|
|
|
____________________
|
|
|
|
First Steps
First Steps in Reducing an ACSIS Cube
The routines and methods required to reduce a raw ACSIS data cube are
described in this section. Most of these steps are also carried out by
the ORAC-DR data reduction pipeline (both the offline and telescope
versions). If you do not have ORAC-DR available on your machine then you
can download it from the Starlink download page here.
|
Visualising the time series cube
The GAIA data visualisation package handles 3D cubes. To visualise any
image (2D or 3D) place the
file name after the gaia
command:
> gaia temp.sdf &
The ampersand runs the program in the background allowing you access to
the command line. Loading a time-series cube into GAIA will give
something like the following:
This example shows a HARP time series giving the response of the
receptors (x-axis) as a function of time (y-axis). Note that at the
time this cube was taken, 3 of HARPs receptors were not functioning.
Simply by clicking onto a pixel on the image, the spectrum recorded at
that time for that receptor will pop-up:
The red line running down the window is not an artifact. Grab it with
the cursor and move it around - see what happens?
|
Masking bad receptors
There will be times when you will want to mask out bad receptors from your analysis. This may be due to a poorly performing HARP receptor or because of poor baselines in your data. It is best to remove these bad receptors from the data before you begin your analysis.
In the example above, an image of the time series is shown, giving the response of each receptor as a function of time. We can set all data from a particular receptor to be BAD using the chpix command in KAPPA:
> chpix in=file_in out=file_out section="',8:8,'" newval=bad
or
> chpix in=file_in out=file_out section=\',8:8,\' newval=bad
Note the use of the \ to prevent the shell from interpreting the quote symbols when entering this in the command line. This command should now have set receptor number 8 (or H07) to be "switched off" and all future data reduction should ignore the information from that receptor. To switch off other receptors, repeat the command with the new receptor number. Of course, if the receptors are contiguous then you can put them into the same command. So, to mask out receptors 7 and 8:
> chpix in=file_in out=file_out section="',7:8,'" newval=bad
It is also possible to use the chpix command to mask out certain sections of the time series of a certain receptor. Time is stored in the third dimension of the data structure (note that this is only relevant for the time series cubes). Perhaps you notice that something went awry with the receptor 5 for just one jiggle pattern (say, time steps 17 to 32 of a 16-point jiggle). To mask out just this section of the time series the command should be:
> chpix in=file_in out=file_out section="',5:5,17:32'" newval=bad
Other ways to identify bad receptors are discussed under
Advanced Methods.
Combination of several maps with masked pixels (or how to omit unmasked receptors) is discussed at the bottom of the next section.
|
Making a cube
To convert the information from the time series cube into a 3D cube of
spatial and spectral dimensions you need to run the SMURF command makecube. For
example:
> makecube a20070113_00022_00_0001.sdf autogrid=true
Projection parameters used:
CRPIX1 = 0
CRPIX2 = 0
CRVAL1 = 51.3958333333333 ( RA = 3:25:35.0 )
CRVAL2 = 30.75 ( Dec = 30:45:00 )
CDELT1 = -0.00166666666666667 ( -6 arcsec )
CDELT2 = 0.00166666666666667 ( 6 arcsec )
CROTA2 = 0
Output cube pixel bounds: ( -57:60, 4:58, -1024:1023 )
Weights cannot be determined for the input spectra.
Each output spectrum will be an unweighted sum of the input spectra.
Output cube WCS bounds:
Right ascension: 3:25:07.3 -> 3:26:02.3
Declination: 30:45:15 -> 30:50:45
Radio velocity (USB): -429.1489 -> 437.7256 km/s
OUT - Output NDF data structure > tempcube1
By specifying autogrid=true,
the software will attempt to calculate the optimal projection
parameters when the image is regridded. This usually works well for grid and jiggle observations but it's a good idea to always check the numbers reported (the CRDELT1 and CRDELT2 values shown above).
For raster maps, the current advice is that this option should be set to false. When doing so then you should provide the pixel scale as a command line argument. For example,
> makecube a20070113_0022_00_0001.sdf autogrid=false pixsize=7.3 out=tempcube1
The files produced can be several hundred MB in size (this is especially true for large rasters). To try
to minimise disk space usage and, more importantly, processing time,
there is an important parameter within makecube which
allows you to select just a specific region in velocity space. For
instance, if you know that your line will lie close to 0 km/s and will
not be exceedingly broad, then you may choose to only process the
central 200 km/s, say. You can do this with the specbounds
paramter:
> makecube a20070113_00022_00_0001.sdf specbounds=\"-100.0 100.0\" autogrid=false
out=tempcube1
In the example here, there are 3 subscans from a single raster
observations. They can be reduced into a single cube by specifying all
three time series cubes. On the command line this can be done with:
> makecube "a20070113_00022_00_000[1-3].sdf" specbounds=\"-100.0 100.0\" autogrid=false
out=tempcube
Note how the various delimeters had to be escaped on the command line
with the \.
