A 24 μm POINT SOURCE CATALOG OF THE GALACTIC PLANE FROM SPITZER/MIPSGAL

The MIPSGAL image tile products are extremely uniform integration depth maps of 24 μm flux density, but robust point source detection is nontrivial because of nonuniform background emission across the Milky Way. For a large survey such as MIPSGAL, automated data analysis techniques are essential. However, many automated point source detection techniques produce substantial numbers of false detections among the filamentary emission structures of the nebulae surrounding recent star forming events. Here, we have adopted the IDL program PhotVis (version 1.10) to robustly identify point-like sources regardless of the complexity of the background (Gutermuth et al. 2008).

PhotVis employs a modified version of the DAOFIND source detection algorithm (Stetson 1987), as implemented in the IDL Astronomy Users' Library (Landsman 1993). In summary, the DAOFIND technique involves convolving each image with a "sunken" two-dimensional Gaussian function sized to match the beam size of the observations (for this work, 625 FWHM). This convolution concentrates the flux of unresolved structure into the central pixels of that structure, while large scale structure effectively convolves to a value near zero. Ideally, the convolved image makes stellar sources easy to identify with a simple threshold search. Unfortunately, numerous false sources are found by this algorithm among filamentary and other nonuniform structure in the bright nebulosity associated with the Galactic plane, and regions of star formation more generally (e.g., Megeath et al. 2004). In PhotVis (v1.10), the standard DAOFIND algorithm has been enhanced to include empirical estimation of a noise map for the Gaussian-convolved source detection image. Specifically, the absolute value of the convolved image is boxcar median-smoothed with a box size of five times the FWHM of the point-spread function (PSF). The original Gaussian-convolved image is then divided by this noise map, effectively converting the search threshold from a signal-based threshold into an approximate local signal-to-noise ratio (S/N)-based threshold. We use a threshold value of seven in the local noise map scale, based on considerable testing on MIPS 24 μm data of star-forming regions (e.g., Gutermuth et al. 2008, 2009; Beerer et al. 2010; Megeath et al. 2012). The resulting algorithm simultaneously achieves excellent sensitivity in dark, uniform, uncrowded regions of images and automatic adaptation to less sensitive local conditions, largely mitigating the production of false sources associated with nebulous structure (Gutermuth et al. 2008).

Once sources are found, their flux is measured using synthetic aperture photometry via aper.pro from the IDL Astronomy Users' Library (Landsman 1993). The MIPSGAL tiles are made of merged observations at a range of spacecraft rotation angles, thus we chose to use aperture photometry rather than PSF-fitting photometry due to its computational simplicity and measurement robustness under that circumstance. We adopt aperture and inner and outer sky annulus radii of 635, 762, and 1778 respectively, and a magnitude zero point of 14.525 mag (Vega standard) for a 1 Digital Number per second (DN s⁻¹) source observed at 24 μm (Gutermuth et al. 2008).² The photometric uncertainty is derived from calculations of the shot noise in the aperture and shot noise and internal variance in the sky annulus pixels that are used to compute the background emission per pixel for subtraction from the aperture flux. An internal noise floor of 0.02 mag is enforced to prevent rare data anomalies from yielding untenable uncertainty estimates.

Finally, as a characterization of source quality, we compute the FWHM of each source. As noted above, calibration of aperture photometry includes a correction for the finite sampling of the PSF set by the choice of aperture and sky radii. If an object is intrinsically resolved beyond the instrument resolution, then the source would be of relatively poor photometric quality in our catalog because the aperture correction would be incorrect. The measurement algorithm used is entirely empirical, extracting the half of peak flux radial distance from a cubic spline interpolation of the radial profile (Barth 2001). We azimuthally average (by median) the radial profile before running this algorithm to improve measurement success probabilities near structured nebulosity.

Once the source lists and photometry have been obtained from all of the individual tiles, we combine them into a unified survey "archive" data product. The tiles were constructed with some degree of overlap, thus duplicate detections near tile edges are common and must be identified and removed. Once astrometry systematics were treated (see Section 2.3), a simple angular offset tolerance of 1'' is used to identify all inter-tile duplicates. For each set, the instance of the source that is furthest from tile edges is selected to represent that source in the final combined source list as this maximizes the coverage of the sky annulus and the surrounding area for the noise map calculation. The resulting tally of detections in the final archive that have $\lt 0.33\;{\rm mag}$ uncertainty at 24 μm is 1,353,228. This requirement is approximately a S/N of 3, significantly lower than the approximate S/N $\gt 7$ limit mentioned above for our empirically derived noise map in the source identification process. The photometrically determined uncertainty is generally somewhat higher because it includes photon shot noise. Ultimately, the photometric S/N limit is a sensitivity limit, but not where the survey is complete, as we will explore in Section 4, below. In Figure 1, the variations of magnitude uncertainties (top) and FWHM (bottom) with magnitude for the archive sources are expressed as two-dimensional histograms. The spread in magnitude uncertainty for a given value of [24] simply reflects the variation of backgrounds throughout the MIPSGAL field.

