A Poly-Cylindrical Bayesian Network for Clustering Oceanographic Data

When circular variables are observed along with linear variables, we refer to them as cylindrical or circular-linear in the bivariate case, and poly-cylindrical in higher dimensions. A salient field of application that has encouraged research into cylindrical statistics is oceanography. The properties and processes of the ocean, such as currents, waves, tides, temperature, salinity, marine life, and ecosystems can be described with a set of circular and linear variables. Clustering of oceanographic data is essential for researchers to understand the complex dynamics of the ocean and predict the behaviour of oceanographic phenomena. In this chapter, a pioneering multivariate cylindrical distribution is used to apply a Bayesian network in combination with the Expectation-Maximization algorithm to learn a finite mixture model for clustering which will ultimately allow us to understand the conditional independences amongst random variables. This proposed methodology will enable efficient clustering of multivariate directional-linear oceanographic data.