Skip to content

sill not reaching 1 in a climatological variogram

5 messages · Edzer Pebesma, subash

Original
subash
# 
Hi All,

I constructed a climatological variogram using 'gstat' package in R at daily
scale using 2 monsoon(JJAS) data. The *exponential* model was with variogram
parameters  

nugget - 0.04243
psill - 0.96633
Range - 16.88 Km

<http://r-sig-geo.2731867.n2.nabble.com/file/n7583465/climat_vario.png> 

I have couple of questions:

(1) Why is the psill not reaching 1, when the each day rain has been
normalized?
(2) Why is there a dip in the variogram at 50Km?

Your help will be highly appreciated.

Thanks & Regards
Subash







--
View this message in context: http://r-sig-geo.2731867.n2.nabble.com/sill-not-reaching-1-in-a-climatological-variogram-tp7583465.html
Sent from the R-sig-geo mailing list archive at Nabble.com.
Edzer Pebesma
#
On 05/02/2013 06:30 PM, subash wrote:
total sill, i.e. psill + nugget are supposed to approach the variance of
the field; they seem to do so. I also believe that this only holds for
the ergodic case, where you would make your spatial domain infinite.
I wouldn't see this as a structural property; consecutive sample
variogram estimates are dependent as they share the same observations
when forming pairs.

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      subash
    

    

      
      #
    

  


  

      
Thanks Edzer  for the prompt reply.

Can you please elaborate on the answer two my 2 question. I could understand
"consecutive sample variogram estimates are dependent as they share the same
observations when forming pairs." I am aware that the same observation are
used when forming pairs but how that would lead to the dip in the variogram.


Regards,
Subash



--
View this message in context: http://r-sig-geo.2731867.n2.nabble.com/sill-not-reaching-1-in-a-climatological-variogram-tp7583465p7583478.html
Sent from the R-sig-geo mailing list archive at Nabble.com.

  
  


  


      

    

    

      
      

        


  

        

      Edzer Pebesma
    

    

      
      #
    

  


  

      
Subash,

We would only expect the sample variogram to be on the variogram model
line when (a) we had an infinite number of observations, taken over (b)
an infinitely large domain.

As we don't have this in practice, sample variogram points vary around
the theoretical line. As these sample variogram points are correlated,
their fluctuation often does not look like random fluctuation -- you
described this as "a dip [...] at 50 km" in your original mail.

        
On 05/03/2013 05:38 PM, subash wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    
3 days later