5.2 Bias Subtraction
5.2.1 Bias Calibration Issues
Bias reference frames are acquired daily for scientific calibration purposes and for monitoring the detector performance. Every week multiple bias frames are combined together into a reference superbias image. The combination removes the cosmic rays accumulated during the readout time and enhances the signal-to-noise ratio of the fixed pattern noise.
CALACS performs the bias correction in two steps (see Section 3.4.1): doBlev subtracts the bias level from the overscan region and doBias subtracts a reference bias image.
Each quadrant of the WFC focal plane (A&B for WFC1 and C&D for WFC2) has two overscan regions: a 24-pixel-wide leading physical overscan at columns 1-24 and a 20-row-wide virtual overscan at rows 2049-2068 obtained by overclocking after the readout of the last CCD row.
All four WFC amplifiers produce a horizontal ramp in the leading overscan which extends up to 18 columns toward the active area (see Sirianni, M., Martel, A.R., & Hartig, G.F., 2001, JHU-ACS internal report, available on-line at:
http://acs.pha.jhu.edu/instrument/calibration/results/by_item/detector/wfc/build4/overscan/).
The amplitude of the ramp is different for each amplifier, but it is generally larger in the A&D quadrants than in the B&C quadrants. The perturbation associated with the ramp fades gradually and disappears within approximately 15 columns. The bias level from the physical overscan can be safely measured using the last six columns adjacent to the active area [columns 19-24 for A&D and columns 4121-4126 for B&C].
The virtual overscan is not used to estimate the bias level. This region shows large scale structure which is quadrant and gain dependent. This structure, due to the bias fixed pattern noise, makes using the virtual overscan in an automatic routine problematic. Moreover, this region can be contaminated by deferred charge due to degradation in the parallel charge transfer efficiency.
Analysis of the active area in the bias frames shows small differences in the bias level between the leading physical overscan and the active area (Sirianni, et al., 2002, "HST Calibration Workshop" STScI, page 82). This offset varies from amplifier to amplifier and it can be as large as 3.5 DN. Ideally, no offset would be seen. In principle, if the residual offset between the imaging area and the overscan region were always of the same amplitude, a full frame bias subtraction (doBias) would remove any residual differences. Unfortunately, the offset is not constant but shows random variations on the order of few tenths of a DN. These variations are likely due to interference between the WFC integrated electronics module and the telescope and/or other science instruments. The accuracy of the bias level subtraction in a single quadrant is limited by this random effect.
HRC images are read out using only one amplifier. Each image has three overscan regions: the 19-pixel-wide leading and trailing physical overscans at columns 1-19 and 1044-1062, respectively, and the 20-row-wide virtual overscan at rows 1025-1044. The readout amplifier produces a horizontal ramp in the leading overscan which extends up to 10 columns toward the active area. As in the case of WFC images, the bias level is therefore measured in the last six columns of the leading physical overscan. There is not a significant difference between the bias level in the overscan area and in the active area.
Even after the subtraction of the overscan bias level, some structure remains in the bias frames. This structure includes real bias pattern noise and also some dark current accumulated during the time required to read out the detector. To remove this structure a superbias is subtracted from each science image.
In principle, a perfect bias subtraction would give an image, that once flat fielded and converted to electrons should not present any discontinuity at the boundaries between quadrants.
However, in the case of WFC images the amplitude of the residual offset level varies between each quadrant. This results from the bias level being measured in the leading physical overscan and then subtracted from the active area of each quadrant, causing the center of the resulting image to show a jump between the two adjacent quadrants, A&B or C&D (Sirianni, et al., 2002, "HST Calibration Workshop" STScI, page 82). This jump is not removed by the superbias subtraction. Indeed, since this offset (usually with a maximum amplitude of 0.3 DN) is present in all calibration and scientific frames, the final product of CALACS, after bias and dark subtraction, may show a quadrant-to-quadrant jump as large as few DNs (Figure 5.1).
Figure 5.1: Calibrated WFC1 image showing the quadrant-to-quadrant jump
For point source photometry, the local background in an annular region is typically subtracted so that the residual offset has essentially no impact on the integrated magnitudes. However, if the object falls across the quadrant jump the photometry should be considered suspect. Similarly, extended objects may spread over two or more quadrants, so their surface brightness profile will suffer from the residual offset. More studies are in progress to better characterize this problem and to develop a correction that can be applied directly to the calibrated images. At the moment we suggest fitting the sky level in each quadrant separately.
A detailed study of the on-orbit characteristics of bias frames, the overscan regions and superbiases used in the pipeline will be published in a future ISR.
5.2.2 Bias Jump
There is some evidence for intermittent variations at the sub-DN level in the WFC bias level during the readout. On a few occasions some frames show horizontal bias jumps in one or more quadrants that can run for several hundreds of rows (Figure 5.2). These effects occur at a few tenths of a DN level and they might appear in one or more quadrants simultaneously. The probable source is interference from other activities on the spacecraft and/or scientific instruments with the electronics. At the moment there is no automatic detection of bias jumps within the calibration pipeline. Though bias jumps at the sub-DN level are not important for most applications, users should be cautioned that they might exist in processed data and we recommended a careful inspection of the data for their presence.
Figure 5.2: Bias Jump in the WFC1 Quadrant B only. The vertical stripes in the data are hot columns and are unrelated to bias. 
5.2.3 Bias Subarrays for the WFC and HRC
When science data are obtained with use of subarrays, dark and flat field images will be extracted from full-frame images. Bias images will also be extracted from full-frame images, if the subarray is contained in a single quadrant. Tests have shown that this does not degrade the quality of the dark, flat field or bias corrections as compared to full-frame data. It is strongly advised that subarrays not be defined such that they span amplifier quadrants. In the case that the subarray extends into the adjacent amplifier quadrant, users must obtain their own subarray bias images (with one-amp readout), otherwise the bias level difference between adjacent amplifier quadrants would be applied inappropriately.