Zoom In Zoom Out Reset image size

Via automated queries to the Vizier online catalog service, we obtain all of the 2MASS, GLIMPSE, and WISE sources that fall within each tile. These are matched to our MIPSGAL archive such that the closest match within an angular tolerance of $2^{\prime\prime}$ is linked to each 24 μm source. The matching tallies for each data source are summarized in Table 1. A counterpart is found in at least one of these catalogs for over 94% of sources in the archive. To gauge mismatch rates for each catalog, we performed simple Monte Carlo tests across the entire archive product. Taking the number of objects within 635 diameter reported for characterization of potential beam contamination for each archive source (reported in the source table), we compute the mean density of sources near each object. Note that this includes the matched source. Thus if this is a true match, we will be overestimating the field density somewhat (10–50% for GLIMPSE and 2MASS, 100% for WISE, typically). We then multiply that density by the area corresponding to the smaller 2'' matching radius to determine a mean number of contaminators to expect for that source. Using the mean contaminator rate, we pull random Poisson deviate numbers of potential contaminators for each object. For each object with a non-zero synthetic contaminator count ${{n}_{i}}$ in a given realization, we draw that number of uniform, area-weighted deviates (i.e., radius PDF = 2r, per the classic dartboard problem) and compare the smallest value to the radial separation, ${{r}_{{\rm match}}}$, of the actual match for the archive source. If the nearest false source is within ${{r}_{{\rm match}}}$ + 01 we count that as a possible contamination event in the test. We then integrate contamination counts over the entire archive, over 1000 trials. The resulting estimated mismatch rate is ∼0.1% for each catalog (see Table 1).

Source quality flags are compiled for each source, including the FWHM (described above), a binary flag to note sky annulus overlap with image edges, coverage edges, or saturated pixels, and source proximity among nearest neighbor archive members, in arcseconds. An internal confusion flag based on the nearest neighbor distance and the difference in 24 μm magnitude between source and neighbor is described in Section 2.3. We also tabulate the number of objects in each external catalog that fall within 635 of the sourceʼs centroid position.

Initial efforts to incorporate publicly available external catalogs with our 24 μm archive revealed systematic offsets in the astrometric calibration of the MIPSGAL tiles. These offsets are shown in Figure 2 as astrometric residuals between the GLIMPSE and 24 μm centroid positions (${\Delta\ R}{\rm .A}{\rm .}(i)=[{\rm R}{\rm .A}{{{\rm .}}_{{\rm MG}}}(i)-{\rm R}{\rm .A}{{{\rm .}}_{{\rm GL}}}(i)]$ ${\rm cos} \,({\rm decl}{{.}_{{\rm GL}}}(i));{\Delta\ decl}.(i)\,=\,$ ${\rm decl}{{.}_{{\rm MG}}}(i)-{\rm decl}{{.}_{{\rm GL}}}(i))$ for all unconfused matches (specifically, we require one unique GLIMPSE source within $2^{\prime\prime}$, and no other GLIMPSE sources within 635) in the archive as a function of Galactic Longitude. Many of these offsets are greater than 05 and much larger than the expected random error between centroid differences. The bulk of the deviations can be cured with a constant R.A.–decl. offset corresponding to the median of the offsets in each tile. These tile-specific offsets have been applied to each tileʼs source catalog. Figure 2 shows the residuals after the application of the offset. The applied offsets and the final rms residuals in R.A. and decl. for each tile are recorded in Table 2. A similar issue was reported in the Galactic center MIPS coverage in Hinz et al. (2009), and was addressed in a similar manner, using 2MASS for reference astrometry instead of GLIMPSE. Additional astrometric systematics internal to many tiles are present, but treating these would most likely require rebuilding the tiles from the BCD data products.

