These are the gory details of image processing and combining. The pixel scale is 0.257 arcsec per pixel. Individual exposures are 600s in B,V and z', and 900s in R. MSCRED has been used to process the data through defringing, with custom software deployed after that. Where any variation from the standard procedure was required to adjust for conditions, we note it, but only to give the flavor of this variation. We do not believe that any variation propagates to the final images. The processing steps performed on the individual images are:
WCS correction. Each individual image WCS is matched to the USNO-A2 star catalog (using the MSCGETCATALOG and MSCCMATCH tasks) in order that the proper RA,DEC coordinates for each object in each frame be established. After this step, the uncertainty in the GLOBAL RA,DEC coordinate system relative to the USNO-A2 are less than 0.01 arcsec.
Crosstalk correction. We have used the NOAO-determined crosstalk corrections.
Overscan subtraction. We the relevant parameters from the IRAF task CCDPROC are:
(function = "legendre") Fitting function
(order = 3) Number of polynomial terms or spline pieces
These parameters have been used for all the runs, with the exception
of the December 1999 KPNO run, which was partially compromised by bias
jumps in the MOSAIC CCDs 5 and 6. For that run, the overscan was
computed and subtracted line-by-line. The differences in the final
overscan subtracted data are quite small, although the December 1999
data is slightly noisier as a result of the less accurate overscan
subtraction (but still dominated by sky noise).
Zero correction. The zero image is constructed for each night of the run, from an average of 20 individual zero frames taken before the beginning of observing (and/or immediately after). Stability of the zero pattern has been found to be very good, so the second version images (not available for this release) will be constructed using the zero made from the entire run's worth of zero frames (which will result in a minimal increase in S/N). The zero images are combined using MSCRED's zerocombine task, with the following combine parameters.
(combine = "average") Type of combine operation
(reject = "minmax") Type of rejection
(nlow = 1) minmax: Number of low pixels to reject
(nhigh = 2) minmax: Number of high pixels to reject
Flatfield pupil removal. For KPNO z-band data, the Mosaic corrector produces a ghost image of the 4m primary on the image. This ghost image is present for both the dome flats and the on-sky images. In order to preserve the photometric integrity of the data, we have chosen to model and subtract the ghost pupil from each z-band image rather than treating it as a flatfield feature. This is accomplished using the MSCRED tasks MSCPUPIL to model the pupil and RMPUPIL to subtract it.
Flat field correction. Each night during the run, 7 dome flats are taken for each filter. Images from the night are processed using the dome flats for that night, in order to remove the pixel-to-pixel sensitivity variations. Before being applied to the data, the dome flats are normalized so that the large-scale gradients are identical among the nights, so that meaningful dark skyflats can be constructed. This first version data has been processed using the individual night's dome flats, as has all the data that is fed into the transient pipeline. The final release data will use combined dome flats from each run, resulting in a 7-10% increase in S/N. Dome flats are used rather than twilight flats because of the extreme difficulty in obtaining sufficient twilight flats for each filter over a single run, let alone a single night. The large scale residual features are removed in the next step. The domeflats are combined using MSCRED's flatcombine task, with the following parameters set:
(combine = "average") Type of combine operation
(reject = "avsigclip") Type of rejection
(lsigma = 2.5) Lower sigma clipping factor
(hsigma = 2.5) Upper sigma clipping factor
Sky flat correction. The images in each filter for the entire run (that are not affected by significant moonlight) are combined to make a skyflat. In general, there are several nights for which only 5-10 images in a given filter are taken. Therefore, it is not in general possible to generate a night-sky flat for each filter each night. We have chosen to use the images for the entire run to generate the skyflats. Because the images have already been dome-flattened, the sky flats serve primarily to remove the large-scale features due to the varying chromatic sensitivity across regions of the CCDs. The relevant parameters for MSCRED's SFLATCOMBINE are:
(combine = "average") Type of combine operation
(reject = "minmax") Type of rejection
(nlow = N/6) minmax: Number of low pixels to reject
(nhigh = N/2) minmax: Number of high pixels to reject
Where N is the number of images to be combined. Although this has
been the general rule, skyflats have been generated iteratively until
no spurious features are detectable in the images.
Pupil removal. For the KPNO z-band images, this first-pass skyflat is used to generate a model pupil image, using the MSCRED task MSCPUPIL. This is then subtracted from each of the z-band data images (using RMPUPIL). A second pass skyflat is generated from the pupil-subtracted images, and this is used to generate the fringe image.
Defringing. For both KPNO and CTIO z-band data, the very significant fringing needs to be removed prior to combining. This is done using the following method: First, a smoothed version of the skyflat image is generated, using IRAF's fmedian task and a medianing box of 257 pixels (which is larger than the fringe frame pattern scale!). This image is then subtracted from the skyflat image, yielding a fringe pattern image. This fringe pattern is then subtracted from all the (pupil-subtracted) data images, which are then combined to make a final z-band skyflat for the run. With this iterative procedure, small errors in fringe image construction are incorporated into the skyflat. When this is applied, the fringe pattern is completely removed (at the cost of a typically very small photometric error of <0.002 magnitudes).
Masking. Regions affected by scattered light (especially important for the 1999 KPNO data, and March 2000 CTIO V data, where there was a crack in the V filter above chip 4)) or bright satellite trails are masked out. The areas are marked manually with SAOimage or DS9, and a region file written. This region file is then combined with the default mask for the run (the NOAO bad pixel masks or a similar mask produced by us) to produce a custom mask for that image. Most images don't have scattered light problems, and simply use the default mask. Note that the affected areas are not erased or fixpixed, because pixels that are in the BPM file are just not considered in the coaddition (see below). Note that not all satellite trails are masked out, just the ones that come to our attention when SExtractor (below) fails because of it.
Sky subtraction. We use SExtractor to make a catalog and a sky-subtracted version of each chip. The sky subtraction parameters are -BACK_SIZE 256 -BACK_FILTERSIZE 7. The idea here is to remove any small variations in the sky which would introduce artifacts into a combined image. The catalogs are matched to each other to provide photometric offsets and coordinate fits for registration.
Coaddition. dlscombine combines multiple images by averaging with a 3-sigma clip. Before the actual combine at each output pixel, it applies the coordinate fit to look up the appropriate location in each input image, applies the appropriate bad pixel mask, does a bilinear interpolation if all the surrounding pixels are good, and applies the photometric offset. It also enforces a saturation limit of 25000 counts in the following way. The output is the average of all inputs which were less than 25000 before sky subtraction. If all inputs were 25000 or more, the output is 25000. This means that some areas of sky normally under bleeds can be recovered from those input images where the bleed was stopped by a CCD boundary.
Any questions about the application of the mscred tasks should be directed to Ian Dell'Antonio and any questions about the custom software should be directed to David Wittman.
Go Back to the Data Release documentation.