Classification: Using the Data

STScI ACDSD MASTCASB GSC II Calibrations

Products
GSC
DSS
GSPC
Science

Publications

Data Access
Related Science 
Missions
HST
GEMINI
VLT
NGST
Virtual Observatory
XMM
Facilities
Plate Scanning
COMPASS ooDB 
Staff Pages

Last Updated Jan 2001

Copyright © 2001 The Association of Universities for Research in Astronomy, Inc. All Rights Reserved.

Using the Data Performance

The most important thing to remember about the GSC2.2 classification data is that it discriminates between stars and nonstars - not between stars and galaxies. The nonstar classification includes objects composed of overlapping images on the Schimdt plates. These may be star-star, galaxy-galaxy, or star-galaxy blends.

In principle, nonstars at high galactic latitudes may be taken as galaxies, and nonstars at low galactic latitudes may be taken as blends. In practice, recall that even the highest latitude fields contain a small percentage of blends, and the occasional galaxy can be observed even at low latitudes.

Performance Biases

We have observed the following performance biases:

  • The classifier has a bias towards stars, especially for bright objects (m<16).
  • Deblended "child" objects (identifiable by a nonzero value in the 10th digit of the source status flag) do not perform as well as clean single objects. In examining the output probablities for the single plate classifier, child objects are assigned probabilities close to a "tie" in numbers far out of proportion to their percentage of the plate population. Poor performance of the fractional pixel allocation algorithm has been observed to cause misclassification of deblended stars as nonstars.
  • The classifier was tuned on 15 micron scan data, so its performance is degraded on 25 micron data. This data was excluded from the final export catalog whenever possible, but if only 25um data was available, then it was used. Such objects can be identified by a value of 2 or 3 in the 8th digit of the source status flag.

Interpreting the FITS Table

Classification values are encoded as 0=star, 3=nonstar, for consistency with the GSC1 encoding.

The sourceStatus column of the FITS table also contains information relevant to the classification. This field encodes information about the object one digit at a time. The digits relevant to classification are the 7th and 8th digits of the flag, counting from the rightmost (least significant) digit.

Digit 7 encodes information about the unanimity of the voting process. If this digit is set to 1, this indicates that all counted votes agreed on the classification of the object. A value of 2 indicates that all counted votes agreed that the object was a plate defect. On the assumption that a matched object is real, the classification value for such objects is set to nonstar, but the unanimous defect classification is preserved here as an indication that the object has "defect like" characteristics, for example, high entropy or irregular shape. A value of 0 in this digit indicates a mixed vote.

Digit 8 encodes information about the number and quality of counted votes. This digit will most often be set to 0, indicating that the voting was based on more than one plate observation at 15um resolution. A value of one indicates that only 1 15um observation was available. A value of 2 indicates that the only votes available were based on 25um data, for which the classifier was badly tuned, and a value of 3 indicates that only 1 such vote was available.

Taken together, the information in these digits give some indication of the reliability of the final classification. Objects with unanimous agreement in the voting are likely to be more reliable than objects for which there is disagreement. Similarly, objects whose classifications are based on a single plate are likely to be less reliable than those which had several plates contributing.

A value of 9 in both digits indicates that the object is a Tycho star, in which case no GSC2 data were used.