arXiv:0901.1159v1 [astro-ph.GA] 9 Jan 2009
To Appear in ApJS
GASS: The Parkes Galactic All-Sky Survey. I. Survey Description, Goals, and
Initial Data Release
N. M. McClure-Griffiths,1D. J. Pisano,2,3M. R. Calabretta,1H. Alyson Ford,1,4Felix J.
Lockman,2L. Staveley-Smith,5P. M. W. Kalberla,6J. Bailin,7L. Dedes,6S. Janowiecki,2,8B. K.
Gibson,9T. Murphy,10,11H. Nakanishi,12K. Newton-McGee1,10
The Parkes Galactic All-Sky Survey (GASS) is a survey of Galactic atomic hydrogen
(H i) emission in the Southern sky covering declinations δ ≤ 1◦using the Parkes Radio
Telescope. The survey covers 2π steradians with an effective angular resolution of ∼ 16′,
at a velocity resolution of 1.0 km s−1, and with an rms brightness temperature noise
of 57 mK. GASS is the most sensitive, highest angular resolution survey of Galactic
H i emission ever made in the Southern sky. In this paper we outline the survey goals,
describe the observations and data analysis, and present the first-stage data release.
The data product is a single cube at full resolution, not corrected for stray radiation.
Spectra from the survey and other data products are publicly available online.
1Australia Telescope National Facility, CSIRO, Marsfield NSW 2122, Australia; email@example.com,
2National Radio Astronomy Observatory, Green Bank, WV 24944; firstname.lastname@example.org, email@example.com
3present address: Department of Physics, West Virginia University, Morgantown, WV 26506
4Centre for Astrophysics and Supercomputing, Swinburne University of Technology, Hawthorn VIC 3122, Aus-
5School of Physics, University of Western Australia, Crawley WA 6009, Australia; firstname.lastname@example.org
6Argelander-Institut f¨ ur Astronomie, Universit¨ at Bonn, 53121 Bonn, Germany; email@example.com,
7Department of Physics and Astronomy, McMaster University, Hamilton, ON L8S 4M1, Canada; bail-
8Department of Astronomy, Case Western Reserve University, Cleveland, OH 44106; present address: Department
of Astronomy, Indiana University Bloomington, IN 47405; firstname.lastname@example.org
9Centre for Astrophysics, University of Central Lancashire, Preston, PR1 2HE, UK; email@example.com
10School of Physics, The University of Sydney, NSW 2006, Australia; firstname.lastname@example.org
11School of Information Technologies, The University of Sydney, NSW 2006, Australia
12Faculty of Science, Kagoshima University, Kagoshima 890-0068, Japan; email@example.com
– 2 –
Subject headings: surveys — ISM: general — radio lines: ISM — galaxies: interactions
— Galaxy: structure — Magellanic Clouds
Atomic hydrogen (H i) is a ubiquitous component of disk galaxies. Most readily traced by
the λ = 21 cm spectral line, H i from the Milky Way is observed in all directions of the sky.
It is possible to trace Galactic H i emission to the far side of the Milky Way, probing Galac-
tic and interstellar processes in regions of the Galaxy that are inaccessible at many other wave-
lengths. As the dominant component of the interstellar medium (ISM) by number, H i allows us
to trace a wide variety of Galactic processes including the impact of massive stars on the ISM
(e.g., Heiles 1984; McClure-Griffiths et al. 2002); the interaction of the Galactic disk and halo (e.g.,
McClure-Griffiths et al. 2006; Lockman et al. 2008); the ISM life-cycle; and the formation of cold
clouds (e.g., Gibson et al. 2000; Kavars et al. 2005). Since the discovery of the λ = 21 cm spectral
line in 1951 by Ewen & Purcell (1951), it has been used repeatedly, and with continuous refine-
ments, to explore the rotation curve and map the global structure of the Galaxy (e.g., Kerr 1962;
Henderson et al. 1982; Levine et al. 2006; McClure-Griffiths & Dickey 2007).
Not long after the discovery of the H i spectral line, H i emission was found at velocities in
excess of |VLSR| ? 100 km s−1that could not be explained by Galactic rotation (Muller et al. 1963;
Smith 1963). These high velocity clouds (HVCs) are now known to cover a significant fraction
of the sky (Wakker & van Woerden 1997). HVCs are observed at both positive and negative LSR
velocities with magnitudes up to |VLSR| ? 500 km s−1. There have been extensive searches for high
velocity gas (see Wakker & van Woerden 1997 for a review, or recent surveys by Putman et al. 2002
and Lockman et al. 2002b). It now seems certain that HVCs represent a variety of phenomena.
Some HVCs may be related to a Galactic fountain (e.g., Bregman 1980); some are tidal debris,
such as those connected to the Magellanic Stream (e.g., Putman et al. 2003) or other satellites
(e.g., Lockman 2003); some may be infalling intergalactic gas (e.g., Complex C; Wakker et al.
