[R-sig-Geo] Bayesian Kriging

[Paulo Justiniano Ribeiro Jr]

Stefano

> Dear List,
> 	I would like to try (for the first time!) the Bayesian Kriging Approach using
> *geoR* package, but there is a point that is not so clear to me: since I've
> to transform my data in order to obtain a Gaussian dataset (it's right?),
> 	i) variography analysis has to be taken on this transformed dataset?

if you are using the Bayesian approach inference on the model parameters
will be besed on the posteriors, and not on the variogram fitting.
This way the variogram analysis can be useful as an exploratory analysis,
and yes, in using the same transformation intended for the Bayesian
analysis.
I tend to use boxcox() to find the suitable transformation

> 	ii) if I let krige.bayes() function to perform the transformation by the
> means of "lambda" parameter, the output about model's parameters are referred
> to the transformed data or orginal data?

If you enter with data in the original scale and a value lambda, the
results will be given in the original scale, i.e. back transformed towards
the end. This is because the simulations from the predictive distribution
(conditional simulations) obtained in the transformed scale and
backtransformed

hopethios helps

Best
P.J.


>
> Thanks a lot for any suggestion or advice!
> Have a nice day,
>
> stefano
>
> --
> Stefano Pegoretti, PhD
> Università degli Studi di Trento (Italy)
> Dipartimento di Fisica (Physics dept.)
> email: pegorett at science.unitn.it
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>