IRS: Image and Spectral Features

Contents

A. IRS Peak-Up Imaging

1. Peak-Up Latents
2. Rogue Pixels
   

B. All Spectral Modules

1. Warm/Bad Pixels and NaNs in IRS Data
2. Radhits on rogue pixels
   
3. Severe Jailbars
   

C. Low-Resolution (SL and LL)

1. Low-res non-repeatability
2. Curvature in SL1
   
3. LL1 24 micron Deficit
4. Incorrect Droop and Rowdroop corrections following peakup saturation
5. Wiggles in Optimally Extracted SL spectra
6. Residual Fringes in SL and LL
7. SL14 micron Teardrop
   

D. High-Resolution (SH and LH)

1. LH Order Tilts
2. LH vs. LL 5% offset

E. Solved Issues

1. Low-Res Nod 1 vs. Nod 2 Flux Offset (solved in S15)
2. LL Nonlinearity Correction Too Large (solved in S14)
3. Tilted SH and LH Flatfields (Solved in S14)
4. Bumps in LH Near 20 Microns (Solved in S15)
5. Mismatched LH Orders (Solved in S17)
6. LL Salt and Pepper Rogues (Solved in S17)

A. IRS Peak-Up Imaging

1. Peak-Up Latents

Bright objects falling on the IRS SL detector can result in persistent charge that appears as a latent image of the object in subsequent exposures. The magnitude of this effect is less than 2% of the source flux, but has a relatively long decay time. A second source of latent charge on the detector is simply the background. As a result, during very long PUI AORs or those in very high background regions, the background level may be seen to rise slowly over time.

Mitigation. Proper dithering strategies will mitigate the latent image problem, although with some loss of signal-to-noise on the affected pixels. Latents due to the background are stable and vary smoothly, so users may fit the (small) rise in the background and subtract it. See the "Report on ultradeep IRS spectroscopy of faint sources" at this page.

2. Rogue Pixels

A rogue pixel is a pixel with abnormally high dark current and/or photon responsivity (a "hot" pixel) that manifests as pattern noise in an IRS BCD image. At current bias levels there are very few rogue pixels in the peak-up windows, however a few are present in some AORs.

Mitigation. Proper dithering will mitigate the effect of rogue pixels, but they should be masked before mosaicking. The users should also consult a web page about rogue pixels for more information on how to deal with them.

B. All Spectral Modules

1. Warm/Bad Pixels and NaNs in IRS Data

Early in the mission, all four IRS arrays were subjected to powerful solar proton events that deposited the equivalent dose of protons expected over 2.5 years in only two days. This damaged 1% of the SL and SH pixels, and 4% of the LL and LH pixels. These damaged pixels are masked from data processing, by assigning NaN status to them. By default, EXTRACT interpolates over the NaN pixels. Occasionally, it is possible for warm/unstable pixels to propagate through the pipeline without being flagged as NaNs. In this case, these spurious pixel values will be extracted and appear as sharp spikes in the final spectrum. Such features should be readily recognizable as spectral features which are too sharp to be real.

Mitigation. The IRSCLEAN MASK routine which has been released by the SSC enables observers to visually inspect such pixels and interpolate over them. If there is any doubt about the reality of a given spectral feature seen in an extracted spectrum, we recommend that the observer always examines the 2D BCD images to confirm that the feature shows the expected spatial and spectral dispersion on the array.

2. Radhits on rogue pixels

Low and high resolution spectra may show a large spikes in places cosmic radiation hits (radhits) fall on highly-nonlinear (rogue) pixels. This effect is made more evident by the pipeline's (S15 and onward) rejection of samples following a radhit event. The early samples in nonlinear (rogue) pixels generally have steeper slope than the samples rejected after the radhit, increasing the flux after radhit rejection.

Mitigation. Rogue pixel data have incorrect values and should generally be filtered out before or during spectral extraction. Two methods can be used to remove the offending pixels:

1. Use SPICE to re-extract the spectra (using bcd.fits files as a starting point). Within SPICE you can use the Extract 'Mask' pulldown menu to mark bit 3. This will cause SPICE to ignore pixels with radhits. Then run Extract.
2. Use the IRSCLEAN MASK routine to filter and replace the offending pixels.

