HRS Data Reduction Tips
The FITS Extensions
The HRS CCD is actually 2 CCDs saved within a single fits image with
fits image multi-extensions. The zeroth extension, test.fits
is the header that contains most of the telescope and instrument information.
The first extension, test.fits, contains a short header with CCD
specific information and the red HRS (MM1) CCD.
The second extension, test.fits, contains a short header with CCD
specific information and the blue HRS (MM1) CCD. IRAF has a package
for dealing with fits extensions: mscred. Most useful of this
package is the task mscsplit, which allows the extended image
to be split into 3 separate iraf images with 3 separate headers. This allows
each CCD to be reduced separately. Keep in mind that the telescope and
instrument information will not be in the headers of each data frame.
There is now a script, hsplit.cl which will split the sections and translate the header
information from the  header into the  and  headers. This task
can also correct some errors in the CCDSEC which will allow CCDPROC to
work on HRS files. Please note that after a recent ICE upgrade there may
be a small change required to the hsplit.cl code (look at the code
for the documentation).
Bias frames are very useful for the removal of roughly half of the bad
columns found on the red HRS CCD. We suggest that the 5 taken on
your night be compared with a Master bias that you could create from all of
your bias frames to see if there is any change from night to night.
BIAS should be subtracted from the red HRS CCD.
The blue HRS CCD is very clean and thus bias subtraction is of little
assistance in the data reduction process and can actually add noise if
an unclean (non master) BIAS is used.
Dark frames are very useful for the removal of a quarter of the remaining
bad columns found on the red HRS CCD. However, the overhead involved
in creating a dark frame for each binning and exposure length is
too much of a burden for the current calibration plan. We suggest using
fixpix to remove these bad columns.
Using the iraf task fixpix or other such tasks can aid in the
reduction process particularly in the tracing of apertures on the red
HRS CCD. However, this should be done with care since it is cooking
your data (replacing bad columns with dubious good data).
Scattered Light Removal
In standard long slit or even slit spectroscopy some people
do not find it necessary
to remove scattered light because the scattered light will be removed
with the sky subtraction. THIS IS NOT ALWAYS TRUE FOR FIBER
SPECTROSCOPY AND WHEN FRINGING IS PRESENT IS NOT TRUE FOR LONG SLIT
SPECTROSCOPY. Because of the potentially different fiber throughput
differences there may be a scaling factor in the removal of the sky light
which will inappropriately scale the scattered light. In addition,
if the scattered light is not monochromatic it will not produce fringing.
Thus scattered light
removal should be done for all HRS spectra and done BEFORE flat field
correction. In the IRAF echelle package
the task for the removal of scattered light is apscatter. This
task will fit a 2d surface to the scattered light. It first fits in the
cross-dispersion direction leaving out any defined apertures. It then
fits the surface to the dispersion direction. Earlier versions of IRAF
sometimes fail to subtract the scattered light correctly unless the task
is run interactively. I suggest testing your version in both the
interactive and non-interactive modes.
NOTE: Flat calibrations taken before Feb 6 2005
were contaminated by emission lines in the lamp
itself. In the flat fields Li and Na emission lines are obvious. These
should be removed using fixpix or a similar task. I have found that the
first flat is often not contaminated and can be used exclusively for
Creating flats fields with the HRS is difficult and confusing. The
first thing that you should realize is that these flats are internal
to the HRS (a separate calibration fiber) and not taken through the
science fibers. The flat will cover the same aperture as the science
and sky fibers. There are several methods for creating a flat field
and we will cover three:
apnormalize: This task requires that the user set apertures and
trace the orders to the flat field. It can remove the blaze function
but does not normalize in the cross-dispersion direction. Because the
calibration fiber does not fill the slit in the same way as the science fibers,
the flat will often have an inappropriate shape in the cross-dispersion
direction in comparison to the science fibers. This has the effect of
weighting you science data in an inappropriate way. We do not
recommend using apnormalize.
flat1d: This task fits a spline to every line or column (column in our
case). If you set the order of the spline to a very high value, say 91, then
you will create a reasonable flat. This task has the advantage of being
able to flatten near the edge of the CCD where only partial orders are
seen. This flat will have some strange
ringing at the edges of the flat aperture but is preferable to apnormalize.
The task also has the disadvantage of smoothing out correlated pixel
variations such as fringing at wavelengths above 670nm.
apflatten: This task requires that the user set apertures and
trace the orders to the flat field. It can remove the blaze function
and normalizes in the cross-dispersion direction. When the parameters are
optimized this task can produce superb flats.
For all of the above it is important to pay attention to the flux levels
and the S/N that will be derived. Because the pixel to pixel variation
in the blue is small, a few tenths of a percent, and the throughput of
the flat field lamp in the blue is low it is quite possible to actually
harm your data by flatfielding with a noise flat. I suggest that the
threshold parameter in the all of the above tasks be set such only data
above 10,000 electrons in the summed flat be included. This threshold
can be lowered in the red where the fringing can be worse than a 1% effect.
