MIPS: Image Features

This page summarizes the most common MIPS image features, including image artifacts. We describe these features, show images with representative examples, and provide a recommended mitigation method for their removal from the data. Please be sure to also examine Chapters 5 and 6 of the MIPS Data Handbook for more information on these features.

Image Features: MIPS 24

1. Pick-off Mirror Contamination

There is some contamination on the pick-off mirror (POM) that feeds MIPS. These dark spots appear to move with celestial objects during scan mirror moves (dithers). The darkest, most well-defined spots are 3-4 pixels across and are ~15-20% dimmer than surrounding pixels; they are located primarily in the upper right of the 24 µm array. There are also diffuse features that have a ~1-5% effect. A map of all of the spots seen during normal scan mirror moves based on 24 micron observations has been created (see figure below). These spots appear to be stable; there is no evidence for "growing" or "jumping" spots. As a result of these spots, the 24 µm flats depend on the scan mirror position and scan rate; the specific pattern of the spots depends on exactly what observing mode was used.

We note that 24 micron data are obtained even during Ge photometry observations; these "non-prime" data necessarily use different scan mirror positions than when the 24 micron array is prime. Although scan mirror position-dependent flats are used for primary data, until recently, non-prime 24-micron data typically were not correctly flat-fielded because our library of scan-mirror dependent flats did not include these different scan mirror positions. We are working on improving this library.

Mitigation. We started the process of obtaining a complete library of flats for each scan mirror position, but this is complicated by the fact that the scan mirror does not reproduce its position exactly from campaign to campaign; the spots appear to move slightly (<0.5 arcsec) for each campaign. There is also limited evidence for changes within a campaign. Scan mirror position-dependent flat fields are applied to the data (when PRIMEARR=1) as part of the pipeline, but until we build up a sufficiently large library of flat fields for each scan mirror position in each mode, the spots may not be in the correct location for your data. The figure below shows the results of dividing a mismatched position-dependent flat from the data; if this happens to you, please notify the SSC Helpdesk (help@spitzer.caltech.edu). It is likely that the whole rest of the campaign is also affected, but it may very well be that you will have to wait for formal reprocessing for the spots to be removed properly. If you have enough data, you can create your own flat field for every scan mirror position (using keywords CSM_PRED and CSM_RATE), and this may be the best solution in general.

Pointing: The fact that the scan mirror does not exactly reproduce its position has implications for the pointing; the pointing for any one campaign can be systematically offset by about 0.5 arcseconds in either direction. We are monitoring this systematic pointing offset; you can fix this effect by manually comparing a source list to 2MASS sources and editing the CRVAL1 and 2 header keywords. Note that a list of 2MASS sources is one of the products served to you by the Archive. You can also use Spot to overlay a source list on your image to check for small systematic offsets in pointing.

2. Jailbars

When a pixel is saturated, a responsivity change occurs in the corresponding readout. This results in a "jailbar" pattern occurring every 4 rows/columns for saturated sources (including some cosmic rays). Only the portion of the array read out after the saturated source is read is affected. Typical effects of jailbars from point sources saturated only in the core produce a decrease in the flux of affected pixels at about the 2 MJy/sr (50 microJy/arcsec) level, but the magnitude of the effect may vary with source flux. For extended sources, often the saturated regions will cover all four or more contiguous readouts, and thus an altered responsivity level is seen but no obvious jailbars. Occasionally, one may see strong jailbars where there are no apparent sources. We believe this effect is due to cosmic ray hits. Faint jailbars are often seen in lower-coverage regions of photometry mosaics where the dither pattern stacks the jailbars.

Mitigation. Many of these effects can be mitigated in the post-BCD stage using outlier rejection. The fix for strong jailbars caused by saturated sources is additive - we recommend that you adjust the appropriate column (or portion of the column) additively to match the median level of the unaffected region of the column. However, weak jailbars appear to be multiplicative. Weak jailbars should flatfield out, but they often do not; self-calibration via dividing by the median of the affected frames often decreases the jailbar contrast.

