Bugs in GSC2.2.1

We are aware of the following small problems remaining with the present release of GSC2.2:

• 1 database region (out of 32768, or 0.003% of the sky) is omitted from the export, due to problems with the database.
• As a temporary fix for problems observed at the celestial poles, the plates actually containing the poles were re-reduced with a gnomic projection, and no correction for refraction. This corrects the observed problems with extra and missing objects, but does put these few plates on a very slightly different system than the rest of the sky.
• Photometric and astrometric errors as reported in the GSC2.2.01 are unrealistically large. These errors are intended as indicators for operational use only, and cannot be used for scientific studies. An understanding of the actual size of the errors in the catalog can be obtained by comparing the results to external catalogs.
• Very bright objects, ie, in the bright extrapolated area of the photometric solution, were assigned a magnitude of 99.9, rather than an extrapolated magnitude. Since this is the same value that was used as the NULL value, these magnitudes are indistinguishable from magnitudes that were fainter than the magnitude limit or missing altogether. Most of these objects are replaced by Tycho catalogue stars, but some few remain.

Please note that some object names, and selected exported values, have changed between GSC2.2.0 and GSC2.2.1. This is an unavoidable (and indeed generally desirable) consequence of fixing many of the known problems with GSC2.2.0. For this reason, CASB does not recommend attempting to use data extracted from 2.2.0 together with data extracted from 2.2.1.

For historical reasons, we retain the description of the known problems with the GSC2.2.0, which was removed on July 16, 2001 and replaced with GSC2.2.1.

Bugs in GSC2.2.0

We are aware of the following small bugs in the production of the first release of GSC2.2. These are typically related to problems with the manipulation of the data, rather than substantive issues. We plan to fix most of these problems over the next few months, at which time we will release GSC2.2.1.

• Fewer than 5% of the Tycho stars are listed with an epoch of approximately 1.67. Add 1990 to this epoch value to get the actual epoch.
• The Tycho catalog objects in the southern hemisphere are incomplete or appear with incorrect declinations due to a database loading problem.
• In 90% of the sky, database objects that did not appear on at least two plates were inadvertently omitted from the exported data. This affects only about 1% of the objects in the magnitude range of the GSC2.2, since the database contains the entire sky twice (once in F and once in J).
• IV-N survey plates have been loaded into the database for less than 2% of the sky. Objects which occurred on these plates, but had magnitudes fainter than the F and J cutoffs, were improperly exported.
• Positions were improperly exported from one 1950 epoch POSS-I plate, rather than the corresponding POSS-II plate.
• 63 database regions (out of 32768, or 0.2%) have not yet been exported due to database problems.
• Photometric and astrometric errors as reported in the GSC2.2 are unrealistically large. These errors are intended as indicators for operational use only, and cannot be used for scientific studies. An understanding of the actual size of the errors in the catalog can be obtained by comparing the results to external catalogs.
• Some problems have been been observed at the celestial poles.
• Very bright objects, ie, in the bright extrapolated area of the photometric solution, were assigned a magnitude of 99.9, rather than an extrapolated magnitude. Since this is the same value that was used as the NULL value, these magnitudes are indistinguishable from magnitudes that were fainter than the magnitude limit or missing altogether. Most of these objects are replaced by Tycho catalogue stars, but some few remain.

Known limitations of GSC2.2 Calibrations

We are aware of the following issues with the GSC2.2 calibrations. These are more substantive issues that may not be fixed until the release of GSC2.3.

• There is a radial magnitude effect in the astrometry. The positional errors vary with magnitude as follows:

• 0.25 - 0.40 arcsec	11 to 16 mag (errors increase at plate edges)
  0.20 - 0.65 arcsec	fainter than 16 mag (errors increase at plate edges

The worst cases are omitted from the GSC2.2, since we preferentially selected positions from the plate on which the object was closest to the plate center.

• The GSPC2 reference catalog is not yet complete. Approximately 10% of the sky, primarily in the north, has been photometrically calibrated without any deep sequence stars available. These plates were calibrated using the GSPC1 and the solution was extrapolated to fainter magnitudes. Extrapolated magnitudes from these and other plates can be identified by the Source Status flag.
• Galaxy magnitudes are systematically too bright because no separate galaxy photometry calibration with proper handling of extended sources was applied.
• The classifier has a bias towards stars, especially for brighter objects (m<16).
• A small fraction of the objects in the database contain classifications obtained from an early version of the classifier, before multi-tree voting was implemented. This version of the classifier also sometimes exhibited an unphysical sharpness at the subplate region boundaries in the distribution of stars and nonstars. This was repaired in June 1999 and the majority of the plates were recalibrated after that time.
• All the calibration tasks (astrometry, photometry, classification) are less reliable on deblended objects (that is, objects that were deblended from a single overlapping image) than on clean single objects. These objects can be identified by a nonzero value in the 10th digit of the source status flag.

Known limitations of GSC II image processing

The following section describes the known limitations and problems with the GSC II image processing pipeline. These problems are unlikely to be fixed in a future version of the GSC II.

Sky Processor:

Follows poorly around bright stars, halos, nebulosity, and in very crowded regions. This can affect object detection and photometry.

Object Detection:

Misses small objects that are connected pixelwise to large objects that are too big to be cut out and further processed. This includes bright stars, bright galaxies, and satellite trails. The large objects themselves are incompletely processed (since most of the image processing occurs in image cutouts) and will therefore have inaccurate calibrated values. These objects can be identified by a nonzero value in the ninth digit of the source status flag.

Can detect false objects around bright stars or in areas of the plate where the background is particularly noisy. Very few of these false objects are included in the GSC2.2, since they are fainter than the magnitude cut. Additionally, objects that occur on only one plate and have been classified as a plate defect (which most of these false objects are) have been omitted from the GSC2.2.

Deblending:

There are two known failure modes for the deblender. First, it sometimes incorrectly breaks up an object that should not be deblended: this typically occurs with bright stars but can occur with quite faint objects as well. Second, it sometimes fails to deblend close objects with a large magnitude difference.

The sky background handling for the deblender is inconsistent with the sky background handling from the object detection. Rather than go back to the sky map to determine the sky background, the deblender takes a single constant local sky value as reported from the object detection. This can contribute to missed detections and result in slightly different shape and intensity parameters than inventory would have produced for the object alone.

We have some indications from the photometric calibration task that there is something wrong with the intensity estimates for deblended objects: whether this is a result of the sky handling, or the fractional pixel allocation algorithm, or both, is unclear.

Centroiding:

Due to resource limitations, we did not attempt to centroid small objects that occurred in the same cutout as a large object. These objects, and other objects for which the centroider failed to converge, have a calibrated position based on the barycenter reported by the object detection or deblending tasks, rather than from an elliptical Gaussian fit.

The convergence criteria for the centroider is slightly unphysical (delta P/P for each parameter P) for some parameters. This results in the convergence criteria depending slightly on the location of each object in the cutout (since position is measured from the corner of the cutout). Stray objects inside the cutout array are not zeroed out before the centroider is run and may skew the centroid somewhat.