There are certainly as many variants of acquiring and reducing near infrared images as there are for CCD data. While this chapter is by no means the final word on handling infrared imaging data, it is useful information to the novice and may provide insight into the key issues surrounding infrared imaging. Interested Redeye users are encouraged to read other descriptions of data acquisition and processing, including the papers by McCaughrean (1989) and Hodapp et al. (1992).
There are several types of frames that must be recorded on a nightly basis if reliable, photometrically accurate images are to be attained. These types include (beyond the obvious science images) flat fields, dark frames, and sky frames. In many cases flat fields and/or sky frames can be derived directly from science images, which greatly increases observing efficiency.
For reasons described in Chapter 2, the dark current seen in blanked-off framesis not a simple linear function of integration time. Furthermore, extremely low level light leaks contribute to the signal when the filter wheel is in a blanked-off position. These two factors demand that dark frames of identical length as those used during science observations be acquired. Since the Gen III controller used by Redeye has an array temperature control circuit, the temperature of the array is extremely stable over the course of a night, as long as the dewar was topped off with LN2 in the evening. In practical terms this means the dark current is stable over the course of the night and observers can take a set of dark frames in the morning, before topping off the dewar and heading back down the summit, with exposures that are identical to those used over the course of the night's observations. Cautious observers might take dark frames at the beginning and in the middle of the night, but the gains of these extra dark frames are minimal and one must balance their acquisition in light of the lost observing time. Also, remember to always flush the array of residual charge to the maximum extent practical before recording dark frames. The easiest way to flush residual charge is to take 3-5 bias frames with the filter wheels in the blank-off mode before recording dark frames. Also, note that because of the double correlated sampling used by Redeye, bias frames are not needed as part of the nightly routine. Unlike CCDs, a frame with a 0 sec integration time with Redeye yields a frame with an average signal of 0 ADU, hence subtracting bias frames off object frames will only add noise to processed images. Finally, do not expect to see a significant amount of dark current in exposures < 30 sec. Even though the camera specifications list a ~1 e-/sec dark current rate, the e-/ADU conversion settings of the controller, intrinsic read noise of the array, and non-linear dark current generation mechanism all work to conceal a significant dark current signature in short dark frames. Nonetheless, these short dark frames are extremely useful for removing fixed pattern noise in images, as well as help minimize the impact of moderately defective pixels, e.g., photosensitive pixels with an unusually high dark current that is repeatable.
There are a couple of simple techniques for acquiring flat fields. These include dome flats using incandescent lamps and deep exposures of the sky. The literature is filled with examples of infrared images being processed with both types of flat fields, and the respective advantages of these techniques tend to depend on the goals of science programs.
The simplest way to acquire flat fields is to image the inside of the CFHT dome while it is being illuminated with incandescent lights. This is the same technique often used to acquire FOCAM dome flats. Such flats have the distinct advantage of requiring only a few seconds per exposure (for broad band filters) to acquire. Furthermore the ~2000 *K temperature of the filaments in the incandescent bulbs is a reasonably close match to the color of many infrared objects. An alternative strategy for making flats is to create a median sky flat from dithered images of the sky. In this case the same frames used for imaging science targets can also be used to create flats as long as the target is (1) small with respect to the field of view and (2) the dither pattern is larger than the size of the target so flux from the object does not interfere with the median sky calculation. Figure 7.1 illustrates this technique, which would be commonly used on deep imaging of distant (few arcsec) galaxies, or creating mosaics of open clusters. Basically, after the images are stacked into a data cube, they are normalized (generally to the mode of each frame), then a median value is calculated for each pixel down the cube. Though bright objects like stars may be in the individual frames, they will be thrown out of the resulting sky flat since they do not overlap in the stacked data cube. Note that this technique fails if the vast majority of the field of view is not filled with sky flux, so using it on objects like extended galaxies or the cores of globular clusters will lead to significant artifacts in processed flats.
