The initial data reduction for the HDF was done rapidly, with many
procedures and calibration files that had not been tested previously
on HST data. While this preliminary set of reduced data will probably
serve most of the immediate scientific needs, the final quality
of the images and the accuracy of the photometry can be improved with
some careful reprocessing. The data will be reprocessed over the
coming months with improved
calibration files, and with slightly different procedures for stacking
the images. These differences are outlined below.
Copyright © 1997 The Association of Universities for
Research in Astronomy, Inc. All Rights Reserved.
New Superbias Frame:
The preliminary superbias frame was constructed from individual
bias frames taken from January to November, 1995. The final superbias
will include bias frames taken during and after the HDF observations.
This will probably have a negligible effect on the noise in the
combined images, but is something that might as well be done as long
as the data are being reprocessed.
New Superdark Frame:
The preliminary superdark frame was constructed from individual
bias frames taken from January to November, 1995. The final superdark
will include frames taken during and after the HDF, and will eliminate
frames taken more than 6 months before the HDF. This should reduce
the number of hot pixels.
Time-dependent Hot Pixel Masks:
In the initial processing we masked all pixels that became hot
(or stopped being hot) at any time during the observations (hot
being defined as more than 5-sigma deviant from the mean dark current).
With more time, we can do a better job of finding all the hot
pixels, perhaps to a lower sigma cut, and can mask them in
a time-dependent fashion (probably using a different mask for each
24 hour period).
Moving Target Masks:
In the first pass, moving targets were only eliminated as well
as the crrej task can do it. As many of the moving objects are
resolved, this actually does not work very well. In the reprocessed
data, these objects will be masked before the images are combined.
We are currently experimenting with different ways to remove the
scattered-light X pattern. There are very few frames where the
scattered light background significantly affects the rms pixel-to-pixel
noise. However, the varying backgrounds increases the pixel-to-pixel
noise in the final subsampled (drizzled) images, and is best removed
before the drizzling stage. Our tentative plan is to drip the
images that have no scattered light, geometrically transform them
to the positions of the frames that are affected by scattered light,
scale by the exposure time, and subtract. This will remove the sources.
The resulting image can be median filtered to produce a fairly smooth
background model that can be subtracted to remove the X pattern.
True Inverse-Variance Weighting:
In the combining the data for the preliminary products,
the individual data frames were weighted by exposure time.
This is close to optimal for the F606W and F814W images,
but not for F450W and F300W, where readout noise makes a significant
contribution. Version 2 of the data will have a more optimal
weighting, which takes into account the readout noise and dark
current, but still uses a constant weight for each frame.
For version 3 we plan to let the weight vary pixel by pixel
depending on the brightness of the source and the number of
cosmic rays rejected.
Harry Ferguson firstname.lastname@example.org 1/14/96
Copyright © 1997 The Association of Universities for Research in Astronomy, Inc. All Rights Reserved.