IRAC Cryogenic Image Features and Caveats
This page summarizes the most common IRAC image features, including image artifacts and radiation hits. We describe the features, show images which have representative examples of them, and provide a recommended mitigation method for their removal from the data.
Chapter 7 in the IRAC Instrument Handbook has more information on these features.
Please note that the IRAC pipeline now mitigates several features in IRAC data, including column pulldown, muxbleed, muxstripe, banding and (mild) saturation. These are noted below. To take advantage of the artifact-mitigated data, one should start the data analysis with the "..._cbcd.fits" files in the bcd/ directory. If a user is not happy with pipeline artifact mitigation, then there is always the option of going back to the "..._bcd.fits" files.
Demonstration of artifact removal in channels 1 and 2. In this case, the column pulldown, muxbleed and muxstripe have been mitigated with the pipeline algorithms.
1. Bias Levels (Darks) and Bad Pixels
2. Electronic Artifacts
3. Optical Artifacts
4. Cosmic Rays and Solar Protons
1. Bias Levels (Darks) and Bad Pixels
The true median dark current is very small compared to the current from the sky background at the darkest part of the celestial sphere. The dark bias in labdarks, (measured in ground tests with the cold IRAC shutter closed and therefore with no photon flux) is not zero, but has a significant pixel-dependent offset, usually positive, which depends on the frame time and the Fowler number, as well as the history of readouts and array idling over the previous several hours. Due to this "first-frame effect" the first frame of every commanded sequence of observations will have a very different bias offset from the rest of the observations (see Section 5.1.10 in the IRAC Instrument Handbook). The bias offset in channel 3 is by far the largest. In order to mitigate this effect, the first frame in the sequence is taken in the HDR mode. This means that two frames are generated: a very short exposure followed by a long one.
While the effect is largest for the first frame of an observation, a residual bias pattern exists for all data but at a lower level. The residual pattern may be significant in channels 3 and 4 for observations using repeats and/or fixed mapping patterns and low coverage (< 4 images per sky location). One possible symptom of significant residual pattern noise is a large-scale (and artificial) gradient in the pipeline-produced mosaics.
We can break the offset down into contributions beginning with the largest spatial scale down to the smallest. The largest part of the offset is uniform over the array, followed by the contribution of a few spatial gradients, and some pinstriping that repeats every four columns (due to the four readouts), including a few columns with odd offsets (due to hot pixels or parts of the mux), and weakest of all, pixel-to-pixel dependent offsets.
A comparison of a mosaic for an observation using repeats. The left-hand image is the channel 3 mosaic using the first frame in a set of two repeats. The right-hand image is the mosaic using the second frame (second repeat). Note the large left-to-right gradient in the right-hand image due to incorrect first-frame subtraction.
"Dead" pixels are really just very hot pixels, so hot that they saturate before the first pedestal sample (see Section 2.4.2 in the for an explanation of Fowler sampling). Hot pixels do not appear in CBCD (or BCD) images because they have been canceled by the labdark or skydark subtraction. Most hot pixels appeared after launch and are the result of hits by energetic particle nuclei. By annealing the arrays, we restore most pixels that are hit by radiation. Some of them cannot be restored, and thus they become "permanent" hot pixels. Some pixels jump randomly from normal to high dark current and back, dwelling in one state for anywhere from a few minutes to weeks, so they may not be canceled by a skydark subtraction. These are IRAC's "rogue pixels." The IRAC "static" bad pixel masks (pmasks) during the cryogenic mission were updated when significant changes in the permanent bad and/or hot pixels occur, usually every 6 months or so.
For more information, see Section 7.1.1 in the IRAC Instrument Handbook.
Mitigation. We recommend that the first short-exposure frame (taken in the HDR mode) is not used for building a mosaic (it is not included in the pipeline mosaic either). The first long frame may also suffer from a large bias offset. This can usually be corrected with a delta-bias as explained below. If there is enough data the first long frame may also be discarded. Noticeable bias patterns in the remainder of the data can be mitigated by removing a delta-bias frame from the data. As the pattern is a function of image cadence (time between subsequent frames), only the data at a particular cadence should be stacked together in creating the delta-bias frames. The delta-bias frame can be constructed by forming a robust (outlier-rejected) median stack of the data at a particular cadence. An example delta-bias code is available. Any remaining large-scale biases can be removed by overlap-correcting the BCDs before mosaicking using e.g. MOPEX.