Certainly one of the greatest complications associated with infrared vs. optical imaging is the variability of the sky on short time scales. This variability is due in large part to changes in the emission of OH* in the upper atmosphere (see Figure 7.2). Ramsay et al. (1992) give an excellent account of this emission source and illustrate how fast it can vary (as quickly as a few minutes). While OH- emission is the dominant sky component in the J and H bands, it is only part of the problem in the K-band, where thermal emission from the environment is beginning to dominate. Since the thermal flux encroaching on the K-band is on the exponential side of a Planck curve, small changes in the ambient temperature can lead to significant changes in the sky background. This effect is most noticeable in winter, when the coldest temperatures are reached at observing sites and the deepest K-band imaging is generally achieved.
With the variability of the sky in mind, sampling the sky at frequent intervals can be crucial for programs that fill most of the Redeye field with source flux (e.g., extended galaxies). In such programs the sky should be sampled no less frequently than every ~15-30 minutes. A manual telescope beam switch to a clear field is something that can be easily configured into the TCS by the telescope operator. If the source only occupies a small portion of the field, then to a good approximation the sky can be treated as an unresolved background component in images and subtracted off as a constant during reductions. Specifically, the median value in a box off the target of interest in object frames can be evaluated and subtracted off after dark frame subtraction and flat field division. Beyond the obvious boost in observing efficiency this simple technique implies, it also has the advantage of not injecting noise into the processed frames through a separate independent sky-frame subtraction step.
Like observing there are a number of proven strategies for reducing infrared images. Many of these techniques stem from well established CCD reduction procedures. Illustrated below in equation 7.1 is one such technique:
This is the algorithm used in the Redeye on-line preprocessing option available in Pegasus. In this relation, a dark of identical integration time to a raw image is subtracted from a raw image. This first step serves to (1) subtract the dark current component from raw object frames, (2) minimize the impact of bad pixels that are effected by abnormally high dark current values but still are photosensitive, and (3) helps reduce the problem of amplifier glow from the corners of the array. Next, this difference is divided by a flat field that has been normalized to unity in order to compensate for variations in responsivity (both pixel-to-pixel and broad changes across the field) in the array. These steps also serve to reduce the impact of fixed-pattern noise that can be strong in raw images. Such noise is in the form of alternate columns have high and low signal offsets. Next the frame has the sky flux removed by simply evaluating the median signal level in the field and subtracting this constant from the image. This technique is only effective if the field is dominated by sky flux. If the field is filled with stars or an extended object the median sky value will be over estimated. This technique also assumes that there is no residual structure in sky flux, either from the sky or the telescope environment. Finally, the image is passed through a bad pixel filter designed to substitute systematically bad pixels with median values of nearest-neighbor pixels. This bad pixel map remains on disk and is updated periodically as part of a long term program of monitoring the "health" of the Redeye detectors. In the case of the Pegasus algorithm, a bad pixel is defined to be one that differs in responsivity from the mean across the array by more than a factor of 2. In almost all cases this corresponds to dead (insensitive) pixels.
A more rigorous technique for image processing involves explicit handling of separate sky frames, dome flats, and object frames in final reductions. Specifically, after object frames have been dark subtracted and divided by a normalized dome flat, a sky frame from a clear field taken close in time to the object frame being processed is subtracted off. This technique (rather than simply subtracting the sky as a DC offset) handles spatially resolved thermal components in the background, in particular emission from the telescope and dust particles that have collected on the Redeye window (and therefore emit light) and cold internal optics (that can absorb light).
It is well known that, unlike visible light observations, those made at wavelengths beyond ~1 *m are not particularly susceptible to scattered moonlight. While in general observers need not be concerned about the effects of moonlight during Redeye observations, some programs may suffer from moonlight contamination if targets are too close to a nearly full moon. For this reason the brightness of the moonlit sky as a function of a variety of parameters are described in this section. Users who are interested in a detailed description of the moonlit sky should see Krisciunas and Schaefer (1991; hereafter KS) and references therein.
