Image persistence is a phenomenon commonly observed in HgCdTe IR detectors.
It is an afterglow of earlier images that in the case of WFC3 is present when individual pixels are exposed to values greater than about 50,000 electrons. It has been observed both in dithered exposures in a single orbit and in some cases in subsequent orbits. A very obvious example is shown in Figure 6.12
. The image shown is of a fairly sparse, high latitude field, but the observation occurred shortly after several observations of fields containing very bright stars that had been dithered. Refer to WFC3 ISR 2010-17
for detailed information.
Persistence is most likely due to imperfections in the detector material, traps,
which become available to electrons or holes when voltage levels change as a pixel saturates. A detailed theory of persistence is described by Smith (SPIE 7021-22, Marseille 2008-06-24). In WFC3, a fully saturated pixel (one filled to > 100,000 e) produces a signal of about 0.2 elec/s 1000 seconds afterwards; the charge decays roughly as a power law with a slope of about -2. An example of the amount of persistence produced as a function of the amount of exposure in an earlier image is shown in Figure 6.13
Given a pixel that has been exposed to a depth z at a time t previous to an
observation, the persistence P(z,t) is roughly described by a Fermi-like distribution of the form:
is the persistence at a fiducial time to
and characteristic exposure zo
measured in electrons. Here dz describes the scale of exposures over which the persistence rises from a minimal to maximal value.
From the perspective of the data analyst, the primary questions about persistence
are how to find it in ones images and how to mitigate its effects if it is there.
Within a visit, persistence is usually not important unless the images are dithered
by steps that are larger than the size of the psf. This is because the flux due to persistence is a very small fraction of the original rate (usually considerably less than a percent). However, some observations will have been planned with large dither steps in order to avoid large features, e.g. the “Death Star”, and these images can contain regions with visible persistence. The easiest way to locate regions of the detector where persistence could be important in, for example, the last exposure of an orbit, is to multiply each of the science extensions in the preceding flt
images by their respective exposure times to recover the depth to which each pixel was exposed, and then to construct a “hit map” containing the maximum value of the individual images. Regions with values greater than about 50,000 e–
may have noticeable persistence (if as noted above the images have been dithered by large amounts). A fairly straightforward way to see the persistence is to use MultiDrizzle, comparing the so-called single_sci.fits files, which are the individual exposures projected onto a common coordinate frame. Subtracting the first image from subsequent images (with the same exposure and filter settings) will reveal the persistence in the later images.
Persistence from previous visits is often easier to recognize than self-induced
persistence because the persistence image often shows multiple dither offsets superposed or a very different star field, e.g. a cluster of stars in a field that was not expected to contain one. The easiest way to see if your image has persistence is often simply to display the first image in an orbit with a stretch designed to bring up faint features (or in histogram equilibration mode) and look for features and/or patterns that are unexpected. Using the single_sci.fits files, as described above, is another way to look for persistence from previous visits, particularly if the last exposure is a significant amount of time later than the first exposure.
An example of this is shown in Figure 6.14
. The top image shows the first exposure in an observation of a high latitude field with WFC3-IR. The image contains a fairly large number of objects which appear to be stars, although in fact an examination of the “objects” would show that they all appear extended. Many of these stars are persistence images of bright stars in the outskirts of 47 Tuc which had been observed in the previous orbit. The difference of the first and last single_sci.fits images is shown in the bottom image. It clearly reveals the persistence.
If you suspect persistence, or if you want to assure yourself persistence from earlier
visits is unlikely to be a problem, it is also useful to establish what the history of WFC3-IR usage was prior to your visit.
If there are no visits within 4 hours of your observation, then it is unlikely there is any persistence in your images. If there are earlier images, and the images are public, you can learn a lot more about the problem by downloading the images and determining which portions of the image have been saturated. If the images have not yet been released to the public, you cannot download the images, but you can learn more about the field looking at the proposal abstract, and in some cases by looking at information that is available on the program status page (which you reach from the results from the search above by following the link on proposal ID, and then the link labeled “about this proposal”.) Downloading the phase II proposal for the “offending” proposal, and displaying the DSS or 2MASS field, will give you a pretty good idea of the magnitude of the problem you are facing.
Basically, there are two ways to mitigate persistence. One can exclude the regions
affected by your data from your analysis, or one can try to subtract the persistence signal from your data and use the modified data in your data analysis.
About 1% of the pixels in the IR array have data quality problems (e.g. IR blobs),
and if one needs to exclude another 1% of the pixels due to persistence this may well be the best course. Simple IRAF procedures can be used to mark bad regions and add a flag in the data quality extensions of the flt
files. At this point, down-stream analysis proceeds as it normally would if the tools that are used take data quality into account. This is the case, for MultiDrizzle, which most observers use to analyze their data.
If on the other hand, it is important to use the data in the region affected by
persistence in data analysis, then one needs a procedure for modeling the persistence, and then subtracting the data from the images. The STScI is currently proto-typing such a procedure. It assumes a Fermi-like distribution for the persistence signature described above and a power law time decay of persistence. An estimate of the persistence affecting Figure 6.12
is shown in Figure 6.15
, and the persistence subtracted image is shown in Figure 6.16
The image appears much cleaner now, though a careful inspection shows some
residual signatures of the persistence. At present, we are usually able to remove 90% of the persistence signatures in typical images, but we have to tune some of the parameters in our model for the overall signature and the decay in order to do so. Most likely this is due to the fact that we have not included some of the characteristics of persistence in our model, including for example the length of time at which pixels are filled to a high level. Calibration observations have been undertaken to attempt to characterize the persistence better.
At some point, we expect to release the tools we have for general use. In the meantime, if persistence is a significant problem in the analysis of your images, please contact the help desk. We can provide estimates of the persistence in flt
files and work with you to either mask out the bad regions or to subtract at least partially the effects of persistence in your images.