[R] problems with variogram

Jon Olav Skoien j.skoien at geo.uu.nl
Wed Jun 4 15:03:32 CEST 2008

Laura Saltyte wrote:
> Hello,
> I have data at 10 locations, in each location there are time series
> (T=56). Question is: when I’m fitting variogram what happens with those
> measures in each location? Are they taken as repeated measures? It's very
> important for my to know this
> Thanks a lot
Hi Laura,

I think this depends on which package you use for your variogram 
function. In variogram in package gstat, you can define your wanted 
behavior by manipulating the dX-parameter. Generally, the easiest way 
you can find an answer to your question if you dont get any help from 
the help-file of your variogram function, is to look at the number of 
pairs in the calculated sample variogram;
- if you have relatively few pairs altogether (45 or less for your 10 
locations, depending on the cutoff for the variogram function), the 
variogram was calculated from the average of the observations, or from 
one of the values (probably the first)
- if the number of pairs is lower than 45*56 (the length of your time 
series), the variogram was probably calculated as an average of the 
variograms from each separate time step
- if the number of pairs is "high", the variogram was probably 
calculated by comparing all observations independent of time steps with 
each other.

If you realize that your function uses the last method, I would 
recommend you to switch to one of the first two. Unless the temporal 
correlation is very high, the spatial correlation is likely to be 
(strongly) underestimated in this case.

For questions about spatial statistics, the list r-sig-geo will often be 
more useful than r-help.

Good luck,

More information about the R-help mailing list