A1019
Title: Hierarchical spatial copula models for large spatial data
Authors: Alan Pearse - University of Wollongong (Australia)
David Gunawan - University of Wollongong (Australia) [presenting]
Noel Cressie - University of Wollongong (Australia)
Abstract: A fully Bayesian, hierarchical spatial statistical model that incorporates copulas in the process model is presented. These models have several novel features compared to other spatial copula models: The spatial copula models explicitly account for measurement error and allow the noisy and incomplete spatial data to be distinguished from the underlying process; spatial change-of-support is addressed by modelling the spatial process at the level of basic areal units (BAUs) that discretise a spatial domain into a finite number of fine resolution areas; and spatial copula models are developed for large spatial datasets using ideas from fixed rank Kriging. Additionally, it is shown that, when building spatial copula models on point-level spatial support, some choices of copula may lead to violations of Kolmogorov consistency and thus fail to define a valid spatial stochastic process. The models are constructed to avoid such pitfalls for all choices of copula function. It is confirmed that full Bayesian inference on these models is feasible and yields accurate and valid inferences via a simulation study. An illustrative application to remotely sensed atmospheric methane is also presented.
