From Galex Wiki
Contents |
Chapter 8: Changes to the Imaging Pipeline for Data in GR7 and Beyond
This chapter describes the changes made to the GALEX pipeline mainly for data taken after May 4, 2010. Anyone making use of data taken after that point should read this chapter.
Introduction
After a "Coarse Sun Point" event on 2010 May 4, it was noticed that point sources on the reduced NUV direct images were spread out (streaked) in the Y detector direction, resulting in a degradation of resolution of a factor of >5 in Y (the X direction was not affected). This behavior is likely due to the partial failure of a capacitor associated with an A/D converter. An adjustment to the NUV digitizer TAC (Time-to-Amplitude-Converter) scale on 2010 June 23, significantly improved the Y direction photon spread, although still left a factor of >2 Y resolution degradation.
Fortunately, the photon position deviations were found to be correlated with the YA value (found in the raw photon position data). After much testing and some changes to the GALEX pipeline code, we have recovered the previous resolution of the reduced GALEX images to within about 0.1 arcseconds FWHM on average (for the data after June 23rd).
Unfortunately, we were unable to completely correct a small number of the photons, which results in 'ghost' images of bright sources. These ghosts generally amount to less than 1% of the flux of the primary source and appear between 30 and 60 arcseconds above and below the primary source in the Y detector direction. The detector X position of the primary and ghost are the same (within the error). Although the ghosts will appear on the GALEX images, they do not appear, generally, in the '-xd-mcat.fits' catalog, but are instead absorbed into the flux of the primary source.
For data taken between 2010 May 4 and 2010 June 23, sources generally have about 25% larger FWHMs and brighter ghosts. Most of the data during this time period have been given a FAIL grade during the QA process.
Image Comparison
Below are shown three GALEX images near NGC 0895 (12 arcminutes on a side).
The first image shows data before the May 4th, 2010
CSP event reduced with the old GALEX pipeline.
The second image shows the same area of sky observed
with GALEX after the CSP event reduced with the old
pipeline. And the third image shows the same data
reduced with the new GALEX pipeline.
| 2009-11-06 (pipeline version 7.0) | 2010-10-25 (pipeline version 7.0) | 2010-10-25 (pipeline version 7.1) |
|---|---|---|
| 
| 
|
Current Status of PSF
Over the course of the mission (since 2003), the FWHM of the PSF has increased at a rate of about 0.03 arcseconds per year. The plot below shows FWHM data covering the period from 2009-2010 averaged in approximately one week intervals. The data between 2010 May 4 and 2010 June 23 are notably higher and afterwards the FWHM returned to within 0.1 arcseconds of the trend line.
The plot below shows the distribution of FWHMs for point sources averaged over each eclipse in two time periods, before and after the CSP event in May. The median of the distribution is about 0.12 arcseconds higher in the recent data.
The plot below shows the distribution of FWHMs in the X and Y detector directions. Over the entire mission, the Y direction FWHM has been about 0.8 arcseconds (on average) larger than the FWHM in the X direction. In the most recent data, the Y FWHM has been degraded after the CSP event, while the X FWHM remains approximately the same.
The plot below shows the X (in black) and Y (in red) FWHMs as a function of detector X position (averaged over Y).
The plot below shows the FWHMs as a function of detector Y position (averaged over X). In both of these plots the FWHMs degrade near the edge of the detector, especially at high detector Y positions. The most recent data shows a slight degradation in Y FWHM near the center and upper center (X~0,Y>0) of the detector, although the differences are small compared to the variation in FWHM across the detector. Note that sources appear most oblong at high and low Y detector positions and most circular at high and low X detector positions.
Ghosts
As mentioned abvove, GALEX data taken after 2010 May 4 exhibit 'ghost' images of bright sources that appear about 30-60 arcseconds above and below the primary source in the Y detector direction. Fortunately, the upper ghost image flux is less than 0.05% of the primary flux over most of the detector in GALEX data taken after 2010 June 23. The lower ghost image is generally a factor of 2-4 fainter than the upper image.
In the plot below, the upper ghost image flux percentage is shown as a function of Y detector position for 3 time periods in the year 2010. The first time period shown in red is after the initial CSP event on May 4th until the TAC scale adjustment on June 23rd. The ghost images are significantly brighter during this period. The FWHM is also much worse during this period and most of this GALEX data has been given a FAIL grade. The second time period, June 23rd to September 30th is shown in black, and the third period, October 1st to December 31st is shown in green.
These last two time periods only show significant ghosts between Y detector positions -250 to +50 pixels (see 'nuv_det_y' in the -xd-mcat.fits files). The average ghost flux percentage in this region is 0.5%, but can vary by a factor of 2 with the X detector position. Over most of the detector the upper ghost flux will be less than 0.05%. The lower ghost flux can be as high as 0.5%, but mostly is less than 0.03%.
In the first time period the ghost image appears about 60 arcseconds from the primary source. In the second period, the ghost is about 32 arcseconds from the source, and about 45 arcseconds from the source in the most recent period. In almost all cases, these ghost images are either too faint to be found by the GALEX pipeline source extractor, or will be absorbed into the flux of the primary source as part of the PSF wing.
The plot below shows the spatial profile of the ghosts in the wings of the primary source. These data were created by combining hundreds of bright sources in the region with Y detector positions between -100 and 0 pixels during the second time period.