3. Severe Jailbars

Very strong vertical stripes (jailbars) are seen in a single BCD image of an AOR. This may lead to very noisy or corrupted spectra.

A few rows of data in one plane of the corresponding cube will be seen to have anomalously high values (can be several tens of thousands of DN). These rows are not tagged as missing data, and in cubes of 4 samples, they are not recognized as outliers. Even before slope estimation, these rows affect the entire array because droop is impacted for all pixels of that plane, leading to severe jailbars.

The severe jailbars discussed here occur very infrequently (~1/700 DCEs). The root cause of the corrupted rows in the cube leading to severe jailbars is unknown. More mild jailbars may observed as a result of other causes.

Mitigation. There is currently no fix for this problem in the pipeline. It would have to be fixed at the raw datacube level. If multiple DCEs are available, throw out the offending DCE and recompute the spectrum.

C. Low-Resolution (SL and LL)

1. Low-Res Nonrepeatability

When measured using high accuracy self peak-up, gaussian fits to standard star fluxes give the following standard deviations: 1% for SL1, 2% for SL2, 3% for LL1, and 1% for LL2. This is based on 43 observations of the standard star HD 173511 taken from campaigns 17 to 37. The variation is currently attributed to a combination of inaccuracies in telescope pointing (which can place the source away from the center of the slit) and flat field errors. The distribution of fluxes in SL1 is non-Gaussian, with a tail towards low fluxes. Roughly 23% of all visits give fluxes which are low with respect to the median by more than 3%, and 2% of all visits give fluxes which are low by more than 10% (see next item 2. Curvature in SL1). The difference between flux values obtained in the two nods average ~1%. These values are only valid for observations using high accuracy peak-up.

Spectral curvature (concave) may be induced in SL1 when the source is not centered in the slit. The curvature increases with apparent flux deficit.

Mitigation. All spectrographs are susceptible to pointing errors. Flux calibrations are based on the average of all HR 7341 observations, so individual observations may give fluxes above or below the 'truth' value. Flux calibrations for S15 use the median of all observations within 7% of the mean flux, so are less affected by outliers.

2. Curvature in SL1

Spectral curvature (concave) may be induced in SL1 when the source is not centered in the slit, due to chromatic PSF losses. The curvature increases with apparent flux deficit. 

Warning: The effects of the spectral curvature and order mismatch may be mistaken as broad absorption features. In addition, for the S14 pipeline the non-linearity coefficient was adjusted, resulting in a 5% drop in flux at 17-21 microns (LL2) (See _Solved issues: LL Nonlinearity Correction Too Large_). This last effect has been corrected in S15 but for users with S14 data the combination of order mismatch, curvature and LL2 drop may be mistaken by a redshifted silicate absorption feature.

3. LL1 24 micron Deficit

LL1 spectra of faint sources show a dip at 24 micron relative to standard models. The amplitude of the effect increases from 0% for bright (>250 mJy at 24 micron) sources to >5% for faint (<50 mJy sources). Some sources which show the deficit apparently have nonlinear ramps. Faint stellar standards such as eta1 Dor and HR 5467 show the effect prominently. LL1 is calibrated to HR 7341 (250 mJy at 24 micron).

Mitigation. Under investigation.

4. Incorrect Droop and Rowdroop corrections following peakup saturation

Spurious broad absorption or emission features or incorrect spectral slopes or curvature are seen in SL1 or SL2 spectra. This may be due to saturation of the peakup arrays. As mentioned in the IRS Section of the SOM, pg 208, SL exposures with long exposure times may saturate in the peakup arrays. In cases of strong saturation, the droop and rowdroop corrections will be incorrect because the total charge on the chip (which determines the droop correction) and on each row (which determines the rowdroop correction) cannot be measured. The effect becomes worse as more pixels are saturated for a greater portion of the ramp. It may not be obvious from the image that the peakup arrays are saturated.

Mitigation.

