New UCD1+ for SIA protocol

[Doug Tody]

On Thu, 25 Mar 2004, Anita Richards wrote:

> > 1) VOX:Image_Scale --> instr.scale
> >
> > A much better alternative could be the definition of the new ucd1+:
> > obs.image.scale
> 
> That still suggest to me that the scale is inherant to the observations -
> but it is better

This is a general issue with what we are calling the "observation"
data model.  This may be just a naming issue but I worry that we may try
to describe actual observations.

What we really need to do for VO is characterize the physical attributes
of a dataset.  A dataset may be a calibrated observation, or it may be
the result of an arbitrary amount of processing of multiple observations,
or it may be synthetic data.  It does not matter how the dataset was
generated if we describe only the physical attributes of the actual final
dataset we are dealing with.  Scale and resolution are good examples of
such physical attributes.

For VO data analysis where we may need to deal uniformly with data from
many origins, physical dataset characterization is what is needed.  At this
level we should not have any information about the actual observations,
instrument characteristics, etc., (if any) used to produce the dataset.
Such information may be present in each individual dataset, and can be
useful to fully understand individual datasets, but is not very useful
for automated processing of data from many origins.

A simple example is exposure time.  While a typical attribute of an
individual original observation, it tells us almost nothing about an
arbitrary dataset.  To understand what this means we would have to
understand the full instrumental model and configuration, and all the
post-processing done to get to the final dataset we are actually looking at.
The related physical dataset attributes are the sampling and coverage in
time of the final (possibly aggregate) dataset, and some physical measure
of the limiting flux of a signal detected by the dataset.