Space Telescope Science Institute
DrizzlePac 2012 Handbook
help@stsci.edu
Table of Contents Previous Next Index Print


The DrizzlePac Handbook > Chapter 1: Introduction to AstroDrizzle and DrizzlePac > 1.3 What’s Different about Data from the Archive

1.3
As mentioned earlier, flt.fits images from the HST Archive now contain information that describes an image's full astrometric solution. As a result, image headers have several new keywords (listed in Section 3.4), and some image files, such as ACS data, have new FITS extensions that carry small-scale distortion corrections. This format includes the ability to give each astrometric solution a unique name, and to send it to another user as a FITS headerlet.
The PyRAF command catfits can be used to display the new FITS extensions structure for ACS flt.fits images, as shown in the example below for ACS/WFC:
The first seven extensions ([0] thru [6]) should be familiar to ACS users:
Extensions [7] through [11] are new. As noted in Section 1.2, these extra extensions incorporate distortion correction information beyond that captured in the world coordinate system (WCS).
The D2IMARR extension in group [7], in the case of ACS/WFC, is a one-dimensional vector containing corrections for physical distortions in the detector columns.
The WCSDVARR1 extensions, groups [8], [9], [10], and [11], describe residual geometric distortion corrections imposed by the optics, that is, corrections to distortions not modeled by the polynomial fit. For ACS/WFC, these are four 65x33 interpolation arrays.
The tabular information in FITS extensions [7] through [11] were previously represented by the DGEOFILE reference image that was used by MultiDrizzle, which was well over 250 MB in size.
A comparison of output drizzled images between the new AstroDrizzle pipeline and the old MultiDrizzle pipeline may also show small file size differences, or differences in the pixel number placement of a particular position in the output image. Nonetheless, astrometry of images with the default orientation, as used in the pipeline, are identical for AstroDrizzle and MultiDrizzle.
However, when AstroDrizzle was being tested, a problem was found for cases where MultiDrizzle was re-run outside the pipeline using a non-default orientation (the final_rot parameter set to a value other than INDEF). In these instances, MultiDrizzle introduces a small (typically of order 0.1 arcseconds) error in the astrometry of output drizzled images. This bug has been fixed in AstroDrizzle. But it will not be corrected in any version of MultiDrizzle because development for MultiDrizzle has been stopped in favor of finishing AstroDrizzle in a timely fashion. This MultiDrizzle bug does not affect previously-retrieved MultiDrizzled archive data because those drizzled images were produced using default orientations.
During AstroDrizzle development and testing, several other problems were found in MultiDrizzle that affected data from the old pipeline. These bugs have been fixed in the new pipeline running AstroDrizzle:
-
For WFC3/UVIS binned data, MultiDrizzle does not properly account for binning when determining the distortion-corrected chip centers. Among other effects, this produces an extremely large gap between the two chips in the final drizzled image.
-
The ACS polarizer distortion model was not properly applied by MultiDrizzle; it was incorrectly specified in the IDCTAB reference table (which has since been fixed) and not correctly interpreted by MultiDrizzle software. Polarizer images, which are treated as sub-arrays, were not properly corrected to align with a distortion-corrected full-frame image at the same pointing. This error induced a scale difference of nearly 3%, and a rotation of nearly a full degree, relative to a properly distortion-corrected direct image of the same field. Drizzled polarizer images from the AstroDrizzle pipeline are now properly de-distorted, and distortion corrections are correctly represented in the flt.fits header SIP coefficients.
Identification of cosmic rays during MultiDrizzle processing can be slightly skewed by high sky values in the image. Bad pixels and cosmic rays are identified by comparing each distorted input image with a comparison image created by applying the distortion model to an undistorted median-combined “clean” image. In MultiDrizzle, this comparison occurs after subtracting the sky from the input image. Pixels flagged as bad (cosmic rays and other defects) are identified based on a statistical noise model that does not account for sky values that have already been removed. This often results in pixels being flagged as “bad” by MultiDrizzle when they were actually fine. Such overzealous flagging of bad pixels is particularly noticeable in the diffraction spikes of bright stars.
When MultiDrizzle computes the parameters necessary for running drizzle using PyDrizzle (an older implementation of the Drizzle code), some pixels get reported as “dropped” during the single drizzle step and final drizzle step. This happens because the computed size of the output image is not large enough, or the final image is not centered as expected by the code. This computation has been dramatically improved in AstroDrizzle to reduce (if not entirely eliminate) the number of pixels that end up outside the drizzled image.

1
Its name is in compliance with a FITS convention.


Table of Contents Previous Next Index Print