The charge transfer efficiency (CTE) of the UVIS detector has inevitably been
declining over time as on-orbit radiation damage creates charge traps in the CCDs. Faint sources in particular can suffer large flux losses or even be lost entirely if observations are not planned and analyzed carefully. In this section, we describe the effect of CTE losses on data, observational strategies for minimizing losses, and data analysis techniques which can to some extent correct for CTE losses.
The flux of energetic particles in low-Earth orbit, mostly relativistic protons and
electrons encountered during HST’s frequent passages through the South Atlantic Anomaly, continually damages the silicon lattice of the CCD detectors. This damage manifests itself as an increase in the number of hot pixels, an increase in the dark current, and an increase in the charge trap population.
The effect of hot pixels, about 1000 new/day/chip using a threshold of 54e-
/hr/pix, is addressed with anneal procedures, dark calibration files, and dithering. The anneal procedures, performed monthly, warm the detectors to +20C and restore 80-90% of the hot pixels to their original levels. Dark calibration files (running averages of daily dark images) can provide reasonable identification of hot pixels as well as a calibration for overall dark current (see Section 5.4.8
for the median level and growth rate). The calibration darks allow hot pixels and dark current to be subtracted from science images in the calibration pipeline. Due to the time-variable behavior, the corrections are imperfect, but the dark current levels are low and dithering can help reduce any residual impact of hot pixels in final image stacks.
The effect of charge traps is more difficult to address, as the traps do not respond to
the anneals and the damage appears to be cumulative and irreversible based upon both flight and ground test experience. The traps cause a loss in source flux as well as a systematic shift in the object centroid as the charge is trapped and slowly released during readout. The majority of the trapped charge is released during the readout within ~1/2 dozen pixel shifts, as evidenced by the charge trails which follow hot pixels, cosmic rays, and bright stars. A low percentage of the initial signal can be seen extending out to ~50 pixels in length (see Figure 6.15
|The image background: a higher background fills some of the charge traps,
thereby minimizing flux losses during readout of the source signal. WFC3/UVIS images can have very low intrinsic backgrounds due to the low detector readnoise and dark current as well as the small pixels of the CCDs. Furthermore, the WFC3 UV and narrowband filters have exceptionally low sky backgrounds.
Thus, the CTE loss will depend on the morphology of the source, the distribution of
electrons in the field of view (from sources, background, cosmic rays, and hot pixels) and the population of charge traps in the detector column between the source and the transfer register. And, of course, the magnitude of the CTE loss increases continuously with time as new charge traps form. Further details of the current understanding of the state of the WFC3/UVIS charge transfer efficiency (CTE) are presented in Section 5.4.11
and can be found at:
The remainder of this section will discuss the available options for mitigating the
impact of CTE losses and their associated costs. Broadly, the options fall into two categories: those applied before data acquisition, i.e., optimizing the observing strategy during the proposal planning stage, possibly including the use of post-flash, and those applied during image analysis after the images have been taken, i.e., formula-based corrections or image reconstruction.
1) Consider the placement of the target within the field of view
. For example, when possible, place a target close to a readout amplifier - a viable option when the target is small. This reduces the number of transfers during readout, thereby minimizing CTE losses. Naturally, this will only be appropriate for a small subset of programs but should also be considered e.g. for observations of compact targets where the scientific interest in the surrounding field is limited or where inferior CTE for those surrounding observations is acceptable.
Another approach, suitable for sparsely populated fields in which the sources of
scientific interest are relatively bright, involves obtaining observations at multiple spacecraft roll angles. In this case, the different roll angles (ideally at or near 90 degrees) will result in sources having large variations in the number of pixels over which they must be transferred during readout. This permits a direct assessment of the reliability of the available formulaic photometric CTE calibrations which can be applied during post-processing (discussed in more detail below in Section 6.9.3
If observations are being taken on a field larger than the instantaneous field of view
of the cameras, then stepping in the Y direction (i.e. along the CCD columns) with a small degree of overlap will place some sources at both small and large distances from the transfer register again permitting a direct assessment of the photometric reliability of the CTE corrections applied during data processing (see section on formula-based corrections below in Section 6.9.3
2) Increase the image background by lengthening exposure times, using a broader filter, and/or applying an internal background (post-flash)
. Ensuring that images with faint sources contain a minimum of 12e-/pix background is expected to be a key CTE mitigation strategy for many WFC3/UVIS science proposals in 2012 and beyond.
