Obtaining ultra-deep imaging of the sky is plagued with difficulties. For this reason going beyond the 30 mag arcsec−2 frontier (approximately 1500 times fainter than the darkest sky on Earth) has remained rather elusive. In this Section, we review the most important challenges that need to be addressed carefully if one desires to obtain ultra-deep imaging.
Professional astronomical observatories are located on the darkest spots on Earth. Even at these locations (where light pollution caused by human activity is minimal) the night sky brightness is substantial: µV ∼ 22 mag arcsec−2. This brightness is mainly due to various processes in the upper atmosphere, such as the recombination of atoms which were photoionized by the Sun during the day, a phenomenon known as airglow. With long enough integrations we are able to reduce the noise in this brightness which results from the intrinsic variability of the night sky, and easily obtain images in which we can measure features which are much fainter than the intrinsic sky brightness. However, the sky brightness also contains other components. Beyond our atmosphere, a diffuse light component is caused by the reflection of sunlight on the dust plane of our Solar system. This is the zodiacal light, with a brightness of around µV ∼ 23.5 mag arcsec−2. This brightness affects particularly those regions of the sky around the ecliptic plane and it contaminates all observations, including those obtained with space telescopes. The intensity of the zodiacal light is variable and depends on the Solar activity.
Internal reflections are due to the structure of the telescope and the dome. These reflections can appear at different surface brightness levels but are quite common when one reaches levels of µV ≳ 26 mag arcsec−2. There are two approaches to minimize these reflections. One way is to use telescopes with simple optics (Abraham and van Dokkum 2014; see also Abraham et al, this volume), another is to implement clever observing strategies which avoid repetition of similar orientation of the camera on the sky (e.g., Trujillo and Fliri 2016).
To obtain reliable ultra-deep imaging, an exquisite flat field correction of the images needs to be performed. This correction is needed to ensure that a uniform illumination of the CCD leads to a uniform output, or uniform counts in the image. In a CCD camera, this is not the case by default as the gain and dark current change across the face of the detector, and distortions due to optics can cause non-uniformities. The process of flat fielding removes all these pixel-to-pixel variations in sensitivity and the effects of distortions in the optical path. Flat field images are often obtained by exposing on uniformly illuminated surfaces.
Any artefact (gradient, pattern, etc.) left behind during the process of creating the flat field image used to correct the raw science images will introduce a systematic error in the final image, which in turn will prevent reaching the expected surface brightness limit of the observation. The key to creating a good flat field image is to have a uniform illumination of the CCD of the camera. For most observing cases, a twilight (or even a dome) flat is good enough for this purpose. However, when the goal is to reach very faint details of the image a different approach is needed, consisting in creating a flat field using the night sky imaging itself. Such a flat field, using a set of science images, is sometimes referred to as a master flat.
To create a good master flat, the set of science images must be taken at different locations on the sky. In fact, depending on the apparent size of the galaxy, the displacement between one science image and the next should be at least as large as the size of the object. Ideally, not all the science images of a specific galaxy should be located at the same position on the CCD (and, ideally, should also not have the same position angle on the sky). That means preparing an observing scheme that includes both a dithering and a rotation pattern (see the example of this procedure in Trujillo and Fliri 2016). In addition, if the observations are taken over different nights (or observing blocks), the best approach is to create a master flat for each night (or observing block). The use of all the science images in a run rather than those of a single night or observing block is not generally a good idea as slight differences from night to night in the focus and the vignetting correction hinder such an approach.
Once all the science images have been acquired, the process of building a master flat is as follows. First, all the objects in each individual science image are generously masked (for instance using SExtractor; Bertin and Arnouts 1996). Only the pixels outside the masked areas are used to create the final master flat. Second, to guarantee that all the science images are appropriately weighted during their combination to create the master flat, every individual science image of a given night is normalized. Finally, the normalized and masked individual science images are median-combined into a single master flat.
