Abstract
Probabilistic modelling of relevant environmental variables are crucial for the safe design and operation of marine structures. Using metocean data, a joint model of several variables can be estimated, including their dependence structure. Often, a conditional model is assumed for this, but recently the non-parametric Bernstein copula has been suggested as an alternative tool to model such dependencies. As a non-parametric technique, it is very flexible and often provides excellent goodness-of-fit to data with different dependencies. However, non-parametric techniques are prone to over-fitting and generalizability might be challenging. Moreover, care should be taken when using such models for extrapolation. In this paper, a simple simulation study will be presented that has investigated the usefulness of the Bernstein copula in modeling joint metocean variables. First, data have been generated from a known parametric joint distribution model. Then, a joint model based on the Bernstein copula is fitted to a subset of these data. Data simulated from the Bernstein-based models are then compared to data from the initial model. A particular focus will be put on how the model captures the dependencies in the extremes.
