The WFPC2 analog-to-digital conversion (ADC) has a bias towards DN values which correspond to the setting of all low-order bits in the digitization, with larger signal levels being more strongly affected. The correction for this bias involves replacing the output value with a value corresponding to the scaled average input signal for that output. DN values in the range of our data have an average additive correction of 0.86 DN, with the correction smoothly varying for levels in this range. The same correction is used for each of the four CCDs in WFPC2. The data used to determine this correction were taken pre- and post-dynamic testing, with only 0.02DN variation. The correction is thought to be stable under normal usage and insensitive to temperature fluctuations (STScI Technical Memo RSB-85-01). At the signal levels of our data, the error after the ADC correction applied is 0.02 DN.
The bias level was removed with a two-step process of overscan subtraction and bias-image subtraction. We employed the usual method established for WFPC2 data of subtracting overscan from odd and even columns using the overscan from each individual exposure. For bias-image subtraction, we used the "superbias" frames produced for the reduction of the Hubble Deep Field (HDF, see Williams et al. 1996). This bias frame was produced from the average of 200 frames, and has lower read noise than we could achieve using the 30 bias frames taken during our orbits. Instead, bias frames from our orbits were test-reduced (overscan and bias subtraction using the HDF superbias) to verify that the superbias produced corrected levels consistent with zero, with a scatter of 0.002 DN. As this error is inseparable from our estimate of the error in the dark subtraction, it is included with the dark subtraction in our error budget.
We used the pipeline flat-fielding images provided by the WFPC2 GTO team in May 1996. The pixel-to-pixel flux error in those images is reported in the WFPC2 Data Handbook (V3.0) to be roughly 0.3% (rms) for the WF chips, and 0.5% for the PC1. The errors over spatial scales greater than 10 arcsec are less than 0.5% for all four chips. As we excluded data within 75 pixels of the edge of each chip from our analysis, issues of geometrical distortion and vignetting are avoided. The photometric calibration of WFPC2 is tied to the mean level of the flat-fielding images, so that no systematic error is introduced due to flat-fielding.