3

I am working with the build in dataset meuse, which has 155 measurements of Zinc and the distance to the river "Meuse".(http://rspatial.r-forge.r-project.org/gallery/).

Now I am trying to imitate universal kriging. In R, you first model a variogram, which is used to model the spatial correlation between the observations.

Is there a difference in doing the following:

  • First make a variogram of the zinc measured values, and use universal kriging;
  • First regress zinc on distance, and make a variogram with the residuals.

Is somebody able to explain the deeper thoughts about these two different approaches? Can this be linked with the difference between conditionally autoregressive and simultaneous autoregressive models?

enter image description here

enter image description here

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Kasper
  • 3,059
  • 2
  • 22
  • 37
  • I suppose it is somehow similar to calculating autocorrelation of the dependent variable or on the residues after detrending the original data. Analyzing the residuals often gives a better insight on the spatial dependance because if the autocorrelation is strong the original data are probably spatially trended. – ZKB Sep 03 '15 at 10:48
  • There is a *huge* difference between using the data and using their residuals to estimate a variogram. For a discussion of this, please see the thread at http://stats.stackexchange.com/questions/35510. – whuber Sep 03 '15 at 15:14

0 Answers0