You are on page 1of 4

P1: GCR

Mathematical Geology [mg] pp963-matg-470280 October 14, 2003 15:16 Style file version June 25th, 2002

Mathematical Geology, Vol. 35, No. 5, July 2003 (°


C 2003)

Letter to the Editor


Comment on “Understanding Anisotropy Computations” By
M. Eriksson and P. P. Siska
In a recent interesting paper, Eriksson and Siska (2000) critically examine the clas-
sical notion of variogram model geometric anisotropy (calling it range-anisotropy).
The authors propose extending the concept of geometric anisotropy commonly
used to characterize the range parameter to all other variogram parameters; for
example, the sill, the nugget effect, and the slope and exponent of power mod-
els. The proposal is novel and simple to apply. It possibly defines a wider and
more flexible class of models that would be useful for modeling various types
of zonal anisotropy. It could lead to more parsimonious variogram model fitting
than is practised in the classical approach that uses nested models with geometric
anisotropy. However, one point the authors did not examine is the important ques-
tion of admissibility of the models resulting from their proposal. It is shown below,
using simple numerical examples, that the models are not necessarily admissible.
Thus, except for the range, the use of models having parameters with directional
dependence (defining an ellipse) is not straightforward. The difficult verification of
model admissibility must be done before using the model for further geostatistical
computations.

CONDITION OF ADMISSIBILITY FOR A VARIOGRAM FUNCTION

A variogram function γ (h) is said to be admissible if −γ (h) is conditionally


positive definite (Chilès and Delfiner, 1999, p. 61). That is

P
n P
n
− λi λ j γ (h i j ) ≥ 0,
i=1 j=1
Pn
s.t. λi = 0.
i=1

Normally, verification of the above condition necessitates using the spectral rep-
resentation of variograms. Details can be found in Yaglom (1962), Christakos
(1984), Cressie (1991), and Chilès and Delfiner (1999).

643
0882-8121/03/0700-0643/1 °
C 2003 International Association for Mathematical Geology
P1: GCR
Mathematical Geology [mg] pp963-matg-470280 October 14, 2003 15:16 Style file version June 25th, 2002

644 Letter to the Editor

SILLS ON AN ELLIPSE (Eriksson and Siska, 2000, p. 691–692)

Consider the simple configuration: x1 = (0, 0), x2 = (h, 0), x3 = (h, h).
Without loss of generality, a spherical model is chosen with sills varying
directionally on an ellipse. The main axes of the ellipse for the sills, C x and C y ,
are assumed parallel to the x and y axes, with C y ≤ C x . The ranges ax and a y
are taken to be equals (ax = a y = a) and the nugget is set to 0. Consider the case
h > a. The variogram values with this model and configuration are

γ (xi , xi ) = 0, i = 1, 2, 3
γ (x1 , x2 ) = C x ,

2 Cx C y
γ (x1 , x3 ) = C45 = q¡ ¢,
C x2 + C y2

γ (x2 , x3 ) = C y .

Now, let λ1 = 1, λ2 = 1, λ3 = −2.


Pn
Then λi = 0 as required and
i=1

X
n X
n
− λi λ j γ (h i j ) = −(2C x − 4C y − 4C45 ),
i=1 j=1

which is negative for some values of C x and C y . For example, with C x = 1


and C y = 0.1, we obtain C45 = 0.1407 and −(2C x − 4C y − 4C45 ) = −1.0371.
In fact, with C x = 1, any value of C y such that 0 < C y < 0.2097 results in a nega-
tive value indicating a nonadmissible model. Note that having C y > 0.2097 results
in a positive value for the corresponding configuration and the particular choice of
weights λ but this is not sufficient to guarantee the model is admissible. According
to Chilès and Delfiner (1999, p. 97) there exist no valid variogram model allowing
the sill to vary continuously with the direction.

NUGGETS ON AN ELLIPSE

Eriksson and Siska (2000, p. 692) mention the possibility of modeling a


directionally varying nugget effect with an ellipse. In this case, using the same
example as in the previous section, an expression identical to the previous ones
for the sills is obtained. Thus, there is no guarantee that a given choice of ellipse
for a nugget effect provides an admissible model.
P1: GCR
Mathematical Geology [mg] pp963-matg-470280 October 14, 2003 15:16 Style file version June 25th, 2002

Letter to the Editor 645

POWER MODEL WITH “SLOPES” ON AN ELLIPSE

Again, consider the simple configuration: x1 = (0, 0), x2 = (h, 0), x3 =


(h, h).
Eriksson and Siska (2000, p. 692, Eq. (9)) proposed power models with
anisotropic “slope” or “exponent” parameters on an ellipse. Consider a slope pa-
rameter on an ellipse:

γ (h) = Cθ h a with 0 < a < 2.

Using the same example as previously,

X
n X
n √
− λi λ j γ (h i j ) = −(2C x h a − 4C y h a − 4C45 ( 2h)a ).
i=1 j=1

The reader will verify that choosing a = 1, h = 1, C y = 0.1, C x = 1, gives C45 =


0.1407 and

−(2C x h a − 4C y h a − 4C45 ( 2h)a ) = −0.8040.

Thus, the model is not admissible with this choice of parameters. Similar counter-
examples can be constructed for the power a on an ellipse that lead to inadmissible
models.

DISCUSSION AND CONCLUSION

Some choices of parameters for the ellipse could possibly provide admissible
variogram models, the isotropic case being a trivial example. The counter-examples
presented simply stress the need to develop simple sufficient admissibility condi-
tions ensuring correct specification of the elliptic parameter model. This task is not
trivial. In the meantime, except for the range parameter, care must be exercised in
adopting the author’s proposal of models having parameters varying directionally
on an ellipse.

REFERENCES

Chilès, J. P., and Delfiner, P., 1999, Geostatistics: Modeling spatial uncertainty: Wiley, New York,
695 p.
Christakos, G., 1984, On the problem of permissible covariance and variogram models: Water Resour.
Res., v. 20, no. 2, p. 251–265.
P1: GCR
Mathematical Geology [mg] pp963-matg-470280 October 14, 2003 15:16 Style file version June 25th, 2002

646 Letter to the Editor

Cressie, N. A. C., 1991, Statistics for spatial data: Wiley, New York, 900 p.
Eriksson, M., and Siska, P. P., 2000, Understanding anisotropy computations: Math. Geol., v. 32, no. 6,
p. 683–700.
Yaglom, A. M., 1962, An introduction to the theory of stationary random functions: Prentice-Hall,
Englewood Cliffs, NJ, 235 p.

Denis Marcotte
Department of Civil, Geological and Mining Engineering
École Polytechnique de Montréal
C.P. 6079, Succ. Centre-ville
Montréal, Canada, H3C 3A7
e-mail: denis.marcotte@polymtl.ca

You might also like