[CF-metadata] Fwd: Re: Question from NODC about interplay of standard name modifiers, cell_methods, etc.

From: Jonathan Gregory <j.m.gregory>
Date: Sat, 30 Mar 2013 13:19:56 +0000

Dear Ken

This email is about your specific question relating to the use of the CF
standard as it currently is. By the way, I applaud your efforts to provide
CF-compliant datasets from NODC and to help others in the ocean data world to
make CF-compliant datasets.

I am about to make another posting about the more general question.

I don't think the CF standard gives any instruction about what the
cell_methods should be when the standard_name contains a modifier. I would say
that, because the standard_error values belong with the data values, the
ancillary variable that contains the standard_error should have all the same
metadata as the data variable, except for the addition of the standard_name
modifier. So I would put

  t_an:standard_name = "sea_water_temperature standard_error" ;
  t_an:cell_methods = "area: mean depth: mean time: mean" ;

for the ancillary variable of standard error, if the data variable has "area:
mean depth: mean time: mean". This means the SE variable is completely
self-describing.

If the ancillary variable were seen as entirely dependent on the data
variable, you could argue that it didn't need any metadata such as
standard_name and cell_methods, because it would implicitly inherit all the
metadata of its data variable. For this kind of reason, bounds variables
actually do not need units or standard names; they are regarded as dependent
on the coordinate variables they belong to. However in the case of ancillary
variables, CF doesn't make such a strong link. The ancillary_variables
attribute is a kind of convenience or courtesy feature, which allows the user
to locate the ancillary variables more easily, but those variables are
themselves independent variables, rather than subservient. One reason for
that design is that the ancillary variables might be in different netCDF
files from the variables to which they are ancillary e.g. you might have a
file for the temperatures, another file for their standard errors, and another
for the number of observations. If this is the case, they need to be able to
describe themselves properly.

By the way, in my version of udunits, degrees_celsius is not legal. It will
accept only degree_Celsius out of the four variants singular/plural c/C. Maybe
this is a difference between udunits 1 (which I have) and udunits 2.

Best wishes

Jonathan
Received on Sat Mar 30 2013 - 07:19:56 GMT