ABSTRACT
As with many disciplines, spatial statistical applications have transitioned from a “data poor” setting to a “data rich” setting in recent years. For example, when the primary interest is in predicting ore reserves, one might only have a limited number of observations available, due to the expense of ore extraction. In such cases, the standard geostatistical prediction formulas are easy to implement with basic computer programs, requiring little more than a few low-dimensional matrix (i.e., multiplication and inverse) manipulations. As more data have become available through increases in the number and extent of automated observation platforms (e.g., global weather station networks, remote sensing satellites, ground penetrating radar, lidar, medical imagery, etc.), the issues related to practical implementation of spatial prediction algorithms become significant. These issues are present regardless of whether one takes a Bayesian or frequentist perspective when performing inference and prediction.
