The Living Caveat Files

User beware! The 2MASS Extended Source Catalog (hereafter, Catalog) has many known problems and features that if not understood by the USER may compromise the science return. Most of these problems are related to data anomalies and image artifacts associated with environmental (atmospheric & thermal effects), instrumental (detector and telescope), software and astronomical (bright stars) phenomena. The first place to start is Appendix H: Data Anomalies.

Additional issues include variable reliability, software bugs, and data quality. See Cautionary Notes.


The Extended Source Catalog is constructed to be reliable (that is, contain real extended sources) for most areas of the sky, including most of the high Galactic latitude sky where the stellar source density is not a critical factor. It fails to be reliable under some circumstances, include

In the vicinity of bright stars (K < 7th mag)

During severe transient phenomena events (e.g., meteor streaks, heavy atmospheric "airglow")

Variable atmospheric seeing

Detector/Instrument variability and telescope focus instabilities

When the stellar source density is high (i.e., the confusion noise is appreciable), including within or near clusters of stars and in the Galactic plane

During the catalog generation phase, we attempt to minimize these sources of unreliability (e.g., masking large areas around bright stars). It is, however, impossible to eliminate all artifacts and data corruptions. Hence, the catalog will inevitably contain false extended sources.

High Source Density

When the stellar source density is large (>103.3 stars deg-2 brighter than 14th at K), and the confusion noise is large (see Appendix C, and Figure C.1), the probability that an extended source is in fact a multiple grouping of stars (e.g., double stars, triple stars, etc) rises dramatically (see Figure 12, e.g.). In constructing the catalog, we have opted to be more complete at the expense of reliability. The Galactic plane represents a challenging yet exciting zone to find new uncharted galaxies and extended sources, which requires "completeness" to have any chance of detecting these highly confused objects.

Minimization of multiple-stellar sources by the USER is possible by thresholding on the "G" score (see Section 4, and the FAQ page for explanation). Other parameters that may be helpful toward eliminating multiple stars include the symmetry measures (see Appendix B for a full list of parameters).

Classification Scores

The user should also understand that the Catalog is really a summed composite of two different catalogs, constructed from two independent "classification scores", the "E" score and "G" score; discussed in Section 4, Star-Galaxy Discrimination and Section 6. The "E" classification is designed to find "extended" sources, which may include galaxies, H II regions, planetary nebulae, and multiple stars (doubles, triples, etc). The "G" classification, on the other hand, is designed to correspond to galaxies, whose colors are predominently redder than stars. The level-1 science requirements are relevant to the "G" score only (hence, it is the most reliable indicator of given source being a real galaxy). Hence, the USER should apply a threshold on the "G" score if galaxies are what they want. Note: even though the "G" score is sensitive to galaxies, it is also an excellent tracer of any extended object; thus, there is appreciable overlap between "G" and "E" sources.


Bright stars are the major source of incompleteness in the Catalog. Large areas around bright stars are masked, including a circular "confusion" radius, vertical/horizontal diffraction spikes, horizontal stripes, persistence ghosts and reflection glints. Discussion of bright stars and their artifacts is given in Section 5. We calculate that approximately 5% of the sky is lost around bright stars in low stellar source density fields, and >20% in high stellar source density fields (the Galactic plane, e.g.).

Loss of sensitivity due to high levels of background emission is another source of incompleteness. This effect is generally minimized in Catalog generation, but can be significant under certain conditions: severe and highly variable "airglow" emission events (particularly at H band this is a problem) which may go undetected during Catalog generation.

Source Confusion

Sources that are in close proximity to each other, including galaxy-galaxy pairs, and more commonly, galaxy-star pairs, may not have "de-blended" photometry or separate identifications. The extended source candidate, in this case, is really an extended source system (galaxy-galaxy) or a contaminated extended source (galaxy + star). Not only will the photometry be compromised, but the source position is liable to be incorrect (or it simply may reflect the position of the "system" itself). A good indicator of "confusion" are the photometry flag parameters (see Appendix B), which indicate masking due to stellar contamination (or from some other means, such as a bright star being nearby).

Sky Coverage & Pixel Masking

For any given extended source, there may be local image pixels that were masked due to nearby bright stars, detected point sources, or previously processed sources (see further down). We attempt to minimize contamination from neighboring stars by masking their emission (masked area is a function of the PSF and the brightness of the contaminent source). These masked pixels are later replaced with an "isophotal"-derived flux in an attempt to recover the flux of the galaxy lost in the masking procedure; see Sections 2 & 3 for further explanation. After an extended source has been extracted, an elliptical area encompassing the source is masked from the images, leaving a hole in the images; these are referred to as "previously processed sources".

There is currently no way to tell how much area in and around the extended source is masked. The photometry flags, however, do indicate if the given photometry aperture contains at least one masked pixel (in which case the flag is tripped and it given a non-zero value, coded according to the nature of the masked pixel). In future versions of the Catalog, there will contain "mask" images mapping the exact area around extended sources that were masked (and subsequently recovered with isophotal substitution).

Image Background Problems

Backgrounds are tracked on spatial scales of ~3 arcminutes, which eliminates most kinds of background variations seen in the infrared (see Section 3 for detailed explanation of the background fitting procedure). There still remains cases in which the background varies on higher frequency scales and thus the background is not fully removed from the images. Atmospheric airglow is the primary culpret, and instrumental (i.e., array detector) variability is another factor. These data anomalies are described in Appendix H: Data Anomalies. Improperly subtracted background emission has a very large effect on the photometry, effectively adding a very large systematic flux (bright emission "bumps" are more likely to pass flux thresholding, so normally faint objects will have systematically larger integrated fluxes). Background "features" can also induce detection or mis-classification of false sources (both "artifacts" and sources that are not extended in reality). The Catalog does contain sources that have additional flux due to an incorrectly subtracted image backgrounds, many of which are systematically brighter than their true flux would have them.


Apertures from which integrated fluxes are derived are vulnerable to a number of data anomalies and image artifacts, including source confusion, pixel masking, background variation (both real and instrumental), and sky (background) noise. Discussion of the photometric error tree is given in Appendix A, which describes some of these components. Some components, however, are difficult to quantify and fold into the estimated flux uncertainties. The instrumental "banding" phenomenon (see Data Anomalies), for example, is difficult to measure and even more difficult to detect automatically (that is, know that it is there and correct for it). For the first Catalog, no attempt is made to correct for this particular data feature. There will be corrections for this systematic in future Catalog releases. Another phenomenon that is difficult to measure is the airglow power and frequency. Again, we do not attempt to further correct 'airglow' other than the standard background fitting procedure. For the future, we hope to identify cases in which severe airglow events are identified and re-scheduled for observation.

 The estimated flux uncertainty does include the background "poisson" component, the measured flux calibration uncertainty and flat-fielding uncertainties. Additional systematics (e.g., global calibrations variations) are not included in the flux uncertainties. For future release, we will attempt to fold in known systematics and random errors.