[CF-metadata] Units: absolute vs difference

From: Jonathan Gregory <j.m.gregory>
Date: Wed, 18 Aug 2004 19:36:46 +0100

I tend to think that these are two different kinds of problem:

* Calculating anomalies wrt climatology is a "processing" operation that we
could describe with attributes, perhaps analogous to computing temporal
standard deviations from a set of time-means. You might well perform the
same operation on an entire file of quantities. It would be undesirable to
have "anomaly" standard names for every conceivable quantity. This is the
same kind of reason as the one for using standard name qualifiers for
uncertainties and data quality flags. In this case, I agree with Roy that
factorising or atomising the information is helpful.

* Grouping quantities which are scientifically related such as

> tendency_of_air_temperature_due_to_diabatic_processes
> tendency_of_air_temperature_due_to_dry_convection
> tendency_of_air_temperature_due_to_large_scale_precipitation

Indeed they do belong together, but the *way* in which they should be grouped
is not unique - it depends on the scientific description adopted by the person
who wrote the data. One model might distinguish large_scale_precipitation from
convective_precipitation and another lump them together, for instance. There
will be many ways of grouping the most precisely defined quantities into more
"inclusive" quantities. I don't really think that the way this is done by a
particular model or observer is something that belongs in the metadata of the
data. It is something that ought to be recorded in the metadata of the model
or observing system. The metadata of the data serves only to describe what the
data *is*, not to tell the user what should or could be done with it
scientifically. To take a simpler example, we have standard names of
convective_precipitation_flux (kg m-2 s-1), convective_rainfall_amount (kg m-2)
and convective_rainfall_rate (m s-1). These are related quantities, but the
standard names do not indicate the relations (though the help info might give
some guidance). I think that to indicate how they are related as part of the
construction of standard names would be inappropriate and would make the job
an N^2 problem rather than just N (and N is already big enough).

Cheers

Jonathan
Received on Wed Aug 18 2004 - 12:36:46 BST