If strong pinstriping is seen in the data, there is a jailbar corrector that can be found in the SSC contributed software pages and which can be used to correct strong pinstriping.
The permanently bad or hot pixels are masked in the pmasks and should have been masked out in the CBCDs (or BCDs). However, sometimes new hot pixels are created during an observing campaign. Single hot pixels will be removed if a proper outlier rejection is performed when making a mosaic if the coverage is adequate (at least 4). In low coverage observations the hot pixels need to be masked or edited individually in each frame by an appropriate program, depending on the software used.
2. Electronic Artifacts
A. Nonlinearity and Saturation
The IRAC arrays are slightly nonlinear at all signal levels. At levels above 30,000 DN (in the raw data) the detector response tends to be below a linear response by several percents. As part of pipeline processing, the data are linearized, based on ground calibrations (which have been verified in flight) of this effect. The cryogenic CBCD (and BCD) data are linear to better than 1% up to about 90% of full well, which is defined to be the level where we no longer can fully linearize the data, and at which saturation, by definition, begins. We flag pixels which are above the range of our linearization correction.
A comparison of a channel 1 BCD (left) showing a saturated star core and a pipeline-corrected CBCD image (right) showing the restored core.
The IRAC pipeline now performs a saturated star (point-source) recovery. The pipeline identifies candidate saturated stars, and rectifies the pixels of those stars that are lost due to saturation and/or non-linearity near the saturation limit. The 2MASS Point Source Catalog is used to help identify stars that may be saturated. Therefore, the saturated source correction does not work for sources that are extremely red and are not bright in the 2MASS Point Source Catalog. The saturation recover does not work for extended sources either. The wings of the candidate star light distribution are then fitted with the appropriate PSF to determine which pixels require replacement with the fitted function. The "recovered" pixels are then placed into the CBCD image. It is possible though unlikely that a source can be added to the CBCD by this procedure in rare occasions; therefore, observers should check instances of single band bright sources in CBCDs with the corresponding BCD. Additionally, for IRAC channels 3 and 4, an additional replacement of pixels is carried out for the locations of the so-called "bandwidth effect," which represent non-physical (i.e., artifact) contamination of the stellar profile.
For observations of very bright sources when the pedestal reads (see Section 2.4.2 in the IRAC Instrument Handbook for an explanation of Fowler sampling) are saturated, the DN value for a particular pixel, frame time, and Fowler number is double-valued for fluxes between zero and some very high value. Thus, for a bright, strongly saturated point source, the DN value will increase from some low number away from the source to some maximum value of 35,000 to 47,000 DN, and then decrease to a small, usually negative number, at the center. The source looks like a doughnut.
For more information, see IRAC Instrument Handbook Sections 5.1.13, 5.1.21 and 7.2.1.
Mitigation. There is contributed software for performing manual saturation fitting if working on the BCDs. Pixels which saturate in the pedestal reads are not flagged by the SSC pipeline and need to be identified.
B. Muxbleed and Muxstripe
Multiplexer bleed, or "muxbleed," appears in IRAC channels 1 and 2 (3.6 and 4.5 microns). It looks like a decaying trail of pixels, repeating every 4th column, with enhanced output level trailing a bright source on the same array row. Muxbleed trails from left to right in BCDs and CBCDs, and then continues at the leftmost column of the row below. It does not wrap from the bottom row to the top row.
For more significant instances, muxbleed is accompanied by a pinstripe pattern ("muxstripe"; every 4th column) that may extend over part of the image preceeding or following the bright pixel. Stars, hot pixels, and particle/radiation hits can generate muxbleed and muxstripe, and the characteristics depend on frame time and Fowler number. Hot pixels may show muxbleed in a raw image, but in the BCD the muxbleed induced by hot pixels may not be present because it was canceled in either the labdark subtraction or in the skydark subtraction.
Images showing the muxbleed effect (the horizontal dark line on both sides of bright stellar images). The dark pixels on the left side of the bright source are pixels on rows below the row where the bright source was located (and have wrapped around in the readout order of the array). The vertical (white) lines are due to the so-called "column pull-down" effect, discussed below. The muxstripe effect is also clearly seen, especially in the image on the right between the two bright stars.
