Skip to content
Prev 23966 / 29559 Next

Ordinary kriging variance of the prediction with error structure

Hello Ruben,

Thanks for your answer,
I tried to check your reference, but still I'm having the same issue,

Cressie 1993, he says in 3.2.27:
var(S|Y) = Sum(w_i)*V_io + m - c_me

Being:
[w_i, vector of weights]
[V_io, Semivariogram value from the sampled point to the predictied]
[m, lagrange multiplier]
[c_me, nugget (due to measurement error)]

This is equivalent to the notation I posted previously as C(h>0)= partial
variance + nugget - V(h>0)
[C, Covariance function]
[V, Semivariogram function]

Substituting V_io and knowing that Sum(w_i)= 1, this gives:

Var(S| Y) = partial variance - Sum(w_i) * C_oi - m
[C_oi, vector Covariance between predicted and sampled coordinates]

This is the expression that I'm working with, in which the nugget should
not appear (unless x_o=x_i) isn't it?

I'm working with the covariance matrix as the nugget will only affect
C(h=0) and not C(h>0) which is a property I need. And then the layout of
the kriging system should be the same yet adding a nugget term in the
diagonal of the covariance matrix.

Am I misunderstanding something?

Thanks and kind regards
Antonio
On 5 February 2016 at 10:56, rubenfcasal <rubenfcasal at gmail.com> wrote: