[R-sig-Geo] lognormal kriging and backtransfomation

[Patrick Giraudoux]

Dear Edzer,

We are revising a paper and it appears that one of the referee ask 
(innocently ?) about how we have back transformed the predicted values 
obtained from a lognormal kriging. Quite naively we initially did it 
using a simple exponential, but the question drove us to think this 
question was not that innocent, and to some more reading... It happened 
that things are not so simple and that a quite complex formula has been 
proposed in Cressie (2006) Block Kriging for Lognormal Spatial 
Processes. Mathematical geology 38(4): 413-443 and used at least in 
Rawlins et al (2006). It is however a far beyond my understanding using 
"Lagrange multipliers" and so on...

Do you know if this back transformation has been implemented in gstat or 
elsewhere ?

If not, any though about how to do it from gstat outputs ?

Thanks in advance for any hint,

Patrick