When taking an astronomical image, it is important to account for any spatial variability in the CCD's sensitivity. The variability can be due to a non-uniform response across the chips surface or particles of dust on the surface of the CCD window, amongst other things. These things are accounted for using an image called a flat. A flat is an image which is taken of a blank field of sky, or an illuminated white screen. The important thing is to have a field free of variation. The image is divided by this flat field to remove the effects of the sensitivity variations on the chip, as well as dust doughnuts.
When taking flats, it is important to get as high of an exposure as possible, without the pixel response becoming non linear. The large number of counts will reduce the effects of the random noise in the flat on the statistics of the final image. To find this point, an experiment was performed in which flat fields of incrementally greater exposure lengths were taken. The pixel values were then plotted versus the exposure time to give a graphical representation of what point the pixels become non-linear (see Fig. 1). Most of the pixels become non linear somewhere above 40,000 counts.
Fig. 1 The number of counts recorded as a function of the exposure duration, plotted for a large number of chip pixels, randomly selected. The data were generated using a flat field, so only the relative exposure time is important. The images (taken at a variety of exposure lengths) were unbinned and no dark current was subtracted. A small fraction of the pixels start to go non-linear at 40,000 counts. Many are linear to 45,000 counts. We also see that saturation occurs at a variety of levels between 50,000 and 60,000 counts.
A source of uniform illumination is required. For this pupose, a flat-screen has been installed in the dome. The screen is illuminated by a ring of lights mounted on the frontpiece of the telescope. A ring of light located parallel to a plane provides constant illumination within the radius of the ring projected onto the plane. Well, that's the theory.
To set up the dome flat:
Fig. 2 The telescope and dome position for taking flat-fields.
Now you are prepared to take the flats. Simply take exposures as you normally would, but name the output files Flat_ddC_f.fits where dd is the temperature of the CCD and f is the filter.
When using the white screen to take flats, the ideal exposure lengths for a few of the filters are:
These give pixel values around 40,000 counts, give or take a thousand counts or so.