[vsnet-outburst 11447] Re: [AVSON] APASS Astrometry Awareness

[arne]

The zip file John refers to has not been formally released,
and therefore a descriptive document detailing the constraints of
the system is not yet available.  About the best that you can work
with is the APASS webpage:
http://www.aavso.org/apass

One of the nice things about CCD imaging is that you can reprocess,
as you learn more about your system or have new techniques to use.
We already know that the astrometry can be improved by a factor of
two; we may use psf-fitting rather than aperture photometry in the
final solution to handle some of the blending issues.  Handling
the entire sky can be done at fine resolution with a
very expensive system that has lots of pixels (think: PanSTARRs),
at fine resolution with a less expensive system but taking a decade
to complete (as with CMC in only one passband), or at coarser resolution.
We took the latter route so as to get reasonable data available
to the community in the shortest possible time.  For now, you
need to take these data releases as interim products: better than
what is available elsewhere, but likely to be improved as more
photometric nights are added to the solution, and during the
final processing in a year or so.

I'm happy to see people like John testing these releases, as then
we have guidelines for their improvement on the next round.
Arne
------------------
On 8/15/10 8:35 AM, isthisthingon at Safe-mail.net wrote:
> Source data : ftp://ftp.aavso.org/public/calib/master_apass.zip
>
> Now, there is a general thing about wide field surveys inherent to all of them, and not just specific to APASS.
>
> Users of ASAS3 and NSVS data to find variable stars will mostly be aware of it.  And indeed in papers from time to time will be seen comments that it is not possible to tell which star of two in the field is represented by the ASAS3 object from the ASAS3 data alone, and the same is true for NSVS.  Chart makers who have used USNO A2.0 will also see similar, with a single USNO A2.0 star appearing from time to time between and in the middle of two distinct roughly equal magnitude CMC14 stars, for instance.  Unresolved on POSS plates, or at least by the USNO A2.0 digitisation of said, but split by CMC14.  Often split by 2MASS too.
>
> ASAS3 and NSVS had systems where the arcsecs per pixel resolution was 14.4 arcsecs.
>
> Despite this value equal brightness stars near the fainter end of these surveys that are as much as 48 to 50 arcsecs apart are often seen as one slightly brighter "star" apparently lying roughly midway between the true positions of the two real stars.  That's nearly four times the arcsec per pixel resolution.  Or, if thought of in pixels, nearly four pixels across.
>
> APASS has a roughly two arcsecond per pixel situation it seems.
>
> Yesterday I placed a plot to an url of CMC14 r' versus APASS r' for fifty thousand stars in the KEPLER field to vsnet chat, but I'm not sure if the email got there.  If it did not, let me simply say that there was an increasing trend in wider scatter towards the fainter end when the two sources were differenced.  The KEPLER field isn't a sparse field, nor is it a paticularly very dense field.
>
> Accordingy, as I check today, I see quite a few cases of APASS objects cropping up midway between two CMC14, or even UCAC3, objects in the fifteenth to sixteenth magnitude range in V.  And the distance apart of these faint 16th magnitude true pairs?  About 8 to 9 arcsecs.  Or roughly four times the arcsec per pixel resolution of APASS.
>
> Now, this doesn't necessarily matter, depending on how the photometric software works.  If an astrometric solution is done first, and an astrometric solution against an independent astrometric catalogue, then when an observer uses their image, which will have usually readily split such pairs, then when the photometric software tries to match stars to APASS catalogue objects then these floating midway objects are not going to be matched to any real object.
>
> Except, so far we've been examining the simple and generalised situation.
>
> Data analysts who have examined ASAS3 and NSVS data looking for variable stars are well aware that the above are just the extreme cases.
>
> If you have a fifteenth and a sixteenth magnitude star (or in the case of ASAS3 that would be a twelfth and thirteenth) of about half a dozen arcsecs (half an arcmin or so for asas3 / nsvs) separation, then the APASS object appears nearest to the brighter star, but not coincident with it, and possesses a magnitude that is that of the two stars combined (and same for asas3 / nsvs cases).
>
> In the case of APASS we're going to have a situation where matching of catalogue objects to image objects can easily mean using a crossmatch radius of 2 arcsecs, and in sparse fields and in most normal circumstances, this can be quite safe.  But in the situation of two stars about a magnitude or so apart and about half a dozen arcsecs or so apart in normal to slightly dense fields, the APASS catalogue object will appear to be about two arcsecs away from the brighter star on the line between the two true stars and with a magnitude one to a few tenths brighter than true.  Thus the plot I noted yesterday had increasingly larger scatter (and pretty much symmetric around zero), with a trend to the increase, as the magnitudes got fainter.  CMC14 is a safe one arcsecond resolution system in terms of _astrometry_, in that it will split two faint end similar magnitude stars as close together as being just over one arcsecond apart.
>
> There is no avoiding this, it is inherent.
>
> It only really matters if the images the observer is using also safely split the pair of real objects.  Some observer using the same system, or a very simliar system, to APASS will be able to use APASS unreservedly.
>
> There are partway solutions to these sorts of things.  One is to take quality over quantity.  As UCAC3 is broken, CMC14 is not all sky, and few other catalogues of suitable resolution, the just about remaining allsky solution is to match the APASS catalogue with 2MASS.
