|
Activity Number:
|
265
|
|
Type:
|
Invited
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
WNAR
|
| Abstract - #305071 |
|
Title:
|
Spatially Varying AR Processes Based on Discrete Convolutions
|
|
Author(s):
|
Bruno Sanso*+ and Alexandra Schmidt and Aline Nobre
|
|
Companies:
|
University of California, Santa Cruz and Universidade Federal de Rio de Janeiro and Universidade Federal de Rio de Janeiro
|
|
Address:
|
Applied Mathematics and Statistics, Santa Cruz , CA, 95064,
|
|
Keywords:
|
Bayesian methods ; spatio-temporal models ; process convolutions
|
|
Abstract:
|
We consider a class of models for spatio-temporal processes based on convolving independent processes with a kernel represented by a lower triangular matrix. We first consider a family obtained by convolving spatial Gaussian processes with isotropic correlations where the kernel is obtained from the covariance of an AR(p). A second family is based on considering convolutions of AR(p) processes and using the kernel to provide spatial interactions. The resulting random field corresponds to an AR(p) with spatially varying coefficients. Suitable priors on the parameters of the AR(p) guarantee the parameters satisfy the conditions for stationarity at each location. The proposed modeling framework provides a rich variety of covariance structures. We consider applications to the problem of detecting trends in environmental variables.
|
