Dear GRASS folks,
I am interested in ways to perform model validation
on various spatial interpolation schemes.
In particular, I would like to be able to do a
resampling (drop one) analysis of prediction error.
That means that the j-th point in the data set is
dropped, then the interpolation run again to see
what is predicted for the location of point j in the
absence of data for point j. The procedure is repeated
for each point in the data set. Thus, N interpolations
on N-1 points must be performed. If the interpolation
method is reasonable for the data set being interpolated,
then the actual value of point j and the predicted value
for its location should be reasonably close. When graphed,
they would fall near the line y = x.
Where should I be looking?
Eric Kostello