1999; Tripp et al. 2003; Lockman et al. 2008)l; and some may be associated with gas condensing
from a massive, hot halo (Maller & Bullock 2004; Sommer-Larsen 2006). Studies of the structure
and distribution of high velocity gas provide critical information on the evolution of the Milky Way
All-sky surveys of Galactic H i with broad bandwidths and high spectral and angular resolution
allow astronomers to simultaneously explore Milky Way structure, the ISM and HVCs. The most
recent such survey is the Leiden-Argentine-Bonn survey (LAB; Kalberla et al. 2005), which has
produced a sensitive (σ TB = 70 − 90 mK) database of the entire sky with a beam size of ∼
30′sampled on a 30′grid, giving an effective resolution of ∼ 1◦. However, we know from high
resolution studies of targeted areas of the Galactic halo H i emission that there is a wealth of
– 3 –
interesting structure on scales smaller than 30′(Lockman 2002; Stanimirovi´ c et al. 2006; Peek et al.
2007). For example, Green Bank Telescope (GBT) results from Lockman (2002) show that the
halo of the Galaxy contains a significant population of small, cold clouds, possibly products of
the Galactic fountain, that are not detectable with the angular resolution of the LAB survey.
Similarly, high angular and spectral resolution studies have resolved physical and spectral structure
in many HVCs, giving clues to their nature (Br¨ uns et al. 2005; Westmeier et al. 2005). The HIPASS
survey (Barnes et al. 2001, Putman et al. 2002) covered the sky at ∼ 15′resolution, but was not
designed for Galactic H i, and therefore did not accurately measure low velocity gas, nor have the
velocity resolution needed to resolve interstellar H i lines. While high angular resolution Galactic
Plane surveys (Taylor et al. 2003; McClure-Griffiths et al. 2005; Stil et al. 2006) have allowed us to
carefully explore Galactic H i emission in the disk, they are not very sensitive (σ TB∼ 1 K) and
are restricted to a few degrees around the Galactic Plane, leaving most of the volume of the Galaxy
unexplored on scales less than a degree. To fully understand the nature and origin of H i structure
in the halo and of high velocity clouds there is a need for a sensitive, unbiased high resolution
survey of the entire sky.
We have made a new sensitive, fully sampled, high resolution sky survey of Galactic H i
emission south of declination δ = 1◦, with the Parkes 64 m radio telescope: the Galactic All-
Sky Survey (GASS). It was designed with the primary goals of studying the interaction of the
Milky Way disk and halo and the nature of HVCs and intermediate velocity clouds (IVCs; e.g.,
Kuntz & Danly 1996). Secondary science goals include Milky Way structure and the extended
structure of the Magellanic system. Here we describe the GASS project, focusing on the survey
goals and techniques. In §2.1 we describe the main design goals of the survey and the details about
the observations and data reduction techniques are given in §2.2 and §2.3, respectively. In §3 we
present the data products and discuss the limitations of this first stage of our two-stage data release.
Finally, in §4 we briefly discuss some applications of the first release data and describe some initial
2. Observations and Data Reduction
2.1. Survey Design
The parameters of the survey were chosen to match specific criteria relating to the spatial and
velocity extents and velocity widths of known H i emission. Accurate measurement of Galactic H i
at all velocities requires that the data be taken by frequency-switching, as H i emission, especially at
VLSR≈ 0 km s−1, is always very extended. The survey had to cover at least −400 < VLSR< +400
km s−1, which is the range of high-velocity H i in the Southern hemisphere (Putman et al. 2002).
Throughout this work all velocities are given in the kinematic Local Standard of Rest (LSRK),
also commonly referred to simply as LSR, and defined from an average of the velocities of stars in
the Solar neighbourhood (Delhaye 1965; Gordon 1976). The velocity resolution had to be sufficient
– 4 –
to measure the narrow lines (a few km s−1) found in HVC cores and disk H i (Haud & Kalberla
2007; Kalberla & Haud 2006; Lockman et al. 2002; Br¨ uns et al. 2000). The sensitivity goal was set
to match or exceed the LAB survey, but with higher angular resolution. The survey completely
covers the sky with Nyquist sampling so that the data can be used to provide short-spacings for
GASS was designed to meet these criteria by fully sampling Galactic H i in the velocity range
−400 ≤ v ≤ 500 km s−1over the sky south of δ = 1◦with angular resolution of 16′and a
velocity resolution of 1 km s−1. The survey used in-band frequency-switching, described below, to
maximize observing efficiency while preserving sensitivity to extended emission. Integration times
were chosen to achieve a brightness temperature noise (1σ) ≤ 70 mK per channel thus matching
the LAB sensitivity for extended sources while having four times the sensitivity of the LAB survey
for unresolved sources.
GASS observations were made with the 21 cm multibeam receiver system on the Parkes 64
m radio telescope. The 21 cm multibeam has 13 dual linear polarization receivers in the single
cooled dewar. The 13 feed horns are arranged in a hexagonal pattern at the focal plane, consisting
of a central feed, an inner ring of six feeds and an outer ring with an additional six feeds (see
Staveley-Smith et al. 1996). At λ = 21 cm, the average measured beamwidths for the multibeam
are: 14.0′for the central beam, 14.1′for beams two to seven with an ellipticity of 0.03, and 14.5′for
beams eight to 13 with an ellipticity of 0.06 (Staveley-Smith et al. 1996). The mean beamwidth is
therefore 14.3′with a separation between adjacent beam centers of 29.1′. If the receiver is oriented
at an angle of 19.1◦with respect to the scan direction it produces equally spaced tracks with the
inner seven beams.