1. Avoid saturating SL exposures. The limits that the current pipeline can handle are given in the SOM, pg 208
2. In the future, it may be possible to truncate the data cubes after peakup saturation, to salvage the pre-saturation data. This will require new pipeline software and reprocessing of the data.

5. Wiggles in Optimally Extracted SL spectra

Sinusoidal oscillations are seen in some high signal-to-noise SL spectra extracted using SPICE's 'Optimal' extraction option.

This effect is caused by a mismatch between the source and the standard star spatial profile (rectempl) file. The mismatch may be due to an incorrect 'Ridge' percentage or to undersampling of the PSF. The undersampling of the PSF in SL is evident in the BCD images as the source jumps from one column of pixels to the next, along the spectral trace. The location of these jumps must match between source and template to get a clean optimal extraction. SL is the module most affected by undersampling.

Mitigation.

1. Use Regular extraction for high signal-to-noise data. Generally, if the S/N is high enough to see the wiggles there is little or no gain from using Optimal extraction.
2. Compare Regular and Optimal SPICE extractions to verify the source of the wiggles.
3. Manually change the percentage in 'Ridge' to minimize the wiggles. In general, automatic ridge finding is preferred for accurate ridge determination, however it is possible for the automatic procedure to miss the correct peak.
4. Generate and use a standard star template (rectempl) that is a better match to the source. The template will match best if the position along the slit is the same (within 0.1 to 0.2 pixels) as the source position, modulo one pixel. For example, if the template is shifted by exactly one pixel, optimal extraction will work well, while if it is off by 0.5 pixel, the match will be poor.

6. Residual Fringes in SL and LL

SL and LL spectra show residual (non-sinusoidal) wiggles after flat fielding, at the 1-2% level. The detector fringing pattern is sensitive to source location relative to the slit, which can be affected by pointing inaccuracies. In particular, the phase of the fringes changes with source distance from the slit center. The flat field has been constructed to accurately remove the fringes for the mean pointing at the standard nod positions. The residual fringes are non-sinusoidal, so IRSFRINGE does not do a good job at fitting and removing them.

7. SL 14 micron Teardrop

Excess emission is found in the spectral 2D images, centered at 13.2-15 micron, to the left of SL1 spectral trace. The amplitude of the feature is ~10% using standard point source extraction, greater for full slit extraction. The flux in the teardrop appears to correlate with source brightness. The tear drop is present in all of the pipeline products beginning as early as the Linearize cube. Therefore the tear drop shape is due to some light leakage, optics, or detector response. The tear drop shape changes slightly with position indicating that it is most likely correlated with the angle at which the light enters the slit, consistent with an internal reflection. There is no conclusive evidence whether or not the magnitude of the effect depends on source color. More information can be found on the Teardrop page.

Mitigation. There does not appear to be a simple way of quantifying the teardrop excess, and hence users should use extreme caution before interpreting features between 13.2-14 microns and correspondingly beyond 6.5 microns in SL1. Flux calibrations (S13 onward) ignore the teardrop region, so that the feature appears at full strength in extracted spectra.

D. High-Resolution (SH and LH)

1. LH Order Tilts

Flux-calibrated LH spectra of faint sources have orders tilted to the blue.BCDs show temporal-spatial dark-current variations of order ~50 mJy, visible in-between orders. The magnitude of the effect appears to be independent of source brightness. No similar effect has been conclusively identified in SL, LL, or SH modules

Mitigation. Memo sent to Spitzer Users describing the effect recommends throwing out data from the first few DCEs, if necessary.

2. LH vs. LL 5% offset

LH spectra are on average systematically higher than LL spectra by 5%. There may also be a tilt across LL relative to LH. Different standard stars and Decin models are used to calibrate LL and LH. HR 7341 (0.97 Jy at 12.0 micron) is used to calibrate LL. Ksi Dra (11.2 Jy at 12.0 micron) is used to calibrate LH. The overall 5% difference may be attributed to uncertainties in the absolute stellar diameters, which are of this order.

Mitigation. Users may want to use broad band photometry to scale their spectra to the correct level.

E. Solved Issues

We include here issues that may be relevant for users that have data processed with older pipeline versions.

