1. Overview

IRAC as designed relied heavily on it's shutter for calibration. In regards to darks, the shutter was used to seal off the inside of IRAC so that the arrays could not see the sky (or any other light source). Images were then taken which would be used as the calibration for the IRAC bias and dark current response. In this respect the data reduction for IRAC would most strongly resemble the methods used for optical CCD data.

Due to the decision not to use the shutter in flight, such data will now never be taken. IRAC will always be exposed to the sky, and hence all IRAC images will contain flux from the diffuse celestial background (such as zodiacal light) as well as compact objects such as stars and galaxies. As a result, we will now turn to a data reduction method that more strongly resembles traditional ground-based data reduction techniques for infrared data.

"Dark Sky" images will be taken in pre-determined regions which have been specifically chosen to be the darkest parts of the sky and relatively free of bright objects such as stars. Just as when we had the shutter, we will take these darks using all the modes and exposure times of IRAC. This data will be combined into "sky darks" using outlier rejection which will include single-frame object detection and flagging.

These sky darks will be subtracted from all the data in the same fashion that the shutter-based darks would have been. Since the sky darks contain any additive term from the bias or dark current, these are subtracted out from the data. This is nearly the same as how "median sky subtraction" is done for ground-based data. Because this naturally oversubtracts the real data, we will include an estimate of the sky brightness (based on a model) in the header for later use by the science thread, which will try to add this back as a constant.

Another change is that the data will now be taken using AORs instead of IERs. On the one hand this should simplify processing because each AOR will use only a single exposure time, unlike the IERs where any number of modes could have been mixed together. On the other hand, this also means that something will have to set the pipeline script number, since now all calibration data will look like science data.

2. Detailed Description Of Thread

The following are the detailed steps to be taken in the thread. For each incoming DCE the following steps will be performed. They closely resemble the current processing. I have left out modules that do not yet exist, like the first-frame effect correction module.

1. TRANHEAD will process the headers and add keywords.
2. INSBPOSDOM will flip the sense of channels 1 and 2.
3. CVTI2R4 will convert to floating point.
4. IRACWRAPDET1 will detect wraparound.
5. IRACWRAPDET2 will detect doughnuts.
6. IRACWRAPCORR will fix the wraparound.
7. IRACNORM will correct for barrel-shift and fowler sampling.
8. Channels 3 and 4 will be corrected for bandwidth effects.
9. A "lab dark" will be subtracted from the data. A library of these will be supplied by the IST, and will be based on ground testing. They are needed to remove as best as possible the absolute bias level prior to linearization. The also eliminate the muxbleed from hot pixels.
10. FOWLINEARIZE will linearize the data. This linearized frame will be saved as input to the image combination step.
11. The linearized frame will be divided by whatever currently existing flat-field exists. Initially this will be a ground-test flat.
12. DETECT is used to generate a mask file that flags all the pixels containing objects in the flattened frame.

Then, once all the data in an AOR for a given channel has been processed in this fashion,

1. The linearized frames and the object-masks are used as input to FLATFIELD, which combines the input data with outlier rejection. Note that the output image must be in units of DN, so we use the -o4 output from FLATFIELD.
2. An estimate of the sky brightness in MJy/sr will be inserted in the header as SKYMODSB. This will be via a module that accepts as input the mean RA, DEC, wavelength, and UT of the input images, and outputs a model brightness. This module does not yet exist, but should be based on the modeling work being done by Bidushi Bhattacharya and Bill Reach.

A few extra notes:

• For simplicity, IMFLIPROT is not used in this thread. This way the darks are in exactly the form they need to be when they are subtracted from the science data in the science thread.
• Radhit rejection will be done via the outlier rejection in FLATFIELD, so the single-frame radhit detector does not appear here. The dark AORs will generally contain 18 or so frames. DETECT is tuned to trigger on real sources, not radhits.
• MUXBLEED was left out of this pipeline. Because the input is heavily dithered, muxbleed from celestial objects will be rejected by the outlier rejection during image combination, or will be rejected immediately by by the object detector. As for muxbleed arising from hot pixels, these should be considered "features" of the darks. Since they will be in all data frames, the appropriate thing to do is leave them and allow them to be directly subtracted, rather than trying to remove them with an algorithmic approximation.

3. Demonstration of Prototype Thread

A prototype of this thread was created using the actual S6 pipeline modules and a pair of CSH wrappers. The first csh wrapper is called iracdarkprep.csh and it performs all of the initial steps needed to make the linearized data and the object masks. The other wrapper is called iracdark.csh. It scans the current directory for data, repeatedly calls iracdarkprep, and then runs FLATFIELD to make the dark.

Input data for this test was generated with the IRAC Science Data Simulator (ISDS). The science truth image was a model of the zodiacal background scaled to the "low" level described in the PICD. The background objects are based on the SWIRE extragalactic model. This is basically the same model used in previous tests. Here, a new simulator version was used. Relevant changes for this version include a crude first-frame effect model that inputs a fluctuating bias and a more complete cosmic ray model. Also, a different dither strategy was used. The dither strategy and times were selected to allow for good dithering with a sufficient number of frames for object rejection. There were 9 mapping positions arranged in a 3x3 grid with 100" throws. The large cycling dither pattern with 3 dither positions was used, giving a total of 27 pointings per channel. Exposure times were 100 seconds in this case. Total AOR time was one hour, which is roughly the optimal number derived previously given a 3-hour flat observation (which is required to meet the 1% accuracy requirement). The raw data is shown in figure 1.

Figure 2 illustrates a single DCE passing through the thread. Basic data reduction steps are taken, a mask is generated, and then all the data is combined. The final combined sky darks are shown in figure 3. Note that the sky darks don't actually resemble the "darks" as we know them. This is because they are dominated by the term which comes from the sky background multiplied by the flat-field. Because a lab dark was subtracted from all the data, these sky darks actually contain a dark term which is the delta from the lab dark.

Fig.1 - Raw simulated data of a low zodiacal background region. Channels 1 and 2 are at top, and 3 and 4 are at bottom.

Fig.3 - Derived "sky" darks for all channels. The sky darks are dominated by signature of the bright sky multiplied by the flat-field.
Channel 1	Channel 2	Channel 3	Channel 4