[mpas-developers] attributes in grid generation code

[Michael Duda]

Hi, Everyone.

I've updated the grid generation programs to add the attributes
on_a_sphere and sphere_radius to grid.nc files. I've also updated
the existing grid.nc files available from
http://www.mmm.ucar.edu/people/duda/files/mpas/ and
http://www.mmm.ucar.edu/people/duda/files/mpas/single-tracer.

If you'd like to add the two new attributes to any grid.nc files
that you might have already, I've placed a simple Fortran program
to do this at
http://www.mmm.ucar.edu/people/duda/files/mpas/add_sphere_atts.f90

I think the next step will be to add code in the MPAS
infrastructure to read and write these attributes from and to
files. Ideally, we would have some general mechanism -- perhaps
through additional registry syntax -- to specify that arbitrary
attributes be read and written to MPAS files. However, this might
get a bit involved, and the most direction solution would be to
just add code to handle these two attributes specifically. How
often we expect to need to add attributes might determine which
approach we take. Do any of you have any thoughts on the need to
add and subtract attributes more than infrequently (e.g., as
frequently as you might add fields in the registry)?

Cheers,
Michael


On Mon, May 17, 2010 at 07:28:34PM -0600, Michael Duda wrote:
> Hi, All.
> 
> Toward the goal of being able to distinguish between grids in the
> plane and grids on the sphere in MPAS, I'm proposing that we add
> two attributes to the grid.nc file, and make changes to our grid
> generation codes (global_scvt, periodic_hex, and periodic_quad)
> that will add these attributes to future grid.nc files. The two
> attributes are
> 
>    on_a_sphere -- a character string, either 'YES' or 'NO'
>    sphere_radius -- the radius of the sphere
> 
> I'd prefer on_a_sphere to be a logical, but it appears that having
> a logical global attribute in netCDF is not possible (which isn't
> to say that global attributes must be illogical!).
> 
> The code changes to module_write_netcdf.F in the three grid
> generation codes mentioned above would essentially look like
> 
> 95a96,97
> >       real (kind=8) :: sphere_radius
> >       character (len=16) :: on_a_sphere
> 104a107,109
> >       on_a_sphere = 'YES             '
> >       sphere_radius = 1.0
> > 
> 110a116,117
> >       nferr = nf_put_att_text(wr_ncid, NF_GLOBAL, 'on_a_sphere', 16, on_a_sphere)
> >       nferr = nf_put_att_double(wr_ncid, NF_GLOBAL, 'sphere_radius', NF_DOUBLE, 1, sphere_radius)
> 
> with YES being replaced by NO and 1.0 being replaced by 0.0 in the
> case of the doubly-periodic codes.
> 
> Although 'YES' requires only three characters and 'NO' requires
> only two, I wanted both to be the same length, and I figured
> making the attribute be 16 characters rather than three would
> allow for some potential flexibility in future without adding an
> unreasonable amount of storage. For example, if not on the sphere,
> we might in future find it useful to be able to set on_a_sphere to
> something like 'NO (X-Y PLANE)'. Who knows.
> 
> If there are no suggestions for improvements (or requests to see
> the changes as they appear in the module_write_netcdf.F files),
> I'll commit these changes. I can also update all of the existing
> grid.nc files available from
> http://www.mmm.ucar.edu/people/duda/files/mpas/ as well as provide 
> a utility program to update any other grid.nc file with the 
> new attributes.
> 
> Cheers,
> Michael