An alternative approach which has been used to achieve high-quality flat-fielding is referred to as drift scanning, or a variant on this called time delay and integration (TDI; McGraw et al 1980, Wright and Mackay 1981). In drift scanning, the reading of the CCD is done at the same slow rate as the CCD is moved across the sky. In TDI, the CCD does not move, but the readout is timed to coincide with the sidereal rate at which the sky passes by. In both cases, an object is sampled by every pixel in a column, thus averaging out all defects and achieving an extremely efficient flat fielding. As in the case of the masterflat described above, the background itself is used for flat fielding, assuring a perfect colour match. Further details, as well as a more complete historical overview, are given by Howell (2006).
In practice, in spite of these significant advantages, drift scanning or TDI have not been used much in the literature. The reasons for this vary, but include the difficulty of adjusting the relative movement of sky and CCD with the readout, the loss of efficient observing during the ramp-up and ramp-down phases at the start and end of an exposure, image elongation effects, and the fact that the exposure time is fixed by the telescope+CCD setup, and often rather short.
In the field of imaging nearby galaxies, the most notably exception to this general dearth of TDI results is the Sloan Digital Sky Survey (SDSS, York et al 2000). As described by Gunn et al (1998) 1, the SDSS large-format mosaic CCD camera has been designed to image strips of the sky simultaneously in five colour bands using the TDI technique. The approach chosen by the SDSS team has proven to be very successful in terms of imaging to low surface brightness levels. Even though the exposure time of SDSS images is less than one minute (53.9 s) and the telescope of modest size (2.5 m), the exquisite flat fielding and sky background allow one to reach very low surface brightness levels indeed, down to 26.5 mag arcsec−2 (3σ in an area of 10 × 10 arcsec; Trujillo and Fliri 2016) or down to 27.5 mag arcsec−2 when analyzing elliptically averaged surface brightness profiles (see, e.g., Pohlen and Trujillo 2006). As illustrated in Sect. 3, co-adding series of SDSS images, as is possible in the Stripe 82 survey area, brings that level down by another 2 mag arcsec−2, allowing ground-breaking science to be performed.
2.4. Masking and Background Subtraction
Deep images often reveal coherent structure at low levels, which can be
due to, e.g., imperfect flat fielding, spatial variations in the
background sky, or residual point spread function (PSF) effects. Deep
images of galaxies also show foreground stars, as well as background
galaxies. All these components must be identified, taken into account,
and/or subtracted before a deep galaxy image can be analysed. As an
example of the procedures followed, we show in
Fig. 1 a SDSS Stripe 82 image of NGC 941 as analysed by
Peters et al
(2017).
Stars and background galaxies are generously masked, after which a
polynomial two-dimensional fit is made to the remaining background
pixels. In our example, this is dominated by a left-right gradient in
the background level, but the fit also shows smaller-scale
structure. The latter could, in principle, be real structure, either
related to the galaxy (e.g., tidal streams) or not (e.g., Galactic
cirrus, see Sect. 2.6). If one looks for structure
like tidal streams a different background modelling technique must be
used, for instance only modelling large-scale fluctuations or
gradients. This neatly highlights the difficult nature of this kind of
analysis.
Figure 1. Example of residual background
modelling and subtraction, for a Stripe 82 image of the galaxy
NGC 941. Top left panel: original image,
showing a gradient in
the background level. White areas are masked-out images of foreground
stars and background galaxies. Image labels are in pixels, with size
0.396 arcsec. North is right, East to the bottom. Top right:
model for the background, excluding the area of the galaxy. Lower
left: background model, extrapolated over the area of the
galaxy. Lower right: final image, with the background model
subtracted from the original. Reproduced with permission from
Peters et al
(2017).
In our example, the gradient in the background in all probability is
also present at the location of the galaxy, which is why we extrapolate
the background model into the galaxy region (lower left panel of
Fig. 1). This model is then subtracted from the
original image, and the result can be used for scientific analysis, in
this case studying the shape of the outer regions through analysis of
azimuthally averaged radial profiles (see
Sect. 4.2; this is why the small-scale
background structure described in the previous paragraph could safely be
subtracted off in this case). The uncertainty limit, down to which these
radial profiles can be trusted, is just below 30 mag
arcsec−2 for the image shown in
Fig. 1
(data taken from the IAC Stripe82 Legacy Project;
Fliri and Trujillo
2016).
