AAO technical note (originally 9 March 1999, updated 1 Nov 1999)
Summary. A couple of observers have noticed large-scale dips in the spectra for objects near one edge of CCD 2, using the engineering grade Tek 1024 CCD prior to August 1999. Two large dark blobs are apparent on corresponding fibre flat fields, near the middle of the top edge. Examination of data taken since January 1997 indicates that this is not a new feature. However, its manifestation changes markedly with wavelength; while at blue wavelengths it appears as a local drop in sensitivity by up to 30%, at red and infrared wavelengths it appears as a minor enhancement at only the 2% level. The implication is that this is probably not dirt or a blemish on either the field flattener lens or the CCD itself, but more likely a manufacturing flaw on the CCD (CCD 2 is after all only an engineering grade device), which probably arose in the thinning or UV sensitising processes. The problem affects only the last ten or so fibres on CCD 2 (i.e. numbers 190-200) and is readily removed by using the `divide by fibre flat field' option in 2dfdr. It is probably only worth worrying about for blue spectra taken with the 1200B gratings centred near 4200Å, since at lower dispersions the relevant area of the CCD is seeing much longer wavelengths where the effect is small. However, there may well be other smaller but similar features elsewhere on the CCD, in which case dividing by the fibre flat field may be the optimum strategy in general, at least in the B/UV spectral region and provided that flats are obtained with adequate counts at all wavelengths.
1. The Problem. This effect was noticed by Gary Da Costa in September 1998, when reducing some globular cluster spectra taken with the 1200B gratings. I rediscovered it in some of the Service data taken in January this year (data for LMC4, Cannon et al.), using the 1200V gratings but set to about the same central wavelength of 4250Å. Figure 1 shows a set of spectra with the offending dips at around 4150Å; there is a relatively sharp feature in fibres near 192, becoming broader in fibres 196 to 200.
Figure 2 shows part of the corresponding fibre flat field, after reduction in 2dfdr (i.e. tramline extraction, followed by self-flat fielding: frame 24jan0005red.sdf). The region shown is the central third (in wavelength space) of the top quarter of the full frame (flat5_jan99.ps), displayed at approximately the original aspect ratio (i.e. 5 pixels per fibre vertically). The grey scale ranges from 0.6 (black) to 1.2 (white). The problem is the two large dark blobs near the middle of the top edge, between about pixels 400 and 550 in x and starting at about fibre 185 (corresponding to pixel 950 in y on the original CCD frame). Note that those fibres which are markedly non-grey (e.g. 181, 187 and 190) were all parked and so irrelevant and clearly not correctly flat-fielded, but the rest of the field is very uniform apart from a few much smaller blemishes.
The dark blobs are certainly not an artefact of the data reduction process. The structure can be seen by eye in the raw data frames, although it cannot be displayed clearly in one shot since there are large gradients in the counts across the field. In any case, the dips are too large to be caused by any tramline fitting error, which could not cause such a loss of counts in many adjacent fibres. Figure 3 shows plots of the reduced flat field as pseudo-spectra; as in Fig. 1, large dips are evident in the last few fibres.
The dips in the spectra can be effectively removed, at least to first order, by dividing the extracted spectra by the fibre flat. The result is shown in Fig. 4. Clearly what we are seeing is a real local drop in CCD sensitivity.
2. Has the effect varied with time?
The effect seen in Figs. 1-3 is also present in data taken for Da Costa et al. in September 1998. In particular, flat field 22sep0026red.sdf for CCD 2 shows almost the same pattern. The one difference was that those data were taken with a 1200B grating, not 1200V, but the nominal central wavelength was the same, 4250Å. The small differences in detail may be due to having a slightly different central wavelength.
Going further back, some flat fields taken in January 1998 (from data on LMC carbon stars for Cannon et al.) and January 1997 (from some data being analysed by Terry Bridges for Mike Irwin) also show structure in the same region of CCD 2 (in January 1997 the same engineering CCD was in use, but was probably then installed in spectrograph 1). However, there is a crucial difference: the regions covered by the dark blobs in Fig. 2 appear lighter in the older data, i.e. their sensitivity was apparently higher, although only by about 2 or 3%, rather than being lower by 15-20%. Those earlier flat fields were taken towards the red end of the spectrum (1200R gratings, central wavelength 6100Å in 1998 and 8500Å in 1997) and raised the possibility that the effect is wavelength, rather than time, dependent.
3. Wavelength dependence.
An unambiguous test is afforded by the Service data taken on 24 January this year, since data were taken at both the blue and red ends of the spectrum using the same 1200V gratings on the same night. The result is shown in Fig. 5, from flat field 24jan0022red.sdf (taken for Peter Wood). The format is the same as for Fig. 2, but the central wavelength here is 6800Å and the grey scale is much reduced, ranging only from 0.98 (black) to 1.03 (white). Again, the regions that are down by about 15% in Fig. 2 are here enhanced by about 2%. (The curved near-vertical white line is probably the result of an arc lamp being inadvertently left on when the flat field was taken.)
Figure 6 shows the extracted flat field `spectra' from Fig. 5, for comparison with Fig. 3. There is an accurate match in wavelength between the `emission' features near pixel 500 in the `spectra' in Fig. 6, especially for fibres 190-200, and the `absorption' features seen in Fig. 3. However, the intensities in Fig. 6 are much lower, and the bumps have become comparable with many other features on the CCD.
These two plots are very similar to corresponding plots for red and near-IR data taken in January 1998 and January 1997. There is no evidence for any significant time variation in the blobs; clearly the wavelength dependence is the dominant effect.
4. Lower dispersion spectra.
The strong wavelength-dependence of the sensitivity of the dubious region on CCD 2 is fortunate because it implies that the effect is likely to be small for data taken at lower dispersion. Such data will be taken at longer central wavelengths so that the light landing near pixel 500 will then generally be much redder than Fig. 2. In particular, data taken for the main galaxy and quasar redshift surveys are unlikely to be seriously affected, whether or not the data are flat-fielded.
This has been confirmed by inspection of data taken in January 1999 using the same set-up as for the 2dFGRS. The `spectra' from a reduced flat field are shown in Fig. 7 (frame 23jan0016red.sdf, a fibre flat taken with CCD 2 using a 300B grating centred at 5800Å, for some measurements of photometric standards). As for Fig. 6, there may be relatively small effects near pixel 500 related to the blemish, but the variations are no larger than many others across the CCD.
Clearly the low-sensitivity region in blue exposures with CCD 2 is not due to obscuration by dirt on the CCD or field flattener lens, since it has slightly enhanced sensitivity at longer wavelengths. It is much more likely to be due to a manufacturing fault on this engineering-grade CCD.
Since it seems to have remained stable for more than two years there is probably little point in opening the camera or checking the CCD at this stage; the problem should go away when the new high-quality CCD is finally installed in camera 2, once the halation problem has been cured.
The effect on most programs will be no worse than (or as bad as, depending on your viewpoint) for many other minor blemishes and sensitivity-variations across the CCDs. However, users who work with the 1200B or 1200V gratings at central wavelengths shorter than say 5000Å should be advised that at least the data taken in the final 15 fibres on CCD 2 should be flat-fielded using a fibre flat.
Note that this feature on CCD 2 is no longer relevant for 2dF data taken from August 1999 onwards, after the engineering grade CCD was replaced by the second science grade CCD.