Bayesian Analysis of Multifidelity Computer Models With Local Features and Nonnested Experimental Designs: Application to the WRF Model

Motivated by a multi-fidelity Weather Research and Forecasting (WRF) climate model application where the available simulations are not generated based on hierarchically nested experimental design, we develop new co-kriging procedure called augmented Bayesian treed co-kriging. The proposed extends scope of in two major ways. We introduce binary partition latent process multifidelity setting to account for nonstationary potential discontinuities outputs at different fidelity levels. Moreover, an efficient imputation mechanism which allows practical implementation when design is nonhierarchically enabling specification semiconjugate priors. Our strategy reversible jump Markov chain Monte Carlo that involves collapsed blocks direct simulation from conditional distributions. recursive emulator provides proxy full predictive distribution output each level, computationally feasible manner. performance our method demonstrated benchmark examples used analysis large-scale modeling WRF model.