Observations were conducted between 28 January 2005 and 1 November 2006 in eight observing
sessions of typically two weeks in duration. Observing sessions were organised to cover a contiguous
region of the sky, usually several hours in right ascension and all declinations. All observations were
conducted at night and at elevations greater than 30◦, which is the Parkes elevation limit.
GASS consists of two complete surveys of the sky: one scanned in declination, one scanned in
right ascension. The receiver angle was adjusted every 5 s to ensure that the individual beams of the
multibeam produce approximately equally spaced tracks on the sky parallel to the scan direction.
GASS was designed to be fully sampled with only the inner seven beams to simplify future stray
radiation corrections, but data from all 13 beams are used for the results presented here.
Spectra were obtained using a special purpose correlator mode, combining the Multibeam
correlator and the Wideband correlator to achieve 8 MHz of bandwidth divided into 2048 channels
for both polarizations on all 13 beams. The data were recorded with “in-band” frequency-switching,
in which spectra are recorded at two closely spaced frequencies (IFs) centered at 1418.8435 and
– 5 –
1421.9685 MHz with a 5 s duty cycle. Frequency-switching allows one to effectively remove the
continuum signal and time variations in the system gain. In-band frequency switching, where both
bands contain the H i line, has the added benefit of maximising on-source time because the two
spectra can be combined to yield a total about 4.5 MHz or 950 km s−1of continuous bandwidth.
The channel spacing is ∆v = 0.82 km s−1and the effective channel width is 1.0 km s−1. Doppler
tracking was not applied on-line so the LSRK velocity range accessible varies slightly for each
spectrum. The frequency switch of 3.125 MHz corresponds to 660 km s−1, so every real emission
line feature has an associated, spurious, negative image displaced by ±660 km s−1, but most of
these negative images fall outside the velocity coverage of the survey. We discuss this further in
A scan consists of all the data obtained while driving the telescope through 8 degrees of right
ascension or declination at a rate of 1 deg min−1. It is composed of 26 independent sub-scans
from the 13 beams and two polarizations with data samples every 5 s. With these scanning and
sampling rates the spacing between adjacent samples is 5′, which compares favourably with the
Nyquist sampling, λ/2D = 5.6′for an observing wavelength of λ = 21 cm and the dish diameter of
D = 64 m. The separation between adjacent beam tracks with a 19.1◦receiver rotation angle is 9.4′.
For Nyquist sampling interleaved scans were required. Consecutive scans were offset by 32′so that
adjacent beam tracks were observed with different receivers. An example of the scanning pattern for
three interleaved scans is shown in Figure 1. After three scans the spacing between adjacent tracks
is 3.1′. On-the-fly observing and subsequent gridding broadens the effective telescope beamwidth
to 16′. The integration time per spectrum (pixel) was 30 seconds for the final data product.
2.3. Data Reduction
The majority of the data reduction was carried out with Livedata and Gridzilla software1
specifically designed to process multibeam data. Livedata performs bandpass correction and flux
calibration and Gridzilla produces gridded images from the calibrated, corrected data.
2.3.1. Bandpass Correction: GASS Mode of Livedata
The Livedata processing pipeline was developed and used for the HIPASS survey as described
by Barnes et al. (2001). In its original form, Livedata derived a bandpass solution from the emission-
free portions of a scan. However, neither this nor similar techniques are suitable for GASS because
Galactic H i emission covers the whole sky.
From its origin, Livedata established the use of robust estimation, based mainly on the use of
1Binaries and source code are available from http://www.atnf.csiro.au/computing/software/livedata.html
– 6 –
median statistics, and this was significant in ameliorating the effects of radio frequency interference
(RFI) in the HIPASS survey. For GASS processing, Livedata was adapted to use similar robust
techniques for bandpass calibration of frequency-switched Galactic H i data. It works with the
quotient of each frequency-switched pair, carefully masking any emission, both spectrally and in
the time domain, before determining the average bandpass solution for each scan. The bandpass
is calibrated separately for each beam and polarization.
In this section we describe in detail the procedures used to perform the bandpass correction
for each beam and polarization, including the formation of quotient spectra, masking emission
spectrally and temporally and fitting the average bandpasses. These steps are also graphically
outlined in Figure 2. We first define some variables. Let subscripts i1and i2denote the first and
second integration in pair number i. The spectrum centred at 1418.8435 MHz is Si1(ν), and the
next integration, Si2(ν), is centered at 1421.9685 MHz. The first integration, Si1(ν), is divided by
Si2(ν) to produce a quotient spectrum qi1(ν), and vice versa to produce qi2(ν). In general we use
the variable q to denote quotient spectra, ˆ q to denote time averages of quotient spectra, and Q to
denote estimations of the baseline, which may be medians or polynomial fits. Steps in the first and
second passes, as described below, are distinguished by the use of prime such as, q′and q′′.