Muxbleed is corrected in both the BCD and CBCD files. We fit a polynomial to the intensity of muxbleed in the trailing pixels, and scale it appropriately, depending on the brightness of the offending source, in both IRAC channels 1 and 2. The muxbleed mitigation may leave some residual, as a component of the muxbleed is stochastic. Very low level (below 10,000 DN in the raw frames) muxbleed is not corrected, neither is it corrected in the subarray images. Muxbleed is not mitigated in BCDs for strongly saturating sources as it is not flagged by the BCD pipeline. Muxbleed from saturated sources is appropriately flagged as part of the CBCD pipeline (but not yet corrected).
For more information, see the IRAC Instrument Handbook.
Mitigation. If you want to work on the BCD files instead of the CBCD files, then the best method to mitigate muxbleed is incorporated in the IRAC artifact mitigation script. Another code is available to remove the pinstriping (or "muxstriping") that is associated with muxbleed, from the BCDs.
C. Bandwidth Effect
The bandwidth effect appears in IRAC channels 3 and 4 (5.8 and 8.0 um). It looks like a decaying trail of pixels 4, 8, and 12 columns to the right of a bright or saturated spot. Only in the most highly saturated cases is the effect visible 12 columns to the right. Due to the design of the IRAC electronics, there is a maximum rate at which they can react to changes in pixel intensity during readout. Thus, it is not possible to go from full to zero pixel intensity between two adjacent pixels in a single readout channel. The effect is nonlinear except in the weakest cases.
A comparison of a 0.6 second (left) and a 12 second (right) BCD of the same region in channel 4. The 12 second BCD shows many instances of the bandwidth effect (the trailing black dots to the right of a source).
For more information, see IRAC Data Instrument Handbook Section 7.2.3.
Currently there is no correction for this effect. Affected pixels should be masked before performing PRF-fitting. We recommend using small apertures (< 4 pixel radius) for photometry of bright sources at 5.8 and 8 microns.
D. Column Pull-down/Pull-up
When a bright star or cosmic ray on the array reaches a level of approximately 35,000 DN, there is a change in the intensity of the column in which the signal is found. In channels 1 and 2, the intensity is reduced throughout the column (thus the term "column pull-down"). When the effect occurs, it shifts the intensities of the pixels above and below the position of the "guilty" source, within the same column. This effect is limited to the brightest sources. The amplitude of the column pull-down does not scale linearly with the flux of the source or the brightest pixel.
IRAC channel 1 (left) and channel 2 (right) observations of a crowded field with column pull-down apparent from the brightest sources. Note that the brighter sources affect a larger number of columns.
Column pulldown has been corrected in the cryogenic IRAC data CBCD files. The code estimates the "true" sky value for the affected pixels and fits a DC offset, which is in general a different constant above or below the offending bright source.
For more information, see IRAC Instrument Handbook Section 7.2.4.
Mitigation. The effect appears to be (a different) constant on either side of the source, and algorithms which fit separate DC offsets above and below the source should be effective. If one wants to work with BCDs instead of CBCDs, one such corrective algorithm is implemented in the IRAC artifact mitigation script.
E. Full Array Pull-up
In all four arrays, the background level is increased because of internal scattering in the array when a bright source is on the array. For channels 3 and 4, the background offset becomes noticeable for sources brighter than 1 Jy. To first order, the background offset appears to be uniform across the array. The offset is probably also generated by diffuse emission. For channel 4, the typical zodiacal background (~10 MJy/sr) integrated over the array is of order 20 Jy which would generate an offset of order 2 MJy/sr.
For more information, see IRAC Instrument Handbook Section 7.2.5.
Mitigation. Full array pull-up can be mitigated by matching the background level of the affected BCD with that of the overlapping, unaffected BCDs. For example the overlap module in the
MOPEX package usually corrects this effect well. For analysis of very extended sources (size of the array or greater), some thought must be given to applying an overlap correction and how this will effect the extended source aperture correction. The best method is a function of the observing strategy and source morphology.
F. Persistent Images
The terms "persistent image," "residual image," and "latent image" are used interchangeably to describe the contamination of an IRAC image by a bright source from a previous exposure. Tests performed during In-Orbit Checkout (IOC) revealed that there are both short-term persistent images, with time scales of order minutes and which are present in all four arrays, and longer-term persistent images in channels 1 and 4. The short-term persistent images were known before launch, and extensive calibrations and data analysis were performed to characterize them. The IRAC pipeline produces a mask (bit 10 of imask) for each image that indicates whether a bright source seen by a previous exposure would have left a short-term persistent image above three times the predicted noise in the present frame.
