ACA MAG FITS from ACA DATA

Paul J. Green & John Grimes
10 Jan 2000

See /data/mp2/ACA_cal/dec99/ for data, code, and notes. This file is /data/mp2/ACA_cal/dec99/AGASCupdate.html

Occasionally, a poorly estimated AGASC magnitude results in failure to acquire a star. A larger problem is that the padding on star magnitude errors in the OFLS Characteristics, means that too many stars are being rejected from consideration, so that targets are rejected, or have a number of guide stars that is less than optimal. In Aug 1999 PCAD and aspect camera (ACA) exercises (for IRU Cal observations and M7 ACA Field Distortion) provided about 100 realtime observed ACA magnitudes with which to compare the current star catalog AGASC1.3

All the tracked stars are compared below to AGASC1.3, which shows that a mean offset of about 0.26mag in the sense that AGASC1.3 magnitudes (MAGACA) are too bright. 

	  ACA-MAGACA COLOR1
 NUM          103      103
 MED        0.187    0.573
 MEAN       0.259    0.657
 SIGMA      0.353    0.572
 MIN       -1.214   -0.126
 MAX        1.432    1.832
Here are further details. In reality, this is a strong function of color (usually cataloged B-V): the mean offset is 0.13mag for bluer (B-V < 0.7) stars, but is 0.41mag for red stars. Here is a plot of the difference between observed ACA mag and AGASC1.4 MAG_ACA values, as a function of (B-V) color.

Here I plot only stars with a measured color or spectral type. Note that many stars in AGASC have no cataloged colors, for which we assume (B-V)=0.7mag. The mean colors for the 103 stars here is 0.66, bearing out this assumption.

NEW ACA DATA
More recently, Jeff Shirer et al. produced, using GRETA, data for 876 star observations while in Normal Pointing Mode and while on Kalman filtering. Magnitudes more than +/- 0.5 from the mean were excluded as outliers. The y and z angles for the stars are corrected for dither, but they are not statistical averages over the entire observation. The angles are instantaneous values, 60 seconds after going to Kalman filtering. John Grimes used the Y and Z angles, along with the OBS timing information to compare to SOE files, to find the correct corresponding AGASC star.

Star data is now easily found by AGASC_ID using a tool that spews out chosen columns for each AGASC_ID:
/data/mpcrit1/bin/getagasc.pl <AGASC_ID> Removal of repeat observations of the same star cull the list down to 271 unique stars ACA observations, matched to cataloged AGASC stars. For a primer on how MAG_ACA values are calculated, see MAG_ACA update pseudocode. Essentially, I fit (ACA - V) as a function of (B-V). Based on this new ACA data, I've done polynomial fits to predict ACA mags for all stars, to recalibrate the MAG_ACA values. New fit results are as follows: 

 C0      C1      C2      C3      C4      Reduced Chisq
-0.079   .       .       .       .       0.193
 0.542   -0.730  .       .       .       7.829E-02
 0.355   -0.022  -0.442  .       .       6.918E-02     CHOSEN FIT
 0.359   -0.060  -0.382  -0.025  .       6.943E-02
 0.355    0.162  -1.113   0.704  -0.224  6.955E-02
Note that the older fit was similar: 
 C0      C1      C2      C3      C4      Reduced Chisq
0.347    0.099  -0.482  .       .       7.84321E-02     CHOSEN FIT

So the most reasonable fit is a quadratic. The difference between AGASC1.4 magnitudes and observed ACA magnitudes as a function of color for this sample is shown below, both for a simple offset (0th order fit) and quadratic (2d order fit).

While the mean difference is 0 above, the 1-sigma dispersion in the plot is 0.26mag. This figure illustrates the inescapable dispersion involved in a transformation based only on B,V colors to a camera system with a much redder bandpass.

Finally, the new fit shows no signs of being magnitude dependent:

Three important points:

1) A few percent of stars will be rather extreme outliers (a magnitude or more off) even when colors are known. This may be due to extinction, or unusual spectral features, or variability. The outliers in this dataset have Observed-Predicted extremes of -1.2 and +1.7 mag.
2) I made the fit and plot above using stars with cataloged (B-V) colors. Many stars in AGASC (nearly all that are fainter than 10th) have no cataloged colors, so a color of (B-V)=0.7 is assumed. The dispersion in a observed-predicted plot for such stars will be significantly larger. However, some 90% of stars bright enough for guide star selection are Tycho stars with measured colors.
3) The new calibration will improve things in 2 ways. The OFLS will not tend to pick as many red stars as it does now, because the effect of color has exaggerated the predicted brightness of stars in AGASC1.3. The predicted magnitudes of stars picked will be closer to the actual magnitudes, ESPECIALLY for red stars.
4) We may now change the values in the OFLS CHARACTERISTICS file from their preflight values: 
  
       ODB_MAG_ACA_NSIG  = 3,                    
       ODB_MAG_ACA_ERR_RAN = 0.34,               
       ODB_MAG_ACA_ERR_SYS = 0.5,                
to
       ODB_MAG_ACA_NSIG  = 3,                    
       ODB_MAG_ACA_ERR_RAN = 0.26,               
       ODB_MAG_ACA_ERR_SYS = 0,                

The OFLS SSA MAG_ACA bright and faint thresholds are calculated as
     MAG_ACA_BRT_THR= MAG_ACA - DELTA_THRESH
     MAG_ACA_FNT_THR= MAG_ACA + DELTA_THRESH
where
	DELTA_THRESH   =  MAG_ACA_ERR_SYS + 
  MAG_ACA_NSIG*sqrt(MAG_ACA_ERR*MAG_ACA_ERR + MAG_ACA_ERR_RAN*MAG_ACA_ERR_RAN)

