|
Abstract:
|
Co-kriging methods utilize data from a lower fidelity level source to improve the estimation and prediction, or compensate the missing variables of data from a higher fidelity level. In remote sensing studies, applying regular co-kriging method is usually computationally expensive and sometimes impossible because these studies usually analysis datasets with large numbers of spatial locations. We developed a new model named nearest neighbor co-kriging Gaussian process model. By embedding the sparsity-introducing prior in the Bayesian Hierarchical co-kriging model, a computationally efficient Markov chain Monte Carlo(MCMC) algorithm can be executed for estimation and prediction. Our simulation studies demonstrate this model can jointly estimate correlated datasets from different sources, and the result is more accurate than from estimate separately. The model can be applied in both nested and non-nested design.
|
