Wenli-
you are right, an (n-x) D case always can be embedded into n-D space for
x>0. However, from a data structure viewpoint I would distinguish: while
a 1-D timeseries can be seen as { (x,y,t,v) | x=x0, y=y0 } in practice
one still would store a 1-D array plus the location of the sensor, that
is: (x0,y0).
In the end, as you know I am most sympathetic with this your notion of
n-D arrays (ie, coverages with an n-D domain) as it allows to unify (and
correlate) all these different data.
-Peter
Wenli Yang wrote:
Peter and Kerian,
I think that it's a good idea to start collecting requirements and to investigate what WCS can and can't do. I believe that until this point, WCS deals with "grid coverage" only but not more general coverage types including feature coverage (vector data types). Thus, WCS essentially is for N-dimensional data arrays, especially those having two of the N dimensions associated to two near horizontal geographic/projected coordinate reference system.
A 1-D time series data array, e.g., Data(time) at a fixed location can be considered as a special case of 3-D, with the two horizontal Ds fixed at one point, e.g., Data(time,lat,lon) where the dimension sizes of lat and lon are one.
A 2-D time series vertical profile data array, e.g., Data(time,pressure) at a fixed location can be considered as a special case of 4-D, with the two horizontal Ds fixed at one point, e.g., Data(time,pressure, lat,lon) where the dimension sizes of lat and lon are one.
Similarly, a data array of any n-D fixed at one location can be considered as
(n+2)-D by adding two spatial Ds having dimension sizes being one.
Multiple n-D data arrays at multiple locations or one (n+1)-D data array, where none of
the "n-D" is horizontal spatial dimension, can be considered a (n+1)-D coverage
with one of the dimension indicating a spatial location. This dimension that indicates
spatial location may be an location ID, lookup table, etc which can be associated with a
2-D near horizontal position. This dimension can be indicated using an Engineering
coordinate axis such a locationID, profilePosition, etc. How can such n-D or (n+1)-D
data array, or coverage, be handled in WCS? How about more complicated data structures
such as meshes described in Kerian's email? I guess that the best way is to list use
cases/requirements as Peter suggested.
Wenli
