Dear Jonathan,
>> * I would prefer if lat and lon are not necessary for some CF-minor
>> compliance.
> As you and John have recognised, this makes the file less useful to anyone
> else. There's nothing to stop you writing non-compliant files for your own
> use or use within a project, obviously.
The main issue here is that the label "non-compliant" puts the file back
into the realm of just about any netCDF file. So, if you cannot indicate
where a dataset is located on the earth, you loose basically all rights to
CF compliancy. Anyway I will continue to support such ungeoreferenced "Local
CF" files without lat/lon for visualisation. Let's see whether anyone else
has comments on this topic. I expect so, since I noticed that on the Wiki
discussion page
http://cf-pcmdi.llnl.gov/trac/wiki/GlobalAttributes a
reference is made to coordinate types and units associated with idealized
cases and real world simulations. Only the latter one could be
georeferenced.
>> * I would prefer an extension of the "grid mappings and projections" with
>> the attribute "mapped_coordinates" for models using curvilinear cartesian
>> coordinates.
> I think it is fine to have auxiliary coordinates of x y lat lon all at the
> same time, as in your example. You say that
> "What is lacking in this example is some clue that x and y are the
> coordinates in the Lambert projection."
> but that should be clear from their standard names of
> projection_x_coordinate
> and projection_y_coordinate, shouldn't it?
It might work like that. One could search all variables in the file for
projection_x/y_coordinate defined on a subset of the dimensions of the data
array. Hm, why was the coordinates attribute introduced if one can do it
like that (just scan for variables with names latitude and longitude)? The
only problem can occur when there are two sets of x/y projection coordinates
with the same dimensions. That can happen if there are two data sets in the
file with the same dimensions defined at different locations, e.g. staggered
data. However, this case is circumvented by defining different dimension
variables instead of reusing them. Another problem would be the case in
which you may have two projections lat/lon coordinates (e.g. at the border
of two countries with different national coordinate systems). I must admit
this is a rare case, but not completely impossible (I am involved in such a
project, but it is not yet using netCDF). However, in such case there is a
larger problem namely that the association between projection_x/y_coordinate
and related grid_mapping is lost. My previous solution doesn't help to solve
that case either. I would suggest that the attribute "grid_mapping" should
not be associated with the data variables as presented in section 5.6 of
Conventions 1.0 but with the auxiliary coordinates. Any pair of
projection_x/y_coordinates with appropriate dimensions and consistent
grid_mapping could be used for plotting.
>
>> * I would like to learn that missing_values are allowed for auxiliary
>> coordinates.
> That is an interesting question and I think you are right that we have not
> discussed it in the standard. I think we should allow it. Missing data is
> not allowed for 1D coord variables as they have to be usable for
> subsetting
> and ordering the axis.
Thanks, I hope that others agree. It is not a standard feature in all
postprocessing software.
Best regards,
Bert
Received on Fri Nov 24 2006 - 12:10:03 GMT