3. Latents

There are three types of MIPS-24 latents: (a) ~0.8% bright latents that last from seconds to 10s of seconds, decay exponentially, look like PSFs, are easily recognizable by eye, and are flagged (but not corrected) by the pipeline; (b) ~2% dark splotchy effects from very bright objects (>50 Jy) that last for hours; and (c) 0.5% bright latents from extremely bright objects that can last for days. (Saturated sources will also produce cosmetically ugly "jailbars" (see above), but these are easily recognizable.)

A. Short-Lived Bright Latents

These bright latents are the easiest to recognize by eye or automatically because they always occur in the in-scan direction.

Mitigation. At present there is no easy correction for this effect, as most of these bright latents are not removed by outlier rejection in scan mode because the scan mirror produces repeated overlapping patterns. This is why you may see a trail of latents (often on both sides of a bright source) along the scan direction, particularly for fast scans. The pipeline does flag them in the mask file but does not attempt to remove them.

B. Dark Latents

Examples of dark latents from bright sources can be found in the figures below. This reduction in response of 1%-2% comes from very bright sources, those >50 Jy. Low-level dark latents have been seen for sources as low as 18 Jy. They can last a long time, with a timescale of about 10 hours. If your observations have these features, it may be a result of a bright object seen in observations prior to yours. The dark latents are removed by thermal anneals, which are a rare event for the 24 micron array, and are generally only done at the start of a campaign. Currently, we try to limit the impact of these latents by manually scheduling already-known bright sources (known to be bright from 25 micron IRAS data) at the ends of campaigns.

Mitigation. The impact of this is usually reduced by redundancy (e.g., additional dithered frames). The effects of dark latents AND any residual but static imperfections in the flat fielding can be removed quite effectively by division of each BCD affected by dark latents by a normalized median of all of the BCDs affected by the dark latents within the AOR (excluding very bright pixels); see MIPS Data Handbook for more information. This effectively gives an improved flat field and is generally recommended in regions of low background where subtle (1%-2%) effects may be important to remove. After correcting the BCDs, you then have to make a new mosaic (using the SSC's MOPEX or other software).

C. Long-Lived Bright Latents

These bright 0.5% latents remained for the duration of the week-long instrument campaign.

Mitigation. Just like for long-lived dark latents, the impact of these latents is usually reduced by redundancy (e.g., additional dithered frames, inherent to MIPS observing modes). If you encounter these kinds of latents in scan mode, because the latents last longer than the maximum length of an AOR, they are stable and can be removed using your own data, called self-calibration (see MIPS Data Handbook). If you have these kind of latents in data taken in photometry mode, the pattern of latents may be different (because the dithers are different), and you might not have enough data to self-correct it. Just as above, you can remove these effects by dividing each of the affected BCDs by a normalized median of all of the affected BCDs (excluding very bright pixels). This effectively gives an improved flat field, and is generally recommended in regions of low background where subtle (1%-2%) effects may be important to remove. After correcting the BCDs, you then have to make a new mosaic (using the SSC's MOPEX or other software).

4. First and Second DCE Effects

The first DCEs of every commanded sequence of observations (e.g., data with the keyword DCENUM=0) have a shorter exposure time and are depressed in response by 10%-15%. The photometry AOTs are designed such that these "extra" frames can be discarded. Although there could be useful information in these frames, they are omitted from the automated post-BCD mosaics. You will be sent all BCDs (including those with DCENUM=0), so if you reconstruct the mosaics, you should avoid using the DCENUM=0 frames.

Mitigation. If you get a mosaic that looks like this, then you have indeed accidentally included these first DCEs, and you should omit the DCENUM=0 frames. You can identify these from the filenames alone; see Section 3.2.1 of the MIPS Data Handbook. In addition, the second and even third DCE/BCD (DCENUM=1, 2) appears to have a somewhat reduced signal (~2%) and often enhanced jailbars compared to the rest. For now, these second DCEs are included in the automatic mosaics, but if you have enough data, you might consider rejecting these as well. We are anticipating that a future version of the pipeline will account for DCENUM dependence so that all frames are well-calibrated.