Zoom In Zoom Out Reset image size

We identified a secondary issue related to astrometry in the archiveʼs nearest neighbor distance (${{d}_{{\rm NN}}}$) distribution shown in Figure 3. The functional form of the distribution is approximately log-normal, with a narrow true normal excess centered on 10'' angular separation. That distance corresponds to the central radius of the first diffraction ring outside of the Gaussian core of the MIPS 24 μm PSF, suggesting that one of the pair could be a false identification. Moreover, such a feature can skew the photometry and astrometry of faint sources that fall near the feature. The magnitude difference between each source and its nearest neighbor in the archive versus ${{d}_{{\rm NN}}}$ is displayed in Figure 4(a). The same data are plotted for those objects without and with GLIMPSE counterparts within 2'' in Figures 4(b) and (c), respectively. The distribution of magnitude differences for sources without GLIMPSE counterparts exhibits clear excess source counts in three distinct locations: $-0.2\lt {\Delta}$[24]$\;\lt \;0.2$ and ${{d}_{{\rm NN}}}\lt 8^{\prime\prime}$, ${\Delta}$[24] $\gt \;0.8$ and 9''$\lt {{d}_{{\rm NN}}}\lt 11^{\prime\prime}$, and ${\Delta}$[24] $\gt \;2.8$ and 25''$\lt {{d}_{{\rm NN}}}\lt 27 \buildrel{\prime\prime}\over{.} 5.$ This excess is further illustrated in Figure 4(d), which shows the ratio of the magnitude differences of sources without and with GLIMPSE counterparts and normalized by the expected ratio uncertainty, assuming Poisson counting statistics. Guided by this figure, where the grayscale has been set to mark $\gt 3\sigma$ regions as black, we define the conditions for the internal confusion flag. The conditions and the source counts affected are listed in Table 3.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 3.  MIPSGAL Confusion Conditions

Region No.	Conditions	No. w/o GLIMPSE	No. w/GLIMPSE	Ratio
1	${{d}_{{\rm NN}}}\lt 5^{\prime\prime}$ or (${{d}_{{\rm NN}}}\lt 8^{\prime\prime}$ and $|{\Delta}$[24]$|\lt 0.2\;{\rm mag}$)	2083	7197	0.290
2	80 $\lt {{d}_{{\rm NN}}}\lt 11 \buildrel{\prime\prime}\over{.} 5$ and ${\Delta}$[24]$\gt 0.8\;{\rm mag}$ and ${\Delta}$[24]$\gt$(0.4(${{d}_{{\rm NN}}}$ −11.5)+1.8) mag	3284	10767	0.305
3	25''$\lt {{d}_{{\rm NN}}}\lt 27 \buildrel{\prime\prime}\over{.} 5$ and ${\Delta}$[24]$\gt 5.4\;{\rm mag}$	355	600	0.591
Ref.	13''$\lt {{d}_{{\rm NN}}}\lt 25^{\prime\prime}$ and $|{\Delta}$[24]$|\lt 5.4\;{\rm mag}$	35,655	393,842	0.091

Download table as:  ASCIITypeset image

The archive data product is meant to be an inclusive list of 24 μm point-like sources extracted from the MIPSGAL survey. A higher reliability subset of the archive sources, the "catalog" data product, is selected to mitigate the systematic issues in the archive discussed in Sections 2.2 and 2.3. First, we impose a more stringent $\lt 0.20\;{\rm mag}$ uncertainty requirement (S/N ∼5). Then, we require a confined range of source FWHM, with thresholds that are flared to a wider range for dimmer sources to allow for their larger FWHM variance. These two stricter limits are drawn in Figure 1, and they yield the vast majority of the archive objects that get culled from the catalog, roughly evenly divided between the two constraints. In addition, we require that the binary confusion and edge flags must be zero to ensure that these relatively rare instances are also culled from the catalog. All of the requirements for catalog inclusion are listed in Table 4. There are 933,818 sources that meet these more restrictive requirements. A counterpart in at least one of WISE, 2MASS, or GLIMPSE is found for over 98% of sources in the catalog, compared to 94% in the archive.