On-orbit testing has shown that CTE losses are a non-linear function of both the
source and image background signals: a faint source in a low-background (<12 e-
) image will lose a significantly larger proportion of signal than a similar source in a high-background image. In some cases, faint sources can even disappear completely during the readout transfers, as illustrated in Figure 6.16
(Anderson et al. 2012
). The left panel is the result of a stack of long exposures minimally impacted by CTE losses, i.e., effectively a ’truth’ image. The middle panel presents the result of a stack of short, very low background exposures; the CTE trails are clearly visible above each source and the charge traps have completely smeared out the signal from the faintest sources (e.g., A and D). The right panel is a stack of short exposures where each image had ~16 e-
total background. The CTE trails have been reduced considerably and stars lost in the stack of very low background images are recovered in the stack of higher-background images, a clear qualitative demonstration of how a small amount of background can preserve even small charge packets of signal.
A portion of the Omega Cen central field far from the readout amplifier. The left panel
shows the result of a stack of eight 700s images, with minimal CTE losses. The middle panel shows a stack of nine 10s exposures with only ~2e-
natural background each; note the charge trails due to CTE loss extending upwards from each source in the field. The right panel is a stack of nine 10s exposures with ~16e-
background total (sky + post-flash) in each image.
A more qualitative measure of how relatively low levels of background can
significantly improve the CTE and increase the S/N of very faint sources is presented in Figure 6.17
. Shown are aperture photometry results for faint stars in very low background (top row) and higher-background data (bottom row), as a function of the number of transfers, i.e., distance from the amplifier. The target sources are faint: 100, 50, and 10e-
total in a 3x3 pixel aperture from left to right. Sources far from the amplifiers (~2000 on the x-axis) in images with little background (top row) effectively disappear. The same faint sources embedded in images with slightly higher background are detectable at the 50-75% level (Anderson et al. 2012
The required increase in background necessary to provide effective CTE-loss
mitigation can sometimes be obtained simply by lengthening exposure times or by using as broad a filter as possible. Sky levels in the UVIS filters are such that with exposure times of 2000 sec or longer, about half the filters will have natural background levels >10e-
(WFC3 ISR 2012-12
). The remaining filters, mostly UV and narrowbands plus some medium band filters, will have relatively low backgrounds and a post-flash is recommended as a means of boosting the total background. While using a post-flash to increase the underlying background in images may seem counter-productive from a S/N perspective, it will significantly increase the CTE for low-level sources and as a result, source signal will accumulate much faster than the noise will increase. Observers with faint sources in e.g. narrowband or UV filters or with science programs requiring the coaddition of multiple images to reach very faint limits should plan to achieve a total
background level of ~12 e-
per exposure. This recommendation represents the optimum value as of late 2012: a balance between the level of CTE mitigation achieved and the additional noise penalty incurred from adding extra background. Note that once backgrounds exceed ~12 e-
/pix, there is minimal improvement in retaining charge from faint sources (Anderson et al., 2012
In order to determine the necessary post-flash level to use for mitigating CTE
losses, observers will first need to estimate the expected natural backgrounds. The Exposure Time Calculator
provides such estimates, including contributions from sky, dark, zodiacal light, earthshine, airglow, and a selected level of post-flash. In addition, empirical backgrounds as measured on all WFC3/UVIS frames in the archive are summarized in the Appendix of WFC3 ISR 2012-12
. If the natural background of the exposure will be higher than 12 e-
/pix then there is no need to add post-flash. For images with very low background levels, enough post-flash should be applied to achieve ~12 e-
/pix total background (natural+post-flash).