5. Post-Anneal Slow Response Drift

The figure below shows the median flux of each BCD (after rejection of pixels affected by cosmics and sources) for the two AORs of the ELAIS-N1 deep 24 micron observations. The background of the region observed is uniform. The observation is in photometric mode, using the offset position just to observe a more extended field. Every ten BCDs, a bias frame has been taken, which also resets the detector. The first AOR shows a clear monotonic increasing trend. The second AOR, which has been taken a few hours later after observing a bright source, is more stable (there are however some latencies in the BCDs from the bright source, which have been corrected with a median stack).

Mitigation. There is no present correction for this effect, which is at the level of less than ~1%, and lasts for ~3 hr timescale.

6. Bright Sources and Droop Correction

In order to correct properly for the droop (the whole frame in which a bright source lies has an elevated surface brightness level) effect, one needs an accurate estimate of the total flux incident on the array. If part of the array is saturated, one does not know the total flux, and the part of the array that is saturated is likely to contribute significantly to the total flux. The pipeline attempts to correct for these kinds of effects as best it can, but in some cases where the array is really saturated, the droop cannot be calculated properly, resulting in a DC offset to the whole array for that frame.

Mitigation. If your observation includes very bright sources, you may have to manually correct the affected frames for the offset by subtracting a constant background value, matching it to surrounding images.

7. Large- and Small-Scale Gradients

For large scan maps, we have noticed a slow large-scale gradient across the entire map, which is along the direction of a scan leg. Since we scan along directions perpendicular to the ecliptic plane, this may very well actually be real, physical variations in the zodiacal light. However, sometimes the gradients do not match in adjacent scan legs. Gradients are also seen within individual 5x5 arcmin BCDs, but despite being in the scan direction, these are not likely to be astronomical in nature.

Mitigation. Some of these gradients may be due to poor flat-fielding, and can be corrected with self-calibration (see MIPS Data Handbook). However, systematic gradients are seen as a function of scan mirror position, especially in photometry mode, which uses the largest range in scan mirror angle. These may be due to scattered light changes as a function of scan mirror position. These are likely to be additive. Additional analysis is pending.

8. Asteroids

Finally, as another common "gotcha" to list here, there are many, many asteroids to be found, even in observations obtained some considerable distance from the ecliptic. Beware of 24 micron sources without short-wavelength counterparts, especially near the ecliptic plane. Most of the point sources in the figure below are asteroids.

Beware of asteroids! Most of the point sources in this frame are asteroids.

Mitigation. Take imaging in more than one epoch; even 6 hours later is usually enough. Asteroids will move, while the background sources will not.

Image Features: MIPS Ge

1. Variations of the Slow Response

The stim flash calibration does not completely remove long-term transients for the MIPS-Ge detectors. The effects are most noticeable before the bias change for MIPS-70 (e.g., dark horizontal stripes in bottom of the figure below). The effects are not as noticeable for MIPS-160 due to short time constants for the stressed 160 micron detectors.

Mitigation. For point source science, this effect can be removed using a temporal median filter (e.g., for scan data). The variation of the drift in the slow response affects the ability to accurately measure the true background level. For point sources, the long-term drifts can be treated as an additive effect (i.e., subtracting off a temporal median). The long-term drifts affect the actual response from the background, suggesting a multiplicative correction may be needed for large extended regions. For large extended sources, we recommend that observers take enough off-source data of the surrounding background so that the total extended source flux can be estimated.

2. Stim Latents

For data taken in IOC/SV or the first 4 MIPS campaigns in nominal operations, the bias setting of the 70 µm array was set at a level that attempted to increase the sensitivity of both sides of the array, but a consequence was that the stimulator flashes left latents in the data. For bright sources, this is less of a concern than for extended regions or long integrations built up from short exposures (e.g., in photometry mode). The amplitude of the stim latents depends on background and appears to increase with the time since anneal. The stim latents linger longer for 70 microns than 160 microns, since the decay time constants associated with the stim latents are larger at 70 microns. Data taken after the bias change (MIPS-5 or later, after 14 March 2004) should still see these effects, but at a much lower level.

Mitigation. The stim latents and variations of the slow response as a function of time can be mitigated by median filtering the data if you are only interested in faint point sources. Be careful, however, about using filtered data for extended sources (see the section below).

