Observations of twilight sky can also be used to generate flat-fields. A sequence of frames is acquired during darkening or brightening twilight sky. The flat-field image is construct by averaging together differences between pairs of images that have a difference in illumination levels that is large compared to the read noise of the array. For instance, if a sequence of 100 frames are taken in evening (darkening) twilight, the frame 1 - frame 51, frame 2 - frame 52, etc.. differences would be taken, multiplicatively normalized to a common modal value, and then median combined to create the flat-field. In principle, this procedure should avoid problems due to possible illumination level dependent bias effects. In practice, twilight sky observations were taken in June of 1994, and a twilight sky flat was applied to the repeated M92 field data. Photometric repeatibility tests, as described in Reference 7, indicate that twilight sky flattening produces improvement the photometric consistency for both psf-fitting and aperture photometry (see Figure 3).
Figure: Average repeatibility of DAOPHOT aperture photometry for point sources detected in 9 or 10 out of 10 Protocam scans of the M92 region in June 1994. The mean standard deviation in observed K magnitudes for all sources within 0.5 wide magnitude bins is plotted against the average K magnitude for the bin. Flat-fielding was done using twilight sky observations.