Table 4.  MIPSGAL Catalog Requirements Summary

$\sigma [24]\lt 0.2\;{\rm mag}$
$|{\rm FWHM}-6 \buildrel{\prime\prime}\over{.} 25|\lt 0 \buildrel{\prime\prime}\over{.} 5$ $(1+0.125\times [24])$
Internal Confusion Flag = 0
Edge Flag = 0

Download table as:  ASCIITypeset image

In order to verify the photometric performance and calibration of our source extraction process, we merged the MIPS 24 μm photometry of red sources provided in Robitaille et al. (2008, R08) with our catalog. The base image data set is the same in both cases, but R08 used the original Spitzer Science Center pipeline-reduced mosaics for their photometry instead of the enhanced MIPSGAL-reduced tiles. Regarding source extraction, they also used PSF fitting photometry by hand, instead of automated aperture photometry as we have done here. Of the 18,949 red GLIMPSE sources in the R08 catalog, 16,469 have reported MIPS 24 μm fluxes and uncertainties. Matches for 16,079 of those sources are made within the archive product (97.6%), and 14,926 matches (92.8%) with the catalog product. The magnitude residuals between the R08 photometry and ours are plotted in Figure 6. The 2D histogram grayscale shows the magnitude residuals to R08 matches in the archive, and the contours represent a similar 2D histogram that uses the catalog product instead. The mean zero-point calibration offsets are $-0.07$ and −0.09 mag, and the rms deviations are 0.19 and 0.14 mag for the archive and catalog products, respectively. In summary, we find that our photometry agrees well with the limited photometric sample of R08.

Zoom In Zoom Out Reset image size

As a secondary check, we merged our archive with the MIPS 24 μm photometry of the Galactic center region from Hinz et al. (2009, H09). As with R08, the image data sets are the same as ours, but the image data treatment and source extraction differ. In this case, the image data were processed with the MIPS instrument teamʼs DAT software (Gordon et al. 2005), and the photometry was extracted via PSF fitting. The benefit of this reference catalog over that of R08 is that it is a complete catalog of 24 μm sources from the region in question, instead of targeted photometry of red 2MASS and GLIMPSE sources across the entire inner Milky Way. As such, it is a good test of our completeness within one of the most challenging parts of the survey. Of the 120,883 sources in Hinz et al. (2009), we have 82,832 and 68,608 coincident sources in our archive and catalog products, respectively. Obviously, this is a substantial miss rate. In Figure 7, we plot the magnitude residuals versus magnitude in the top plot, demonstrating largely consistent photometry among matches. Thus while the photometry appears to agree, the issue of the discrepant sources demands further characterization.

Zoom In Zoom Out Reset image size

In order to fairly examine the sources within a well-covered region, we first crop both the H09 catalog and our archive to an easily defined common coverage region of $-3\lt l\lt 4$ and $|b|\lt 0.5$. Within this region, we find 63,129 and 51,937 sources from the H09 catalog and our archive product, respectively. Among those two source lists, 39,910 sources match. The bottom panel of Figure 7 shows the relative detection fraction per 1 mag bin among the matched sources (solid line), those found in our archive but missed by H09 (dotted–dashed line), and those missed in our archive but found in H09 (dashed line) with the common coverage region. In the brightest bin, we see a clear deficit of H09 sources. Generally, those with marginally detectable peak saturation are rejected by the PSF fitting of H09 but are included in our archive. The range $1\lt [24]\lt 6\;{\rm mag}$ exhibits consistent behavior: 70% matched sources, 10% H09-only sources, and 20% H09-missed archive sources. At $[24]\gt 6\;{\rm mag}$, the fraction of sources rapidly becomes dominated by the H09 source counts, as our archive loses completeness (characterized in detail in Section 4, below). We visually inspected some of the faint H09 source positions in the MIPSGAL tiles and found that the vast majority of those that we viewed are not apparent in those data. This effort was sufficient to cement our confidence in our methodʼs omission of these fainter sources. Further investigation of the veracity of the faint H09 sources is beyond the scope of this paper.

Finally, the merger of the MIPSGAL archive with the all-sky WISE catalog enables a check for general agreement between our photometry and the WISE 22 μm photometry on a larger sample of objects. We find that 368,956 objects have reported $\lt 0.33\;{\rm mag}$ uncertainties in both the MIPSGAL archive and WISE 22 μm catalogs. We plot the magnitude residuals versus WISE 22 μm magnitude in Figure 8. The median residual for bright sources ([24] $\lt 3\;{\rm mag}$) in the "Disk, Off-Plane" field is −0.07 mag, similar to the offset to the MIPSGAL photometry reported in R08 and discussed above. The bias toward brighter values in the faint source WISE photometry is frequently observed in lower relative resolution data, where structured nebular emission is more likely to contaminate the photometric aperture relative to the surrounding sky in some sources, resulting in background flux underestimation and source flux overestimation (e.g., Gutermuth et al. 2009). In this case, the WISE 22 μm fluxes of some sources are found to be as much as 3 mag brighter than the MIPSGAL photometry.

Zoom In Zoom Out Reset image size