Two factors make scattered moonlight a minimal issue in infrared observations. First, the scattering mechanisms for light propagating through the atmosphere are highly wavelength dependent and tend to be much less efficient at longer wavelengths. Second, the intrinsic brightness of the sky is so much brighter in the infrared than it is at visible wavelengths, that the relative contribution from scattered moonlight to the total sky flux is relatively small. The exact brightness of the moonlit sky is a complex function of lunar phase, lunar altitude, target altitude, target-lunar angular separation, local extinction, and wavelength. KS list a series of equations in a model of scattered moonlight that can be used to estimate the sky brightness as a function of these parameters. The model equations are:
Here I* is the brightness of the moon with phase angle *, * is the lunar-target separation, Z is the zenith distance, f(*) is the scattering function, and k is the atmospheric extinction. All angles are in units of degrees. The sky flux, Bmoon, is calculated in the linear unit of nanolamberts. It is important to note that the model of KS (1) only applies to the V-band and (2) has a scattering function derived empirically from Mauna Kea observations. Transforming this model to infrared wavelengths is non-trivial primarily because of uncertainties in the * dependence of the scattering function, which KS describes as a linear sum of the Rayleigh and Mie scattering mechanisms. Specifically,
Short of replicating the measurements of KS at J, H, and K, the wavelength dependent terms in the KS equations can be adjusted (approximately) as follows to work in the near infrared. First, the apparent brightness of the moon must be changed to reflect the solar color, changing equation 7.2 to:
Second, the Rayleigh scattering term is assumed to have the same functional form as equation 5, but a wavelength dependence of *-4, hence can be modified and normalized to the V scattering function:
Finally, the Mie aerosol scattering terms can be rewritten to introduce a *-1.3 dependence (Allen 1976),
With these terms known, the total sky brightness in an infrared band is:
where X is defined in equation 7.4, and Bzenith is the brightness of the moonless sky at zenith. The relation in Garstang (1989) to convert from nL to mag/arcsec2 is,
Using the extinction coefficients plotted in Chapter 5 and measured zenith sky brightnesses in Chapter 2 (measured with the moon below the horizon), it is possible to generate a grid of moonlit sky flux values at various wavelengths. Such models are plotted in Figures 7.3-7.6. All of the plots are for full moon conditions (with a lunar back scattering term added) with contours calibrated in mag arcsec-2. The moon is located on the prime meridian with a 20* zenith angle. This is a representative lunar position from the latitude of Hawaii. Prominent in all of the results is the bimodal nature of the scattering function, with relatively inefficient Rayleigh scattering dominating far from the moon and efficient Mie forward- scattering near the moon. The latter creates aureole and the highly peaked contours near the moon. Along the horizon nocturnal sky glow dominates scattered moon light. The main reason for inserting a V-band model is to demonstrate how scattered moonlight heavily dominates sky glow at visual wavelengths. In contrast, the infrared sky appears stratified, with relatively little scattered moonlight and most of the flux coming from sky glow along the horizon. At K, almost no scattering is seen beyond ~10* from the moon. Note that the typical brightness of the sky at V near the zenith on Mauna Kea on a moonless night is ~21.6 mag arcsec-2 (KS), and the faintest the V sky gets during a full moon is ~18.5 mag arcsec-2. At J, H, and K though, the brightness of the sky at zenith barely changes from its moonless value, even with a full moon ~20* away.
Of course some caution should be used while interpreting this model. While the results are qualitatively correct, the exact brightness of the moonlit sky varies from night-to-night for a number of reasons, including changes in extinction, OH- emission, and air temperature. Nonetheless, the conclusion illustrated here, that moonlight is not a problem for Redeye observations except when targets are within ~10 - 20* of the full moon, is certainly accurate.
Allen, C. W. 1976, Astrophysical Quantities (London, Athlone).
Garstang, R. H. 1989, P. A. S. P., 101, 306.
Hodapp, K-W., Rayner, J., and Irwin, E. 1992, P.A.S.P., 104, 441.
Krisciunas, K., and Schaefer, B. 1991, P. A. S. P., 103, 1033.
McCaughrean, M. 1989, Proceedings of the Third Infrared Detector Technology Workshop, ed. C. R. McCreight (NASA Technical Memorandum, No. 102209), p 201.
Ramsay, S. K., Mountain, C. M., and Geballe, T. R. 1992, M. N. R. A. S., 259, 751