On 11/13/2013 6:04 AM, Hedley, Mark wrote:
Hello Steve
I appears we have very different perspectives on this. I am
interested that you state that:
> Cell_methods, as the name suggests, is the place to document the
transformations that have occurred within grid cells -- beneath the
resolution of the grid.
after making the comment
> I'd like to cast a strong vote *against* the idea of overloading the
cell_methods attribute (trac #108) for the purpose of describing
collapsed axes.
It seems to me that a collapsed axis is an aspect of a grid cell and
the cell_method is exactly the place to capture the extra information,
to be read with the standard_name and unit, defining what the data
values refer to.
Perhaps I am mistaken in this.
The text says:
To describe the characteristic of a field that is represented by
cell values, we define the|cell_methods|attribute of the variable.
This is a string attribute comprising a list of blank-separated
words of the form "/name: method/". Each "/name: method/" pair
indicates that for an axis identified by/name/, the cell values
representing the field have been determined or derived by the
specified/method/.
I have (perhaps incorrectly) assumed that if there is a reference to an
"axis", then it must be an axis that exists in the file. But for an
axis such as "time" (e.g. cell_methods = "time: average") , it could be
unambiguous, even if the axis had been collapsed out of the file.
In which case (following up on Jamie's input), the natural and
consistent syntax for an ensemble average would seemingly be
"realization: average" (like time, not requiring a reference to an
ancillary coordinate because the nature of the "axis" is self-describing).
I'm inclined to defer to JG's views on this, myself. Cell_methods has
largely been a creature of his vision.
- Steve
> What's needed is a consistent mechanism for documenting the details
of how an axis was collapsed.
I think this mechanism is currently the cell_method. I have seen many
CF compliant data files which use it in exactly this way. I have also
seen data processing code which explicitly adds an entry to the
cell_methods string every time it collapses an axis.
However you are challenging my assumptions about the purpose of a
cell_method so I want to make my assumptions clear and let you and
others help me reinterpret the conventions appropriately.
To try and illustrate my assumption another way: I think that a
combination of standard_name, units and cell_methods enables me to
explicitly and unambiguously encode that my 2D (lat lon) data values are:
"the 30 year mean of the seasonal(djf) mean of the daily maximum
air_pressure values at the surface in hPa"
I thought the was the raison d'etre of the cell_methods string
mark
------------------------------------------------------------------------
*From:* Steve Hankin [[email protected]]
*Sent:* 12 November 2013 17:49
*To:* Gregory, Jonathan
*Cc:* Hedley, Mark; CF metadata
*Subject:* Re: [CF-metadata] Cell methods when there are no coordinates
Two quick comments:
1. "We could perhaps ... introduce a new standard name such as
ensemble_member_id (a string) or ..."
* Have you hit on a void in CF that needs to be filled? Unless
I've overlooked it CF has not yet standardized the manner in
which ensemble axes are to be identified -- a standard_name
for the ensemble axis, units, etc.. (The string "ensemble"
does not appear in the CF 1.6 document.)
2. I'd like to cast a strong vote *against* the idea of overloading
the cell_methods attribute (trac #108) for the purpose of
describing collapsed axes. What may appear as an appealing short
cut now, seems to me destined come back to bite us later.
Cell_methods, as the name suggests, is the place to document the
transformations that have occurred within grid cells -- beneath
the resolution of the grid.
The concept that is under discussion here is how to document an
algorithm that was applied across the entire grid domain when
collapsing an axis. The concept is the same that is needed when
documenting an axis collapsed by, say, a long term time average
or a vertical integration. Isn't this the type of content that
normally gets (partially) documented in the standard_name modifier
-- "integral of ..." "change over time of ..."? Should a
starting point for this discussion be a addition of "ensemble
average of ..." to the list of standard_name modifiers? (The cons
of packing transformation descriptions into standard_name deserve
debate as well as the pros, though for reasons that lie outside of
the current discussion.)
What's needed is a consistent mechanism for documenting the
details of how an axis was collapsed. We should consider how
(say) the time limits of an axis that has been collapsed by
performing a long term time average should get documented. The
answer to this question should guide the manner of documenting the
collapse of an ensemble axis. It is unlikely that cell_methods
would be part of this answer.
- Steve
========================================
On 11/9/2013 1:07 PM, Jonathan Gregory wrote:
Dear Mark
Thanks for the clarifications in your last email. If I have understood this
correctly now, there are two needs, which might be distinct.
(1) Indicate that a collapsed (size-one) axis is an ensemble axis, although
it doesn't have a coord var or any aux coord vars. I can see that it might
well not have these vars because there might not be anything useful you could
put in them. In the absence of these vars, the dimension name in the cell
methods isn't informative. We could perhaps solve this by introducing a new
standard name such as ensemble_member_id (a string) or ensemble_member_number
(a number) or perhaps both. These could anyway be useful. These standard names
could appear in your cell_methods, following section 7.3.4, instead of the name
of the size-one dimension, indicating that the statistic applied to all the
available members of an ensemble, without needing any coord information. I
don't think this change would require an alteration to the convention.
(2) Point to the coordinate information which applied to the axis before the
collapse. This could be useful for any sort of collapsed axis, and in fact I
think we have discussed it before at some point in the last 15 years! I agree
with your suggestion that it would be logical to record this in cell_methods
as a standardised comment. An alternative would be to add an attribute to the
collapsed coord var, but you don't have those (as you say), and also the
collapse may apply to a combination of axes. You suggest listing all the coord
or aux coord vars in the cell_methods comment. Would it not be sufficient, and
more economical, to give the name of the uncollapsed dimension(s)?
Best wishes
Jonathan
_______________________________________________
CF-metadata mailing list
[email protected]
http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
_______________________________________________
CF-metadata mailing list
[email protected]
http://mailman.cgd.ucar.edu/mailman/listinfo/cf-metadata
