Activity Number:
|
150
- Methods and Computing for Spatial and Spatio-Temporal Data
|
Type:
|
Contributed
|
Date/Time:
|
Monday, August 8, 2022 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Statistics and the Environment
|
Abstract #322088
|
|
Title:
|
Computationally Efficient Algorithms for Bayesian Nearest Neighbor Co-Kriging Gaussian Processes
|
Author(s):
|
Bledar Alex Konomi* and Si Cheng
|
Companies:
|
University of Cincinnati and BeiGene, Ltd.
|
Keywords:
|
Augmented hierarchically nested design;
Autoregressive Co-kriging ;
Nearest neighbor Gaussian process;
Remote Sensing
|
Abstract:
|
The recently proposed nearest neighbor co-kriging Gaussian process (NNCGP) models make possible intersatellite calibration by combining measurements of different fidelity and accounting for spatially varying bias correction. The inference of the NNCGP is based on a sequential Markov chain Monte Carlo (MCMC) sampler which involves updating a latent random effect vector. Because of the high-dimensional nature of the latent random effect vector this sampler may be sensitive to the initial values and has high auto-correlations with the tendency to converge slowly. Here we propose two alternative inferential procedures which target to reduce high-dimensional parametric space, improve convergence, and reduce computing time. The first alternative procedure reduces the posterior sampling space by integrating out the latent processes. The second alternative procedure is a new MCMC free procedure which significantly decreases the computing time without significantly sacrificing prediction accuracy. The good computational and prediction performance of our algorithms are demonstrated on benchmark examples and the analysis of the High-resolution Infrared Radiation Sounder data.
|
Authors who are presenting talks have a * after their name.