May be worth adding here that, despite algorithms and the
"funny naming conventions" (good descrition Edzer!)
SK and others are diferent in the following way:
SK, as described in Edzer post, assumes you **know** the mean, in other
words, there is no uncertainty about it.
On the other hand, variants such as OK, UK, KED, SKlm uses (explicitly
or implicitly) estimated means.
Therefore, such uncertainty has to be propagated and reflected in the
predictions.
Suposse the fixed mean in SK is the same as the (implicitly) estimated by
OK. The point predictions will be the same, however, the uncertainty
around them will not (and should not) refleting the uncertainty (or lack
of it) in the process mean.
The prediction variance expressions for SK ond OK will reflect this
whatever the kriging neighborhood is used.
best
P.J.
On Tue, 26 Jan 2010, Edzer Pebesma wrote:
Oh, geostatistics and its funny naming conventions!
I see local models vs. global models as a completely different modelling
aspect (model decision, basically) then the SK/OK/UK differences. When
building on the same tradition / body of literature you quote: in that case
KED would be a special form of UK, having only a single non-coordinate
predictor called 'external drift'.
In my eyes (and that of the literature with more mathematical statistical
grounding, such as Cressie 1993 and others), the difference between SK on the
one hand and OK/UK on the other is that SK assumes that you know the mean or
mean structure. SKlm is then residual kriging added to a known mean function.
In the gstat R package you obtain SK by specifying a beta value (for the
mean); SKlm by specifying one or more predictors and passing the (known)
regression coefficients as beta; you obtain OK/UK by not specifying beta; a
formula ending on ~1 results in OK with an unknown mean only.
Ah, and then SK = simple kriging, OK = ordinary kriging, UK = universal
kriging.
--
Edzer
Cutberto Uriel Paredes Hernández wrote:
Dear Edzer,
Would it be correct to say then that if a neighbourhood is specified
in the krige command the result would be that of Kriging with an
External Drift (KED), otherwise it would be that of Simple Kriging
with varying local means (SKlm)?
Apologies for posting on this thread but I was about to post a
similiar question.
Thanks, Cutberto.
2010/1/26 Edzer Pebesma <edzer.pebe...@uni-muenster.de>:
Yes, that is right.
Els Verfaillie wrote:
Dear list,
I want to use Kriging with an external drift for a sedimentological
dataset
of grain-size that has a linear relation with the depth.
Am I correct that when I set a 'maxdist' using the krige command, that a
trend for the primary variable (grain-size) is calculated as a local
linear
function of the secondary variable (depth)? Is this function thus
different
for each interpolation window?
d50.ked.dir50 <- krige(D50F~depth, locations=ds50, newdata=Depth,
model=d50.fit.var.50, nmin=2, nmax=16, maxdist=9000)
Thank you for your help.
Best regards,
Els Verfaillie
______________________________________________
Dr. Els Verfaillie
Carto-GIS cluster
Ghent University (UGent) - Department of Geography
______________________________________________
[[alternative HTML version deleted]]
_______________________________________________
R-sig-Geo mailing list
R-sig-Geo@stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
--
Edzer Pebesma
Institute for Geoinformatics (ifgi), University of Münster Weseler Straße
253, 48151 Münster, Germany. Phone: +49 251 8333081, Fax: +49 251 8339763
http://ifgi.uni-muenster.de http://www.52north.org/geostatistics
e.pebe...@wwu.de
_______________________________________________
R-sig-Geo mailing list
R-sig-Geo@stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
--
Edzer Pebesma
Institute for Geoinformatics (ifgi), University of Münster Weseler Straße
253, 48151 Münster, Germany. Phone: +49 251 8333081, Fax: +49 251 8339763
http://ifgi.uni-muenster.de http://www.52north.org/geostatistics
e.pebe...@wwu.de
_______________________________________________
R-sig-Geo mailing list
R-sig-Geo@stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
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
_______________________________________________
R-sig-Geo mailing list
R-sig-Geo@stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
