08/21/2012: AstroDrizzle Error Affecting Saturated Sources in WFC3/IR Images
Users retrieving WFC3/IR images from the archive after June 7 and until mid-September may find that bright objects which were saturated in one or more reads have total counts which are lower than expected (up to half a magnitude in some cases). Similarly, WFC3/IR data processed offline with AstroDrizzle prior to version 1.0.5 may have this problem. Version 1.0.5 was released on August 13 and is currently only available in IRAFX. For details, see the DrizzlePac release notes.
The WFC3/IR detector is typically instructed to perform multiple non-destructive reads over the course of an exposure. A bright source that saturates the detector over the total length of the exposure may be faint enough that those pixels are not saturated in the early reads. In this case, the calibration software uses the unsaturated reads to calculate the photon flux and reports the exposure time of the pixel as the total time in the unsaturated reads. Thus a bright star may have its central pixel, as well as some of the surrounding pixels saturated in a number of reads, with pixels further out remaining unsaturated throughout the entire length of the exposure. The central saturated pixels will have shorter exposure times recorded than those around them, with those pixels saturated earliest having the shortest exposure times of all.
In the latter stages of AstroDrizzle development, a change was made to use this pixel-based exposure time, rather than the total exposure length when drizzling. This was done to allow users to more accurately compute photometric errors using the output weight map. Unfortunately, the regression tests that were in place at the time did not catch an error that this scheme can introduce. When an image is drizzled, adjacent pixels, one of which may be saturated and the other not, can contribute to the value of the same output pixel. Using pixel-based exposure time weighting (which is employed by the archive), the unsaturated pixel would receive a higher weight since it had a longer exposure time. This will bias the output count value low. We are therefore returning to a flat exposure time weight across the entire image. Users who manually reprocessed saturated WFC3/IR data with AstroDrizzle prior to version 1.0.5 may also encounter this problem when using either exposure time weighting or "inverse variance map" weighting.
Again, this error will only affect users attempting photometry (or very precise astrometry) on sources bright enough to have saturated in one or more reads. Therefore many users can ignore this issue. However, if this bug could affect your science, we recommend reprocessing. To determine whether an IR image contains saturated pixels, users may inspect either the [TIME] or [SAMP] extensions of the image, where the exposure time and number of samples will be lower in pixels that were saturated. We will provide an update on the DrizzlePac webpage when the archive is using a version of AstroDrizzle containing this fix. In the future we will continue to look into other approaches for simplifying the creation of accurate pixel-level final exposure maps for WFC3/IR.