Bayesian Geostatistics Using Predictive Stacking.

Area of Science:

• Geostatistics
• Spatial Statistics
• Computational Statistics

Background:

• Geostatistical models are crucial for spatial prediction.
• Traditional Bayesian inference can be computationally intensive, often requiring Markov chain Monte Carlo (MCMC) methods.
• Accurate spatial predictions are vital across various scientific disciplines.

Purpose of the Study:

• To develop a computationally efficient Bayesian predictive stacking method for geostatistical models.
• To provide accurate inference on latent spatial random fields.
• To enable spatial predictions at arbitrary locations.

Main Methods:

• Bayesian predictive stacking is employed, combining models across hyper-parameter values.
• Analytically tractable posterior distributions are utilized for regression coefficients and spatial process realizations.
• The method avoids iterative algorithms like MCMC, leveraging parallel computations.

Main Results:

• Stacked inference demonstrates comparable accuracy to full sampling-based Bayesian inference.
• The proposed method significantly reduces computational cost.
• Novel theoretical insights are provided within an infill asymptotic paradigm.

Conclusions:

• Bayesian predictive stacking offers an efficient and accurate alternative for geostatistical modeling and spatial prediction.
• This approach reduces computational burden without sacrificing predictive performance.
• The method is suitable for large-scale spatial data analysis.