[R-sig-Geo] variogram using gstat

Edzer Pebesma edzer.pebesma at uni-muenster.de
Thu May 8 08:10:48 CEST 2008

 From the gslib book, I understand that there is a lag, and a lag 
tolerance, and a cutoff. lag 0 is 0, the lag 1 is <0,tol], lag 2  
<lag2-tol,lag2+tol>, lag3 <lag3-tol,lag3+tol>, etc. This allows for 
overlap in the lags, if tolerance is larger than half the difference 
between lags.

gstat uses regular lag (the first one is not shorter than the latter):
lag1 = [0,width>
lag2 = [width,2*width>
lag3 = [2*width,3*width>
up to the cutoff. So, there's no possibility of overlapping lags, and a 
different organization of them.

tol_hor is the horizontal direction tolarance, and has nothing to do 
with width and cutoff.

note that gstat does not have a bandwidth concept, so for comparison 
purposes you should set this to a very large value.

> Dear sir,
> I spent lot of time to figured it out what causing the difference in 
> result of R/gstat and gslib but didn't get any logical explanation. In 
> my opinion the source of difference is the cutoff and width.
> I would be thankful if you give me the in depth explanation of these 
> two parameters used in gstat or you can suggest me the suitable value 
> i should adopt for my case. In my data set the diagonal is 2771.89 so 
> 1/3 of it is 923.96 i believe this is the default cutoff. but what 
> about width, i don't get the logic of this function? where as in gslib 
> we have to define the limit to what level pairs will be compared and 
> the distance between the pairs and its tolerance. for my case in case 
> of directional semivariogram the distance between is 12m and tolerance 
> 6m and the extent is 923.96m that is i have to define numbers of lag 
> as 923.96/12 ~77 nos. i didn't see the tolerance for lag in gstat, i 
> think this has been done with build in script which incorporates with 
> tol.hor command?? One more querry; is it correct to use same cutoff 
> and width for omnidirectional and directional semivariogram 
> (especially in 45 and 135 directions)?? as the lag in 0 and 90 
> directions is 12 where as in 45 and 135 is 16.97m.
> Though my question seems very basic, hope you find sometime to answer them
> Thanking you
> Pujan
> ------------------------------------------------------------------------
> > Date: Wed, 7 May 2008 18:27:44 +0200
> > From: edzer.pebesma at uni-muenster.de
> > To: regmi_pujan at hotmail.com
> > CC: thierry.onkelinx at inbo.be; r-sig-geo at stat.math.ethz.ch
> > Subject: Re: [R-sig-Geo] variogram using gstat
> >
> > PUJAN RAJ REGMI wrote:
> > >
> > > Dear sir,
> > > Ya, Now I am comparing the results from gslib and R/gsat. The 
> shape of
> > > variograms exactly overlaps but the position of points varies, I 
> don't
> > > know why? or it is normal! I first checked the result of
> > > omnidirectional variogram but still to have checked for directional
> > > cases. I hope the plot will overlap exactly on these cases too!!
> > > Thanking you
> > > Pujan
> > >
> > Does gslib also compute average distances over the point pairs, or does
> > it give the middle point of the distance bin?
> > --
> > Edzer
> ------------------------------------------------------------------------
> Windows Live SkyDrive lets you share files with faraway friends. Start 
> sharing. 
> <http://www.windowslive.com/skydrive/overview.html?ocid=TXT_TAGLM_WL_Refresh_skydrive_052008>

More information about the R-sig-Geo mailing list