MAPCOR Initial Purge Parameter Tuning

T. Evans - IPAC

[Note: Much of the work in this effort was contributed by D. Kirkpatrick and S. Wheelock -- Thanks!]
 

Finding Artifacts

To make the first cut at parameter tuning, I selected out all of the stars brighter than a certain limit --12.0 in J, 11.5 in H, and 11.0 in K -- from the MAPCOR output files of the scans in 11 photometric nights of data: However, I ran the science (survey) scans separately from the calibration scans, and most of the science scans were found in 970521n and 970608n, with a few from 971006n and 971009n, since the pipeline has not yet been run on the science scans from most nights.   Each of these bright stars was then tested to see if any other brighter source could be found within +/-10 camera pixels (20 arcsec); if not, the bright star was considered a "parent" source, and all other sources in the file within a box of +/- 60 pix were output as possible artifact sources, along with the parent source.  The output data includes the distance of the possible artifact sources from the parent sources, del_x and del_y in camera pixels, the "default" mag of each source (R2-R1 PSF mag or R1 mag),  and the difference between the possible artifact default mag and the parent default mag, del_m.  The output was split by parent source mag into bins of 0.5 mag each, and del_x/del_y plots were made of all parent sources in each mag bin, with their associated possible artifact sources.  Thus, in each plot, there is a randomly  scattered background of real sources, plus distinct groups of bright-star artifact sources found around the central parent stars at (0,0).  (Hence the need to run the science scans alone -- the calibration scans came from only a few (~6) fields, so the background of real sources was not random in plots that include the calibration scans.)

Here is a sampling of those plots:

From these plots and more zoomed-in plots, various measurements were made of the artifacts around the bright stars, including glint positions, diffraction spike lengths, and confusion radii.

A few notes on this process: 

  1. For the brightest sources, the box of +/- 60 pix was expanded to +/- 150 pix because of the length of the diffraction spikes.  However, not enough of these bright stars were available in this data, so the analysis usually began at m~3, and the plots only needed to include +/- 60 pix.
  2. Unfortunately, the test for any other brighter source within 10 pix, meant to exclude most of the actual artifacts from being considered as parent sources and muddying up the plots, initially neglected the fact that in some cases the R1 mag for a bright source can be fainter than the R2 PSF mag.  Since, for these bright stars, the R1/R2-R1 merge doesn't always happen because of confusion, that meant that the parent sources were sometimes the R2-R1 (PSF mag) sources, so the parent mags were too bright.  This affected the del_m values more than the positional difference values, because even for bright stars the R2-R1 positions are not (usually) too far from the R1 positions.  This situation occurred, apparently, quite often in the H band, but less frequently in the J and K bands.  When I noticed this was happening, it was too late to re-run the program for the initial analysis, so I simply opened up the delta-mag MAPCOR parameters for the H band glints.   It was corrected after the initial parameter measurements were done and MAPCOR was re-run, so the "after" plots do use a R1 source as a parent if one is within the 10 pix box, even if a R2-only source in the area has a brighter mag (as will all future analysis).
  3. As of yet, there really has been very little comparison with actual images of these stars, to determine if sources that seem to be artifacts in the above plots really are artifacts, since this is a very time-consuming process.  From previous experience in looking at images, and after viewing these plots, I believe that the glints are very easy to pin down, the ends of the diffraction spikes are more difficult to determine, and confusion radii are iffy at best.  Hence, the statement that this is only the initial parameter tuning!
After the artifact measurements were made, the actual MAPCOR parameters were determined as described below.  Note that all measurements are in camera pixels.
 

Determining Parameters

 Filter Glint Parameters

 The filter glint parameters were very easy to determine because, as you can see in the plots above, the glints form very compact groups in del_x and del_y compared to the random background, and even compared to diffraction spikes, etc.  This is even more apparent in zoomed-in del_x, del_y plots.  From these plots I located 2 glints in each band.  I used smaller scale plots to measure the edges of each of the glint groups in each mag bin, then pulled out only the sources in the glint regions to examine the del_m ranges.

Glint #1 in each band is very similar across the bands, but glint #2 is different.  The following plots show the ranges of del_x, del_y, and del_m for each of the glints in each band.  Again, remember that most of the del_m values for glint #1 in the H band use the wrong parent mags and thus are incorrect; the only ones using R1 parents are those with del_m ~ 6-7, so I took that as the delta mag and opened up the sigma very large.

 The derived MAPCOR parameters are found in the table below; all positional parameters are in units of camera pixels.  [The glint positions are (ex+f, gy+h), where the parent source is located at (x,y); the search radius about that position is glint_r , and the mag difference is glint_delm, within +/- glint_sigm.]
 
