CFE 2019: Start Registration
View Submission - CMStatistics
B0607
Title: Copula-based segmentation of cylindrical time series Authors:  Francesco Lagona - University Roma Tre (Italy) [presenting]
Abstract: Bivariate sequences of angles and intensities are often referred to as cylindrical time series, because the pair of an angle and an intensity can be represented as a point on a cylinder. In environmental studies, examples of these data include time series of wind directions and pollutant concentrations, wind directions and speeds and wave directions and heights. The analysis of cylindrical time series is complicated by the difficulties in modeling the dependence between angular and linear measurements and the temporal correlation between cylindrical observations. An additional complication that often arises in environmental studies is the multimodality of the marginal distribution of the data because environmental cylindrical data are typically observed under heterogeneous conditions that vary over time. A parsimonious hidden Markov model is introduced to simultaneously account for linear-circular dependence, temporal auto-correlation and multimodality. Under this model, the distribution of cylindrical data is approximated by a mixture of copula-based cylindrical densities, whose parameters depend on the evolution of a latent Markov chain. While the copula-based cylindrical density accommodates linear-circular dependence, a mixture of copula-based densities allows for multimodality and, finally, a latent Markov chain accounts for temporal correlation.