Hierarchical Nearest-Neighbor Gaussian Process Models for Large Geostatistical Datasets.

Area of Science:

• Geostatistics
• Statistical Modeling
• Computational Statistics

Background:

• Traditional spatial process models face computational challenges with increasing numbers of spatial locations.
• Analyzing large geostatistical datasets requires scalable and efficient modeling approaches.

Purpose of the Study:

• To develop highly scalable nearest-neighbor Gaussian process (NNGP) models for large geostatistical datasets.
• To provide a framework for fully model-based inference in massive spatial data analysis.

Main Methods:

• Developed a class of nearest-neighbor Gaussian process (NNGP) models.
• Embedded NNGP as a sparsity-inducing prior in a hierarchical modeling framework.
• Utilized computationally efficient Markov chain Monte Carlo (MCMC) algorithms with linear computational complexity.

Main Results:

• Established NNGP as a well-defined spatial process with sparse precision matrices.
• Demonstrated computational efficiency with linear floating point operations per iteration.
• Showcased scalability for analyzing massive datasets, such as U.S. Forest Inventory data.

Conclusions:

• NNGP models offer substantial scalability for geostatistical data analysis.
• The proposed models provide computational and inferential benefits over existing methods.
• NNGP facilitates analysis of large-scale spatial data previously intractable for other methods.