|
Activity Number:
|
112
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistics and the Environment
|
| Abstract - #309430 |
|
Title:
|
Spatial Modeling for Large Multivariate Environmental Data: Advancing Methods and Applications
|
|
Author(s):
|
Sudipto Banerjee*+ and Andrew Finley
|
|
Companies:
|
The University of Minnesota and The University of Minnesota
|
|
Address:
|
420 Delaware Street SE, Minneapolis, MN, 55455,
|
|
Keywords:
|
Bayesian modelling ; Markov Chain Monte Carlo ; Multivariate Processes ; Spatial Predictions
|
|
Abstract:
|
Hierarchical spatial process models implemented through Markov Chain Monte Carlo (MCMC) for understanding scientific relationships, though flexible and versatile, involve expensive matrix decompositions rendering them infeasible for large spatial datasets. This computational burden is exacerbated in multivariate settings with several spatially dependent response variables. Here we propose to use a predictive process to model multivariate spatial data that projects process realizations to a lower-dimensional subspace thereby reducing the computational burden. The resulting predictive process models enjoy attractive theoretical properties along with greater modeling flexibility. We show how the predictive process adapts to multivariate nonstationary processes, with richer association structures. A computationally feasible template that encompasses these diverse settings will be presented.
|