3. Improper Use of Filtered BCDs for Extended Sources

Currently, the SSC produces two BCD products: (1) bcd.fits which is the standard calibrated BCD and (2) fbcd.fits which is a "filtered" BCD product designed for point sources. The fbcd is produced by subtracting off a median of the surrounding DCEs as a function of time per pixel. This filtering technique significantly mitigates the accumulation of stimflash latents and the residual background drifts due to variations of the slow response as a function of time (see above for further discussion). The application of the median filter removes the background from both the sky and the residual detector effects (i.e., loses information about the extended background level in the field). Tests show that the application of a median filter maintains point source calibration for scan maps of fields with uniform backgrounds, but the fbcds do not preserve calibration for extended sources or for bright (>~0.2 Jy) point sources within complex emission regions.

Mitigation. For extended sources, we recommend you use the default BCDs, NOT the filtered BCDs. The filtering techniques are optimized for point sources and remove a significant fraction of the emission from extended sources. For example, note how much the extended source fluxes change in the figures above. To remove data artifacts for extended sources, observers need to observe enough off-source regions for good background subtraction. We recommend that observers linearly interpolate the measured background levels in the BCDs across the target on a pixel basis. The interpolated background corrections should be subtracted from the BCDs before coadding the data and making the mosaic. The additive background corrections as described here have been shown to yield good results for extended galaxies. To derive the true sky level for extended regions, observers may want to consider TPM observations. Alternatively, for these extended and/or bright sources, you may wish to try offline custom filtering using the GeRT (Germanium Reprocessing Tools; see MIPS Data Handbook, Section 7.4).

4. Extrapolated or Corrupted Stim-Minus-Background Measurements

As mentioned above, long-term transients are tracked using the stim flashes. In scan maps, necessarily the first set of DCEs has an extrapolated stim calibration since there is no background for the first stim DCE. Of course, this implies that these first DCEs are of lower reliability; except for fast scan, these DCEs are generally part of the "overscan" region, and do not count for coverage of your requested area. This also means that you should not just grab all of the DCEs and blindly work with them, especially when you have scans over bright regions or scans that end in bright regions, which show a portion of a fast scan over a bright region. The artifacts apparent in the left strip below are largely omitted in the right strip when the DCEs with extrapolated stims are dropped. Note that the white streak in the right is still a bright object artifact, but the effects of the bright object are substantially reduced in the final product on the right. Conversely, however, the trailing DCEs have plenty of stim flashes, and are quite reliable.

Mitigation. DCEs with extrapolated stim solutions can automatically be rejected by the SSC mosaicking software MOPEX. There is a bit in the bmask which indicates extrapolated stim solutions so that you can tell the software to ignore those frames.

Bad stim calibration can occur not only for extrapolated solutions, but in cases where the background DCE for the stim is on a bright region, which due to the latents from the source can yield a bad stim-background measurement. You should try to avoid this when planning your observations (e.g., by taking a long enough scan to get enough background). You may need to correct these cases by deriving a correction based on the surrounding valid stim-background measurement; in some cases, the observation may not produce optimal results. This is a difficult problem, and if you think you have encountered it, your data may not be easily calibrated. Consider using the GeRT (see Data Handbook, Section 7.4).

5. MIPS 70 Fine Scale or 70/160 Large-Field Off-Source BCDs

The observing strategy implemented for the MIPS-70 Fine Scale AOT and the 70/160 large-field photometry differs from both the MIPS default photometry and scan modes in that alternating on- and off-source DCEs are obtained during each cycle. The distance between the centers of the on- versus off-source positions is roughly 4'. Consequently, the two fields of view overlap and the online pipeline currently creates a single mosaic consisting of both on- and off-source BCDs. The acquisition of the off-source frames provides the opportunity for additional offline post-processing. While in most cases the pipeline reduced BCDs are of good quality, we have found that the data can be improved to varying degrees by different methods of off-source "background" subtraction.

The figure below highlights the case where a serendipitous source is present in the off-source BCDs. The target was positioned so that it appears both in the "on-source" and "off-source" frames, hence neighboring "on-" and "off-source" frames cannot be subtracted. Instead, all frames were median combined and the resulting image subtracted from each BCD.

