It is probably important to distinguish between a theoretical model and the dataand to ask exactly what is meant by "multiscale" vs "multiple sills"

First of all, when the data is from only one realization it is not possible to test for non-stationarity. However the data can be used to look for evidence of

non-stationarity:

1. If the data locations are in two space, plot data values against the

horizontal position coordinate (also fit the least squares line), if

the least squares line has non-zero slope this could be interpreted

as evidence of non-stationarity. (Trend in the data), similarly

plot data values against the y coordinate.

2. Does the sample variogram exhibit quadratic or higher order growth?

There are no valid variogram models with these properties and one

explanation of such behavior in the sample variogram is non-stationarity

Non-stationarity (of the model) and trend (a characteristic of the data)

usually have a directional effect (a single point source or sink might

exhibit radial non-stationarity however. Using only data to determine

non-stationarity and anisotropy may make it difficult to distinguish

between a directional effect in the non-stationarity and an anisotropy

in the variogram but these two are definitely not the same and they do

not necessarily go together.

Multiple scale/multiple sills

A nested variogram model with different ranges of dependence (for example as noted in P. Gooavert's comments on simulation) does not shown multiple

"plateaus" in the graph hence one should be a little careful about referring to

multiple sills. For example, Pierre's nested model would not[D[D[D[Dshow plateaus at 100m and at 1000m (he didn't say that it would).

It is illusory to say that the combination of a cardinal sine model and a linear model (power =1) shows many sills, since by the inclusion of the linear

model you are assured that the variogram is unbounded and has NO sill.

A comment about the scale of the problem having an effect is important,

ordinarily one would only use the sample variogram for lag distances up to

half the distance across the region (after that distance the number of pairs

decrease rapidly and also the possible pairings are severely limited geographically). Hence if one or more of the scales of dependence exceed this distance they

may be hard to detect. Also if the order of the non-stationarity is low, the

non-stationarity may not show or have much effect on the sample variogram.

Obviously the physical phenomenon is important in examining these characteristics. Note also that the pattern of sample locations may induce an apparent

non-stationarity or anisotropy.

See "To Be or Not to Be-Stationary", Math. Geology 1988.

DEMyers

Department of Mathematics

University of Arizona

http://www.u.arizona.edu/~donaldm

--

*To post a message to the list, send it to ai-geostats@....

*As a general service to list users, please remember to post a summary

of any useful responses to your questions.

*To unsubscribe, send email to majordomo@... with no subject and

"unsubscribe ai-geostats" in the message body.

DO NOT SEND Subscribe/Unsubscribe requests to the list!