Mathematical Geology, Vol. 25, No. 3, 1993
Spatial Orthogonality of the Principal Components
Computed from Coregionalized Variables 1
P. Goovaerts 2
Within the frame of the linear model of coregionalization, this paper sets up equations relating the
variogram matrix of the principal components extracted from the variance-covariance matrix to the
diagonal variogram matrices of the regionalized factors. The spatial orthogonality of the principal
components is investigated in three situations: the intrinsic correlation, two basic structures with
independent nugget components, three basic structures with independent nugget components and
uncorrelated subsets of variables. Two examples point out that the correlation between the principal
components may be nonnegligible at short distances, especially if the correlation structure changes
according to the spatial scale considered. For one of the two case studies, an orthogonal varimax
rotation of the first principal components is found to greatly reduce the spatial correlation between
some of them.
KEY WORDS: variogram matrix, linear model of coregionalization, principal component analy-
sis, intrinsic correlation, orthogonal rotation.
INTRODUCTION
Besides its use in mining or petroleum field, geostatistics is now widely applied
in areas such as geochemical exploration or soil science where the information
is mainly multivariate. When dealing with multivariate datasets, difficulties es-
sentially arise from the fitting of a valid model of coregionalization and the
solving of large cokriging systems. Similar problems are encountered in non-
parametric geostatistics where a vector of indicator variables is available at each
sampled point. To reduce the modeling and computational effort, several authors
(Davis and Greenes, 1983; Suro-Prrez and Journel, 1991) have proposed rather
similar methods that involve: computing the principal components of the co-
variance function matrix for h = 0 or small lags h, kriging them and estimating
the initial variables as linear combinations of the kriged principal components.
~Received 18 February 1992; accepted 13 October 1992.
2Unit6 de Biomrtrie, Universit6 Catholique de Louvain, Place Croix du Sud 2 Bte i6, B-1348
Louvain-la-Neuve, Belgique. Present address: Department of Applied Earth Science, Stanford
University, Stanford, California 94305.
281
0882-8121/93/0400"0281,¢[ff}7.00/1 © 1993 International Association for Mathematical Geology