gstat: Variogram with non-Euclidean Distances

Julia,

I can't help with implementation, but Waller&  Gotway discuss non-euclidean
distances in geostatistics briefly (p321).

Same other information on the subject may be found in Shabenberger and Gotway 
("Statistical Methods in Spatial Data Analysis"). The authors state that 
non-euclidean distances may result in non-positive defined covariance matrices, 
hence invalid covariance functions (though multi-dimensional scaling can come to 
the rescue in some cases)... looks like a tricky subject.

Regards,

Luca Morandini
http://www.lucamorandini.it