Dear Jonathan, Bryan,
I think Jonathan is right that we really need to start with some real-world
examples. I will look at some of our datasets for errors/uncertainties (when
I have a few moments) and raise the issue again. Off the top of my head we
had a user last year asking for error data for some of our ECMWF datasets.
It appeared that in some cases an error was quoted as one number for an
entire global 3-D dataset whereas the other extreme is that a specific error
is given for every single data point (as in some instruments).
As I say, I'll come back with some examples. And if anyone else is listening
in to this conversation then please give us some examples of how you record
uncertainty on your datasets.
Ag
-----Original Message-----
From: Jonathan Gregory [mailto:jonathan.gregory at metoffice.com]
Sent: 20 March 2003 09:35
To: cf-metadata at cgd.ucar.edu
Subject: Re: [CF-metadata] Encoding Errors on variables in CF
Dear Bryan
> What then I'd like to do is have an attribute error_variable, which when
> attached to a variable, points at another variable within the file (the
> actual variable containing the error values) .... exactly what sort of
error
> it is can be an attribute of the latter variable itself.
No, we don't have an attribute to do that at present.
I understand your point. There certainly are plenty of situations where you
would want to provide both observations and their uncertainties as data
variables. However, I am not convinced about introducing a linking
attribute,
because I don't think the relation is so close that it needs to be given
this
special status. It is often the case that you want to find two data
variables
which correspond in some way in order to plot them together or combine them
in
some way e.g. zonal and meridional components of wind to plot vectors,
temperature and salinity of water to calculate density. Rather than "lace"
the
data with numerous attributes, I would suggest we should use the existing
metadata to identify what you want. In any one case this is harder, but the
problem with setting up internal links is that they may go wrong in all
sorts
of ways (variables not copied from one file to another, attributes copied
inappropriately from one variable to another, etc.) The links we have
supported
in CF are specifically to be able to locate the spatiotemporal metadata.
In this particular case, you can look for another variable with the same
spatiotemporal metadata and a standard_name which contains the original one
as a pattern. That would be very general. I would have guessed that usually
you would not treat all kinds of uncertainties in the same way, regardless
of
what they said they were, would you? It might be safer to be aware of what
kinds of uncertainty might be in the file and search for their standard
names
explicitly, handling each case in an appropriate way.
I do think it would be useful to discuss some concrete examples. I might not
have grasped the problem properly. Also, other people's views would help.
Best wishes
Jonathan
_______________________________________________
CF-metadata mailing list
CF-metadata at cgd.ucar.edu
http://www.cgd.ucar.edu/mailman/listinfo/cf-metadata
Received on Thu Mar 20 2003 - 03:00:21 GMT