Observers invoke post-flash in APT by choosing the exposure optional parameter
‘FLASH’ and specifying the desired number of electrons per pixel to be added to the image using the LED post-flash lamp. (See Section 13.2.4 in the Phase II Proposal Instructions
.) The additional overhead times required for post-flash are generally very small (typically ~ 6 sec). The flash is performed on WFC3 after the shutter has closed at the end of the exposure: an LED is activated to illuminate the side of the shutter blade facing the CCD detector (WFC3 TIR 2012-01
). The experience so far with these lamps (corroborated by the design analysis) indicates that the illumination pattern is very repeatable (to <<1%); it is similar for the two sides of the shutter blade. The intensity is likewise very repeatable, as expected. The brightness of the flash has shown fluctuations of rms~1.2%; the long-term stability is ~0.1%. Calibration reference files have been delivered to CDBS for the calibration of exposures using post-flash. (See WFC3 ISR 2013-12
The main disadvantage of post-flash is, of course, the increase in the background
noise. In the worst case, a short exposure with low background and dark current would require the addition of about 12 e-
/pix of post-flash. Thus the original readout noise of ~3.1 electrons is effectively increased to 4.6 e-
in un-binned exposures. (See Section 9.6
for S/N equations.) For a noise-limited observation, the exposure time would need to be increased by more than a factor of 2 to retain the same noise floor. In most cases, however, the impact will be significantly less severe as exposures will generally contain some natural background already and will not require a full 12e-
Finally, please note that even with a moderate background, larger charge packets
from brighter stars, hotter pixels, or cosmic rays will still experience some loss and trailing of their initial number of electrons. Thus, even with 12 e-
background, all sources will still suffer some CTE losses (see Figure 6.18
) and it will be necessary to apply an additional correction during data processing.
3) Use charge injection.
For completeness, this mode is included as an observing strategy option for CTE-loss mitigation but, in practice
, it is not considered as useful as e.g. increasing image backgrounds or applying corrections during image processing. Its use will be permitted only in exceptional cases where the science requires it. Observers who wish to use this mode are advised to consult their Contact Scientist or e-mail email@example.com
Charge injection is performed by electronically inserting charge as the chip is
initialized for the exposure, into either all rows or spaced every 10, 17, or 25 rows. Only the 17 row spacing is supported as of mid-2012. The injected signal is ~15000 electrons (not adjustable) and results in about 18 electrons of additional noise in the injected rows (Baggett et al., 2011
). The rows adjacent to the charge-injected rows have between 3 and 7 electrons of effective noise due to CTE effects. The charge injection capability was supported in Cycle 19, but experience has demonstrated that it is useful for very few types of observations. Its primary drawbacks are the uneven degree of protection from charge trapping in the rows between the injected charge rows, an increase in noise in the rows closest to the injected charge, and a very difficult calibration problem posed by the combination of sources in the field and the injected rows, which give rise to different levels of CTE at different places within the image. Furthermore, the strong dependence of CTE losses on image backgrounds makes it challenging to produce a suitable calibration, as typically there will be a mismatch in image backgrounds between the charge injected calibration and science frames (i.e., differing levels of CTE losses).
1) Apply formula-based corrections for aperture photometry.
One way to correct CTE losses after the images have been acquired is to apply an empirical photometric calibration. The current model, based on stellar aperture photometry results, provides corrections for CTE losses as a function of observation date, image background, source flux, and source distance from the amplifiers. Figure 6.18
illustrates the necessary corrections for low, intermediate, and high background images. As expected, larger corrections are required for fainter sources and/or lower image backgrounds. The top left panel represents the worst-case scenario: in short exposures in the narrowband F502N filter, with effectively zero background, corrections approach 50% for the faintest sources. Even a small amount of background (2-3e-
, bottom left panel) significantly improves the CTE, with corrections of ~20% required for the faintest sources. The corrections for high background data are at the few percent level and relatively independent of source level while corrections for brighter sources are <5% regardless of background level. More details, including the tabulated coefficients, can be found in WFC3 ISR 2012-09
. Note that these results are for small apertures; corrections for larger apertures will be smaller as more of the trailed charge is included in the aperture. These, as well as corrections for a wider range of image backgrounds, will be available in the near future.