Considering the first in each pair of quotients, qi1(ν), a time-averaged value in emission-free
regions for the eight minute scan is determined for each channel. This is ultimately fit to provide
the final bandpass solution, Q′′(ν). Because qi1(ν) and qi2(ν) are reciprocals, the bandpass solution
need only be computed for one or the other. Much of the complexity of the algorithm relates to
the identification of line emission and RFI and the construction of masks that this identification
entails. These masks are determined iteratively in two passes.
The first pass begins with the determination of a base level and characteristic deviation in
emission-free regions. The median value over ν of qi1(ν) is computed for each i in the scan. The
median of these medians then gives the median of quotients, Q1, a single number that provides a
zeroth-order approximation to the base-level of qi1(ν) over frequency and time. Normally Q1will
be close to unity. Now for each i, we compute the median value over ν of |qi1(ν)−Q1|. The median
of these medians is the median quotient deviation, D1. This is used together with Q1to identify line
emission or absorption when selecting data to form a time-average quotient value for each channel.
Thus, we consider each channel of qi1(ν) in turn for all i in the scan; i.e. as a function of time.
We reject qi1(ν) for statistical purposes if |qi1(ν) − Q1| > 3D1. The rejected qi1(ν) form a time
mask for the spectral channel indicating where the scan has passed through a source or encountered
The time mask is then subjected to a broadening algorithm that aims to exclude the low-level
wings of the source profile. This step recognizes that masking by means of a discriminant only
accounts for the central part of an emission line or RFI. The choice of time masking broadening
parameters is a careful balance between defining a mask that is too small so that the diffuse wings
of sources are fit as bandpass and the source is clipped or defining a mask that is too large so that
– 7 –
many masks merge together and the majority of a scan is masked. Many values for the time mask
were tested before settling on parameters such that each patch of consecutive false values (rejected
samples) in the mask is extended by one on each side if it is at least 2 elements wide, and by a
further one on each side for every additional 4 elements. Thus a single isolated false value in the
mask is left alone on the basis that it is probably a noise spike. Isolated patches that consist of
1,2,3,4,5,6,7,8,9,... consecutive false values will broaden to 1,4,5,6,7,10,11,12,13,... samples.
As the patch size increases the limiting value of the broadening factor is 150%. In this process it is
not uncommon for neighbouring masked regions to blend into a single, larger masked region. An
example of the first pass time-masked quotient spectrum is shown in the third panel of Figure 2.
After broadening the time mask, or even beforehand, there will usually be some channels for
which too few quotients remain to compute a meaningful time-average value; typically this occurs
at low H i velocities. If fewer than 90% of quotients for a given channel are rejected, a first
approximation to the bandpass solution for the channel, ˆ q′
of those remaining. If more than 90% are rejected, then this channel is masked for the entire scan
forming a channel mask. Channel masks are then subject to the same broadening process as the time
masks, though with more aggressive broadening parameters: the mask is extended by one channel
on each side for the first false value, and by a further one on each side for every additional 2 false
values. Thus an isolated patch of 1,2,3,4,5,6,... false values becomes 3,4,7,8,11,12,... with a
limiting broadening factor of 200%. Once again, these masking values were chosen as a compromise
between channel masks that are so small that real emission in the line wings is included in the fit
and masks that are so large that a significant fraction of the band is masked and as a consequence
the fit is poorly constrained. Livedata also allows for the channel mask to be augmented manually
by specifying up to ten pairs of channel ranges not subject to mask broadening. A single 82 km s−1
wide channel mask centered near 0 km s−1was applied to each IF.
1(ν), is computed as the median value
Broadening of the channel mask is more exaggerated than for the time mask to avoid removing
spectral line wings. As they are of scientific interest, it is also important that the wings not be
included in, and therefore potentially removed by, the polynomial baseline fit which is applied in
the next step. Channel masking results in gaps in ˆ q′
particular interest. At low velocities the H i line occupies the whole scan; baseline information
in these channels is effectively lost and can only be estimated by interpolation of neighbouring
channels. In practice ˆ q′
1(ν) derived from observed data, and shown in Figure 3, deviates by a few
percent from unity. Consequently we found that a robust polynomial fit of degree 15 was required
to interpolate across the masked channels and fit the baseline accurately. Robust polynomial fitting
for GASS is an iterative process whereby the polynomial is fit to the unmasked channels and points
outside 3× the median absolute deviation from the median are excluded from the second and final
iteration. A second approximation, Q′
1(ν), to the bandpass solution is thus obtained, thereby
completing the first pass.
1(ν) which tend to coincide with channels of
We found that the high order polynomial fit to ˆ q′
regions where the line emission is spectrally broad the fit is poorly constrained under the line. This
1(ν) works well for most of the sky, but for
– 8 –
is mainly an issue towards the Galactic plane and results in errors at the 3-5% level as discussed in
§3.1.2. Lower order polynomials were also tested but we found that they were not able to fit the
substructure of the quotient spectrum.
