4.6 Effects of Overexposure
Because each pixel of the NICMOS detectors is read individually, overexposure does not cause "bleeding" along columns or rows, as is seen with CCDs. Exposure to bright sources, however, can result in two other sorts of NICMOS artifacts: persistent images, and electronic ghosts known colloquially as the "Mr. Staypuft" anomaly.
4.6.1 Photon Persistence
HgCdTe detector arrays are subject to image persistence. When pixels collect a large amount of charge, they will tend to "glow" for some time after the end of the exposure. This persistent signal decays exponentially with a time scale of about 160 ± 60 seconds, but there is also a long, roughly linear tail to the decay such that persistence from very bright sources remains detectable as much as 30 to 40 minutes after the initial exposure. Exposures of bright astronomical targets therefore can leave afterimages which appear in subsequent exposures taken within the same orbit, or even into the next (figure 4.7). If you are analyzing dithered exposures of a bright target, you may wish to carefully inspect each image for residual ghosts from the previous exposure(s), and perhaps mask them out when coadding the dithered images. It appears that all sources of illumination probably leave persistent afterimages, but under typical conditions they are most noticeable for sources which have collected 20,000 or more ADU during the previous exposure. It is not unreasonable to expect afterimage signals of ~1 e-/second immediately following a severe over-exposure. Occasionally, persistent images result from bright targets imaged by the previous NICMOS observer.
Figure 4.7: Persistence induced by an exposure of a bright star. The left and right images show dark frames taken 32 and 512 seconds after the exposure of the star.
Cures:
There is no easy cure for persistence in existing data; just be aware that it may be present. For dithered observations you can mask out persistent images from previous exposures before combining into a mosaic. It may be possible to subtract a rescaled version of the previous image that caused the persistent afterglow from the persistence image itself and thus remove it (see the discussion of cures for cosmic ray induced persistence below)
4.6.2 Cosmic Ray Persistence
More insidiously, during regular passages of HST through the South Atlantic Anomaly (SAA), the arrays are bombarded with cosmic rays, which deposit a large signal into nearly every pixel. The persistent signal from these cosmic rays may then be present as a residual pattern during exposures taken after the SAA passage. This appears as a mottled, blotchy or streaky pattern of "noise" (really signal) across the images, something like a large number of faint, unremoved cosmic rays. These persistent features cannot be removed by the MULTIACCUM cosmic ray processing done by the standard pipeline (i.e. the CRIDCALC step of calnica) because they are not transient. Rather, they are a kind of signal, like a slowly decaying, highly structured dark current.
Cosmic ray persistence adds non-Gaussian, spatially correlated noise to images, and can significantly degrade the quality of NICMOS data, especially for exposures taken less than 30 minutes after an SAA passage (see examples in figure 4.8). Count rates from moderately bad cosmic ray persistence can be of order 0.05 ADU/second, with large pixel-to-pixel variations reflecting the spatial structure of the signal. The effective background noise level of an image can be increased by as much as a factor of three in the worst cases, although 10% to 100% is more typical. This "noise" is primarily due to the spatially mottled structure in the persistence, not the added Poisson noise of the persistence signal itself. Because HST passes through the SAA seven or eight times a day, a large fraction of NICMOS images are affected by cosmic ray persistence to one degree or another. Observations of bright objects are hardly affected, since the persistent signal is usually quite faint. Similarly, short exposures are not likely to be badly affected because the count rate from persistence is low and may not exceed the detector readout noise. But deep imaging observations of faint targets can be seriously degraded. The NICMOS Instrument Science Report ISR-98-001 (Najita, Dickinson and Holfeltz 1998) presents a detailed discussion of this phenomenon and its effects on imaging observations.
Figure 4.8: Cosmic ray persistence in three dithered NIC2 MIF1024 images. For each image, the histogram at bottom shows the distribution of pixel values near the sky level. The shoulder of positive values for the persistence-impacted images demonstrates the non-Gaussian nature of CR persistence noise. The "normalized RMS" is the measured value divided by the mean for many "clean" images free of CR persistence.
Cures:
If you have multiple exposures within a given SAA-impacted orbit, the amplitude of the persistence should decay, and you may want to give later exposures higher weight when combining them. Sometimes it may be preferable to discard the worst frames altogether. Well-dithered data lessens the impact of persistence, since objects will move relative to the persistent CR signal and it will not sum coherently when the data are registered. With well-dithered data (at least three positions), one can also take advantage of the drizzling procedure and associated software in the STSDAS dither package to identify and mask the worst effects of persistence, as described, e.g., in the "Drizzling Cookbook," (Gonzaga et al. 1998, STScI ISR WFPC2 98-04).
A web-based tool for determining the time since SAA passage for any given data set is now available from the STScI NICMOS web pages. The tool is also linked to the NICMOS History Tool (see section 5.2). It accepts a list of observation times (specified as modified Julian dates, MJD - these can be obtained from the EXPSTART image header keyword), and generates a list of the elapsed time and durations of the preceding SAA passages, as well as a graphical display showing the observation and SAA times. Because the amplitude of cosmic ray persistence in NICMOS images depends strongly on both the time elapsed since SAA passage and on the duration of the passage itself, this tool provides a useful way of quickly "pre-screening" your data to see which frames are most likely to be contaminated.
