|
Activity Number:
|
110
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 7, 2006 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
| Abstract - #307580 |
|
Title:
|
Circulant Embedded Extended CAR Models for Large Spatial Data
|
|
Author(s):
|
Ernst Linder*+
|
|
Companies:
|
University of New Hampshire
|
|
Address:
|
Kingsbury Hall, Durham, NH, 03824,
|
|
Keywords:
|
Gaussian random fields ; spatio-temporal analysis ; large data
|
|
Abstract:
|
We examine an extension of the usual CAR model for spatial lattice data. The extension incorporates a second spatial parameter that governs smoothness of the underlying spatial field. We show that for a circulant embedded rectangular lattice this model has the same discrete Fourier spectrum as the Matérn class of covariance functions for spatial point data. This extended model can also be applied to point referenced data using appropriately defined distance-based weight functions. Thus we achieve an estimation scheme based on the fast Fourier transformation that is computationally efficient for large spatial data. We illustrate how this model and the circulant embedding is implemented within a hierarchical Bayesian estimation framework for spatial and spatio-temporal data. We give examples from earth systems science related to carbon cycling and large scale hydrological systems.
|
