[R-sig-Geo] geoR: specification of trend.d and trend.l in krige.control()

[Paulo Justiniano Ribeiro Jr]

Andre

Let's see if I can get this.
Some of the caracters I canot read (probably due to the encoding used
in your email)


On Wed, 12 Mar 2008, Andre Schützenmeister wrote:

> Dear Subscribers,
> I have some trouble in specifying the parameters trend.d and trend.l
> correctly.
>
> As global trend I assume row and column effects which translate into the
> formula:
> trend=~factor(Row)+factor(Column)
>

OK. argment "trend" is used in functionhs such as variogram() and
likfit().
When it goes to krige.conv() you nned to separate trend.d for data
locations and trend.l for predictions locations.
If options are for "1st" or "2nd" the coordinates are used as trends and
them both are internally set to the same.

> Consequently, I obtain the parameters of a the model via variofit() and want
> to do OK-estimation
> for each point of a grid. Therefore, I call krige.cov() n times, for my n
> data points. Each time leaving
> out the point which I want to estimate. Actually, I want to smooth the
> surface using information
> of neighboring locations.

is this cross-validation " leaving one out" . xvalid() may be usefull.

>
> parms <- krige.control( type.krige="ok", obj.model=my.model, aniso.pars=c(
> ai.axis,ai.ratio), trend.d =
.?, trend.l =
.? )
>

I cannot read the options here, but hope above clarifys this otherwise leg
me know.
geoR page have some tutoruials with extra examples on this topic.
are you using the tilde caracter? e.g something like  trend.d = ~datatrend

> I always get the error: “..trend.spatial(trend = krige$trend.d, geodata =
> geodata) :
> trend elements not found.

aparently the objects being passed does not exists...or cannot be found

>
> The geodata object I pass to krige.conv() contains the covariates Row and
> Column.
>

if Row and Columns are your data coordinates consider using trend.d="1st"
and so on

> How can I specify the parameters “trend.d” and “trend.l” correctly in order
> to do the OK-estimation
> location by location?
>

see dosc but basicaly you need the values at each prediction locations
passed as matrices or data-frames (not vectors)

hope this helps otherwise you may consider contact me offlist and maybe
sending reproducible code

best
P.J.


> I would appreciate to get a hint how to do that.
>
> Thanks,
>
>    Andre
>

Paulo Justiniano Ribeiro Jr
LEG (Laboratorio de Estatistica e Geoinformacao)
Universidade Federal do Parana
Caixa Postal 19.081
CEP 81.531-990
Curitiba, PR  -  Brasil
Tel: (+55) 41 3361 3573
Fax: (+55) 41 3361 3141
e-mail: paulojus AT  ufpr  br
http://www.leg.ufpr.br/~paulojus

-------------------------------------------------------------------------
53a Reuniao Anual da Regiao Brasileira da Soc. Internacional de Biometria
14 a 16/05/2008, UFLA, Lavras,MG
http://www.rbras.org.br/rbras53