Precision Agric (2008) 9:173–175

DOI 10.1007/s11119-008-9056-2

SHORT DISCUSSION

Variogram or semivariogram? Understanding

the variances in a variogram

Martin Bachmaier Æ Matthias Backes

Published online: 24 February 2008

Ó Springer Science+Business Media, LLC 2008

The theoretical variogram and the confusion in the literature

The definition of the theoretical variogram, c, is based on regionalized random variables

Zðx ~þ ~
~Þ and Zðx hÞ where ~ x ~
x and ~þ h represent the spatial positions separated by a vector ~h:
h i h i
cðh~Þ ¼ 1 E ½Zðx~þ ~ hÞ
Zðx
1
~Þ2 ¼ Var Zðx ~þ ~ hÞ
Zðx~Þ : ð1Þ
2 2
The Zðx~Þ and Zðx~þ ~hÞ denote random variables. According to the intrinsic hypothesis,
cðhÞ is assumed to depend only on the separation vector, the lag ~
~ h; but not on the location
~:
x Further, the increments Zðx ~þ ~hÞ
Zðx ~Þ are assumed to have no drift: E½Zðx ~þ ~hÞ

Zðx ~
~Þ ¼ 0 for all h and all ~;
x otherwise the last identity in Eq. 1 would not hold.
Concerning the terminology, there is great confusion in the geostatistical literature.
Some authors call the function c a ‘variogram’ (Wackernagel 2003; Worboys 1995;
Gneiting et al. 2001), several authors call it a ‘semivariogram’ (Journel and Huijbregts
1978; Cressie 1991; Goovaerts 1997; Burrough and McDonnell 1998; Olea 1999; Stein
1999; Gringarten and Deutsch 2001), stating that a semivariogram is half a variogram, and
others use the terms variogram and semivariogram synonymously (Isaaks and Srivastava
1989; Webster and Oliver 2007). To explain what is depicted in a variogram, authors of
geostatistical books and articles often take refuge in phrases such as ‘‘spatial variability’’ or
terms like ‘‘semivariogram value’’ or ‘‘semivariance’’ without saying of what. Evidently
there is great uncertainty with regard to terminology and the interpretation of variograms.
The confusion concerning the prefix ‘‘semi’’ has arisen because Matheron (1965) in his
seminal thesis had in mind the variance of differences, Var½Zðx ~þ ~ hÞ
Zðx
~Þ; but the
M. Bachmaier (&)
Department of Life Science Engineering, Crop Production Engineering,
Technische Universität München, Am Staudengarten 2,
85354 Freising-Weihenstephan, Germany
e-mail: [email protected]

M. Backes
Institut für Kartographie und Geoinformation, Rheinische Friedrich-Wilhelms-Universität Bonn,
Meckenheimer Allee 172, 53115 Bonn, Germany
e-mail: [email protected]

123
174 Precision Agric (2008) 9:173–175

quantity we want in practice is the half (‘‘semi’’) of this, because this gives the ‘‘variance
per point when the points are considered in pairs’’ (Webster and Oliver 2007, p. 54; cf. also
the expression in Eq. 3). Thus, cðh~Þ can be interpreted as the variance of the variable (e.g.
of the yield data) at the given separation vector ~
h; which means that we consider only pairs
that are spatially separated by the lag ~h: It should not be called a semivariance since this
term originates from the variance of the differences, which is not the actual quantity of
interest. And if it were, one should not compute the half of it, but the whole variance. No
one says that the semiheight of his or her body is 86 cm.

Understanding the empirical variances in a variogram

The empirical variance of measured values zi can be computed in two different ways:
1 X n
1 1 X
s2 ¼ zÞ2 ¼ 
ðzi

ðzj
zi Þ2 ; ð2Þ
n
1 i¼1 2 nðn
1Þ all i6¼j

where n(n-1) is the number of pairs in the sum. This latter could be halved, because it
suffices to consider only all pairs with i \ j since (zi-zj)2 = (zj-zi)2.
The variances in an experimental variogram, ^cðh ~Þ; arise by restricting the latter
expression in Eq. 2 to pairs of measured values, zi ¼ zðx ~i þ ~
~i Þ and zj ¼ zðx hÞ; that are
separated by a spatial vector ~h:
~Þ
1 1 Xh i2
Nðh
~Þ ¼
^cðh  ~i þ ~
zðx hÞ
zðx
~i Þ ; ð3Þ
~Þ i¼1
2 Nðh

where zðx ~i þ ~
hÞ and zðx ~Þ pairs of comparisons, ~
~i Þ are the measured values of Z at Nðh xi þ ~
h
and ~ ~ ~
xi ; separated by the vector h: Thus, ^cðhÞ should simply be called the (empirical)
variance of the measured values, e.g. of the yield data, at the given separation vector ~ h:
When referring to isotropic variation, ^cðhÞ denotes this variance at a given separating
distance h ¼ jjh~jj: We do not need the term semivariance unless we want to cite references
where it is used. But then it should be added that it is the semivariance of the difference of
random variables or measured values.
Finally, Eqs. 2 and 3 also show that the empirical variance of all measured values, s2,
can be computed as a weighted mean of the variogram variances ^cðh ~Þ; where the weighting
~ ~
is according to the number of pairs, NðhÞ; in ^cðhÞ (Bachmaier 2007, Eq. 23).

References

Bachmaier, M. (2007). Using a robust variogram to find an adequate butterfly neighborhood size for one-
step yield mapping using robust fitting paraboloid cones. Precision Agriculture, 8, 75–93.
Burrough, P. A., & McDonnell, R. A. (1998). Principles of geographical information systems. Spatial
information systems and geostatistics. Oxford, New York: Oxford University Press.
Cressie, N. A. C. (1991). Statistics for spatial data. New York: Wiley.
Gneiting, T., Sasvári, Z., & Schlather, M. (2001). Analogies and correspondences between variograms and
covariance functions. Advances in Applied Probability, 33, 617–630.
Goovaerts, P. (1997). Geostatistics for natural resource evaluation. Oxford, New York: Oxford University
Press.
Gringarten, E., & Deutsch, C. V. (2001). Teacher’s aide—variogram interpretation and modeling. Mathe-
matical Geology, 33, 507–534.

123
Precision Agric (2008) 9:173–175 175

Isaaks, E. H., & Srivastava, R. M. (1989). An introduction to applied geostatistics. Oxford, New York:
Oxford University Press.
Journel, A. G., & Huijbregts, C. J. (1978). Mining geostatistics. London: Academic Press.
Matheron, G. (1965). Les Variables re´gionalise´es et leur Estimation. Paris: Masson.
Olea, R. A. (1999). Geostatistics for engineers and earth scientists. Boston: Kluwer Academic Publishers.
Stein, M. L. (1999). Interpolation of spatial data—some theory for kriging. New York: Springer Verlag.
Wackernagel, H. (2003). Multivariate geostatistics. Berlin, Heidelberg: Springer Verlag.
Webster, R., & Oliver, M. A. (2007). Geostatistics for environmental scientists. New York: Wiley.
Worboys, M. F. (1995). GIS—a computing perspective. London: Taylor & Francis Ltd.

123