Note also how the construct <filename>_000[1-3]
means that the last integer is cycled through the range given within
the square backets. This has the effect of sending all three files with
suffixes _0001, _0002 and _0003 into the makecube
program for processing together. The final result is the complete
raster written to the file tempcube.
The output cube can be plotted in GAIA for
inspection. Note that the pop-up window has a number of tabs
which allow the user to manipulate the data in various ways. Click on
the Help
button on the top-right corner of this window to get a description of
what each does. Most of the functionality is described in this
cookbook. Click anywhere on the image to reveal the spectrum at that
position.
Coadding a list of maps
One can coadd different maps by editing an ASCII file filelist.txt with a list of filenames and then e.g.
> makecube in=^filelist.txt specbounds=\"-100.0 100.0\" autogrid=false
out=tempcube
The KAPPA command
> provshow tempcube
will give a list of the files which were used to create a particular data cube.
filelist.txt should be of the format:
(path)file1.sdf
(path)file2.sdf
...
(path)lastfile.sdf
where (path) is the location of the files when they are not in the current directory.
Parameter detectors
If one would like to make a cube where a receptor appears to be bad, but without masking it one can use the detectors parameter: > makecube in=a20070113_00022_00_0001.sdf autogrid=true out=tempcube detectors='"-H04,H10"'
would make a map using all receptors, except H04 and H10.
Similarly,
> makecube in=a20070113_00022_00_0001.sdf autogrid=true out=tempcube detectors='"H10"'
would make a map using only receptor H10.
Parameter badmask
When combining several maps with masked receptors one can use the badmask parameter of makecube.
The best is to use 'badmask=AND', but for large datasets the memory requirements will be excessive in this case. Then it is better to use 'badmask=OR' (which is
the default). Using 'badmask=FIRST' means
that the decisions on which pixels will be flagged as BAD will be made with
the first input spectrum (the rest are ignored).
Gridding options in makecube
There are several re-gridding options available to the user through the makecube command. These are specified with the spread parameter on the command line. To see the complete list of regridding options type the following into the command line:
> smurfhelp makecube para spread
The default method is Nearest which assigns the input pixel completely to the nearest output pixel. This is the fastest method available. There are a number of other methods available including Gauss and Sinc which spread the emission to neighbouring pixels (you will need to provide the size of the convolution function through the PARAMS parameter). These methods are useful in the case of
e.g. HARP raster maps which contain holes because of bad receptors (which are not filled if the method
Nearest is used).
Generating variance
There is a genvar
parameter within makecube
which allows you to generate a variance array for your cube. This array can be displayed with GAIA.
There are three options for genvar:
spread
- the variance values generated are based on the spread of input data values
which contribute to each pixel. Multiple passes over the same region
would be required for this to be effective. At least 3 input pixels are
required to generate an ouput variance value. If only one pixel
contributes to the output then the associated variance will be set to BAD.
Tsys - the output variance is calculated based on the system noise temperature.
None - no variance values are generated.
The default behaviour is to use Tsys.
When using spread, one also must
use badmask=and and use not
spread=nearest (but instead
e.g. spread=linear .
Note that in older data (before early 2007) the off exposure time is missing
in the header and
makecube cannot (yet) generate
variance from Tsys. In those cases there will be a message
Weights cannot be determined for the input spectra.
Each output spectrum will be an unweighted sum of the input spectra.
Sparse array
If one has observed a number of offset positions which are not on a regular grid or are widely spaced, it is of little
use to make a normal cube with makecube, which then would have a large size and which would be mostly empty. Then one can use the makecube parameter
Sparse, which will make a sparse array (when set to true). See
> smurfhelp makecube para sparse
This is a one dimensional array where spectra are displayed by GAIA in the
order where they were observed (from left to right).
Positional information of all spectra is in the WCS header, and an sdf file with a sparse array can be
combined with a normal cube using wcsmosaic - one of its uses
is to fill holes in maps.
At the moment (20071121) spectra in sparse arrays are not yet coadded by makecube, so it contains all e.g. 1 second spectra of e.g. 30 seconds integrations. Hopefully this will be
changed soon.