CTE losses in magnitudes per 2048 detector rows as a function of star flux (within 3-pixel
radius), observation date, and image background. Top row panels show results based on narrowband images with short and long exposure times (left and right, ~0e-
/pix and ~1e-
/pix, respectively) while bottom row panels are for broadband images with short and long exposure times (left and right, 2-3e-
/pix and 20-30e-
/pix, respectively). Open and filled symbols are based on NGC104 (47Tuc) and the sparse cluster NGC6791; lines denote the best fits. MJD 55106 and 55854 correspond to Oct. 2, 2009 and Oct. 20, 2011, respectively.
The formula-based correction method can be quite effective for isolated point
sources on flat backgrounds, but it is less suitable for extended sources or sources in crowded regions. One benefit of this formula-based recalibration is that it is not impacted by the possibility of readnoise amplification, which can be a concern for the pixel-based reconstruction discussed in the next section. Photometric corrections are also useful for planning observations: they allow an estimate of the CTE losses for point-like sources that can be expected in a near-future observation for a given background and source flux. The expected losses should be taken into consideration during observation planning and if necessary, total integration times increased to achieve signal to noise requirements.
2) Apply the empirical pixel-based correction algorithm.
The ACS team recently developed and implemented a post-observation correction algorithm employing the Anderson and Bedin methodology (2010; PASP 122 1035
) and a similar capability has been available for WFC3 since mid-2013. The algorithm is calibrated using the behavior of hot pixels and their charge trails. In the absence of CTE losses, the full charge of a hot pixel is entirely contained within a single pixel and its noise is the combination of its shot noise and the noise due to readout and background in that one pixel. If some of the hot pixel charge is lost due to imperfect CTE, there will be fewer electrons in the hot pixel itself, and more in the trailing pixels. (see Figure 6.15
.) To obtain the original value of the hot pixel, the correction algorithm must determine how many electrons the original hot pixel would have to have in order to be read out as the observed number, given the number of traps left full and empty by the preceding pixels. The resulting correction essentially redistributes the counts in the image, “putting the electrons back where they belong”, i.e., undoing the effects of degraded CTE (Anderson et al., 2012
While the pixel-based algorithm in ACS has been successful at removing trails
behind stars, cosmic rays, and hot pixels, it has one serious and fundamental limitation: it cannot restore any lost S/N in the image
. Faint sources, and faint features of extended sources, may be so strongly affected by CTE losses that they become undetectable and cannot be recovered (e.g., see Figure 6.16
and Figure 6.17
). In addition, this pixel-based method is effectively a deconvolution algorithm, and it can amplify noise or sometimes generate artifacts. Even so, despite the limitations, the reconstruction algorithm provides the best estimate of the original image before it was read out and also aids in understanding how the value of each pixel may have been modified by the transfer process.
We end this Mitigation section by noting that, depending on the science goals, a
single mitigation method may not be sufficient for some programs. Observers, particularly those with faint sources, may need to consider applying both pre- and post-observation CTE-loss mitigation strategies, e.g., increasing the image background to ~12e-
/pix to reduce CTE effects followed by an application of either the formulaic photometric or pixel-based corrections. We note that the pixel-based correction algorithms are not able to operate on binned data, but binning is not an effective way of increasing the detectability of faint sources (See Section 6.4.4
For the most current information on the WFC3 CTE and mitigation options, as well
as updates on the availability of a pixel-based correction algorithm for WFC3, please refer to the page CTE webpage at: