Software and algorithms for REA Bayes (2004)

A companion to: Tebaldi, Smith Nychka and Mearns (2004) Quantifying Uncertainty in Projections of Regional Climate Change: A Bayesian Approach to the Analysis of Multimodel Ensembles

The form of full conditional distributions and the steps of the Gibbs sampler are presented in the Supplement PDF 100K

The Markov Chain Monte Carlo was implemented in the R statistical language. This is a Matlab-like, high level langauage well sutied to statistics and probability. It is freely available from Comprehensive R Archive for a variety of platforms. (Our calculations were run on a LINUX PC.)

To reproduce the analysis and to persue the programs transfer the two files REA.data.r REA.Gibbs.r from the directory REA.Bayes

Install the R package and in R

source("REA.data.r")
source("REA.Gibbs.r")
You need to do this only once provided you save your work space when quitting.

The naming convention for data tables is:
(scenario).season.variable

The X and Y variables are matrices where rows index regions and columns index models. Thus A2.DJF.Y is a 22X9 matrix of the future projections for the A2 scenario in winter. Part of the reason that the data are grouped together is that the Gibbs sampler is more efficient sampling for all regions simultaneously. (For those familiar with R programming we have eliminated an explicit loop over regions using matrix multiplications.)