The longer-term persistent images were discovered in flight. In channel 1, the persistent images are generated by stars as faint as K = 13 (in very long stares). They can be generated by any long dwell time with a bright star on the array, whether or not the array is being read out. The persistent images do not have the same size as a direct point source; they are significantly more diffuse (looking more like the logarithm of the point-spread function). The channel 1 long-term persistent images have time scales of order 6 hours, and they decay gradually. The cause of these persistent images has been identified as a known feature of the flight array (broken clamp) that cannot be fixed. The longer-term persistent images in channel 4 are induced by bright mid-infrared sources or bright stars (the threshold is around 10 Jy at 12 microns for point sources which corresponds to about 50,000 MJy/sr for extended sources). The channel 4 persistent images have very unusual properties: they have lasted for as long as two weeks, they can survive instrument power cycles, and they do not decay gradually.
Median of channel 1 images from a calibration observation performed after observing Polaris. The five bright spots are persistent images from staring at the star while observing, while the set of criss-crossing lines were generated by slews between the various pointings and are referred to as "slew latents."
For more information, see IRAC Instrument Handbook Section 7.2.8.
Mitigation (or "Prevention"). Sufficient dithering and using the SSC-provided outlier rejection methods in the MOPEX mosaicking software provided effective mitigation of most residual images. Observers were encouraged to plan their observation strategy in such a way that unnecessary exposures of bright sources was avoided as much as possible, and that sufficient dithering or map movements were used to enable correction for any residual images. The long-term residuals were usually mitigated by a combination of judicious scheduling and thermal annealing of the IRAC arrays. Observers should examine the AOR median stack images provided with the BCD data to search for long term latents. In particular, the 8 micron (channel 4) persistent images decay slow enough so that they can be subtracted from the BCDs using the median stack as a template.
3. Optical Artifacts
A. Stray or Scattered Light
Stray or scattered light on the arrays can be produced by illuminating regions off the edges of the arrays. Stray light from outside the IRAC fields of view is scattered into the active region of the IRAC detectors in all four channels. The problem is significantly worse in channels 1 and 2 than in channels 3 and 4. Stars which fall into those regions which scatter light into the detectors produce distinctive patterns of scattered light on the array.
Both point sources and the diffuse background generate stray or scattered light. Stray light due to the diffuse background is removed in the pipeline by assuming the source of illumination is uniform and has a brightness equal to the COBE/DIRBE zodiacal light model. This assumption is not true at low Galactic latitudes or through interstellar clouds, but in the 3.6 - 8 micron wavelength range it is nearly correct.
Channel 1 images showing scattered light on both sides of a bright star.
For more information, see IRAC Instrument Handbook Section 7.3.1.
Mitigation. Scattered light due to stars is flagged in the IRAC pipeline. When mosaicking the IRAC frames together the flagged pixels will not count towards making the final mosaic if the DCE_Status_Mask_Fatal_BitPattern parameter in MOPEX (bit 3) has been set correctly. Observers were encouraged to check the scattered light boxes in Spot when planning their observations to avoid introducing scattered light in the image frames.
B. Optical Banding and Internal Scattering
The banding (or "row/column pull-up") effect manifests itself as rows and columns that contain a bright source having an enhanced level of brightness. This happens only in the Si:As (channel 3 and 4) arrays and has been shown to be due to internal optical scattering (inside the array). Both bright stellar sources and bright extended sources cause banding.
In addition, there is an electronic component to banding. Channel 4 has a strong row pull-up, while channel 3 has a weak column pull-up. The column pull-up is uniform across that part of the row where the source is bright. The optical banding intensity falls off with distance from the bright spot. Radiation/particle hits may cause electronic banding, but not optical banding.
Users should be aware of the uncertainties resulting from banding, specifically when attempting measurements of faint sources near the affected rows or columns. For bright sources with significant banding, aperture photometry may not be successful, and it is better to measure the brightness of these sources using frames of shorter exposure times.
Images showing the banding effect.
