Giovanni De Ferrari wrote:
>
> - I could't exactly understand what s stands for in variogram model c
> Mod(a,p,s) while defining zonal anisotropy. I mean: in Gstat manual it
> is said to use a large anis ratio (s) while defining zonal anisotropy
> but the subsequent example shows a very little ratio: which of the two?
s needs to be in <0,1], for a to be the major range -- if s were allowed
to be larger than 1, a would be the minor range and this would be more
confusing to explain.
zonal anisotropy is `tricked' by geometric anisotropy as follows: use a
very large (that is, practically infinite) range in the direction where
the zonal structure is absent, and define s such that a times s is the
range in direction where this structure is present (i.e., perpendicular to
where it's absent).
> Can s be interpreted as sill ratio instead of range ratio in this case?
No -- sills are equal in all directions, but they are never reached within
your study area in the direction where the structure is not present.
> - Is it possible to model both zonal and geometric anisotropy in Gstat,
> having both of them in my data set, sometimes in the same direction,
> some others in different ones? Which is the way to accomplish it?
Yes: use the sum of two simple models, e.g.
1 sph(30, 45, 0.5) + 1 sph(30, 2e5, 1e-5); # geometric + zonal
> - Is it correct to fit the variogram model once I found out which type
> of anisotropy I have and established model parameters (a,p,s known)? If
> the case, should I fit in total direction or along p direction (I think
> total direction, because the model already contain an angular definition
> of main continuity -p)? If not the case, should I use it as it is to
> kriging?
Gstat does not fit anisotropic variograms automatically. The only
way to go is transform your coordinates such that the variogram
becomes isotropic, and then fit the variogram parameters (ranges
and sills only).
You can try and find the major and minor directions, fit variograms
in these directions, and rework the fitted parameters into an
anisotropic variogram.
> - In the last case, if I don't have to fit the model and I've a
> geometric anisotropy, how can I choose the correct sill, given that
> they're not the same in both main range direction and minor direction?
This case sounds like zonal anisotropy to me.
> Should I use the sill of major direction?
In case you want to ignore zonal anisotropy (sills are not the most important
aspect in kriging), I'd suggest to use the average of sills in all
directions.
Hope this helps,
--
Edzer
