[CF-metadata] ungridded data

From: Bob Drach <drach>
Date: Wed, 19 Jun 2002 12:02:18 -0700

Jonathan,

>
>
> > I guess part of the exercise is to reach an agreed upon terminology!
> Indeed. As Brian's reply shows, an alternative definition of "gridded" includes
> anything which is discretised. In that case, gridded doesn't imply independent
> sets of x and y coordinates, with the valid points being at any choice of x
> and y from the sets.

I wish there were a better term than 'grid', which has several definitions. The
term is used here in the sense of a model grid, which relates to model-generated
data as well as your example of the UM timeseries with associated gridboxes.

>
>
> Rather than introducing more metadata at a "higher level", which would be
> redundant in principle, I think it would be better to deduce what you need
> from the metadata we already have. It's not quite clear to me what the
> distinction is between a GriddedVariable and other variables, but if we're
> talking about lon-lat-vertical, it sounds as though a GriddedVariable is
> one which has dimensions of lat, lon and vertical. Station data will not have
> lat and lon dimensions; it will have some "site" dimension instead, with
> lat and lon as auxiliary coordinate variables. A trajectory in 3D will not
> have lat lon or vertical dimensions, but probably all three will be
> auxiliary coordinates with a time dimension. It would be quite easy to make
> tests like this. Is that the right distinction?
>

This gets to the heart of the question, because it doesn't seem easy to make such
a distinction. For example, suppose two variables are defined as in section 5.3
and 5.4 of the CF document:

variables:
  float humidity(time, station) ;
    humidity:coordinates = "lat lon" ;
  float lon(station) ;
  float lat(station) ;

variables:
  float ps(rgrid) ;
    ps:coordinates = "lat lon" ;
  lon(rgrid) ;
  lat(rgrid) ;

The humidity variable represents station data which may or may not have a related
grid. The ps variable does have a related reduced grid, but no boundary variables
are defined, so it is 'implicit'. Structurally the two are identical.

In some contexts the distinction may be unimportant. If a user passes variable ps
to a visualization app, the app may be smart enough to make a reasonable
approximation of the grid cells for plotting purposes, and this is fine since the
user 'knows' that it makes sense. Typically this is how rectilinear grids are
handled now. But as we move to support more server side functionality (as in ESG)
where the user may not know as much about the data, this kind of distinction will
become important.

So I'm left with the operating definition, not entirely satisfactory, that a
variable is gridded if it has associated coordinate and/or auxiliary coordinate
variables with boundaries.

Bob
Received on Wed Jun 19 2002 - 13:02:18 BST