Star-Galaxy Discrimination in the Galactic Plane:
Memo 1: 9th mag Galaxy Placed within MSX

T. Jarrett, IPAC

For the MSX field, stars and galaxies are not easily separated using routine parametric tests, such as the image moments, radial profile and central surface brightness. The stellar number density is so high that doubles & triples completely overwhelm the distribution (see e.g. my memo on expected number of multiples for the galactic plane, Expected Number of Multiple Star Systems as a Function of Galactic Latitude ). GALWORKS has a number of parametric tests designed to weed out double stars and a large variety of triple stars.

These "triple killer" parameters require, however, some careful tuning to minimize their (unwanted) affect upon real extended sources -- galaxies. For low stellar density fields (e.g., Coma) the triple killers have very low thresholds to the point that they are effectively turned off. As the stellar density increases, the thresholds are correspondingly increased until the density is so high that the triple killers are the primary star - galaxy dicrimanent. This is the case for the MSX field which is located within a degree or so of the plane at a galactic longitude of 30 degrees.

Inspection of the MSX coadds reveals a multitude of star/star systems with any number of relative brightness combinations. Finding any galaxies amongst this tangled mess of stars will be a great challenge. In order to perform some preliminary analysis of star-galaxy discrimination with these coadds, it is therefore necessary to "add" real galaxies to the field since there are plenty of stars but no galaxies to compare (note: we do not expect to see many galaxies in the plane due to the innumerable complications associated with looking down the throat of the Milky Way; however, we will have the advantage of shear area of coverage to find any bright galaxies that may be hiding behind our very own).

An excellent field to "add" to our MSX field would be the COMA core. COMA is located near the NGP, so the stellar density is low (thus we do not contaminate the MSX field). It has a high density of galaxies, ranging from very bright (11th mag) to the very faint all located with one coadd if need be. A Coma coadd, after applying a uniform attenuation to mimic the extinction from the plane could be added to each msx coadd, thus providing a means at repeatability: 23 coadd images, thus 23 repeats of one Coma coadd. At this time, the COMA data is not yet ready for use in this experiment. At a later data, I will add the Coma data to MSX and perform the analysis. ** UPDATE !!! See Star-Galaxy Discrimination in the Galactic Plane: Memo 2: MSX + Coma. **

For now, I will instead add one bright galaxy to each MSX coadd. The galaxy I have chosen comes from the M51 scan and is located a few degrees south of the M51 system. It has a K mag of about 8.9 and a J mag of 9.9:

To mimic extinction, I will apply an attenuation factor to the J and H images, leaving K as is. Assuming a uniform visual extinction of 5 magnitudes, the J image is attentuated by 0.86 mag compared to the K image, and H is attentuated by 0.32 mag compared to K (e.g., A(H) - A(k) = 0.20 * A(v) / R ). I do not attempt to model variable extinction (which undoubtedly exists in the MSX field) or the change in extinction as we move from the bottom of the scan (glat = 1 degree) to the top of the scan (glat = 3 degrees). These sorts of complications are second-order effects in a admittedly first-order experiment. Our goal is to measure the parametric characteristics of the galaxy and compare with stars and multiples. In this way we can better tune the thresholds to eliminate triples stars without eliminating the galaxy. Obviously having only one galaxy to compare with is very limiting, but it is a start (and it is better than nothing, which we currently are facing). Later, when the Coma data becomes available, will have much better statistics and range in parameters to help in setting the triple-star killers. In addition, we will have repeatability information which can be used to measure variation in photometry and other important parameters (e.g., elliptical parameter: axial ratios and positition angles).

In order to increase our repeatability information, I will also add the galaxy twice to each MSX coadd, one located in the upper right part of a msx coadd, and the other located in the lower left part of the same msx coadd. The following image is the first K-band coadd in the scan with the addition of the galaxy.

The galaxy is easily visible in the image (upper right, just above the bright star, and lower left, above another brightish star). Remember that is is about 9th mag, which if we factor in extintion (assuming it is 5 mags of visual extinction -- just a guess, but not particularly important to the outcome of this experiment), would imply that it is about 8.5 mag or so in reality -- a very bright galaxy indeed. Given it size and brightness, we should detect this galaxy every time -- EXCEPT when it is located near a very bright star in which case it can be "blanked" away along with the star. Bright star blanking is unavoidable and devastating to anything in its vicinity as will be shown below. Although we should detect this galaxy every time (twice every coadd), except as noted above, we do not expect that it will "look" the same everytime due to star contamination. Sometimes it will look brighter, sometimes more lumpy, which should be reflected in the parametric information.

Bright Star Blanking

One of the first operations that GALWORKS performs is bright star blanking, which is necessary in order to accurately compute the background and to eliminate false detections associated with the numerous artifacts generated by bright stars. See GALWORKS Bright Star Cleansing for more information on this topic.

MSX has many bright stars in each coadd; thus a large fraction of each coadd is blanked away, particularly at K band. The following images show typical blankings in the MSX scan (note: I have applied a blanking algorithm in which the thresholds for blanking have been increased according to the stellar density, in this case, the thresholds have been raised about 1 mag, thus fewer stars are blanked; for more information on this step, see Bright Stars in the MSX Field .

Notice that the "bright" star in the lower right corner (in which the horizontal diffract spike is blanked, as well as persistence ghosts) is not really all that bright. The mag value assigned to this star (not by GALWORKS) is apparently quite a bit off -- a large fraction of area is unnessarily blanked away. This is one of the hazards of bright star blanking: if the input mag is wrong (either to faint or too bright) it can have a wide impact on the processing within GALWORKS. This sort of finding is probably due to initial problems with the pipeline processing that have yet to be shaken from the system. Remember that this 3-channel data is fresh off the telescope and thus is not yet processed in an optimum fashion.