>
> Only accept any APASS object as being real if it lies within 2 arcsecs of a 2MASS source.  However, there is a significant difference in epochs between 2MASS and APASS now, there are such things as proper motions.  Also the astrometric resolution of APASS is likely to be of the order of two arcsecs at worst, probably normally around one arcsec, but we've already seen the case where an APASS object within two arcsecs can be out in r' by a tenth or few in magnitude.  Possibly for ensemble photometry this will not be a problem, however for even ensemble photometry, in some fields, and not rare fields either, there won't only be one case of this happening, but several.
>
> Reduction of the crossmatch with 2MASS to one arcsecond level would be safer still, but then so much would have to be thrown away, probably that is being too cautious.  Or at least felt to be too cautious.
>
> Or two catalogues can be provided, similar exists elsewhere.  Taking a hint from some SPITZER survey programmes, the smaller scrubbed catalogue of more certain objects would be the catalogue, whilst the whole dataset from APASS would be the archive, for example, and which was used would have to be firmly noted.
>
> In the end, in practice, it may matter _relatively little_.  But this is not tested, it has to be checked and measured and "scienced", not assured and claimed and "theorised".
>
> It is an inherent problem, consequent on how the survey is, and a general problem applicable to all such situations and not just unique to APASS.
>
> However APASS will be served as a photometric resource, with already predicted claims of much better than one tenth of a magnitude.
>
> Earlier indications of this likely to result were i) obvious from the outset due to the very nature of the proposed APASS equipment setup and ii) APASS data release 0 comparison with past NOFS and SRO sequence fields taken by Arne Henden, wherein magnitude discrepancies were such that either the old secondary standards or the new APASS data had to be somewhat scattered.  Given the nature of these past photometry exercises, telescopes used, resolution, arcsec per pixel resolution etc, it seems likely that that scatter resulted from the same issue.
>
> Now, as I get accused of negativity and of twisting facts (often repudiated, but rarely refuted), let us take this in context.  APASS will be presented as a serious photometric data source.  Basically, all the above is telling you is user be aware.  Not "beware" but "be aware".  There will be claims made upon APASS, possibly by the providers but certainly by users, that are simply inherently incapable of being claimed, due to the reasons outlined above.  For instance, it cannot be an allsky source of magnitudes at the +/-0.02 level down to magnitude 16 no matter how many passes and measures are made because it simply has not got the same resolution capabilities as most CCD photometrists will have when examining a not insignificant amount of the sky.
>
> Can't be done, can't be fixed, can't be patched, can't be tweaked.
>
> It's like a 'planck limit' thing.  Nothing can be said with certainty about the data within several to eight or nine arcseconds of near same brightness true stars that are also either equallish in magnitude or with a magnitude or so of each other once APASS gets below magnitude 13 or so, and note that such pairs increase in number immensely as magnitudes progress downward to 14, then 15 then 16th magnitude.
>
> It just is.  It was bound to happen.  It has happened.  And yes, I predicted it, and yes I have tested the prediction, and yes the prediction is confirmed.  And yes that is the scientific method.
>
> At a very, very rought guesstimate, based on experience of star densities for different magnitude ranges, a good visual scan of the APASS data plotted up over the KEPLER field (in tandem with CMC14 and other astrometric sources), and eyeballing the general trends of astrometric offset for APASS from true in paired cases (and they are not uncommon), then APASS could well be safe as a comparison star source from this issue down to magnitude 14 for certain and 14.5 or so if lucky for fields of the density of the KEPLER fields.  Given the dynamic range of CCD photometry this may well be sufficient for most people if scatter of the order of +/- 0.1 in comparison star magnitude is not too problematic.  However, large pinches of salt will likely need to be stored and saved for future use with respect to claims in the fifteenth to sixteenth magnitude range for all but the sparsest fields.
>
> All these sorts of things have to be assessed, and checked in practice (not just debated) by serious CCD photometrists, who have to realise that this is a science that they are taking part in, and not some form of religious devotion where handed down doctrine, from whatever ostensibly authoritative source, is merely to be accepted in blind faith.
>
> John
>
> NB use of the Kepler fields allows for some testing of any problem objects against KEPLER and/or SuperWASP epoch photometry to ascertain as to whether any problems are due to significant stellar variability.
>
> Note : some of the astrometric assessment of the APASS data was facilitated by plotting up the master catalogue using Guide 8 via tdf scripting.
>
> Example : http://i33.tinypic.com/s1mavr.jpg shows two stars, plotted to their CMC14 and UCAC3 positions with overlaid POSS II F plate image, which lie 11+ arcseconds apart.  The black + marks the APASS object position lying midway between the stars that represents these two stars, V 15.302 and B-V 0.840 and r' 14.978 from 3 observations according to APASS.  The two stars CMC14 r' 15.084 and 15.817 respectively.  Calculating the combined CMC14 r' magnitude for these two combined star gives 14.637.  The APASS r' magnitude appears to lie near that of the brighter star, five arcsecs away, and not the fainter one, six arcsecs away, nor any combined magnitude.
>