|
Activity Number:
|
276
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #307623 |
|
Title:
|
Priors for High-Dimensional Covariance Models
|
|
Author(s):
|
Charles Curry*+
|
|
Companies:
|
University of California, Santa Cruz
|
|
Address:
|
265 Union Ave., C1062, Santa Cruz, CA, 95008,
|
|
Keywords:
|
covariance modeling ; Bayesian statistics ; MCMC ; reference prior
|
|
Abstract:
|
We extend empirical orthogonal function analysis and optimal fingerprint detection with Bayesian probability models and Markov chain Monte Carlo computational techniques. Using our technique, we fully characterize the uncertainty in climate system properties, as modeled by the MIT 2DLO model. To improve performance of our methodology, we approximate the response of the climate model with a fast, statistically equivalent model and incorporate the additional error added by this approximation into the analysis. Additionally, we include in our parameter set the signal and noise covariance structure of the diagnostics. This goes beyond the maximum likelihood and significance testing strategies employed by optimal fingerprint detection, as it includes the uncertainty in the noise components in the final parameter distributions.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2006 program |