A further and important source of background contamination is produced
by all the emitting sources in the image (or even just outside the
imaged area). The light of these sources is scattered by the PSF of the
instrument across the entire image.
Slater et al
(2009)
have shown that at µV ∼ 29.5 mag
arcsec−2, an image taken from the ground has all its
pixels affected by scattered light from nearby bright sources. Properly
removing this contamination is quite challenging, and requires an
extremely accurate characterization (< 1% at large
distances) of the PSF of the camera.
The excess light redistributed by the PSF from both the object of
interest and the nearby surrounding sources creates a two-dimensional
and highly structured surface that is the main contributor to the
background of the image at the faintest surface brightness levels
(µV ≳ 29 mag arcsec−2,
Slater et al
2009).
All astronomical images are affected by this scattered light background
which is the result of the convolution of the PSF with the light of the
sources. The scattered light background of an astronomical image will be
more intense if the number of bright sources in the image is large and
also if the PSF has significant wings. In this sense, the best option
(if feasible) is to select a target within a field devoid of nearby
bright surrounding objects. A typical scattered light background is
illustrated in Fig. 2.
Figure 2. The scattered light around the
galaxy UGC 00180 produced by all the stars brighter than
R = 17 mag in its vicinity. Top row: original field
(left) and a zoom-in of the galaxy. Middle row: scattered
light. The position of the galaxy is illustrated with a green
cross. Contours of surface brightness are 28.5, 29, and 29.5 mag
arcsec−2 (left) and 28.8, 29, and 29.2 mag
arcsec−2 (right). Lower row: original
field after subtraction of the scattered light produced by the
brightest sources. Figure taken from
Trujillo and
Fliri (2016),
reproduced with permission of the AAS.
Figure 2 illustrates how to deal with the
background of scattered light. First, the PSF must be characterized as
perfectly as possible by using a reliable extended PSF. An extended PSF
with high a signal-to-noise ratio in its outer wings will allow the
exploration of the distribution of the light of the nearby brightest
sources up to the position of the galaxy under exploration. Second,
using the extended PSF a background scattered light map is created. If
the main contributor to the scattered light background is the presence
of bright stars then the scattered light map is created by simply
locating the model PSF at the position of the bright stars and scaling
the flux of the PSF to these bright sources. Finally, the scattered
light map is subtracted from the observed image.
Above, we have discussed the effect of the scattered light created by
the surrounding sources around the galaxy of interest. Naturally, the
light distribution of the object itself is also convolved with the
PSF. In this sense, the scattered light of the targeted galaxy is also
creating an artificial excess of light in the outermost region of the
galaxy that needs to be addressed in order to explore the properties of
the object in its outer regions. We illustrate how to deal with this in
Sect. 4.4.1 of this Chapter.
Finally, if one is interested in exploring the fainter structures of
galaxies, the presence of Galactic cirrus in the images
needs to be considered. These filamentary structures are located
everywhere on the sky, even at higher Galactic latitudes. For this
reason, a careful preselection of the fields to be observed is necessary
to minimize the contamination by cirrus. A good illustration of the
perils of Galactic cirrus is presented by
Davies et al
(2010)
for the case of the M81 group. They found that far-infrared emission
measured by the Herschel satellite correlates very well spatially
with narrow-velocity Galactic
Hi, without any evidence
that this far-infrared emission originates in the M81 group. They thus
inferred that the optical streams and structures seen in the M81 group
are not in fact part of the group, but are rather due to light from our
own Galaxy which is back-scattered off Galactic dust.
1 For a more
detailed description of the flat field procedures used by the SDSS, see
http://classic.sdss.org/dr5/algorithms/flatfield.html.
Back.