|
Activity Number:
|
100
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #303740 |
|
Title:
|
Bayesian Modeling of Non-Gaussian Geostatistical Data via Copulas
|
|
Author(s):
|
Souparno Ghosh*+ and Bani K. Mallick
|
|
Companies:
|
Texas A&M University and Texas A&M University
|
|
Address:
|
Department of Statistics, College Station, TX, 77843-3143,
|
|
Keywords:
|
Non-Gaussian characteristics ; Kriging ; Copula ; Mixture model
|
|
Abstract:
|
Real Spatial often data display non-Gaussian features like skewness, heavy-tails or even multi-modes. We propose a general class of models to handle non-Gaussian spatial data. Unlike earlier methods, this approach begins by modeling the marginal distributions and achieves the spatial dependence via elliptical copula. That way we confirm that a valid stochastic process is obtained for spatial prediction. Diagnostic measures and cross-validation show that our model has a better predictive performance than several kriging variants. The proposed model is then extended with a mixture of elliptical copula. Results from simulations and real data analysis demonstrate that this model has the ability to accommodate non-stationary data. Finally a non-elliptical copula model is developed that can be used to model extreme observations recorded over a spatial domain.
|
