This dataset contains the code and all relevant data and of the paper "Bayesian inversion of hierarchical geostatistical models using a parallel-tempering sequential Gibbs MCMC" by Sebastian Reuschen, Teng Xu and Wolfgang Nowak. Always cite the paper together with this dataset because this dataset is not self-explanatory.

The code.tar.gz file contains an implementation of the parallel-tempering sequential Gibbs MCMC and performs Bayesian inversion of hierarchical geostatistical models. The data.tar.gz file contains samples from the posterior distribution of a Bayesian inversion of two (highly informative and weakly informative) test cases, which are presented in the related publication.

To access the data, download the data.tar.gz file and unzip it. To access the MATLAB implementation of the MCMC code, (which produced the data) download the code.tar.gz file and unzip it.

The data.tar.gz file includes the results of 5 independent MCMC runs using highly informative data (transient) with 1.1 million samples each and the results of 5 independent MCMC runs using weakly informative data (steady state) with 1 million samples each. The high memory usage of the MCMC runs forced us to only save every 10th MCMC sample to the data set. The high autocorrelation suggest that the introduced errors by doing so are neglectable. The 2500 columns of the sample files represent the hydraulic conductivity discretized at 2500 spatial locations. Hence, each row represents one sample.

Further, the acceptance rates of each tempered chain, the acceptance rate of swap proposals between chains and the the log-likelihood of the T=1 chain (of all samples) are published as well.

The README.txt file in the code folder explains how to set up the MATLAB implementation.