regularly-spaced coordinates

Harry Jenter (
Wed, 25 Mar 92 9:55:30 EST writes:
> I suggest we
> approach the question carefully and look around before fixing this
> problem locally. Quick fire-fighting work always turns out to be
> painful.

This is certainly my opinion also.  I just thought that I would throw
out a couple of examples to "stir the pot".

                        < Text deleted >

> Harry Jentner said he would not like to have assumptions. Me either.
> But netCDF already does have one. It is assumed all data are given
> over rectangular grid.  

I think that netCDF does not require this assumption to be made.  It is
just more intuitive for storing rectangular or, more generally,
quadrilateral grids, without introducing one's own conventions. (Note:
I stick to the 2-d grid analogy for the rest of this post.  The
arguments can be extended to 3 or more dimensions, but I don't know the
words for "equivalent of quadrilateral" in 3 or more dimensions.)
There seem to be no assumptions regarding grid shape.  There are also
seem to be no conventions regarding grid shape.

I think, however, that the issue of storing non-quadrilateral grids is
a more difficult issue to address than the regularly-spaced-coordinate
issue. In fact, it is, in a sense, the opposite problem.  For
regularly-spaced coordinates, one can reduce the information stored in
a netCDF file relative to the "intuitive" method of storing a 1-d array for
each coordinate, and still can reproduce the grid.  For non-quadrilateral
grids or even quadrilateral-but-not-rectangular grids, one must
increase the amount of information stored in the netCDF file in order
to reproduce the grid.  For non-quadrilateral grids, information about
what nodes are connected must be stored.  For curvilinear grids, the
coordinate arrays must have more than one dimension.

                        < Text deleted >

>   What Harry suggests is a way to indicate if the given grid in one
> dimension (that may happen to be TIME) is equally spaced. But what
> about the other functions? What if a dimension is logarithmical or
> derived with any other function?

This is a good point. I forgot about grids that are stretched according
to a function.

>   And further, data points are not necessarily given over a
> rectangular grid. The grid could be just genarally quadrilateral, or
> may be triangular as often is. Or what about a general polygonal grid?

I do not know much about non-rectangular grids, but I have heard that
the HDF Vset has a mechanism for storing them.  I have also heard that
the DX format for IBM's Data Explorer product has provisions for
storing non-rectangular grids (I think that Lloyd Treinish, a frequent
contributor to netcdfgroup is helping to design this format.).  Does
anyone know more about these?

> Will we introduce more and more assumptions?  What we need is to think
> over the whole issue, introduce a CONCEPT instead of AD-HOC attribute
> or whatever solutions. 

You don't have to YELL. I agree with you.  I also think that the
netcdfgroup is an appropriate forum for discussing these ideas.

Harry L. Jenter              
U.S. Geological Survey                 COM: (703) 648-5916 FTS: 959-5916
Mailstop 430, National Center          "Sometimes you're the bug.
Reston, Virginia 22092                   Sometimes you're the windshield."