A0695
Title: An axial regression model to detect vegetation growth patterns
Authors: Marco Mingione - Roma Tre (Italy) [presenting]
Abstract: A mixture of axial regression models is proposed for bivariate observations of axial and circular variables. It is noted that an axial variable is defined on the semicircle due to the lack of information about the direction of the propagation. Therefore, the data lies on the Cartesian product of a circle and a semicircle. A valid density function is defined with periodic behavior on this complex manifold by exploiting the bivariate wrapped Cauchy density to induce dependence between the two angular measurements. The proposed parametrization allows for the derivation of conditional distributions in closed form, and straightforward regression models are built. Estimation is obtained via maximum likelihood by exploiting the EM algorithm. The model's performances are illustrated on an original dataset of stripe and wind directions recorded on Marion Island (South Africa).
