Hierarchical Bayesian nearest neighbor co-kriging Gaussian process models; an application to intersatellite calibration

Recent advancements in remote sensing technology and the increasing size of satellite constellations allow for massive geophysical information to be gathered daily on a global scale by numerous platforms different fidelity. The auto-regressive co-kriging model provides suitable framework analysis such data sets as it is able account cross-dependencies among fidelity outputs. However, its implementation multifidelity large spatial practically infeasible because computational complexity increases cubically with total number observations. In this paper, we propose nearest neighbor Gaussian process (GP) that couples GP using augmentation ideas. Our reduces linear spatially observed locations. random effects are augmented manner which allows specification semi-conjugate priors. This facilitates design an efficient MCMC sampler involving mostly direct sampling updates. good predictive performance proposed method demonstrated simulation study. We use analyze High-resolution Infrared Radiation Sounder from two NOAA polar orbiting satellites.