Pipeline processing of IRAC data also included more advanced processing of many individual IRAC frames together to form more advanced data products. Known by the generic title of “post-BCD” processing, this extended pipeline refined the telescope pointing, attempted to adjust background offsets of CBCDs and produced mosaicked images. We did not attempt to improve (relative to the BCD) the point source or extended emission flux calibration by automatically comparing to a reference source catalog. The mosaic so created only included data from within a single AOR.
5.3.1 Pointing Refinement
All IRAC BCD images contain a pointing estimate based on the output of the Spitzer pointing control system (star tracker and gyros), i.e., the boresight pointing history file. This initial pointing estimate is accurate to about 0.5 arcseconds. To improve the ≈ 0.5 arcseconds blind pointing, a pointing refinement was run in which the point sources were identified in the IRAC frames and astrometrically correlated with stars near the source position in the 2MASS catalog. The pointing refinement typically improved the positional error to < 0.3 arcseconds and removed any systematic offsets.
First, point sources were extracted from the pipeline-processed mosaics and transformed to R.A. and Dec. using the transformations derived from the current pointing information. If there were fewer than five sources in an image, then there would be no refinement for that BCD. Second, a comparison was made of the coordinates of each point-source for which a match was found in the 2MASS point source catalog. Then a translational and rotational reference frame could be computed from the positional differences between the raw coordinates and the 2MASS point source catalog positions, and a refinement was made to the celestial coordinates and angles of each BCD in the observation or AOR used for the mosaic. These refined values were written to the FITS headers as RARFND, DECRFND, including many others with “RFND” as an indicator of refined pointing. More pointing related information can be found in Section 4.12.
5.3.2 Superboresight Pointing Refinement
Pointing refinement operated on each IRAC channel independently. This often resulted in poor pointing solutions for channels 3 and 4, in which stellar fluxes were lower and the background higher than in channels 1 and 2. We have therefore developed a technique which combined the results of pointing refinement in channels 1 and 2 and applied it to all four channels using the known offsets between the IRAC fields of view. This improved pointing solution was derived during campaign reprocessing. The results of the pointing refinement from the first run of the post-BCD pipelines were averaged, and the correction derived from this was applied to the boresight pointing history file (which contained the pointing estimate derived from the spacecraft telemetry and which provided our initial pointing estimate). This corrected pointing history file (the “superboresight” file) was then applied to the BCD at the pointing transfer stage of the BCD pipeline. The superboresight R.A. and Dec. estimates were recorded in the CRVAL1 and CRVAL2 FITS keywords, and the position angle estimate was recorded in the CD matrix keywords. The uncorrected R.A. and Dec. were retained, but called ORIG_RA, ORIG_DEC, as was the pointing refinement solution for each frame (as RARFND and DECRFND). Note that to use the superboresight solution, USE REFINED POINTING = 0 should be set in the MOPEX namelists.
Superboresight was implemented as a patch to the S13 software build (see Appendix A for more information on software builds), thus most (but not all) data processed with S13 or subsequent pipeline versions will have it. Users should check for the presence of the ORIG_RA, ORIG_DEC keywords to see if it has been applied to their data. From S14 onwards, the HDR data have the long frame R.A., Dec. solution copied to the short frames, as the short frame pointing solutions are less accurate. If the BPHFNAME value in the (C)BCD FITS header does no start with “S,” then the superboresight did not run. Neither superboresight nor pointing refinement were run on the subarray data.
In a post-mission effort to improve Spitzer pointing refinement, proper motions of 2MASS stars were updated using the more recent and more accurate measurements made by the Gaia spacecraft. In the course of this effort an apparent bug was discovered in the superboresight code that reduced the magnitude of the correction applied to the original pointing solution to bring the (C)BCD FITS keyword CRVAL1 into agreement with the 2MASS pointing refinement. Comparing the differences between CRVAL1 and ORIG_RA, and RARFND and ORIG_RA (C)BCD header FITS keywords revealed that the CRVAL1 keyword values were underestimated by a factor of cos2 (declination) over much of the sky. This behavior was seen in both the cryogenic and warm mission data.
To correct for this error in the CRVAL1 keyword values, we performed a scaling by 1/cos2(dec) to the superboresight-derived corrections for (C)BCDs with |Declination| < 75°, and wrote the new, corrected (C)BCD CRVAL1 values to the FITS headers. For |Declination| > 75° CRVAL1 was instead replaced by the original pointing correction RARFND value in the (C)BCD FITS headers. After correcting for the superboresight CRVAL1 issue, pointing refinement improves the Spitzer IRAC (C)BCD positional information (CRVAL1 and CRVAL2) to better than ±0.16 arsec (one sigma).
5.3.3 Mosaicking and Coaddition
The post-BCD pipeline modules have been made available for general public use as part of the MOPEX tool. They consist of a number of C and C++ modules connected via Perl wrapper scripts. Namelists (plain text lists of parameters and values) are used for input. In most cases, the module operation simply consists of supplying the software with a list of input image files; by default it reads and understands the IRAC image headers.
Using the refined coordinates, individual IRAC (C)BCDs from a given observation (AOR) in a given channel are reconstructed onto a larger field (mosaicking), and overlapping frames are averaged together to achieve a higher SNR. The pipeline SSC mosaicker produces a single image, one per channel (and one per exposure time in case of the HDR observing mode) from many input (C)BCD images (the first long[er] frame[s] in non-HDR observations is [are] not included as it [they] is [are] taken in a different mode, the HDR-mode). First, the BCDs are corrected for overlap consistency. The images that overlap are forced to have the same background value via addition of an offset derived from a least-squares fit. Then a “fiducial frame” is derived. This is the definition for the output frame in terms of its physical size, projection, and orientation. Because IRAC has such a large field of view, projection effects are non-negligible, and the mosaicking and coadding process must reproject the data onto a common grid with a technique similar to “drizzle” (Fruchter & Hook 2002). The fiducial frame finder seeks to minimize the amount of “blank” area in the output mosaic by rotating the output projection such that it is aligned with the map axes. This is useful for long thin maps, where potentially the output mosaic could be very large, but with a great deal of empty space. The mosaicker then reprojects all of the input data onto the output projection. It reads the SSC WCS, which contains the field pointing center, rotation, scale, and instrument distortion, and reprojects this onto a standard TAN FITS projection. In the process, the data are undistorted. The reprojected images are interpolated onto the fiducial image frame with outlier rejection, rejecting radiation hits in overlapping observations. Various masks (mostly the imask and pmask files, see Section 7.1, Table 7.1 and Table 7.2 for which mask bits are set in these masks) are used in the coaddition in such a way that the pixels previously flagged as bad (for example, hot or dead pixels) are rejected before the averaging process. The outlier rejection scheme is specifically designed to work well in the case of intermediate coverage and may not be adequate for all observations and science programs (it used the radhit, dual outlier, multiframe, and box outlier modules, see the MOPEX documentation for more details on these outlier methods) . The pixel size in the mosaics produced by the pipeline was exactly 0.6 arcseconds x 0.6 arcseconds in the final data processing. In addition to a sky map (in units of surface brightness), a noise image and coverage map were also produced.
