Re: cross validation

["Edzer J. Pebesma" <e.pebesma@geog.uu.nl>]

Guillaume Descamps wrote:
> 
> Hi gstat users,
> 
> My first question on this mailing list. Does anyone know where I could find
> documentation describing cross-validation file made by Gstat. I know what
> cross-validation is but I would like to have more info on the file itself
> and how to interpret it.
> 
There probably isn't that much information. However, the cross validation
done by gstat is a simple leave-one-out cross validation: for each data
point do:
  - leave that data point out
  - predict a value at the data point location
  - output coordinates, observed (left out) value, predicted value, and
    prediction variances.

The brief summary report gives descriptive statistics for observed,
predicted, errors (obs-pred), prediction standard errors, and Z-scores
defined by:

Z = (obs - pred)/(pred.std.err)

Z-scores should have zero mean and unit variance.

IIRC, in cokriging situations only the first variable is cross validated,
and secondary variable data at prediction locations are not left out.

Is there more you would like to have documented?
--
Edzer