now I want to sum the answers to my question about indicator kriging (see
>Date: Wed, 29 Oct 2003 14:01:03 +0100
>From: Katharina Buchenau <katharina.buchenau@...>
>Subject: question about Indicator Kriging
>I have used indicator kriging to interpolate data with lots of zero
>values. How can I evaluate the quality of my prediction? Is the RMS useful
>in this case?
>And what does the kriging variance/standard error of indicators say? Is it
>a confidence interval with the unit % as I predict probabilities?
>I would be happy to get some answers!
Here is an abstract of the answers:
The short answer is 'no'.
I am afraid you just have to take the kriging standard
error as a relative measure not an absolute one.
This is one of those places where cross validation is
a useful tool as you can look at the 'true'
distribution of kriging errors with the real data.
If so, take the following for each sample:
(actual value - estimated value)/kriging standard
then draw a histogram of the values. If this histogram
is approximately Normal (Gaussian) you can do
Unless your data is very unusual, you will probably
have a histogram which looks like a 'U'.
Thank you all for your answers!!
Dipl.-Geogr. Katharina Buchenau
Fon: +49 (0)40 - 23 90 95-99
Fax: +49 (0)40 - 23 14 49
[Non-text portions of this message have been removed]