Re: Number of sampling pairs

["Edzer J. Pebesma" <edzer@pangea.Stanford.EDU>]

Anna Jakomulska wrote:
> I produced variograms for remote sensing data (gridascii file, size:
> 128x128, georeferenced, high resolution data 2,5 m) and variograms seem
> to be OK (cutoff: 300, interval width: 10), however the number of used
> samples is suspiciously high: it starts with 383030 pairs and goes over
> 5,500,000 which seems hardly possible to me.
> 
> Could anyone explain to me, how is the number of samples selected? Can I
> change it manually?

Gstat takes the whole map, it does not sample. If you want it to
use less data, then provide it with less data. This explains the
large number of point pairs found -- your data set has size 16384.
The total number of point pairs you have is 16384*16383/2 = 134209536.

It would be fairly trivial to implement some sampling in gstat, but
when you start thinking of it there's a number of options you would
it to have too: random, systematic, stratified,... That's probably
why I never did this into gstat's core.

I suddenly remember that I wrote a sampling class for my students
last year. Try
  gstat -e sample
and read about it at http://www.geog.uu.nl/gstat/sample/
This may do your job. It'll work with any grid map format.

> Another question: how are the default values of cutoff and interval
> width calculated?

One third of the diagonal of the block around your data; this
distance divided by 15. These values can (and should) be adjusted,
and also default behaviour at run time can be adjusted.

There's a manual...
--
Edzer