Dear Jan W Merks
For a set of N input samples I can do by kriging the only ONE estimate and compute kriging variance for this single estimate.
So, kriging variance is a variance (derived from the model) of single unknown true value minus single weighted average.
Kriging variance isnt any variance of a set of weighted averages. We dont need any other single weighted average.
So, kriging variance is a variance (derived from the model) of single unknown true value minus single weighted average.
Kriging variance isnt any variance of a set of weighted averages. We dont need any other single weighted average.
FUNCTIONAL DEPENDENCE
There are the rivers on the Earth.
There are the towns on the Earth.
The man needs a water to live.
The man lives in a town.
The rivers and towns are functionally dependent.
So, we can see the towns at the rivers (or the rivers inside the towns), there is some constraint.
I dont think that kriged estimates are functionally dependent since I can do by kriging the only one estimate at any coordinate I just want. It means that kriged estimates dont see each other, there is no constraint.
There are the towns on the Earth.
The man needs a water to live.
The man lives in a town.
The rivers and towns are functionally dependent.
So, we can see the towns at the rivers (or the rivers inside the towns), there is some constraint.
I dont think that kriged estimates are functionally dependent since I can do by kriging the only one estimate at any coordinate I just want. It means that kriged estimates dont see each other, there is no constraint.
DEGREES OF FREEDOM
For infinite sample the variance in the global case = sum of deviation squares divided by the size of sample (all weights are equal).
For finite sample the deviation squares are weighted by identical weights ONLY in the case of gaussian noise.
Experimental variance = sum of deviation squares that is scaled by degrees of freedom can be applied only in the case of gaussian noise. So, such variance is useful for analysis of grades in the school not in the mine.
Increasing (by degrees of freedom) the denominator of weights we can blow the confidence intervals to infinity. In such case the forecasting always will match the observation. But its not goal of (geo)statistics.
For finite sample the deviation squares are weighted by identical weights ONLY in the case of gaussian noise.
Experimental variance = sum of deviation squares that is scaled by degrees of freedom can be applied only in the case of gaussian noise. So, such variance is useful for analysis of grades in the school not in the mine.
Increasing (by degrees of freedom) the denominator of weights we can blow the confidence intervals to infinity. In such case the forecasting always will match the observation. But its not goal of (geo)statistics.
F-TEST
As above (degrees of freedom, sum of deviation squares), postulated F-TEST for so-called spatial dependence in fact is the test for trend (drift) in high-noised data (gaussian noise) and can be useful to analyse the pupils progress in the school not to analyse spatial dependence in the core.
The famous F-TEST for Clark's hypothetical uranium data in fact is the test for trend in the data under assumption that there is no correlation structure in the data.
The famous F-TEST for Clark's hypothetical uranium data in fact is the test for trend in the data under assumption that there is no correlation structure in the data.
KRIGING ESTIMATOR
Kriging estimator, for gaussian noise, simplifies to the least-squares estimator that was introduced to statistics by Mr. Gauss.
Best Regards
Tomasz Suslo
Tomasz Suslo
How low will we go? Check out Yahoo! Messengers low PC-to-Phone call rates.
