While near-infrared detectors are fundamentally different than CCDs, the general data reduction steps required are essentially the same. Therefore, once you understood how to reduce optical data with THELI, you already have most of the knowledge required for the near-IR regime. However, instrumental and atmospheric effects make reduction of near-IR data tricky. The sections below summarise how you deal with this in THELI.
Crosstalk can in general be corrected well, provided that it is spatially stable. The latter is not always the case for near-IR detector arrays. In particular recent HAWAII2 sensors with multiple parallel readout sections can show crosstalk in form of compact positive and negative ghost images whose amplitude varies between readout sections. THELI assumes that the amplitude is the same, therefore the correction will only partially remove the effect (if at all). If you know in advance that this will be a problem for your science case, then consider choosing different camera rotator angles for your observations.
Some detectors will exhibit a horizontal or vertical brightening towards one edge of the readout quadrant. It is unstable and in general depends on temperature, exposure time, the number of previously executed exposures, and can also have an erratic component. You can correct for it using the Collapse correction.
In addition, if you take n exposures at a certain dither point and then move on to the next dither point where you repeat the same n exposures, piece-wise background correction can become necessary. In other words, even if the physical sky background does not change, the instrumental background pattern can change in a repeated fashion: it is the same for all k-th exposures in the series, but different from the pattern inherent to all (k+1)-th exposures (and so on). To apply such a group-wise background correction, use the spread sequence task and then proceed as normal. THELI will take care of the rest.
Optical and near-infrared imaging differ most when it comes to atmospheric sky background. In the near-IR the background level changes on time scales of a few minutes, and on angular scales of a few arcminutes. Some nights are very stable and one can calculate one background model that will suit all exposures taken within a 20 minute window after a simple rescaling (which can and should be applied with THELI). This would be a static background model. Other nights will be highly unstable and it is necessary to create individual background models for each exposure (dynamic modelling). More about dynamic and static models can be found in the calibration section.
In general, H-band is affected most by this highly variable airglow, as shown by this night sky spectrum:
Exposure times in near-infrared imaging are rather short, for two reasons:
The following table contains typical night sky surface brightnesses in mag/sq. arcsec, and characteristic exposure times. Integration times for Z are for near-IR detectors, for optical detectors they are usually longer (optical and near-IR Z-band filter curves are not necessarily identical):
| FILTER | B | V | R | I | Z | J | H | Ks |
|---|---|---|---|---|---|---|---|---|
| Surface brightness | 22.7 | 21.9 | 21.0 | 20.0 | 19.0 | 16.0 | 14.5 | 13.5 |
| Exposure time [s] | 600 | 600 | 300 | 300 | 60 | 30 | 20 | 10 |
Narrow-band filters
Exposure times with near-IR narrow-band filters should be kept as short as possible, and made as long as necessary in order to reach background limited images. There is a trade-off depending on which narrow-band filter you are using. Exposure times up to 300s are possible for filters where less airglow is present.
Co-averaging
One problem due to short exposure times in the near-infrared is that for high-galactic latitude fields the number density of discernible sources can become very low. This can be critical from an astrometric and photometric point of view, as internal data calibration is difficult. Most observatories offer the possibility to average a sequence of k exposures on-chip before the image is read out. In this manner the S/N is improved without running into problems with non-linearity, saturation or the 16bit dynamic range. This process is sometimes called co-averaging, the individual integration time is called DIT, and the number of co-averaged exposures NDIT.
When you reduce near-IR data the first time, you’ll probably be surprised that after flat-fielding the image appears everything else than flat. The amplitudes of the residual background variations from sky and instrumental contributions are usually much larger than the fluxes from astrophysical objects. Frequently, you won’t even be able to discern any objects at all in the data after flat-fielding.
