CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Multivariate Bayesian change-point model with concurrent breaking points Authors:  Gianluca Mastrantonio - Politecnico of Turin (Italy) [presenting]
Giovanna Jona Lasinio - Sapienza University of Rome (Italy)
Alessio Pollice - University of Bari (Italy)
Giulia Capotorti - Sapienza University of Rome (Italy)
Lorenzo Teodonio - IPAL (Italy)
Giulio Genova - Sapienza University of Rome (Italy)
Carlo Blasi - Sapienza University of Rome (Italy)
Abstract: Extreme temperatures and precipitations have been recorded from 1951 to 2010 in 360 stations across Italy. Motivated by this real data, we present a new multivariate change-point model. The time series are modelled through a spatio-temporal/seasonal Gaussian process, a mean dependent on elevation and with independent trivariate residuals that follow change-point models, one for each station. The change point models take into account possible temporal drifts and parameters and breaking-points can be both shared across stations. Our model is specified using the Dirichlet process in a Bayesian framework. The clusterizations are then compared with the Italian Ecorigion, that are ecologically homogeneous areas of similar potential as regards the climate, physiography, hydrography, vegetation and wildlife, used as geographic framework for interpreting ecological processes, disturbance regimes, and vegetation patterns and dynamics.