Mitigation. Before performing any off-source subtraction, we recommend that you always inspect the mosaic of the off-source BCDs to prevent introduction of off-source structure into the on-source mosaic. There is no general recipe that works equally well for all types of observed sources and the observers are cautioned to carefully investigate the appropriate method for their particular datasets, especially in the case of bright and spatially varying extended emission.

6. Electronic Non-Linearities

The electronic non-linearities for the MIPS-Ge detectors are corrected for in the online pipeline. However, spurious electronic non-linearity solutions for a few unstable pixels on the top row of the 70 micron detector can yield artifacts in the mosaics.

Mitigation. The recommended solution is to make sure you have the most recent pmask downloaded from the SSC website, and make mosaics with this updated pmask, ignoring these pixels (use the pmask rejection parameter=16648).

7. Flux Non-Linearities

8. The 160 µm Short-Wavelength Light Leak

Signals on pure photospheres of stars at 160 µm are stronger than expected by about a factor of five. Review of the instrument design has revealed a weakness in the stray light control that could result in a short wavelength Ge:Ga response being detected in this band (due to scattering off a blocking filter).

For many types of observations, this light leak does not impact the data at all. Stars fainter than m ~ 5.5 mag will not be detectable in the leak above the confusion level. No compact extragalactic sources have m brighter than 5.5 mag. The leak signal for a star of m = -0.5 mag is equivalent to ~ 2.5 Jy at 160 µm. The parameter to consider for any given target is the 160 µm /2 µm flux density ratio. For a Rayleigh-Jeans source, this ratio is 0.0001. Anything with a ratio larger than 0.004 will produce uncorrupted data, so most objects will not be affected. Sources with 160 µm fluxes more than a factor of 40 above that of a Rayleigh-Jeans source from 1 to 1.6 µm will have leak signals <10% at 160 µm. Galactic programs on star formation, ISM, etc., will likely be impacted.

For stellar observations, the strongest signal apparent in a 160 µm observation is likely to be the leak itself. Tests were perfomed using HD 163588, a K2III star which is routinely observed at 70 µm and has also been observed at 160 µm. Similar results are obtained for HD 36673 (Arneb), an F0Ib star, and for HD 87901 (Regulus), a B7V star. The leak is 15±7 times as bright as the photosphere of a star. The figures below show example 160 micron images of a star and an asteroid, respectively. The asymmetry apparent in the stellar image is a result of the short-wavelength light leak. Note that the brightest thing here is actually the leak; the star is the asymmetry to the bottom right.

Mitigation. It has been demonstrated that observations of bright point sources strongly impacted by the leak (stars) can be corrected to a high level using an empirical stellar PSF and careful deconvolution (see, e.g., Stapelfeldt et al. 2004, ApJS, 154, 458). Best results will be obtained with observations for which proper planning and data acquisition was made to allow a good characterization of the leak signature for the object brightness and color. For example, plan to obtain identical observations of "calibrator" stars of similar type and brightness to your target stars, but without long-wavelength excesses.

9. Small Field Point Source Photometry

Since the small field photometry AOTs do not dither with offsets much larger than the size of the MIPS-Ge PSF, temporally filtered fbcd data will underestimate the true point source flux values (by roughly about 10%).

Mitigation. We recommend that you mosaic both the filtered and non-filtered data to determine whether or not additional offline processing is needed. To maintain point source calibration, it is possible to mask out the point sources before filtering. If you do not do that, we recommend that you measure the point sources before and after filtering to check its impact on the flux calibration.

10. Incorrect pmask Bit Settings For SED Mode

Mopex allows the user to create mosaics and extract the corresponding spectra from the BCD products produced by the SSC pipeline. Not using the correct pmask bits can cause the creation of mosaics with spurious bright columns that can confuse the extraction of the spectrum and provide bogus results. This spurious effect will strongly affect only the observation programs for which the background and the spectrum share similar (mostly faint) brightness levels. In all other cases (i.e. observations of bright regions), the effect on the extracted spectrum is negligible.

Mitigation. Be sure to use the most recent 70 micron pmask with the pmask bit setting shown in the example SED namelist templates available on the MIPS Data Handbook webpage. Such a pmask bit setting allows one to correctly mask all the noisy pixels (especially the first column) as well as those affected by electronic non-linearity or by a bad read-out (first row).

Go back to MIPS page