The following example displays a 6x10s exposure (DIT=10, NDIT=6) taken with one of the four 2kx2k detectors of HAWKI@VLT in Ks-band. Shown in the left panel is the appearance after flat-fielding, and the right panel displays the same image after subtraction of a (dynamic) background model:
Which background model is most suitable for the data does not only depend on the atmosphere, but also on the target. If the field is sparse and contains point sources only, then very often a simple (static or dynamic) background model with basic min-max rejection and median combination is sufficient. Leave the DT, DMIN and SIZE thresholds empty.
If the field is crowded and you have many exposures, switching on additional object masking can avoid the background model to be biased by object fluxes. Use some explicit settings for the SExtractor detection thresholds, e.g. DT=10, DMIN=10 and SIZE=256. Thresholds must be chosen high enough such that no background features are masked. The mask images are called *OFC.mask.fits and are collected under
SCIENCE/MASK_IMAGES
once processing is finished. You should eyeball them and see if sky features were detected. In addition, you can use min-max rejection as outlined above.
The main problem with the one-step approach just outlined is that object masking often does not work in the presence of large background gradients. A good mask can only be created from a flat image. This is particularly important if the field contains extended sources, parts of which are too faint to be visible in the flat-fielded exposures. If this hidden flux is not masked, the background model will be biased, leading to significant over-subtraction. Very often this will only be visible in the coadded image, showing up as dark areas or patches scattered around brighter sources. In such cases an iterative approach is advisable:
Create the static or dynamic superflat as normal. You can safely switch off object masking by leaving the DT, DMIN and SIZE parameter fields empty.
If you chose a static background model (superflat in THELI terminology), switch to the Superflatting section and subtract it. If you chose the dynamic approach (setting the window size to a non-zero value), the background model will be subtracted right away. In both cases rescale the model.
Once THELI has finished, your images will have the status string OFCU written into their file names and should look very close to flat (if no iterative modelling is required you can leave here and jump to the weighting process). The following sub-directories are now in the SCIENCE directory:
SCIENCE/SPLIT_IMAGES
SCIENCE/OFC_IMAGES
SCIENCE/MASK_IMAGES
The first contains the raw data after splitting. The second contains the flat-fielded (and possibly dark-subtracted exposures), and the last will only be created if you did not leave DT and DMIN empty, i.e. you did request object masking.
Now that the images are flat, extended objects can be masked. Simply run the collapse correction at the end of the superflatting section. Choose as low detection thresholds as possible without detecting sky features, e.g. DT=1.0 and DMIN=10. The collapse direction can be arbitrary. Corrected images have the status string OFCUC in their file names, and you have two more additional sub-directories:
SCIENCE/OFCU_IMAGES
SCIENCE/MASK_IMAGES
The latter contains the object masks. Previous masks with the same name (from superflat creation) will be overwitten. You should eyeball a few of them and make sure they do not have any sky features masked. If the latter is the case, increase the DT threshold of the collapse correction.
Click on the blue Redo arrow to the left of the Subtract SUPERFLAT task. This moves *OFCUC.fits to a OFCUC_IMAGES subdirectory, and restores the original, flat-fielded exposures (*OFC.fits) in the SCIENCE directory.
Repeat step 1, but this time set DT and DMIN to their default values to trigger object masking. THELI then recognises the SCIENCE/MASK_IMAGES directory with the object masks. Instead of detecting objects in the flat-fielded data present, it will use the previously created masks. The new mask images are called *OFC.mask.fits, and are used to calculate either the static or dynamic superflat. When the process is finished, these final mask images are kept in:
SCIENCE/MASK_IMAGES
The procedure appears a bit entangled as standard reduction tasks developed for other main purposes are used in the process (in particular the collapse correction). However, you will quickly get the hang of it as it is straight forward. The process should provide you with fairly flat images.
If you are not happy, you can repeat the collapse correction (this time keeping the OFCUC images), and optionally apply the usual individual sky background modelling just before coaddition. If your targets are really extended and/or very faint, then you should seriously consider observing blank fields (OFFTARGET in THELI terminology) for background modelling.