anisotropic models vs. detrending

My end goal is to create the best sample spacing for a new
design based on the variogram. ?In the anisotropic case I would create a
rectangular grid with stations in one direction closer than in the other.
?In the other "trend" case I would create a square grid. ?This is the
question I am trying to answer.

Are there phenomenological reasons/analoga to assume an external
drift, a trend or variogram anisotropy? In case of anisotropy, do you
already know the principal axes? Otherwise you cannot design a
sampling oriented in the right way, and this discussion would be
rather pointless.
Furthermore, a kriging of the residuals doesn't imply that those are
isotropic in geostatistical meaning, in my humble opinion. So, just
find the model which best fits your problem.

After all, estimation variance (which you want to minimize for the
sampling design) depends on data locations and spatial model, so noone
can easily say what is better between kriging with external drift or
for example with zonal anisotropy with linear model. This is actually
a typical exercise for undergrads.

Scion
Thr