The IRAC pipeline makes an attempt to correct for banding, using an algorithm similar to column pulldown correction (estimating the true sky background at the location of the affected pixels and interpolating over them). The code does not model the flaring of banding towards the edges of the array, and the data user is encouraged to carefully consider banding effects in the data that have not been fully corrected.
For more information, see IRAC Instrument Handbook Section 4.4.2.
Mitigation. IRAC artifact mitigation script will mitigate optical/electronic banding in BCD files but does not model the flaring of banding towards the edges of the array.
C. Optical Ghosts
There are two common types of optical ghosts visible in the IRAC images. The brightest and most common ghosts are produced by internal reflections within the filters. The first-order filter ghosts (one pair of internal reflections) in channels 1 and 2 are triangular, and in the BCD images they appear above and/or to the left of the star in channel 1, and above and/or to the right of the star in channel 2. The separation between the star and its ghosts increases with distance from the optical axis of the telescope. The channel 3 and 4 filter ghosts appear as small crosses at a larger distance, mostly to the left or right of the star, respectively.
Similar ghosts are created by internal reflections within the beamsplitters. These only affect channels 3 and 4 for which the incoming photons are transmitted through the beamsplitters. They appear as a very faint, short, horizontal bar about 36 pixels below a bright star. They are slightly fainter than the filter ghosts.
Filter and beamsplitter ghosts.
For more information, see IRAC Instrument Handbook Section 7.3.3.
Mitigation. Filter ghosts are flagged in the IRAC data pipeline. Modeling the position of the beamsplitter ghosts has not been done. However, because the relative locations of the ghosts do vary with position on the array, sufficiently large dithering can help to reduce or eliminate their effects. The PRFs currently provided by the SSC include all ghosts, and the apertures used in calibrating channels 1 and 2 include the filter ghosts. When performing point source photometry in channels 1 and 2, the filter ghosts should be included to obtain the correct absolute calibration.
4. Cosmic Rays and Solar Protons
Each IRAC array receives approximately 1.5 radiation hits per second, with about two pixels per hit affected in channels 1 and 2, and about six pixels per hit affected in channels 3 and 4. The cosmic ray flux varies randomly by up to a factor of a few over time scales of minutes but does not undergo increases larger than that. Also, the cosmic ray flux is normally about a factor of two higher on average around solar minimum compared with solar maximum.
Radiation hits for channels 3 and 4 affect more pixels than the channel 1 and 2 radiation hits due to the larger width of the active layer of the Si:As detectors. Some tuning of cosmic ray detection parameters may be necessary when working with deep integrations, especially for channels 3 and 4.
Some high energy cosmic rays cause persistent images, column pull-down, and muxbleed effects.
The central 128 X 128 pixels of IRAC 12-second images taken on January 20, 2005 during a major solar proton event. Channels 1 and 2 are top left and top right; channels 3 and 4 are bottom left and bottom right. Except for the bright star in channels 1 and 3, almost every other source in these images is a radiation hit. This is an extreme case and images taken under normal circumstances do not have nearly as many radiation hits.
For more information, see IRAC Instrument Handbook Section 7.4.
Mitigation (and prevention). The SSC mosaicker, MOPEX, identifies energetic particle hits as follows. All pixels in BCDs that contribute to a given pixel in the final mosaic are identified, and significant outliers (a user-specified number of sigmas above or below the filtered mean of all overlapping pixels of overlapping BCDs) are rejected. This method is very similar to the outlier rejection performed by shifting and adding ground based images. The rejected pixels can be inspected in the "Rmask" output files (one per input image). Outlier rejection in MOPEX can be adjusted. The parameters used in the online pipeline-generated mosaics rely on multiple (>3) sightings of each sky pixel. In general, a coverage of at least five is necessary to produce optimal results with the multi-frame (standard) outlier rejection. If doing improved cosmic ray rejection with MOPEX, in general observations with more than 10 sightings of the source are well handled by the multiframe rejection module alone, whereas sources with 5-10 sightings should probably use both the dual outlier and the multiframe rejection methods. In addition, or alternatively, users may want to experiment with the box outlier method for data with less than 10 overlapping pixels.
Note that there is a continuum of brightnesses of cosmic ray hits. Therefore, there are faint cosmic ray hits in the images that will not be flagged by even the most restrictive outlier rejection. As a result, in any single band, faint source detections should be verified by examining the image stack to look for the detection in multiple BCDs.