1. Low-Res Nod 1 vs. Nod 2 Flux Offset (solved in S15)

SL Nod 2 fluxes are consistently 4% lower than Nod 1. This is the result of a spatial tilt in the flat field. LL2 Nod 1 fluxes are 4% lower than Nod 2 at 14-18 micron, decreasing to 1% at 21 micron. As a result, Nod 2 and Nod 1 are tilted relative to one another.

Mitigation. Average the nods to compute the source flux and determine the flux calibration. After S15, the flat field has been adjusted to remove the Nod1-Nod2 difference.

2. LL Nonlinearity Correction Too Large (solved in S14)

Slopes are overestimated at high count levels in LL. The effect is >5% for a 1 Jy source, and >20% for 2 Jy. The effect is worst for long exposure-times. SL nonlinearity correction appears to be OK.

Red sources appear redder and blue sources appear bluer due to this effect. However, curvature and other high order effects can also appear, reflecting the detector response. For example, fringes in the flat-field may be amplified, leading to extra 'noise' in LL spectra. The primary LL calibration source, HR 7341 (0.97 Jy at 12 micron) is affected by this problem at a low level. Reducing the nonlinearity coefficient causes a 5% drop in flux at 14-17 microns (LL2). The effect is <2% at 17-21 microns (LL2) and <1% at 26-36 microns (LL1).

The effect of the decrease in the non-linearity coefficient.

Mitigation. The non-linearity coefficient was adjusted for S14. New flatfields and flux calibration were derived for S15.

3. Tilted SH and LH Flatfields (Solved in S14)

Before applying the fluxcon, the SH and LH orders appear tilted to the red. This is due to a difference between the standard extraction, which forces the wavsamp rectangle height to 1.0 pixels, and the custom extraction used to make the flatfield, which uses a variable height rectangle as defined by the wavsamp.

Mitigation. Pre-S15, the tilts are removed by applying the fluxcon table. From S15 onward, we will use the variable height rectangles defined via the wavsamp for all extractions. This has the added benefit of matching the instrumental resolution.

4. Bumps in LH Near 20 Microns (Solved in S15)

Bumps are seen in LH orders 19 and 20, at ~19, 20, and 21 microns. They are attributable to bright features at the order edges which are accentuated by the S13-S14 LH flatfield. Features were not present pre-S13, and are fixed in the S15 flatfield. The pipeline S13.2 reduction shows three new peaks around 20 microns that are spurious. The features show in a full slit extraction but when the extraction aperture is shrunk to 2 pixels, the feature vanishes in one nod position but remains in the other. It is because of pixels X=13, Y=12 to 22 which all look unusually bright in the 2D frame even when the source is off on the other side of the slit.

LH spectrum. Notice the peaks at 19, 20, and 21 microns in the S13.2 spectrum.

Mitigation. New LH flat field was created for S15, with bad regions suppressed. For S13, trimming the orders can remove the strongest of the three peaks but not its two weaker neighbors.

5. Mismatched LH Orders (Solved in S17)

(S15.0): Before flux calibration (in extract.tbl spectra), LH orders are mismatched and appear tilted to the red. This is caused by the increasing wavsamp height with wavelength in each order to accomodate the spectrograph resolution. The effect is not apparent in the flux calibrated (spect.tbl) spectra, as it is corrected by the fluxcon polynomial.

Mitigation. For S17 onward, this effect is corrected at the spectral extraction stage. The flux in each wavelength bin is normalized to a wavsamp height of 1 pixel.

6. LL Salt and Pepper Rogues (Solved in S17)

In S15, LL 'Super Darks' averaged over many campaigns were used to subtract off the instrumental bias signature. Due to the variable nature of rogue pixels, this introduced salt (new rogues not in the super dark), and pepper (rogues in the super dark, but not in the current campaign) into the data.

Mitigation. From S17 onward, campaign-dependent 'moving window darks' are used to better match the rogues in each campaign and to track the changing minimum zodiacal light level. The salt and pepper in S15 BCDs can be very effectively removed by subtracting an appropriate background observation.

Go back to IRS page