== SAS Astrometry & Photometry Test (2010.03.10) ==

~ Eugene Magnier

I've done some work to test out the relative photometry and astrometry
calibration system using the SAS data.  I show here some example plots
from that analysis, and the script used to run relastro and relphot
the database.  Hopefully this will provide more light than smoke...

[[Image(sas.field.png)]]

The first figure shows the SAS field: black boxes are the outer bounds
of the GPC1 focal plane, red is just the sky grid, blue is the region
used to generate the rest of the plots.

I ran the attached script (run.relastro) which has a section for
'relastro' and a second one for 'relphot'.  These analysis were used
for the entire SAS DVO database.  The dvo shell script (input) was
used to generate the plots, for which I restricted the analysis to the
blue box (partly to avoid edge effects and partly just for speed and
memory footprint).

Some notes about the analysis: I did not tune any of the clipping
parameters.  The relative astrometry analysis only iterates over the
two steps of [update the object positions, update the chip-level
portion of the astrometry model].  I suspect the parameters for
choosing the set of calibration stars (S/N limit, instrumental
magnitudes) is poorly chosen because a number of these fields has
surprisingly few reference stars (according to the log).  Also, I
suspect the choices do not do a good job of tossing out the saturated
measurements (which are not all labeled with the SATURATED bit).
There is also something very sub-optimal about the database level
interactions for joining the detections to their images: relastro ran
very slowly, and much of the time was spent in making associations
which should be must faster (the indexes are still being ignored).

[[Image(sas.test.photom.g.png)]]
[[Image(sas.test.photom.r.png)]]
[[Image(sas.test.photom.i.png)]]
[[Image(sas.test.photom.z.png)]]
[[Image(sas.test.photom.y.png)]]

Aside from those issues, this analysis looks very encouraging.  The
plots above show the magnitude residuals for the bright, unsaturated
portion of the data (15.5.-17.0 mags for griz, 12.5-15.0 for y-band).
You will notice that the photometric residuals are all in the range
1.1 - 1.3%, which look really good.  Note: no chip-to-chip color terms
have been applied, though each chip has been allowed to float.  This
last point is interesting: in general, if the weather is photometric
and we have solved for the flat-field correction, we should fix the
chips since there is not enough overlap to constrain them very well.
Only if we think the weather is non-photometric for some images should
we let the chips float (or fit a higher-order model).

Note that the weather was good, except on part of one of the nights.
We recover this last feature very well in the last plot
(stormy.weather.png), where I show zero point offsets vs time.

[[Image(sas.test.astrom.g.png)]]
[[Image(sas.test.astrom.r.png)]]
[[Image(sas.test.astrom.i.png)]]
[[Image(sas.test.astrom.z.png)]]
[[Image(sas.test.astrom.y.png)]]

As for astrometry, the residuals are OK (20-27 mas), but I suspect we
can do better by tweaking the parameters.  The nice result here is
that we can get quite good measurements with almost no effort in
tuning the analysis.