UCDs, metadata and AstroOntology

[Ray Plante]

Hi,

I agree that with Guy's supposition that UCDs are molecules in which it 
would be useful to have access to atoms and some refactoring would be 
helpful.  I'd like to jump in with my own take on how UCDs might be 
revamped as well as float a roadmap for how we might migrate.  

A general concern that I have had with UCDs is their limited 
role as tags to simple data.  In addition to attaching meaning to table 
columns, we have need to express more complex information that doesn't 
fit well (and often does not appear) in a table column.  The most obvious 
is a coordinate system description.  Arnold Rots has been working up a 
data model for space-time coordinates including some XML markup that can 
be used to describe a system.  He has also recommended some UCDs that 
map into components of this model.  I find it inelegant that we have two 
separate systems to tag this information.  

Last week, a number of us attended a workshop at SAO on developing data 
models for the VO.  One of the issues I'm interested in here is how 
metadata definitions can drop out of a data model.  At this workshop, we 
played with a data model to describe a bandpass that led to some possible 
XML markup for encoding the information.  I was struck that specific 
"atoms" of information could be identified by its XPath location, which is 
normally quite simple.  For example "Bandpass/EffectiveWavelength".  If 
This atom is part of a filter description, we might refer to it as 
"Filter/Bandpass/EffectiveWavelength".  This, of course, is remarkably 
like UCDs where / replaces _.  The nice thing in my mind is the direct 
relationship between this UCD-like tag and XML markup and the data model; 
no explicit mapping needed.

As many of you know, I'm partial to a data-model-based metadata 
definition; however, I don't believe in building a comprehensive model 
that covers everything at once.  This is done a bit at a time.  That 
means if we move to this model, we have to do it in a way that plays well 
with the UCDs we are already using.  Some keys to doing this:
  * namespaces associated with: 
      parts of a data model
      related metdata
      UCDs
  * one should be able to define and experiment with any of the above 
      three types without doing going through the full modeling process.
  * any model, metadata, or UCDs recommended for wide (IVOA) adoption 
      should be required to explicitly define the mappings to the current 
      set of UCDs.

cheers,
Ray