates a normal
distribution (Hill et al., 2012). For significant wave height, it is
necessary to remove a fit of monthly mean values and then apply a
logarithmic transformation on the data as shown in Eq. (3) (Cunha
and Guedes Soares, 1999). For wave period, it is only necessary to
remove seasonal trends. For the multivariate case presented in this
work, a modified Box-Cox transformation has been used in place
of a logarithmic transformation, shown in Eq. (4). The value of the
transform coefficient Λ, can be tuned iteratively to capture the
observed level of correlation between the wind and wave values in
the data while preserving individual wave climate characteristics.