The 2dfdr data reduction system used for processing AAOmega data is designed to automatically process and wavelength calibrate spectroscopic data. This process is driven by keywords in the image header. This page discusses some of problems that can occur and the strategies available for improving the wavelength solution. Details of how to reduce data with 2dfdr can be found in the cookbook.
- Some problems with the automatic wavelength solution
- Example symptoms
- Final output wavelength solution
It is critically important that the wavelength solution is derived accurately. With AAOmega data reduced by 2dfdr RMS solutions of ~0.1 of a pixel are the norm. However, thermal variations within the spectrograph can occasionally result in the initial guess solution, being too far from the true value. This can lead to a failure of the wavelength calibration. This page describes the simplest solutions to the problem.
This error most commonly occurs in the blue arm of the AAOmega system. The practical result of this error is twofold:
It is usually first noticed as poor sky subtraction in blue AAOmega data, since the inaccurate wavelength solution leads to complications with the preparation and scaling of the sky spectrum, particularly at low resolution with the 580V grating.
The wavelength range covered by the reduced spectrum can be mismatched to the wavelength range covered by the CCD. Spectral curvature at the detector means that there will also be some missing spectral data which is padded in the reduced frame by dummy values. However, if the solution is not ideal, these dummy values can make up to 20-30 pixels at one end of the wavelength range, at the expense of real data at the other end (usually it is the important 3727 [OII] doublet that is lost in the blue at low resolution.
The two most common wavelength solution problems are outlined below:
If the initial estimate wavelength solution is not close enough to the true solution, the arc line-fitting routine can fail to find the correct stable solution, resulting in fuzzy regions of the reduced arc line frame as seen in Figure 2 below. This is usually best fixed by adjusting the GRATLPMM (grating lines-per-mm) FITS header keyword for the data, see Modifying the initial estimate solution below.
Note: Some fibres are marked as Type=U, and are deemed Unusable by 2dfdr. These are usually broken fibres. No new wavelength solution is fitted for these fibres and hence they will often show a sharp discontinuity in the arc solution. Since they contain no science data (even though they will sometimes contain arc line light) no information is lost here - this is illustrated in Figure 3.
Figure 1: A correct AAOmega arc solution is shown, with the arc lines perfectly vertical. Note that bad columns on the CCD have constant x(pixel) and so now exhibit an opposite spectral curvature in this reduced arc frame.
Figure 2: For data with the incorrect value of the GRATLPMM header keyword parameter, the initial guess solution fails to find the correct solution for some fibres. Here we see a fuzzy region of the arc solution from the bottom right of the arc solution (redder wavelengths).
Figure 3: An example Unusable (Type=U) fibre for which no wavelength solution has been derived. This fibre is not available to AAOmega, but has transmitted some arc light to the spectrograph and so shows a sharp discontinuity in the arc solution. This spectrum does not contain any information and so can be safely ignored.
Figure 4 below shows data which has been correctly reduced and wavelength calibrated, but which has been rebinned to a sub-optimal output solution. This rebinning has corrected for spectral curvature, and placed all spectra on a common wavelength solution, but in doing so it has rebinned the data to a wavelength solution which required a large number of filler or padding pixels to be added to all spectra at one end of the data. At the same time real spectral data have been clipped from the other end of the spectra. In this unfortunate example the valuable 3727 [OII] doublet has been clipped from these planetry nebula spectra.
The solution to the Final output wavelength problem is discussed below.
Figure 4: In this example, a poor choice of central wavelength by the data reduction software has resulted in a large number of blank padding pixels at the red end of the spectrum, and the important 3727 [OII] doublet has been clipped from the reduced spectra even though it is clearly visible at the blue end of the raw data frame for these Planetry Nebulae.
The 2dfdr software places all reduced data onto a common output wavelength solution, after correctly fitting and rebinning the observed solution for each fibre. While maximum signal-to-noise with minimum correlation of noise between adjacent pixels would be achieved by preserving the as observed wavelength solution for each fibre at all times, the utility of placing the data onto a common output solution, with a single rebinning step, is usually the better solution. That this solution is defined in advance, rather than on a frame-by-frame basis, means that repeated observations of the same target over multiple nights (or even multiple months) can be easily combined (although we note that 2dfdr correctly rebins data on different solutions if required). This definition of the wavelength solution in advance does have the side effect that if the true wavelength solution for an observation differs markedly from the predicted solution then valuable information may fall outside the output solution wavelength range and hence be lost. For example, with AAOmega in low resolution mode, this can result in the valuable 3727 [OII] doublet being clipped from the final output data even though it was successfully observed.
The AAOmega wavelength solution is controlled by the header keyword parameters outlined in the table below. If the wavelength calibration does not succeed, the user may need to manually intervene. The most common solution is to modify the initial guess parameters in the image headers to provide an improved initial solution. This modification can be achieved easily in a number of ways, e.g. in IRAF with the hedit task:
low resolution value
|GRATLPMM||VPH grating lines per millimeter. The ruling separation of the virtual facets in the VPH grating. Small thermal expansions and contractions with seasonal variations in the temperature within the spectrograph cause small physical shifts in this value with time.||582|
|GRATANGL||The grating angle. This is usually half of the camera angle, although not if the grating is to be used at a blaze different from the central wavelength.||8.0|
|CAMANGL||The camera angle. This is typically twice the grating angle.||16.0|
|ORDER||The VPH gratings for AAOmega are always used in first order||1|
In the absence of any other information from the user in the start up .idx file (see below) the 2dfdr system uses the initial wavelength solution estimate to determine the final output solution onto which each spectrum is re-binned. This solution is derived as follows:
BETA = (Camera angle - Grating angle ) in radians
ORDER is obtained from the header keyword ORDER, but is always 1 for AAOmega
LPMM is obtained from the header keyword GRATLPMM
If an alternate central wavelength is deemed more appropriate for the data, the user can calculate the modified values, usually small adjustments to the Angles and the Grating lines-per-mm, and update the header values accordingly. Small errors in the wavelength solution are typically fixed by small adjustments the the GRATLPMM parameter (over the range +/-3 to the value stored in the image header). This corresponds to small seasonal variations in the true ruling spacing of the grating due to thermal expansion.
It is possible to modify the final output wavelength range via the 2dfdr .idx input file:
|.idx file parameter||Meaning||Typical 580V blue values|
|DISTX||X centre of distortion||1000|
|DISTY||Y centre of distortion||2048.5|
|SKYSCRUNCH||Wavelength calibrate from sky lines||TRUE|
|SKYFITORDER||Polynomial order for sky line fitting||1 (i.e. a small linear offset)|
Sarah Brough email@example.com