[R-sig-Geo] krige.bayes output
Edzer J. Pebesma
e.pebesma at geo.uu.nl
Mon Jun 25 10:44:14 CEST 2007
Is it possible to compute them experimentally, by computing the variance
of the back-transformed simulations for each prediction location?
Paulo Justiniano Ribeiro Jr wrote:
> Dear Giovanna
> For certain values of the Box-Cox transformation it is not possible to
> compute kriging variances.
> Therefore a measure of the uncertainty of the predictions is computed by
> where Q3 and Q1 are the 3rd and 1st quartiles
> Paulo Justiniano Ribeiro Jr
> LEG (Laboratório de Estatística e Geoinformação)
> Universidade Federal do Paraná
> Caixa Postal 19.081
> CEP 81.531-990
> Curitiba, PR - Brasil
> Tel: (+55) 41 3361 3573
> Fax: (+55) 41 3361 3141
> e-mail: paulojus AT ufpr br
> On Thu, 21 Jun 2007, giovanna jona lasinio wrote:
>> Dear all,
>> I'm running a bayesian kriging using krige.bayes in the geoR library.
>> I'm using a linux machine with R 2.4. I'm running a chain of annual
>> surfaces estimation and I'm using every year's output to build the prior
>> for the next one through the function post2prior (always, geoR library).
>> My predictive distribution is back-transformed (Box-Cox) from Gaussian
>> (log transformation).
>> My problem concerns the krige.bayes output, in the predictive list I
>> find an element named uncertainty of the same length as my surface. I
>> was not able to find any explanation of what is in it searching the help
>> and the mailing list archive.
>> Does anyone know what is it?
>> Giovanna Jona Lasinio
>> R-sig-Geo mailing list
>> R-sig-Geo at stat.math.ethz.ch
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
More information about the R-sig-Geo