On Thu 2007/08/16 11:54:16 -1000, Maren Purves wrote
in a message to: Doug Tody
and copied to: William Pence ,
FITSBITS

Where do you want to draw the line?

The interpolation test may help to draw the line. Simply put, an array
is an "image" if it makes sense to interpolate adjacent array elements.
While a collection of 1-D spectra packed into a 2-D array does pass this
test on one axis it fails on the other so fails it overall.

Many historical FITS "images" are also not images according to this
test. For example the conventional axis types of COMPLEX, STOKES and
CUBEFACE, and the ones proposed for colour - RGB, CMYK and HSB.

Conventional axes were a useful storage mechanism in the days when there
was no legitimate alternative (and once FITS...), especially since the
WCS for the true image axes imprints itself on the conventional axis
(except for CUBEFACE).

One characteristic of the existing conventional axes is that they are
all short, of lengths 2, 3, 4 or 6. They are also all defined once
and for all in the literature, and straightforwardly so. However,
a new conventional axis type for 1-D spectra packed into a 2-D array
could be as long as the number of fibres in your IFU. Its associated
WCS, e.g. the celestial coordinates of each fibre, would have to be
recorded in header cards for each point on the axis. There is no
standard mechanism for doing this.

However, there is a much better alternative now. That is to store the
spectra as bintable image arrays with a static (i.e. not using the
Greenbank convention), column-specific WCS. Celestial coordinates can
easily be stored in a separate column or two.

Mark Calabretta