[R-sig-Geo] (residual) variograms and trend adjustment

Tim Richter-Heitmann trichter at uni-bremen.de
Tue Sep 16 15:16:27 CEST 2014


I just had a talk with a reviewer of my thesis. He strongly criticized 
the way i was interpreting variograms without previous trend adjustment.
In fact, i am following the route in "Applied Spatial Data Analysis with 
R", chapter 8. On page 218, i think its stated the basic assumption of 
the variogram is that the variance of the random function Z is solely
based on the separation distances. However, it was said that variograms 
should only display the residuals of Z (y=mx+b+e), never the complete 
term. I first should identify non-spatial trends in the data.
I find this very difficult without knowing the autocorrelation of a 
dataset. So, the real question is, what comes first: Autocorrelation or 
pairwise correlations of observations?
Starting with page 230 of the same book, there is a tutorial for 
residual variogramming, but i think this assumes that i already know the 
functions i want to test, or?
How would you deal with this remark?

Thanks, Tim

More information about the R-sig-Geo mailing list