Both apparitions of the galaxy due survive in this coadd, however. A closer look at the processing steps for analsis of this galaxy are shown in the next image:

Another example of a coadd after bright stars have been blanked:

Notice that the galaxy in the lower left part of the K coadd has now been blanked due to a nearby bright star. The upper right galaxy is nearly zapped by the diffraction spike blanking of yet another bright star. Fortunately, in the J band because the stars are fainter the bright star blanking is much less intrusive and the galaxy(s) survive this process. It clearly appears that our K-band bright star blanking is working too hard. The thresholds should probably be raised at least another 0.5 mag or so. This finding is of course preliminary pending a more mature version of the pipeline processor.

A final example of a coadd after bright stars have been blanked:

The galaxy survives the blanking process in both J and K. A closer look at the processing steps for analysis of the bright galaxy in this coadd set:

Star - Galaxy Discrimination

A brief description of the various scoring parameters to discriminate stars from galaxies can be found in Star - Galaxy Discrimination Parameters and additional information may also be found in the GALWORKS SDS.

The following plots show the multiple detections of the bright galaxy as a function of its integrated flux (circular fix aperture, radius = 10) and the various discrimination parameters. The typical K flux (radius = 10) of the galaxy is 9.6, which is about 0.7 mag fainter than its actual total flux (i.e., the galaxy is much larger than a radius of 10 pixels). The plots also show the candidate extended sources -- for MSX these are all multiple star systems (typicall triples or worse). Remember this scan has thousands of stars, most of which have been culled away with the star - galaxy discrimination parameters. The stars that remain have a score(s) that are above the thresholds in at least one band. The plots can be used to "tune" the scoring thesholds in order to eliminate the star systems with minimal effect upon the real galaxies.

"sh" Score

The real galaxy is clearly brighter than any of the false detections, but its "sh" score does not stand out in any way. This result is to be expected since the false detections are all multiple star systems, which always appear extended in "sh". The same result is seen with the "mxdn" score and the image moment(s) scores.

Notice that the galaxy flux seems to vary from 10 to 20%. This is due to star contamination. There are at least two very deviant points, generally on the faint side -- this is due to too much flux being subtracted (star contamination again) during the GALWORKS processing star subtraction step. The opposite can also occur in which stars are not fully subtracted in which case the galaxy appears to bright (see the J results).

Also notice that there are few points beyond K = 13 mag due to a flux threshold set early on in GALWORKS (it is tied to the stellar density; the GALWORKS SDS explains this step).

"wsh" Score

The "wedge" shape is designed to minimize contamination from double stars. The plot shows that most of the false detections (as well as the galaxy) sit well above any reasonable threshold (typical score threshold for high stellar density field would be around 8). Again, this result is consistent with the actual stellar distribution in which we are limited by triple stars and worse. The "wsh" is still a very effective parameter for eliminating nearly all sources early in the GALWORKS processing (remember doubles dominate multiple systems in shear numbers; see Expected Number of Multiple Star Systems as a Function of Galactic Latitude ).

"r23" Score

The "r23" score is very effective against triple stars. As you can see from the plot, most of the false detections have scores less than 10. This parameter, however, can also eliminate plenty of real galaxies (this is based on past experience and simulation results). Note the the galaxy itself has scores ranging from 10 to 20 at K -- not very good considering it is a big bright galaxy. The effectiveness of this parameter can be better judged once I have more galaxies at hand -- this experiment altered to include Coma should provide valuable insight in this matter.

"vint" Score

Similarly "vint" is an effective parameter separating galaxies from multiple stars. However, like "r23" this parameter (threshold) must be tuned carefully to avoid eliminating any real galaxies.

"trip" Score

The "trip" score is a weighted combination of the "wsh", "r23", "vmean" and "vint" scores, with "wsh" having the largest weight. Although the four scores are slightly correlated, the combination does provide a convenient measure of the "extendedness" of an object. The plot shows that the galaxy has a score ranging from 10 to 15 in K (slightly higher in J and H) and most of the false detections are significantly smaller than this range. Note that some false detections have high scores no matter how you measure them, the reason being their multiple nature (i.e., they consist of more than 4 stars each -- any way you look at them they are extended). Again, this parameter (threshold) must be tuned carefully to avoid eliminating any real galaxies. With the addition of the Coma galaxies, we will have a much better handle on the thresholds that can be set (and thus we can then compute reliability of the sample).

Photometric Repeatability

The galaxy was detected 37 times out of a total of 44 (22 coadds X 2 apparitions) possible detections, giving repeatability of 84%. This is rather good considering the density of stars. The following plots are histograms of a few key photometric measures: fixed circular aperture, with radii = 10 and 20, adaptive elliptical and circular apertures, and circular petrosian and isophotal apertures.

The photometry repeats quite well (< 10%) for the fixed circular measures if you ignore the sparse but very deviant points. For the adaptive apertures the repeatability is not nearly so good (10 to 20%, again ignoring the very deviant points) which is not too surprising given the very nature of this method (the adaptive aperture is designed to measure the "total" flux, and thus the aperture can grow quite large for bright galaxies; note that the total mag derived for the galaxy is close to its actual value of 8.9 at K). The adaptive circular photometry has less dispersion than the adaptive elliptical photometry -- consistent with the fact that the galaxy is mostly circular (see below) and adding a ellitpical aperture imparts an uncertainty into the computation.

Finally, the prtrosian and isophotal photometry appears repeat to within 10 - 15% and 10%, respectively. In addition, there are very few deviant points.

These results suggest that the isophotal photometry gives the most precise answer, with minimal gross deviant points.

Although the galaxy appears to be mostly circular, it does have a slight flattened orientation. The mean axial ratio is about 0.85. The histogram below shows the repeatability of the axial ratio measurement.