[CF-metadata] URNs

From: Jonathan Gregory <j.m.gregory>
Date: Wed, 3 Oct 2007 15:19:41 +0100

Dear Phil

> I think (hope?) the day will come when all of the _standard_
> CRS information associated with a netCDF variable will be referenced via
> a single URN-type attribute, e.g.
>
> float temp(lat,lon):
> temp:crs_id = "urn:ogc:def:crs:EPSG:6.3:4326" // OGC-defined URN
> for WGS 1984 CRS.
>
> Clearly it doesn't make sense to repeatedly specify low-level CRS
> information in thousands of netCDF files. ...
>
> However, not all existing netCDF-aware client software is able to
> understand these URNs. So until these applications have been upgraded,
> or replaced in favour of more capable solutions, we need to support both
> the human-readable and computer-readable methods of defining CRS (and
> indeed other) properties.

I think that the description of the coord ref system in terms of projection,
ellipsoid and vertical datum can't be dispensed with until the day when it is
no more difficult to look it up on the net than it is to read it directly from
the file. Until then, the information has to be in the file to make the file
self-describing, and I would say that day is a long way off, when one considers
all the applications, some of them simple user-written programs, that want to
use the netCDF files.

Also, existing authorities will only cater for commonly used CRSs. To have the
flexibility to use any CRS, you need a way to describe it. More specifically,
existing authorities are only interested in the real world, whereas our files
have to deal with various model worlds as well.

To some extent I am being devil's advocate here, but I would like to know what
extra information you get about the CRS if you look up the URN than we are
anyway proposing to store in netCDF attributes. If there is no extra
information, the URN is not needed to describe the data in the file, which is
the primary purpose of CF metadata. You could make an argument that the URN is
useful for other applications which will understand this URN but are not
capable of understanding our equivalent CF description of the CRS. That kind
of interoperability would be a wider kind of purpose for CF. I'm not arguing
against it in principle, but we should be aware that it is a different motive
for any of the rest of the CF convention. It is then essential somehow to
guarantee that this redundancy (between the URN and the description into which
it can be translated) does not lead to inconsistency.

Best wishes

Jonathan
Received on Wed Oct 03 2007 - 08:19:41 BST