[R-sig-Geo] kriging

[Tomislav Hengl]

Dear Dave,

I separate fitting of the deterministic (trend) and residual part of the universal kriging model all
the time. Adding OK of residuals to the trend is fine, as long as the regression model is estimated
using GLS (but many do it even if they use only OLS; the difference is often minor). Both KED and RK
give the same results, but in order to run KED, you need to have the residuals, so you always have
to model the trend-part anyway.

See some code at: https://stat.ethz.ch/pipermail/r-sig-geo/2008-April/003433.html 

In future, please keep the whole history of correspondence.

Tom Hengl
http://spatial-analyst.net 



-----Original Message-----
From: Dave Depew [mailto:ddepew at sciborg.uwaterloo.ca] 
Sent: dinsdag 17 juni 2008 14:57
To: T.Hengl at uva.nl
Cc: r-sig-geo at stat.math.ethz.ch
Subject: re:kriging

Thanks Tom,
I've been able to fit a polynomial function to the data quite well. The 
residuals are behaving (i.e normal distribution and no skewness of 
variance). I'm assuming this means that I could krige the residuals 
(Ordinary K?) and then add the trend back to the predicted residual 
grid? I realize that I won't be able to place confidence limits on the 
predictions, but the data is primarily to show that we might be able to 
use GPS hydroacoustic signals to show macrophyte cover and estimate the 
standing crop. I'm still a newbie when it comes to the theory involved 
in kriging, but I think I am familiar with the basics...the variogram of 
the residuals is a nice spherical model (i.e. it reaches a sill about 
50% or so above the nugget, and there is little scatter). I am assuming 
(perhaps wrongly) that the residuals may be modelled with a variogram 
and then kriged...