Spatial filtering and missing georeferenced data imputation: A comparison of the Getis and Griffith methods

Abstract

Spatial filtering first introduced independently by Getis and by Griffith is beginning to mature, with a third version now being developed by Legendre and his colleagues. Like the Getis formulation, this newest version is distance-based; like the Griffith formulation, it uses eigenfunctions, but extracted from a modified distance matrix - it is a mixture of the other two. Bivand (2002) comments that “the Getis filtering approach . . . seems to admit prediction to new data locations . . . . The Griffith eigenfunction decomposition approach . . . does not . . . .” Missing data prediction equations are presented for each of these two original formulations, and then compared with several popular datasets.