Page 1 of 1

[SOLVED] questions about variogram fitting

PostPosted: Wed Nov 20, 2013 9:18 am
by Maxime Beauchamp
Hi,

I'd like to post on the forum two questions of my colleagues from INERIS.
When fitting the variogram, we obtain several objects. For instance :

data.model=model.auto(data.vario,struct=c('Nugget Effect','Spherical','Exponential'),auth.aniso=FALSE)
In their example, data.model contains a nugget effect and an exponential structure that are two different objects.

So here is their questions :
1) Is the fitting able to introduce two structures of the same kind (I guess this issue comes from an older kriging program developed in gstat they would like to convert in RGeoS and that consider for instance two spherical/exponential structures of small and large range) ?
2) If so, is it possible to fix the maximal number of structure by kind of structure ?

Kind regards,

Maxime

Re: questions about variogram fitting

PostPosted: Wed Nov 20, 2013 6:52 pm
by Didier Renard
Hello Maxime

Concerning model.auto, theu user can define the set of basic structures (exponential, spherical, nugget effect, ...) that should be involved in the global fit.
There is no problem in using several times the same basic structure (say twice a spherical). In fact, internally, these two sphericals will be initialized differently (with different ranges and sills). Then the two basic structures will evolve independently as if they were different in type.

I do not quite understand the second question. Do you mean that you want to specify the maximum number of a given basic structure (say spherical) in the final result. If so, it suffices to enter the name "Spherical"as many times as this maximum. The only possibility is that model_auto will suppress some (or all) of these basic structures (if their sill is considered as negligeable).
I recall that, on the opposite, you can ask not to suppress these negligeable basic structures by specifying the minimum value of the sill beyond which a structure is eliminated: giving a threshold of 0 will cancel the filtering option.