|
Abstract:
|
Gaussian processes have been widely used in spatial statistics but face tremendous computation challenges for big datasets, especially for multivariate spatial data. This paper introduces a multi-resolution approximation of Gaussian processes on an extended space, which facilitates efficient computation for large multivariate spatial data. Within the framework of multi-scale structure, it allows modeling non-stationary covariance in a natural way with spatially varying parameters for different processes.The multi-resolution structure provides great flexibility for the modeling of multivariate random fields, to capture the marginal and cross-covariance among processes at different scales.
|
