relative fluxes

[Doug Tody]

Hi -

As Frank says the original purpose of accuracy.calibration was
simply to roughly characterize the calibration, e.g., for use when
searching for data (as in "find me only data which have an absolute
flux calibration").  As always the full dataset metadata should
hopefully more fully describe the calibrations performed.

ABSOLUTE means we have an absolute flux calibration; such data can
be compared regardless of the source.

RELATIVE means that the data was "calibrated" in some sense (we
have no way to fully specify the details), but is not in the class
of an absolute flux calibration.  A typical case would be an unknown
systematic error or scale term; presumably shape analysis such as an
equivalent width would be possible for such spectra.  Frank's example
is typical of what was intended.

NORMALIZED implies some sort of ratio, continuum normalization
or other.  The UCD can tell more about what kind of ratio so it is
not necessary to limit this to continuum normalization.  For full
generality we need to include the reference spectrum flux vector,
or possibly point to it as a separate spectrum, as Petr suggests.
This has been considered for a future version but is not yet supported.

Improving the photometry model used in SSA is a related high priority
issue.  Adding Provenance could be very useful to better understand the
origin of the data (a theoretical model being a type of "instrument"
by the way, from the perspective of SSA).

 	- Doug



On Mon, 22 Jun 2009, Petr Skoda wrote:

> Hi Frank,
>
>>  adequate for absolute calibration and so we provided spectra that were
>>  calibrated to another (lower resolution) library of spectra continua.
>
> OK it is our reference function for RELATIVE - the other well defined 
> continuum spectrum
>
>
>> 
>> #  Spectrum.Char.FluxAxis.name="Flux"
>> #  Spectrum.Char.FluxAxis.Unit="erg cm**(-2) s**(-1) Angstrom**(-1)"
>> #  Spectrum.Char.FluxAxis.UCD="phot.flux.density;em.wavelength"
>> #  Spectrum.Char.FluxAxis.Accuracy.Calibration="RELATIVE"
>
>
>>  of terms like RELATIVE and CONTINUUM NORMALIZED.
>
> Here I was confused about continuum normalized - but I have just checked the 
> library and all is clear the output spectra are in some physical flux
> units and they conserve the original SED (i.e. the spectrum is falling at 
> Balmer jump and and the red part as if it was absolutely calibrated just with 
> different scaling factor - so it is the example of the Normalization 
> mentioned by Jesus in VOSpec - in you case the reference value at given 
> wavelength (550 nm or so) is put to 1.0  but at blue part it is higher and 
> red part lower (e.g. continuum at 0.4).
> But it is important the shape of curve mimics the real absolute flux SED.
>
>>  continuum shape was relatively correct by calibration to another lower
>>  resolution library, and has numbers which include 1.0 at the tie point
>>  used, the calibration is relative.  Since the continuum shape from the
>>  other library was supposed to be calibrated to standard stars a flux unit
>>  seems appropriate.
>
> I would use the same description and units as you in this case.
>
>
>>  I don't think the particular debate being conducted is only about units
>>  but about how to convey the nature of the calibration.  So people can take
>>  what I considered for whatever it is worth.
>
> I think in this special case yes - but it should not be confused with real 
> continuum normalized spectrum (even you had mentioned it) - as in it the 
> value of continuum EVERYWHERE is set to 1.0 and the curve is horizontal line 
> at 1.0 with protruding spectral lines.
> I would just to emphasize the difference still insisting that the NORMALIZED 
> should be only this case with empty units.
> All other should be UNCALIBRATED including pipeline products
>
> And strictly speaking the pipeline products have to be unitless for echelle 
> spectra if the blaze function is removed (otherwise the spectrum is not 
> practically usable) as you divide by definition (unknown) flux of star (in 
> some physical units like W/-2/s/A) by another unknown flux of flat field 
> lamp. So in fact the result is really only the dimensionless ratio.
>
> In the other case you divide by flat field in 2D (by so called superflat 
> which has artificially set pixels near 1.0) and than you preserve counts - so 
> the result is in integrated counts (or even photoelectrons if gain is known) 
> but the blaze function has to be removed by process simillar to normalization 
> of continua - fitting some high order polynomial....
>
> This is necessary before joining the orders in both cases so again the best 
> choice is to use the ucd and units according to real nature of the spectrum 
> (and so the publisher should know the provenance of his own data - or check 
> pipeline recipes)....
>
>
> Petr
>
> *************************************************************************
> *  Petr Skoda                         Phone : +420-323-649201, ext. 361 * * 
> Stellar Department                         +420-323-620361           *
> *  Astronomical Institute AS CR       Fax   : +420-323-620250           *
> *  251 65 Ondrejov                    e-mail: skoda at sunstel.asu.cas.cz  *
> *  Czech Republic                                                       *
> *************************************************************************
>