Parameter name J band H band K band
glint #1 glint #2 glint #1 glint #2 glint #1 glint #2
glint_e
1.0
 1.0 1.0 1.0 1.0 1.0
glint_f 1.0 2.0 0.5 2.7 0.5 -2.0
glint_g 1.0 1.0 1.0 1.0 1.0 1.0
glint_h -7.0 -14.2 -7.0 -3.0 -7.5 8.0
glint_r 1.0 1.0 1.0 1.0 1.0 1.0
glint_delm 6.7 10.0 6.6 8.2 7.7 7.2
 glint_sigm 1.6 2.4 3.0 3.0 1.6 1.6
 
 

Diffraction Spike Parameters

The diffraction spike lengths and widths were a bit more difficult to work out.  For one thing, the ends of the spikes are less "clumpy" than the glint positions; also, as you can see in the del_x/del_y plots, the 4 spike lengths and densities are not symmetric, as originally supposed.  (This probably indicates that some spikes, for whatever reason, stand out above the background image noise better than others so more sources can be extracted from them.)  Another small concern is that the spikes seem to be at an angle of 1 or 2 deg from the vertical, not 0 deg as originally thought.  Also, in the H band, there is a possible set of extra spikes found only below the parent source and at a ~45 deg angle (H band glint #2 would be part of these spikes).

First I measured the lengths of each of the 4 spikes in a mag bin plot by measuring the points at which the linear patterns seem to peter out into randomness, and took the longest length as the length of the spikes in that mag bin.  I then plotted the logs of the lengths vs. mag, and tried to fit straight lines to those plots.  Unfortunately, linear log(length) vs. mag may not be a good approximation of the actual spike growth, but more data is needed to be sure.  Since it was decided to err on the side of marking more sources as artifacts, if necessary, instead of using a line fit I drew a straight line on the log(length) vs. mag plots that would encompass all of the measured lengths -- so that any calculated spike length would be at least as large as the measured length on one of the plots -- but would still minimize the lengths at both the bright and faint ends as much as possible.  The resulting log(length) vs. mag plots with the encompassing lines are found here:

The diffraction spike widths were measured from both small and large scale del_x/del_y plots, and is a bit larger than absolutely necessary to again be "generous" in selecting artifacts and make sure that the possible small spike angle from vertical will be taken into account.

The derived MAPCOR parameters are found in the table below.  [The spike widths are diff_width camera-pix in half-width, and lengths are l0[10^a(m0-m)] camera-pix; if a source in a spike is brighter than the parent source mag + diffrlsc_dm, it is considered a real source that is contaminated by the spike, instead of a spike source.]
 

Parameter name J band H band K band
diff_width  2.0 2.0 2.0
diff_l0 50.1 38.7 30.0
diff_m0 6.0 6.0 6.0
diff_a 0.25 0.28 0.30
diffrlsc_dm 8.0 8.0 8.0
 

 Confusion Radius Parameters

The confusion radii were very difficult to determine from the plots alone; the inner confused area is quite evident on the del_x/del_y plots, of course, but I know from prior examinations of images that beyond that inner group there are always a few scattered sources, which may or may not be halo artifacts.  (This issue needs to be studied a great deal!)  Therefore, to measure confusion radii I examined not only the del_x/del_y plots of the science scans, but also of the calibration scans, where the non-random background of real sources indicates at least some upper limits on the sizes of the halos and confusion radii.  (Note that "confusion" in MAPCOR has two different components:  artifact sources extracted from the halos of bright stars, and real sources close to bright stars that are thus contaminated by the halos.  More work needs to be done on both of these components to better tune the parameters.)

A few examples of del_x/del_y plots of calibration scans are found here:

Looking at these plots helped a lot, especially for the brighter parent sources.  Measuring the confusion radii was a very subjective process; basically, I used a combination of two methods for each mag bin:
  1. On the calibration scan del_x/del_y plots I tried to find the source with several extractions (thus likely to be a real source) closest to the center but outside any tightly-clumped "inner halo", and used the radius to this source as an upper limit on the confusion radius.
  2. On the science scan plots I measured the distance from the center at which the background source density becomes similar to that of the outer regions of the plot.  (This only works well for the mid-to-faint range of mag bins, where the background density is higher.)
After measuring (or guessing) the radii, as for the diffraction spikes I plotted log(radius) vs. mag and fit straight lines to those plots.  Unlike the diffraction spike case, however, a linear relationship between these two quantities seems adequate, at least at this time.  The log(length) vs. mag plots with the linear fits are found here: The derived MAPCOR parameters are found in the table below.  [The confusion radii are r0[10^b(m0-m)] camera-pix.]
 
Parameter name J band H band K band
conf_r0
 6.14 4.72 3.55
conf_m0
 6.0 6.0 6.0
conf_b
 0.197 0.216 0.265
 
 
 

New MAPCOR Results

After replacing the parameters derived above in the namelist, MAPCOR was re-run on all of the science scans used in the above analysis.  The analysis program was then re-run, and del_x, del_y plots were made for these new results, with different point types indicating different "purge flag" settings.  These are as follows: Not all 12 seemingly possible purge flag settings are found in this list; these only show the combinations of settings that MAPCOR allows.

Some sample del_x/del_y plots are found here:

The sources with purge flags set which are scattered (not always randomly) in areas outside the expected artifact locations are simply artifacts of other parent sources nearby.  (These nearby parent sources usually will not have an * as a point type.)  Note that the diffraction spike lengths, as mentioned earlier, seem too large in most of the plots; again, more work is needed to understand the real behavior  -- length and brightness -- of the spikes compared with the parent magnitude.

Some things on the "to do" list are:

 
 


e-mail: T. Evans
Last modified: 6 Nov 1997