Dear colleagues,
as I have explained in dal group, I think the IVOA spectral data model should pay more attention to the continuum normalized spectra and simple 1D FITS images with CRVAL and CDELT keywords.
So here are my comments to the proposed Spectral Data Model ver 1.01:
in sec. 3.2 Units:
As I have emphasized in dal group, the normalized spectrum is a important fully science ready output and thus it is not equivalent to ignore the flux axis by stating the units as "n/a" and telling the units to be "unitless" - but is is perfect quantitative and measurable information on normalized spectra - so I propose the special notation for such a "unit" as a "1" (one) or if letter required as "U" (unity) or "N" (normalized)
ad 3.5 Packaging model
(1) FITS (standard BINTABLE for Spectrum, defined in this document)
What about to add the simple 1D FITS image as acceptable for simple spectra as well - and thus "defined in this document"
Ad 4.1 Spectral coordinate
for WAVE em.wl should be explicitely written "vacuum" wavelength (it may be source of many errors by optical astronomers who are commonly using air wavelength) - it is just counterpart of AWAV given below in alternative choices...
Ad 4.2 Flux (Spectral Intensity) Object:
Despite the fact the Background model is foreseen, it will not handle properly the continuum normalized spectra - UCD spect.continuum is a continuum flux but it may not describe well what I understand by the request to disply the shape of continuum applied to particular spectrum (it will be mostly kind of polynomial function).
I can imagine the expression of the simple spectrum as a raw spectrum comming from pipeline in arbitrary (summed) ADUs and the normalized version produced on the fly using the description of its contiuum flux (theoretical function) stored separately. But again the is problem with units for normalized spectrum given directly by the pipeline. The FluxAxis.ucd arith.ratio;phot.flux.density is similar to the requirements, but the density is misleading and the second spectrum (continuum) is again only a fit not data, so it is probably not correct... We simply need ucd for dimensionless normalized spectrum as a kind of flux calibration type ;-)
at 4.6.5 Calibration
The UNCALIBRATED is not good (although it tells " they are modified by an
unspecified coordinate-dependent correction") ABSOLUTE is not good either
as it is not absolutely calibrtaed and RELATIVE does not fit well as
unknown systematic error is not present - morever the units of flux are
still expected here, probably.
So I suggest another value of Calibration field :
"NORMALIZED"
Ad Part 9 FITS serialization
I understand that the binary table requirement in VO just comes from this part - "We define a reference serialization of this data model as a FITS binary table"
I am afraid that the keywords here are to far from reality: perhaps representing some space experiment data. I think the 1D FITS image should be somehow treated here and a example be given for more common audience.
Best regards,
Petr Skoda