A0483
Title: Different approaches for modeling multivariate space-time data: A performance-based comparison
Authors: Claudia Cappello - University of Salento (Italy) [presenting]
Sandra De Iaco - University of Salento (Italy)
Monica Palma - University of Salento (Italy)
Klaus Nordhausen - University of Helsinki (Finland)
Abstract: In the last decades, great advances have been made in the multivariate space-time framework for modeling the matrix-valued covariance function. One of the first models proposed in the literature has been the space-time linear coregionalization model (ST-LCM), developed for modeling matrix-valued covariance function as a linear combination of univariate models related to latent uncorrelated processes underlying the investigated phenomenon. Moreover, in recent years, the space-time blind source separation-based model (ST-BSS) has been proposed as an alternative modeling approach based on the univariate analysis of the latent components. In the application stage, each of the above-mentioned techniques is characterized by its own drawbacks and advantages in terms of model parameters to be estimated and/or time-consuming in the modeling procedure. The different modelling steps and the predictive performances of the ST-LCM and the ST-BSS model are focused from a both theoretical and practical points of view, and a comparison of their performances is provided through a simulation study and an application on a space-time environmental data set.