The quotients and fits for an example scan in the direction of the Magellanic Stream are
shown in Figure 3. This scan is a particularly difficult case because the Galactic and Magellanic
Stream emission are spectrally near to each other with very little spectral baseline between. The
top panel of Figure 3 shows the results of the first pass, where the masked, time-averaged quotient
spectrum, ˆ q′
demonstrates clearly the need for the 15th degree polynomial fit. The polynomial fit in this panel
is, however, not perfect, caused to a large extent by the parts of the spectral line that are not yet
1(ν), is plotted together with its polynomial fit, Q′
1(ν). The plotted quotient spectrum
The lower panel in Figure 3 shows the results of a second pass through the data, which
improves the masking of emission. The second pass essentially repeats the first pass except that Q1
is replaced with Q′
1(ν), which provides a channel-specific value. D1is recomputed accordingly but
the discriminant for time masking is set at 2D1(rather than 3D1) because Q′
estimate of the base-level than Q1. The new time average of the more effectively masked quotient
1(ν), is shown in the lower panel of Figure 3. This spectrum is once again fit with a
robust polynomial of degree 15 to give the final bandpass solution, Q′′
Here we can see that the more extensive masking results in a better fit to the off-line portions of
the spectrum, although deviations remain at the ∼ 0.1% level.
The bandpass corrected spectra are thus calculated from Q′′
1(ν) is a more reliable
1(ν), as shown in Figure 3.
norm(qi2(ν) × Q′′
where norm() indicates normalization to unit value, Ti1and Ti2are the system temperatures for
the two spectra, and T1and T2are the median values of Tsysfor the two frequency-switched pairs
evaluated over all i in the scan.
Occasional, weak, narrow-line RFI appears at a fixed topocentric frequency in much or all of
some scans. Once Doppler-shifted to the LSRK, such RFI produces low-level features that appear to
move across the sky in successive velocity channels. Because of frequency-switching, such features
may be negative as well as positive.
Narrow-line RFI appears in only one or two channels, but may cause ringing in adjacent
channels. The first step in flagging RFI is to compute, for each channel, the time-average value of
i1(ν) for all i in the scan. From this, the running mean computed over 21 channels honouring the
channel mask determined previously, is subtracted. The result is converted to a mask with values
– 9 –
-1, 0, or +1, depending on whether the channel value is outside 6× the median absolute deviation;
the magnitude of the departure from zero is not considered, only the sign.
The mask is then scanned for non-zero channels that should be preserved, the remainder being
flagged as likely RFI. Consecutive non-zero channels of the same sign are preserved if there are
more than 2 of them in the sequence, the width test, or if they are close to such a sequence of the
same sign. Here, “close” means that there is no intervening value of the opposite sign, and there is
no intervening sequence of consecutive zeroes of length greater than 5 channels. This closeness test
is intended to protect outliers in the wings of real emission lines. This algorithm proved effective
in removing the majority of the RFI that appeared in spectral datacubes.
Low-level baseline residuals may remain in each spectrum at this point because the original
bandpass solution was computed as a time-average over the whole scan. This residue was removed
separately for each spectrum by subtracting a 10-th order polynomial fit from the spectrum for
high Galactic latitude areas (b > 10◦) and a simple median level fit from areas within 10 degrees
of the Galactic plane. The algorithm was as described previously, using a mask derived separately
for each spectrum, including the user-defined mask as before. We discuss baseline quality in § 3.1.
Finally, the spectra were Doppler-shifted to transform them to the LSRK velocity frame for
gridding and analysis. The scheme used for Doppler correction was to rescale the reference frequency
and channel spacing by the Doppler factor, and then Fourier-shift the spectrum, usually by less than
one channel, so that the reference frequency was an integer factor of the original channel spacing
(3.90625 kHz). This scheme is employed by Gridzilla (see below) and allows the combination of
spectra taken months or years apart without requiring interpolation of the frequency axis.
2.3.4. Brightness Calibration
First order brightness calibration was applied on-line through injection of noise from a diode
switched with a frequency of 500 Hz. Average on-line system temperatures for each beam and
polarization were recorded along with each 5 s spectrum. Typical system temperature values are
between 21 - 23 K.
Before imaging, spectra are converted to beam averaged brightness temperature, TB, from ob-
servations of the IAU standard line calibration regions S6, S8, and S9 (Williams 1973; Kalberla et al.
1982). One of the three standard line regions was observed each day. Observations were conducted
by placing each of the thirteen beams on the region in turn. The peak of the observed line was
used to calculate calibration scaling factors for each beam and each polarization assuming peak
brightness temperatures of TB= 83 K for S9, TB= 53 K for S6 and TB= 76 K for S8 valid for the
– 10 –
Parkes beam (Kalberla et al. 1982; Br¨ uns et al. 2005). One set of calibration factors was calculated
for each observing session and showed rms variations between observing sessions of 1−2% on most
beams. The notable exceptions are the second polarization on beams 10 and 12, which were known
to have unstable low noise amplifiers. The calibration factors for these beams varied by as much
as 6% over the 21 months of the survey. The overall effect on the data was found to be minimal
and these beams were included in the final data cubes. These brightness temperature calibration
factors were applied within Livedata following bandpass calibration.
