We present a non-parametric approach for the estimation of the bivariate distribution of twocircular variables and the modelling of the joint distribution of a circular and a linear variable.We combine nonparametric estimates of the marginal densities of the circular and linearcomponents with the use of class of nonparametric copulas, known as empirical Bernsteincopulas, to model the dependence structure. We derive the necessary conditions to obtaincontinuous distributions defined on the cylinder for the circular-linear model and on the torus forthe circular-circular model. We illustrate these two approaches with two sets of realenvironmental data
展开▼
