Spatial Analysis Using Big Data

4.9. Methods for large data

4.9.1. Outline

Numerous studies have been conducted on large-scale geostatistical data modeling. Most existing approaches are classified into either low-rank or sparse approximation. Low-rank approximation applies a rank reduction to a GP to model global spatial patterns behind data computationally efficiently. On the other hand, the sparse approximation replaces the typically dense covariance matrix or its inverse (i.e., precision matrix) with a sparse matrix considering neighboring sites. Thus low-rank approximation attempts to estimate large-scale spatial variations while the sparse approximation attempts to estimate small-scale spatial variations. This section introduces representative approaches in these ...