To check that our calibration factors were not affected by the observing strategy of pointed
calibrator observations rather than the on-the-fly scans that were used for the full survey, we
observed S9 in on-the-fly mode as well. Calibration factors determined from these scans were fully
consistent with the factors determined from pointed observations.
Imaging was performed by Gridzilla, a statistical gridder developed for the HIPASS survey and
subsequently extended for more general use with Parkes multibeam and other single-dish data. The
algorithm is described by Barnes et al. (2001). Part of Gridzilla’s later development involved adding
full support for FITS celestial and spectral world coordinate systems (Greisen & Calabretta 2002;
Calabretta & Greisen 2002; Greisen et al. 2006). In particular, we used the Zenithal Equal-Area
projection (ZEA) as the most appropriate choice for mapping the hemisphere.
For each pixel in the output data cube, Gridzilla computes a weight for each input spectrum
based on its angular distance from the pixel. For GASS, it then calculated the pixel value from
the spectral values and weights using weighted median estimation, which is robust against RFI and
other artifacts. The weighted median of a set of measurements is the middle-weight value - the
sum-of-weights of all measurements less than it being equal to that of all measurements greater;
pro rata interpolation being used to bisect the sum-of-weights if required.
For the weighting function we used natural beam weighting, with the beam modelled by a
2D Gaussian of FWHM 14′.4, combined with an additional 2D Gaussian of FWHM 14′, with a
cutoff radius of 8′. This combination of weighting functions degrades the angular resolution of
the images slightly below the telescope FWHM of 14′.4 but produces smoother and more sensitive
images. Measurements of unresolved sources inserted into the data prior to gridding show that the
resulting resolution is ∼ 16′.
3. Data Products
The primary data product for this first release is a data cube (α, δ, v) of the entire survey
region and full velocity range without stray radiation correction. The data were gridded in a ZEA
– 11 –
projection covering δ < 1◦with the South Celestial Pole at the center of the image. The velocity
range covered by the gridded cube is −400 km s−1≤ VLSR≤ 500 km s−1.
A comprehensive view of GASS is given in Figure 4, which is a combination of moment maps
created in ∼ 40 km s−1intervals over the full velocity range, color-coded by velocity. Most of
the key features of the Southern sky are clearly visible in this image including the Galactic plane,
Magellanic Clouds, Magellanic Stream and Leading Arm, several high velocity cloud complexes,
and a few galaxies belonging to the Sculptor group.
An image of the total column density, with an Aitoff projection in Galactic coordinates, is
shown in Figure 5. The image has been calculated from the zeroth moment over the velocity range
−400 ≤ VLSR≤ 500 km s−1and converted to units of ×1021cm−2using the usual optically thin
scaling factor 1.8×1018cm−2K−1km−1s. This image is dominated by emission at local velocities.
We show three individual velocity channels of low velocity (|VLSR| ? 30 km s−1) gas in Figures
6 – 8 that highlight the small-scale structure visible in the GASS data. Similarly, Figures 9 and
10 are images of the high and intermediate velocity sky, where high velocity gas is defined as
|VLSR| ≥ 100 km s−1and intermediate velocity gas is defined as 40 ≤ |VLSR| < 100 km s−1. The
negative high velocities are dominated by Galactic emission near to the Galactic center and a
portion of the Magellanic Stream, whereas the positive velocities are dominated by Galactic plane
emission and the Large and Small Magellanic Clouds. The intermediate velocity sky is dominated
by Galactic emission, however there are some interesting filamentary extensions off the Galactic
plane. Small HVCs and IVCs are visible in both figures.
A summary of the survey specifications is given in Table 1. Spectra from the survey and infor-
mationabout other dataproducts available
for downloadcanbe foundat
3.1. Data Quality
The typical rms per channel in the cube is ∼ 57 mK. This is determined from an image of the
rms noise across the full survey region, as calculated from the median of the noise in nine blocks of
∼ 20 line-free channels. The rms image is shown in Figure 11. This approach effectively removes
contamination of the rms measurement from negative residuals (see § 3.1.3) from the Magellanic
system and Galactic Center region in the otherwise line-free channels.
smooth across the majority of the sky, with a mode of 57 mK. As seen in Fig. 11, the rms around
the Galactic plane region is clearly higher than the rest of the sky. This is due to both strong
continuum sources and the large fraction of the band filled by H i emission in the Galactic plane.
Both effects raise the system temperature. Similar increases in rms occur in the region of the LMC.
The rms is lower in the overlap region between scans and also where scans were repeated, such as
the large area near 1 h of right ascension. Assuming a rms brightness temperature of 57 mK, the
1σ column density noise in a single velocity channel is 1 × 1017cm−2. For a typical HVC of width
The noise is relatively
– 12 –
30 km s−1, the 3σ sensitivity limit is NHI= 1.6 × 1018cm−2.
There are three extended areas where portions of the spectra in the range velocities ∼ 230−320
km s−1have increased noise by a factor of
scans that were flagged because they exhibited nearly continuous, broadband RFI of unknown
origin. The flagged scans lie in the regions: 11h15m? α ? 12h15m, −25◦? δ ? 0◦; 11h? α ? 13h,
−65◦? δ ? −45◦; and 13h45m? α ? 15h15m, −78◦? δ ? −70◦. These areas are marginally
visible in Figure 11 as regions of increased noise.
