A0229
Title: Spatio-temporal coregionalization modeling by using simultaneous diagonalization
Authors: Sandra De Iaco - University of Salento (Italy) [presenting]
Abstract: The spatio-temporal linear coregionalization model (ST-LCM) represents one of the most common models applied to describe the correlation of multiple variables which evolve in space-time. Thanks to its computational flexibility, it has been recalled in several studies and some advances support a simplified modeling stage through simultaneous diagonalization of the covariance matrices estimated for different lags as well as the choice of appropriate basic covariance models at the different spatio-temporal variability scales. Without these developments, the detection of the uncorrelated components through the identification of the nested structures from the empirical direct and cross-covariance functions would be a hard step in a spatio-temporal context, since 3D plots must be analyzed. Moreover, the selection of the same class of covariance models for all basic hidden components, usually proposed in the past, can be overcome by enabling each basic component to be modelled based on its own features and then by fitting a proper class of covariance models. The ST-LCM fitting process and some computational tools which improve the definition of the uncorrelated components and the main characteristics of the empirical covariance surfaces of the uncorrelated components (in terms of symmetry, separability/non-separability, type of non-separability) are presented together with an application.
