[R-sig-Geo] kriging with external drift on precipitation data

[Anna Fornasiero]

Hi I'm not a geostatistic expert.
I'm trying to interpolate raingauges precipitation measures using
radar precipitation as covariate.
The resulting field obtained by KED is filled of precipitation even
over the sea where neither raingauges nor radar have measured
precipitation. The result changes if I reduce the maximum distance of
the  variogram computation   (i.e. 20 km), but I have no idea on what
produces this change. Using 20 km of maximum distance the variogram
shape is poor.
The variogram shape is well defined starting from 50 km of maximum
distance, but in this case the precipitation is spread over the sea .
I didn't note any substantial difference changing the fitting model of
the variogram: exponential, spherical or gaussian.
Any suggestion?
Anna