Skip to content

Factorial kriging in GStat - tips for implementation?

3 messages · Tim Peterson, Edzer Pebesma

Tim Peterson
# 
Hi all,

I'm looking to undertake factorial colocated cokriging with GStat but 
factorial kriging does not appear to be an available option in GStat (eg 
help.search("factorial kriging") ). There was a post in 2007 requesting 
this feature (http://marc.info/?l=r-sig-geo&m=119810046000747) and a 
proposed approach in 2000 
(https://www.mail-archive.com/gstat-info at geog.uu.nl/msg00045.html) but 
no updates since. As far as I can tell, the required changes would be to 
'gls.c' lines 412-423, specifically the call to GCV0 for calculating C0 
(right-hand side of the kriging matrix) should be edited so that only 
the covariance for the required factor is calculated.

Anyway, before I consider adding this feature to GStat, I'd be grateful 
if anyone could share tips or thoughts. Lastly, any changes I make will 
(of course) be freely shared.

Many thanks,

Tim

----------------------
Dr. Tim Peterson

The Department of Infrastructure Engineering
The University of Melbourne, 3010 Australia
T: +61 3 8344 9950, M: +61 0438 385 937

Dept. profile : 
http://www.ie.unimelb.edu.au/people/staff.php?person_ID=141135
Research Gate : https://www.researchgate.net/profile/Tim_Peterson7
Google Scholar: 
http://scholar.google.com.au/citations?user=kkYJLF4AAAAJ&hl=en&oi=ao
Edzer Pebesma
#
On 08/27/2015 02:48 AM, Tim Peterson wrote:
Not being an expert in factorial kriging, I also think that this is what
is needed. A challenge may be to make the implementation simple and
clean, both from the perspective of the user as well as from the code
maintainer.

If the only goal is to suppress the nugget effect, as the poster in your
first link mentions, try replacing a "Nug" variogram model with an "Err"
variogram model.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Tim Peterson
    

    

      
      #
    

  


  

      
Thanks Edzer. Good to hear that I seem to be looking at the correct part 
of the code. And, I agree that the implementation must be simple and the 
code changes clear.  The best approach I can see for this is to add an 
option to the 'set' input for krige() that specifies the variogram 
number that is to be used for the calculation of the right hand C0.

        
On 27/08/15 16:38, Edzer Pebesma wrote:

            

      
      
      
    

        
Also, I need factorial kriging for estimating only the large scale 
effect (ie the longer range variogram for the primary variable from a 
set of >=2 variograms). This is for the eventual use to undertake 
indicator simulations for a highly non-stationary  (first order) system, 
specifically basin scale water table elevation.