The OFLS SSA simply uses MAG_ACA_BRT_THR
and MAG_ACA_FNT_THR as filter thresholds for consideration
of guide stars.  No star is accepted for which either
of these conditions holds

	MAG_ACA_BRT_THR < ODB_BRT_MAG_AC
	MAG_ACA_FNT_THR > ODB_FNT_MAG_AC

The fraction of stars from the GRETA sample that have |Predicted-Observed| greater than 3sigma is 3percent. This shows, as expected, that the distribution is not Gaussian. The proposed change to the CHARACTERISTICS will decrease the typical value of DELTA_THRESH from 1.5mag to 0.8mag. A corresponding change must be to reduce the current wide limits from 

  
       ODB_BRT_MAG_AC    = 4.5,     
       ODB_FNT_MAG_AC    = 11.3,   
       ODB_BRT_MAG_G     = 4.5,     
       ODB_FNT_MAG_G     = 11.3,   
to
       ODB_BRT_MAG_AC    = 5.2,     
       ODB_FNT_MAG_AC    = 11.0,   
       ODB_BRT_MAG_G     = 5.2,     
       ODB_FNT_MAG_G     = 11.0,
Narrowing those limits somewhat reduces the number of stars available for use as guide stars, but the change is needed to prevent stars that are too bright or faint from being selected.

UPDATE of AGASC1.3 to AGASC1.4

MAG_ACA updates to a copy of AGASC in place.

See /proj/agasc1/binary/UPDATE for data, code, and notes. 

  
source ~/.ascrc -r daily -s head

/home/ascds/DS.daily/bin/mp_magaca*
AGASC_OUTPUT is not set.
AGASC_OUTPUT_FILEMAP is not set.
Usage:  mp_magaca neighborsfilepath

mp_magaca only uses neighborsfilepath for the first column, as a list
of which files need their MAG_ACA values updated

N.B.  The FITS region file is 1-based, while the
neighborsfile index is 0-based !!

Also looks in ASCDS_DATA=/home/ascds/DS.daily/data
for the alpha values, in
/home/ascds/DS.daily/data/gsc_bandpass_corrections 

Since
ASCDS_USER_PARAM=./param
ASCDS will first look there

mkdir param
cp /home/ascds/DS.release/param/ssa.par param/.
Edit in your new coefficients

EXAMPLE:

setenv AGASC_OUTPUT /proj/agasc1/binary/agasc1p4/agasc/
setenv AGASC_OUTPUT_FILEMAP /proj/agasc1/binary/agasc1p4/tables/outfilemap

mp_magaca /proj/agasc1/binary/agasc1p4/tables/neighbors
SPOILER updates in place to a copy of AGASC.

Dennis' new adaptation of the spoiler code update is mp_spoilers. 

  
mp_spoilers -h
AGASC_OUTPUT is not set.
AGASC_OUTPUT_FILEMAP is not set.
Usage:  mp_spoilers radius neighborhoodfile
          (radius is in arcsec)
          (Region numbers in neighborhoodfile are 0-based)
Radius defines the largest size of a circle around any given 'target' star to be explored, and should be set slightly larger than 321 (arcsec). Successive smaller search radii to be used for successive spoilercodes are defined in SPOILER CODES FOR AGASC1.2 STARS 
  
setenv AGASC_OUTPUT /proj/agasc1/binary/agasc1p4/agasc/
setenv AGASC_OUTPUT_FILEMAP /proj/agasc1/binary/agasc1p4/tables/outfilemap

mp_spoilers 330 /proj/agasc1/binary/agasc1p4/tables/neighbors
I suspect that spoiler codes would change little after an update such as this. After the intended spoiler update is performed, the difference between spoiler quality codes in the updated and original version should be relatively small (e.g., the fraction of spoiled stars ASP1.ne.1 should not change by more than 5%).

OTHER RELEVANT STATISTICS on AGASC1.4

A key measure of the effects of this update is the change in selected stars between AGASC1.3 and AGASC1.4. Roughly 1/3 of the Cycle 1 observations will have different star sets chosen by the CXC MP Star Fidlight Evaluator (mp_sfe) for the new AGASC 1.4 mags. How do the stars chosen in these fields differ? 
  
All Fields	AGASC1.3	AGASC1.4
		---------	---------
mean mag:	8.448		8.630
mean fom:	0.15308		0.31221
min fom:	0.13669		0.13721
max  fom:	0.32571		0.32755

Number of
guide stars:	AGASC1.3	AGASC1.4
		---------	---------
5		634  (94.9%)	 610  (91.3%)
4		 22  ( 3.3%)	  33  ( 4.9%)
3		  9  ( 1.3%)	  17  ( 2.5%)
3		  3  ( 0.4%)	   8  ( 1.2%)

Here's the breakdown on changes for individual target fields:

0 stars changed: 434 ( 65.0%)
1 stars changed: 175 ( 26.2%)
2 stars changed:  47 (  7.0%)
3 stars changed:  12 (  1.8%)
4 stars changed:   0 (  0.0%)
5 stars changed:   0 (  0.0%)

total fields analyzed:  668
total stars analyzed:	3285
The upshot is that FOMs and the number of stars per field brighter than MAG_ACA=10.2 looks WORSE in AGASC1.4 Since AGASC1.4 is simply more realistic, that's nature's fault, not ours.