? 2022 Elsevier LtdThe use of nonparametric copulas for modeling joint probability distributions of metocean variables and building environmental contours is examined. Provided that no assumption is made about a particular functional form, nonparametric copulas may be an alternative to improve modeling dependence patterns, which is critical to develop environmental contours. The transformation method with local log-quadratic likelihood kernel estimator and nearest-neighbor bandwidth (TLL2nn), the mirror-reflection and beta kernel methods, and Bernstein polynomials are used for nonparametric estimation. For comparison purposes, well known parametric copulas are also estimated. Hindcast data from tropical cyclones and storms at three sites in the southern Gulf of Mexico is used. The adequacy of the copula models is assessed first using graphical diagnostics tools. Data fitting is then evaluated using Cramér-von Mises statistics, and the Akaike and Bayesian information criteria are applied for model selection. The estimated copulas are also compared in terms of their adequacy to model upper and lower tail dependence. The results showed that whereas different parametric copulas were needed to model the different data sets, the nonparametric TLL2nn model was able to give adequate results for the three sites. Environmental contours using the nonparametric, parametric, and empirical copulas are compared.
展开▼
