[R] Re:

Edzer J. Pebesma e.pebesma at geog.uu.nl
Sun Aug 17 20:43:24 CEST 2003

Miha STAUT wrote:

> Error in gstat.load.set(object$set) : error occured when parsing 
> command: set average = 1;
> If I do not set average=1 I continue getting the singular matrix error.

"average" is not an option that can be set by "set"; see the full gstat
manual (it is an option of a data statement).

> I see only one solution for this problem. That is o get rid if 
> duplicate data in
> some other way. As I am among the beginners in R and I do not speak other
> programming languages either, I still did not find a successful way to 
> krige
> this data with a matern model varigram (which in my opinion suites best).

The R function in library gstat


may be your friend here. It tells you which observations are spatial
duplicates, after which you can take measures.

> I also tried with the independent version of Gstat (2.4.2 if I remember
> correctly) in combination with GRASS but seems that matern is not 
> implemented in
> it. Is that correct?

There is a way out to this, if you compile gstat linked to the necessary R
libraries to support the Matern variogram. However, it still will not fit
the smoothness parameter present in the Matern model.

> I am not a master of geostatistics either. Is there some other way to 
> fit an S
> shaped experimental variogram. Gaussian model would not fit that good, 
> but at
> least is an approximation.

Matern seems a flexible approach, but gstat does not provide automatic
fitting for it. I do think geoR does provide this, however, although it may
use likelyhood fit for this (all the better, but maybe prohibitively 
slow for
large data sets). Another option is combining a Gaussian model with some
other model (Exp, Sph, Lin, ...) which gstat does support.

More information about the R-help mailing list