[CF-metadata] Encoding Errors on variables in CF

From: Brian Eaton <eaton>
Date: Thu, 17 Apr 2003 13:58:43 -0600

Hi All,

I agree with Jonathan's items (1) and (4).

I think backward links provided by an attribute such as "for_variables"
add convenience but no new functionality. As Jonathan noted all
the links between file variables can be determined by parsing all
variable's forward link attributes. The added convenience of a backward
link must be weighed against the disadvantages of adding complexity to the
convention. I this case I vote against providing backward links.

I think identifying the error types by the standard name is the right
approach rather than by adding a new "intent" attribute. I'm not in favor
of the standard_name prefix "ancillary_data_for_" as it adds no new
information. But standard name prefixes like "standard_error_of_" do add
information. We currently require all standard_name values to be
explicitly listed in the table. As Jonathan has suggested we could specify a
set of prefixes that could be added to any standard name in the table with
the result being a valid standard name that is not explicitly listed in the
table. The prefix definition would include information about the
appropriate units for the quantity, for example it might say that the units
are the same as those of the quantity to which the prefix is attached.

I prefer the use of standard name prefixes to Jonathan's proposal for
"ready-parsed" standard names with intents. In that proposal the standard
name part of standard_name's value is actually the standard name of the
data with which this variable is associated. I think the standard
name should describe the quantity that the attribute attached to.

It has previously been pointed out that it will be difficult to describe
the error types using standard names due to the variety of definitions that
are used for errors. I think we can start by defining names for the most
commonly used errors (or other data quality measures) and expand the list
by request. This is how we've dealt with similar issues related to
standard names and projections for example. Also, if a certain type of
error is easily parameterized, as for example with the number of standard
deviations being reported in a standard error, then new attributes that
give these parameter values can be defined and used in conjunction with the
standard name to precisely define the quantity. That would be analogous to
how we define projections.

Brian
Received on Thu Apr 17 2003 - 13:58:43 BST