Spatiotemporal hierarchical Bayesian modeling of tropical ocean surface winds
Published:
This paper is one of the first examples of fitting a large and nonstationary covariance function to climate model output and successfully simulating a Gaussian process from the estimated model. The work depends on local, windowed covariance estimates, using the LatticeKrig (Markov Random field) model to encode a global model, and embarrassingly parallel computation using the NCAR supercomputer, Cheyenne.