Ghost images can be identified by position (same X detector position, shifted about 30-45 arcseconds from primary in Y) and intensity (less than about 1% of primary). They will also have the same spatial profile as the primary. They can also be seen by 'blinking' exposures of the same field with different roll angles. The ghosts will appear to move around the primary.
For coadds created from exposures with several different spacecraft roll angles,
the ghost images will appear even fainter relative to the primary source,
but could appear as a ring of faint ghost images around bright point sources. However, for fields with enough visits, we are able to mask out the ghost images (see below).
In the following three images, ghost images appear above and below a bright star (4 arcminutes on a side).
The ghosts appear at different position angles depending on the roll angle,
and at different intensities depending on detector position.
In the first image, a single ghost appears about 60 arcseconds directly north of the star.
In the second image, the ghost appears northeast of the star, and a second
ghost appears on the opposite side of the star.
In the third image, the northeast ghost of the star is fainter, and the ghost to the southwest is brighter. These examples represent a worse case (taken June 17-18, 2010), and generally
the ghosts (taken after June 23rd, 2010) will be fainter and closer to the bright star.
| Roll= -2.0, X= -205, Y= -100 | Roll= 19.0, X= -254, Y= -23 | Roll= 12.9, X= -273, Y= -42 |
|---|---|---|
 |  |  |
Setting the Ghost Flag
A new artifact flag (10th bit, value=512) is set in the "-nd-flags.fits" image for eclipses greater than 37460. This flag is set for areas of the image which may contain a significant ghost image due to the stim spreading which started in May 2011. Flagged pixels are masked in the coadded images.
For any visit with eclipse greater than 37460 (May 2010), a subroutine is run which estimates the flux from the ghost for each extracted source in the -xd-mcat.fits catalog. The ghost flag is set when the predicted S/N of a ghost is greater than 3. The prediction is based on the flux of a primary (extracted) source and the ghost ratio read from a "ghost ratio" map. Below is shown one map for the upper ghost (stronger) and one map for the lower ghost (weaker). The scale is 0 to 0.005 (ghost/primary):
The position of the ghost is based on the value of a quantity called the YA stim spread slope (ya_stim_spread_slope) for a specific eclipse and the formula:
Y Detector Offset = (30.85590 * ya_stim_spread_slope) - 3.84777
The value "ya_stim_spread_slope" is taken from second number in the first line of the file "yac_coef_(eclipse).dat' which appears in each visit directory where a stim correction was applied. The stronger ghost appears above the primary source (lower Y detector coordinate value) and the weaker ghost appears at the same offset below the primary.
An artifact flag (512) is set in the '-nd-flags.fits' image within a 12 arcsecond radius of the predicted ghost position. Any object in the "-xd-mcat.fits" file whose center position falls in this flagged area will be assigned the ghost (512) artifact flag. No change is made to the intensity image at the visit level.
Artifact Flags
The revised artifact flags will be as follows:
Artifact 1( 1):(edge) Detector bevel edge reflection (NUV only). Artifact 2( 2):(window) Detector window reflection (NUV only). Artifact 3( 4):(dichroic) Dichroic reflection. Artifact 4( 8):(varpix) Varible pixel based on time slices. Artifact 5( 16):(brtedge) Bright star near field edge (NUV only) Artifact 6( 32):Detector rim(annulus) proximity(>0.6 deg fld ctr) Artifact 7( 64):(dimask) dichroic reflection artifact mask flag Artifact 8(128):(varmask) Masked pixel determined by varpix. Artifact 9(256):(hotmask) Detector hot spots. Artifact 10(512):(yaghost) Possible ghost image from YA slope.
These artifact comments are now included by new galexmerge in the header of the '-xd-mcat.fits' file.
Coadding Changes
Changes were made to the portion of the GALEX pipeline that coadds tile visits so that any ghost flagged pixels in the visit level '-nd-flags.fits' image will not be coadded in the final coadd images (including the coadd intensity image). Only coadds which contain an eclipse greater than 37460 will be affected.
Examples
The following in example is a coadd of visits 15,17,24,32, and 33 from the tile "GI2_141001_HDF_South" (these visits are within the eclipse range 43828 to 43862). Below are the intensity images of visits 17, 24, and 32 near a bright star (levels set to 0 to 0.02). The ghost is apparent to the lower left of the star. The position of the ghost varies with roll angle.
Here is coadd produced with the current pipeline and a coadd produced with the new ghost flagging. In the current pipeline the ghost appears as multiple "sources" to the lower left of star.
Here is the resolution response image (-nd-rrhr.fits) for the same section of the image around the bright star. The large pixels are due to the lower resolution of the flag image. The scale is 0 to 4500.
Another Example
The first 3 images below are from the intensity images for visits 24, 26, and 39 for the tile "GI2_141001_HDF_South". Window and Dichroic ghosts are seen near the bright source. The dichroic ghost moves with roll angle; the window ghost is fixed. There is a YA ghost visible below the last source in the last visit (#39). In the final (4 visit) coadd image, the dichroic ghost is repeated, and YA ghost is removed.
New Calibration
A new response correction has been applied to data taken after the TAC scale change on 2010 June 23 (at eclipse number 38150). The response was found to be 1.8% greater in these data, so a scale correction factor of 1.018 was applied uniformly across the detector. The calibration scale was computed in the same manner as with prior GALEX data (GALEX general release 6)-- using the star LDS 749b and a very large aperture. Therefore, the new calibration includes any contribution from the ghost images (averaged across the detector). However, this contribution is less than 0.1% on average, well within the flux calibration accuracy (about +/-1%) for individual sources.