√2. These regions result from two days’ worth of RA
3.1.1. Stray Radiation
A fundamental limitation of GASS data in the present release is stray radiation: H i emission
that enters the receiver from the sidelobes of the telescope rather than through the main beam.
Kalberla et al. (1980) has shown that stray radiation can make a significant contribution (15 – 50%
of the profile area) to observed Galactic H i emission profiles, mostly at high Galactic latitudes where
emission in the primary beam is weak. The LAB survey has been corrected for stray radiation,
and the second GASS data release will be corrected as well, but at this stage we have simply
estimated the total amount stray radiation in GASS for various regions of the sky. This was done
by convolving the GASS and LAB surveys to 2◦angular resolution, computing the total column
density over their common velocity range (−400 km s−1≤ VLSR≤ 400 km s−1), and interpreting
the difference as stray radiation in GASS. Figure 12 shows the convolved GASS column density,
the estimated stray column, and the stray fraction. For most lines of sight the fraction of the total
column density believed to arise from stray radiation is between 5 and 15%, with some particularly
low column density regions showing fractions as high as 35%.
To demonstrate the spectral behaviour of the stray radiation component we compared LAB
and GASS spectra towards several regions representative of high stray radiation fraction, low stray
fraction, and average stray fraction, as calculated above. The regions are marked on Figure 13
and the spectra are shown in Figure 14. These spectra are averaged over 5 × 5 degrees in both
surveys to minimize any resolution-dependent effects. For an area with average stray fraction it is
clear that the GASS spectrum reproduces the stray radiation corrected LAB spectrum very well,
with small departures in the line wings where the total intensity is small. The high stray fraction
spectrum was extracted towards a region where the fraction of the total column density attributed
to stray radiation was on the order of 30%. In this area the GASS spectrum has a very low peak
brightness temperature of ∼ 1.9 K, and yet the LAB spectrum peak is even lower at ∼ 1.4 K. The
GASS spectrum also shows significant effects of stray radiation in the line wings.
– 13 –
3.1.2. Baseline Quality
Although GASS spectral baselines are generally very good, there are residual ripples in the
final baselines with maximum peak-to-peak variations of ∼ 50 mK. It was the presence of much
larger (∼ 100 − 200 mK) bandpass amplitude ripples and striping in test images that necessitated
the use of high-order polynomials in the bandpass correction algorithm as described in §2. These
were effectively removed by the post-bandpass polynomial fits applied to scans away from the
Galactic Plane. To ensure that the high-order polynomials had not subtracted real emission we
compared spectra obtained from GASS to spectra obtained from the LAB survey. These showed
generally very good agreement at the ∼ 50−70 mK level. Several examples are shown in Figure 15,
demonstrating the baseline quality of GASS.
Baseline quality deteriorates in some regions toward the Galactic Plane and the Magellanic
Clouds. The bandpass correction method described in §2.3.1 is limited in cases where the H i
emission in the quotient spectrum fills ∼ 50% of the band. In these cases the polynomial fit
to the quotient spectrum is poorly constrained and therefore the bandpass under the line is not
well determined. This results in some small, ∼ 3 − 5%, baseline errors where the measured line
temperatures can appear either too large or too small relative to spectra from LAB. Examples are
shown in Figure 16, where the GASS spectra show higher and lower peak values than LAB.
3.1.3. Other Artifacts
As discussed in § 2.2, the in-band frequency-switching will produce a negative copy of a spec-
tral line displaced by ±660 km s−1. As a practical matter, most of these artifacts lie outside the
final data cube, but there are some notable exceptions for features at VLSR> 260 km s−1(with
the inverted feature appearing at negative velocities) and VLSR< −160 km s−1(at positive veloci-
ties). This is particularly evident towards the Galactic Center, the Magellanic Clouds (apparent at
extreme negative velocities) and the Northern tip of the Magellanic Stream (apparent at extreme
positive velocities) as can be seen in the bottom panel of Figure 16.
Another artifact appears as a low amplitude grid-like scanning pattern in channel images with
low-level emission. This is due to residual ripples in the bandpass of individual scans, which can
cause slight offsets between adjacent scans. These are typically within the noise, but because they
are spatially correlated they can be seen in the rms image in Figure 11.
The Parkes Galactic All-Sky Survey (GASS) is a high spectral and angular resolution H i
line survey of the sky south of δ = 1◦. The first-stage data release includes a full cube at 16′
angular resolution, 1.0 km s−1spectral resolution and ∼ 57 mK rms noise over the velocity range
– 14 –
−400 km s−1≤ VLSR≤ 500 km s−1. Spectra from this cube and several other data products are
now publicly available at http://www.atnf.csiro.au/research/GASS. Table 2 gives a compar-
ison of GASS with other large-scale surveys of Galactic H i. At southern declinations GASS is
unsurpassed in sensitivity and, outside of the Galactic plane, in angular and velocity resolution.
The improvement of GASS over the LAB survey is illustrated by the spectra in Figure 15, shown
at the full angular resolution of both surveys.