Finding Orders or Apertures
Because the HRS uses fibers the apertures are quite flat topped and thus
marking the orders in the apfind ( an IRAF echelle package) task
is difficult. We find better success by setting the nsum
parameter to a much higher value, such as 30 to 100.
Tracing Orders or Apertures
Tracing the sky fibers is difficult because of the (hopefully) low flux
levels. If you don't have a sky fiber and have a high S/N spectrum
( S/N > 50) then you probably won't have any problems tracing your orders
particularly if you do a bias subtraction and either subtract a dark
frame or fixpix the remaining bad columns. If you do have a sky
frame then you have two options for tracing the sky fibers:
trace your science fiber and offset the apertures (using the
s keystroke in the apedit task. You should do this in
the ALL mode ( a keystroke) and look for the sky lines. You can
interactively change the row that the apedit uses with a
use the sky flat or dome flat that should have been provided in
the calibrations for your setup and then use those apertures as a
In traditional slit spectroscopy the background region would be set
for the removal of the scattered light and sky light in the extraction
process. Because of the potentially different fiber throughput
differences there may be a scaling factor for the removal of the sky light.
Thus for best affect no background should be subtracted during the
extraction and the sky subtraction should be done later. However, for
quick and dirty first order extraction the background can be used to
remove most of the sky light. Should you subtract sky is a separate
question and will be addressed in the sky subtraction section later.
To set the background region interactively for a single aperture in
the apedit task with the keystroke b. The default parameters
for the background can be set in apdefault. In addition to
redefining the sample region, I suggest
setting a few of the following parameters: b_naver=-1,
b_niter=3, and b_high_=2.5. When you run the extraction,
either apall or apsum, you should set the background
parameter to either fit or median.
Extraction in IRAF is done either in the apall or apsum
packages. We recommend that sky subtraction be done separately so that
proper accounting for the throughput differences in the fibers can be
made, the background should be set to none. There are a number
of bells and whistles that can be set to optimize the extraction, such
as cleaning and weighting. If you chose to use these then the extras
options should also be turned on. This will allow you to look at
the unweighted, non-cleaned data in a separate band of the extracted spectra.
You can help the cleaning of your data if you set the saturation
parameter to be slightly higher than the strongest real feature (often your
spectrum plus a bright sky line).
The first question one should ask is should I be subtracting any sky.
If you do not have a sky fiber then it is moot. If you do have
a sky fiber but do not see any sky then you should not subtract the sky.
If you do have a sky fiber but only see sky emission lines then you may
or may not want to subtract sky. You need to determine if you are in
the read noise or the poisson noise limited regime. Even though the
read noise of the MM1 CCD is low (~4 electrons) there are lots of pixels
to sum over. Unbinned there are 20 pixels across the 3as fiber. That
is a lot of read noise. Thus you should consider if subtracting the
sky to get rid of a few sky lines is going to add a lot of read noise to
your spectrum. If you are working in the blue on faint objects the answer
might well be yes! Sky subtract with caution.
If you do decide to subtract the sky fiber you should extract the sky in
the same way that you extract you spectrum. You should have the sky
subtraction parameter in the apsum (or apall) turned off
and the aperture should be the same width as your science fiber. In fact
I suggest that you use the science fiber as a template. Here is a potential
set of steps:
apedit and aptrace the science aperture in the sky or dome flat
using the flat as a reference apedit the science spectrum and check that the
aperture looks good on the science fiber in all lines
change any of the science aperture widths interactively
extract the science spectrum
using the science spectrum as a reference apedit and aptrace
the flat again and offset to the sky fiber
copy the science spectrum to a new file
using the flat as a reference apedit the renamed science spectrum
and check that the aperture looks good on the sky fiber in all lines.
extract the sky spectrum.
You will probably have to scale the sky spectrum flux to that of the science
fiber. This scaling is the reason why you do not want to use the sky
subtraction to remove things that do not come from the sky, e.g.
scattered light, bias and dark current. To get a good idea of the relative
throughput of the science and sky fibers I suggest that you look at the
flux in the dome or sky flat (NOT THE INTERNAL FLAT). In theory you should
just have to multiply the sky fiber by a constant to get the subtraction
It is possible (although very tricky) to subtract the
sky lines and not the read noise (assuming that there is not continuum
sky light in your sky fiber). To do this you will need to remove the
same blaze function from the sky and the science fiber. Then you copy
the science frame to a second science from where you reject all values below
a certain level and set them to zero. This can be done with imcombine. It
is very delicate procedure with limited uses but it might improve the quality
of your data. I suggest that before you try this, you try out
not subtracting the sky at all!
Getting the wavelength calibration is tedious but quite straightforward.
I use the ecid task and end up with a solution with
xorder=7 and yorder=5,
although I start at a much lower order until I have identified a lot of lines.
See HRS wavelength calibration for a ThAr identifications by spectral order.
JCAM Data Reduction Tips