On rare occasions, it may happen that dark exposures were taken after an SAA passage but before science observations. These darks can provide a "map" of the cosmic ray persistence, and in principle it is possible to scale and subtract them from the subsequent science exposures in order to minimize the effects of the cosmic ray afterglow. This has even been done using short science exposures taken early in the orbit, scaled and subtracted (after masking out objects) from the science images that follow. If you have multiple exposures in an orbit, you may wish to consider trying this, but at present there is no general software available for doing this procedure. Because the CR persistence pattern remains fixed in successive exposures, you may also use the first exposure in the orbit to create a mask for setting the affected pixels to zero weight when combining the images. In HST Cycle 11, when NICMOS is revived, STScI will automatically schedule special dark exposures after SAA passages in order to provide a map of the CR persistence. Software for implementing this persistence correction will be written and tested as soon as possible after SMOV3B, when the NCS is installed.
4.6.3 Amplifier Ringing (The "Mr. Staypuft" Anomaly)
There is another sort of NICMOS ghost image, wholly separate from those induced by persistence. Bright targets which appear in a given detector quadrant can produce an electronic ringing effect in the readout amplifiers, which may induce ghost images at the same pixel locations in the other three quadrants. This is believed to be due to the pull-down of the power supply, which does not completely recover from reading a large number by the time it's asked to read the next number (from the next quadrant). In some cases a whole row (or column, whichever direction the "fast" read clocking runs) can be anomalously high. The amplitude of the ghost images is of order a tenth of a percent of the real image count rate. Inside the STScI NICMOS group, this effect is fondly known as the "Mr. Staypuft anomaly" (figure 4.9).
Figure 4.9: The Mr. Staypuft anomaly. The bright source at lower right produces electronic ghost images in the other three quadrants, plus vertical streaks. In Camera 1 images, these streaks run horizontally.
Cures:
At present there is no real solution to the Mr. Staypuft anomaly other than to be aware that such ghost images may exist. For example, if your image has a bright source at pixel (58,143), then you may see ghost images around (186,143), (186,15), and (58,15). If the observation was made with Camera 3 (so that the fast clocking direction is along columns), there may also be elevated signal at all rows on and around columns 58 and 186. Ordinarily it is not possible to eliminate these ghosts from dithered data by masking, since they will move about on the array along with the astronomical targets. If, for some reason, observations were obtained at multiple roll angles, then it would be possible to mask the ghosts and eliminate them; this was done for the Hubble Deep Field South NICMOS observations.
Some NICMOS observers studying relatively blank fields have dealt with the "elevated columns/rows" phenomenon by fitting medians to the affected columns/rows (or to portions of the columns/rows, avoiding regions affected by bright or extended targets), and subtracting the median values from each column or row of the data. This can at least provide a cosmetic fix. In principle, this correction should probably be applied before flatfielding the images, since it is most likely that the "Staypuft" signal is not responding to the array QE variations, although this has not yet been formally tested. However, if the fit is done before flatfielding, then the sky background itself will not be uniform, and in fact will be improperly subtracted by the procedure. For this reason, in practice it is probably safer (if not strictly correct) to apply a median column (or row, for NIC1) correction after flatfielding instead. The STScI NICMOS group is now experimenting with algorithms to correct the Staypuft column/row effect in post-processing, based on an empirical model for its electronic behavior. A software tool for correcting data may be released sometime in the future, and it is possible that the procedure may be automatically implemented as a step in the calnica pipeline.
4.6.4 Optical Ghost Images
In addition to persistence and electronic ghosts, NICMOS polarimetric images of bright targets are subject to optical ghosts (figure 4.10). These are most evident for the short wavelength polarizers in NIC1. The location of the polarizer ghosts relative to the source changes direction with respect to the principal axes of transmission. For POL0S, the brightest ghost is found roughly (-10,+35) pixels away from the target position, and a second one at (-16,+70) pixels. It is possible that for a brighter object more ghosts would appear in the same angle and direction. POL120S images appear to be ghost-free. For POL240S, ghosts appear (+10,+54) pixels from the source position and (+18,+80) pixels, with the possibility of more appearances for brighter targets.
Figure 4.10: Optical ghosts in NIC1 polarizers.
Ordinary (non-polarimetric) NICMOS images of bright sources may have very faint filter ghosts which result from back-reflections off the faceplate and the filters. The position of the ghosts changes from one filter to another. They are roughly 8 magnitudes fainter than the star which produces them. The ghosts cannot be completely subtracted from the field unless the reference PSF is at the exact same location as the star. They are typically seen as a residual in the PSF subtracted images, and look like large "donuts" because they are out-of-focus reflected images.
Cures:
No true cure is possible: be aware that such ghosts may exist; their positions are predictable. If images are available with multiple spacecraft roll angles, the ghosts may be masked out before combining the data.