The current release of GASS is intended to provide data that are ideal for study of high
velocity (|VLSR| ? 100 km s−1) H i, where the effects of stray radiation are negligible, and small-
scale features, where the angular resolution is a significant advance. A complete catalog of HVCs
and IVCs from GASS is in preparation by D. J. Pisano et al. (2009, in preparation). Because the
current data release has not been corrected for stray radiation, care should be taken when using
it to derive global Milky Way properties, especially at low column densities where there can be
significant amounts of stray radiation. For users interested in total column densities measured
to extragalactic sightlines these measurements should be considered in conjunction with the total
fraction of the column due to stray radiation (Figure 12).
GASS data have already been used for several of the scientific areas described in §1: Using
data from the first year of observations McClure-Griffiths et al. (2006) found that one of the largest
Galactic supershells, GSH 242–03+37, is in fact a chimney, with evidence for breakout on both sides
of the Galactic plane. GSH 242–03+37 appears capped by thin, clumpy filaments of H i emission
at heights of z ∼ 1.5 kpc above the Galactic midplane. Those authors suggested that the clumpy
filaments may be the precursor of halo clouds detected by Lockman (2002).
Ford et al. (2008) used GASS data to extend our knowledge of Galactic halo clouds by con-
structing a catalogue of hundreds of clouds in a 720 deg2GASS pilot region centred on l = 335◦,
b = 0◦. Though restricted to a small range of Galactic longitudes, they found that the distribution
of clouds is significantly peaked at a Galactocentric radius of 3.75 kpc and that the clouds are asso-
ciated with loops and filaments consistent with a chimney origin as suggested for GSH 242–03+37.
The forthcoming complete catalog by H. A. Ford et al. (2009, in preparation) will further explore
the Galactic distribution of these clouds.
Finally, McClure-Griffiths et al. (2008) have used GASS data to study the interaction of an
HVC in the Magellanic Leading Arm with the Galactic disk. They showed that the Leading Arm
crosses the Milky Way disk at a Galactocentric radius of 17 kpc, which is close to the interaction
region predicted in older models (e.g., Connors et al. 2006; Yoshizawa & Noguchi 2003) of the
Magellanic System but somewhat surprising given the revised proper motions for the Large and
Small Magellanic Clouds (van der Marel et al. 2002; Kallivayalil et al. 2006).
As a new, sensitive survey of Galactic H i, GASS is already producing excellent results and
will be a valuable database. Work is underway to produce a stray-radiation corrected version of
GASS for a second data release.
– 15 – Download full-text
We acknowledge the great dedication of the ATNF staff at Parkes and Marsfield towards making
the special observing mode for this project available and supporting our observations. We especially
thank Malte Marquarding for assisting with various software challenges during the data analysis.
Thanks also to Bill Saxton for his assistance in creating Figure 2. D.J.P. acknowledges partial
support for this project from NSF grant AST0104439 and thanks the ATNF for its generosity and
hospitality through the Distinguished Visitor program. S.J. thanks the NSF for support through
the Research Experiences for Undergraduates program at NRAO. T.M. acknowledges the support
of an ARC Australian Postdoctoral Fellowship (DP0665973). P.K. and L.D. acknowledge support
from Deutsche Forschungsgemeinschaft, grant KA1265/5-1. The Parkes Radio Telescope is part
of the Australia Telescope which is funded by the Commonwealth of Australia for operation as a
National Facility managed by CSIRO.
Facilities: Parkes ()
Barnes, D. G. et al. 2001, MNRAS, 322, 486
Blitz, L., Spergel, D. N., Teuben, P. J., Hartmann, D., & Burton, W. B. 1999, ApJ, 514, 818
Braun, R. & Burton, W. B. 1999, A&A, 341, 437
Bregman, J. N. 1980, ApJ, 236, 577
Br¨ uZs, C., Kerp, J., Kalberla, P. M. W., & Mebold, U. 2000, A&A, 357, 120
Br¨ uns, C., Kerp, J., Staveley-Smith, L., Mebold, U., Putman, M. E., Haynes, R. F., Kalberla, P.
M. W., Muller, E., & Filipovic, M. D. 2005, A&A, 432, 45
Calabretta, M. R. & Greisen, E. W. 2002, A&A, 395, 1077
Connors, T. W., Kawata, D., & Gibson, B. K. 2006, MNRAS, 371, 108
de Heij, V., Braun, R., & Burton, W. B. 2002, A&A, 392, 417
Delhaye, J. 1965, Galactic structure. Edited by Adriaan Blaauw and Maarten Schmidt. Published
by the University of Chicago Press, 61
Ewen, H. I. & Purcell, E. M. 1951, Nature, 168, 356
Ford, H. A., McClure-Griffiths, N. M., Lockman, F. J., Bailin, J., Calabretta, M. R., Kalberla, P.
M. W., Murphy, T., & Pisano, D. J. 2008, ApJ, in press, (astro-ph/0807.3550)
Gibson, S. J., Taylor, A. R., Higgs, L. A., & Dewdney, P. E. 2000, ApJ, 540, 851
Gordon, M. A. 1976, Methods of Experimental Physics, 12, 277