Compressing the data
As well as carefully selecting the spectral range for producing cubes,
another method for saving on processing power is to bin up the data
before running makecube.
This can be done using the KAPPA
command, compave,
which will bin up data along a given axis.
For example, to bin up the spectral axis by a factor of
3 you would type into the command line:
> compave a20070113_00022_00_0001.sdf compress=\[3,1,1\] out=test
Note that the first dimension is the spectral axis in the time-series
cube. It becomes the third axis after the makecube
command re-grids the time-series data into a spatial grid.
A cube which has been both compressed and had spectral bounds truncated
will occupy much less disk space and memory. Of course, if you are
looking for small discrete lines or require the highest spectral
resolution, compressing the spectral dimension may not be appropriate
for your needs.
The image below shows the result of compressing and truncating the
spectral axis of a cube observed as part of a raster map.
The spectrum in the cube now spans from -100 to +100 km/s. Note that
the blue cross on the image indicates the location of the spectrum
shown in the pop-up window. Grab that cross with the mouse and see what
happens!
|
Rebinning the spectrum
There is a KAPPA command, sqorst, which will allow you to rebin a spectrum (whether a 1-D spectrum or 3-D spectral cube) to a specific resolution.
> sqorst in=tempcube mode=pixelscale axis=3 pixscale=5 out=tempcube_sqorst5
This will rebin a 3-D cube along its spectral axis (axis=3). In the example shown above the input spectrum is rebinned so that it's resolution element is 5 km/s (pixelscale=5). The figure below shows a before and after comparison.
|
Removing
the baseline
Noise on images may be due to varying or bad baselines. The baselines
in a cube can be fitted to and removed using the mfittrend
command in KAPPA.
Note that it is also possible to do this within GAIA, in an
interactive manner, by selecting the Baseline tab in
the Image sections
window.
The mfittrend
routine will
fit a polynomial (up to 15th order) to all lines of data along a chosen
axis. The fits can be restricted to selected regions so that lines of
emission (or absorption) can be avoided. Fits are then extrapolated and
interpolated across the whole spectral region in the cube. The
resultant baselines can be removed from the input cube, or stored as a
seperate cube and removed later. The default behaviour is for the
baseline cube to be written to disk and not subtracted (which will give
you the freedom to undo your actions if necessary).
> mfittrend test2.sdf ranges=\"-90 -30 30 90\" order=1 axis=3 out=baseline
Fitting NDF axis 3, using pixel ranges:
-72 : -24
22 : 70
> sub test2 baseline test3
In this example, a straight line was used to perform the baseline fits
along the spectral dimension (axis=3),
fitting to the velocity range given in the ranges
parameter. The resultant basline cube is written out and in the next
line is subtracted from the input cube.
The image below shows a before-after comparison of baseline removal for
a single plane within an ACSIS cube.
The top image is that before baseline removal and the image below is
after the baselines have been removed.
Note on 1x1 "cubes"
If your data was collected with either RxA or RxW, then it is likely
that you do not have real cubes but just single, pointed observations
of spectra. Up to this point the data reduction has still been
appropriate for such so-called 1x1 "cubes". The rest of this section will now be more specific for properly 3D cubes. Going ahead to the section on SPLAT and GAIA
is likely to be more appropriate for those with single spectra.
|
Mosaicing
cubes
If more than one time series cube is passed to makecube,
then those cubes will be mosaiced together. This has the advantage of
not generating many intermediate cubes, but has the disadvantage of
using up lots of your computer's processing power. In the example given
above for making
a cube
,
three subscans were passed to makecube
to create a single cube. The principle is the same for passing
different raster maps of neighbouring fields. As the next step, you
should remove the baselines from this cube following the procedure above.
The alternative is to use wcsmosaic
for putting cubes together,
> wcsmosaic 'tempcube_base*'
REF - The reference image /!/ >
LBND - Lower pixel index bounds for output image /[-118,-57,-237]/ >
UBND - Upper pixel index bounds for output image /[61,57,236]/ >
Using sincsinc binning kernel.
OUT - Output image > tempcube_base
The resultant image when viewed in GAIA is shown
below. LBND
and UBND
are pixel bounds for the resultant image which the software will
calculate itself. Just accept these to get back all the data. If you
wish to make further truncations to the output mosaic you can do so
here by adjusting these bounds.
Whichever method you use, you should now be in a position to create
data cubes from your ACSIS data.
<Back to
Contents>
|
|
|
