The procedure for the production of STIS CCD bias reference files has recently been changed. This change was prompted by the observation that hot columns in CCD data are often not well eliminated in the pipeline reduction. Recently, the hot column residuals have been becoming more apparent due to the increasing number of charge traps. Especially short exposures have been suffering from residual hot columns.

Hot columns in STIS CCD bias frames are due to charge traps induced by hot pixels during read out. As the number of hot pixels changes with time, so does the number of hot columns. To correct for hot columns in the STIS pipeline reduction, we have been using the following approach in the past. We replaced the hot columns present in STIS CCD superbiases by the values in a median-filtered version of the superbiases, and had the weekly darks take care of hot pixel and hot column subtraction. However, the darks do not show the hot columns at a S/N as high as those of the superbiases, and the hot columns are different for the GAIN=1 vs. the GAIN=4 superbiases. It was therefore deemed better in principle to include the correction for hot columns in the bias file subtraction step. However, not enough biases have been taken to make superbiases with adequate S/N on a weekly basis.

We devised the following solution. In our new bias reference file production procedure, weekly superbiases are created from which the columns hotter than 5*sigma are extracted and added into a "baseline", high S/N superbias from which all hot columns were removed (more detailed information on the weekly bias construction is given at the bottom of this page). The resulting "weekly superbiases" are then subsequently used to create weekly superdarks as before. We have tested this new procedure on a suite of images and spectra taken using different instrument/detector combinations, and found the results to be a major improvement in that the great majority of the hot columns do indeed disappear. The residuals at the locations of hot columns after bias and dark correction are typically consistent with Poisson noise. While this constitutes a major cosmetic improvement to STIS CCD images, it should be noted that for many analysis purposes it is probably not worth reprocessing data merely for this improvement.

As this report is issued (first week of August, 1999), all weekly bias and dark reference files for the period August 1998 - May 1999 created using the new procedure have been delivered to the operational STIS pipeline and the HST archive. Creation of the bias and dark reference files for the periods March 1997 - August 1998 and June - July 1999 is in process.

Here are the details on how the weekly bias reference files are constructed:

The "weekbias" procedure works as follows. After bias overscan subtraction for every individual bias frame, the bias frames in a subset were combined using calstis2, which did a three-step iterative cosmic-ray rejection with rejection thresholds of 5, 4, and 3 sigma, respectively. After cosmic ray rejection, the combined bias was divided by the number of frames combined (calstis2 adds images instead of averaging them) so that the image represents a single zero-second exposure. Subsequently, hot columns were identified as follows. A median-filtered version of the averaged bias was created (kernel = 15 x 2 pixels) and subtracted from the averaged bias, leaving a "residual" image containing hot pixels and columns. The lower 20% of the rows in this "residual" image were averaged together into a 1-D image from which the columns hotter than 5 sigma above the effective RMS noise on the averaged bias were identified. Finally, the pixel values in those identified columns of the "residual" image were added into a high-S/N "baseline" superbias image, from which the hot columns had been eliminated using cl script "basebias".

A side effect of this procedure is that the noise of the weekly bias frame is slightly higher in the columns with hot-pixel trails than it is in the other columns of the image.

Finally, on a related note: We have changed the bias monitoring program for STIS in Cycle 8 in that we are now taking ~100 unbinned biases per week in CCDGAIN=1 (plus ~20 in CCDGAIN=4 and 14 in often-used binned modes in CCDGAIN=1) to enable one to create "straightforward" weekly unbinned CCDGAIN=1 bias reference files (thus eliminating the need to update the hot columns and thus the need to run the "weekbias" script).