[STATSGRASS] gstat: problem with prediction value at point location

[Edzer J. Pebesma]

My guess is that the technique you're referring to (and used by 
Geostatistical Analyst, afaik) is [kriging in] cross 
validation---observations are left out to obtain a kriging estimate for 
which it makes sense to compare it with the observation. Definitely 
distinct from kriging, which does not leave observations out.
--
Edzer

Jarosław Jasiewicz wrote:
> Edzer J. Pebesma napisał(a):
>> This is a well-known property of kriging if locations coincide 
>> exactly: if measured, the best estimate is the measurement itself. 
> Hmm. yes and no. As I remember form other program (like geostatistic 
> analist in arc info) in point value wasn't exact. the difference 
> between exact and predicted values in point location were used to 
> calculate the RMS total (I can be wrong here, I didn't used Arc Info 
> for over 2 years) before final calculation.
>
> I I attached samples, how it looks like with point location.
>
> It was found  during disscution when I presended some data. I was 
> asked about RMS Total for my prediction (in arc info mode  - as sum of 
> differences between exact and predicted values in point location). I 
> tried to do this with sampling points at predicted map. In that way 
> the RMS Total was 0!. I consult result with others and it seems 
> variogram is OK,
> (it was automatically fitted expotential vaiogram, for gaussian 
> variogram result was worse).
>> It's also cause of the problem that you can't have duplicate 
>> observations in kriging. Even with a nugget, gamma(0)=0. You may want 
>> to shift your points or your grid e.g. a few centimeters to get rid 
>> of it.
> it gives the same results
> regards
> Jarek
>
>
> ------------------------------------------------------------------------
>
>
> ------------------------------------------------------------------------
>