Siena Galaxy Atlas 2020

Hyperleda Catalog

To construct the initial galaxy catalog, we query the Hyperleda extragalactic database for galaxies with angular diameter \(D(25)>0.2\) arcmin, where \(D(25)\) is the diameter at the \(25\ \mathrm{mag\ arcsec}^{-2}\) surface brightness isophote (in the optical, typically the B-band), a traditional measure of the "size" of the galaxy popularized by the Third Reference Catalog of Bright Galaxies (RC3).

We execute the following query on the 2018 November 14 version of the Hyperleda database, resulting in a catalog of 1,436,176 galaxies:

WITH
"R50" AS (
  SELECT pgc, avg(lax) AS lax, avg(sax) AS sax
  FROM rawdia
  WHERE quality=0 and dcode=5 and band between 4400 and 4499 GROUP BY pgc
),
"IR" AS (
  SELECT pgc, avg(lax) AS lax, avg(sax) AS sax
  FROM rawdia
  WHERE quality=0 and iref in (27129) and dcode=7 and band=0 GROUP BY pgc
)

SELECT
  m.pgc, m.objname, m.objtype, m.al2000, m.de2000, m.type, m.bar, m.ring, m.multiple, m.compactness, m.t,
  m.logd25, m.logr25, m.pa, m.bt, m.it, m.kt, m.v, m.modbest, "R50".lax, "R50".sax, "IR".lax, "IR".sax,
FROM
  m000 AS m
  LEFT JOIN "R50" USING (pgc)
  LEFT JOIN "IR" USING (pgc)
WHERE
  objtype='G' and (m.logd25>0.2 or "R50".lax>0.2 or "IR".lax>0.2)

Based on a large number of visual inspections and both quantitative and qualitative tests, we cull the resulting sample by applying the following additional cuts:

1. We limit the sample to \(0.333<D(25)<180\) arcmin, which removes roughly 900,000 galaxies (approximately 65% of the original sample), including the Magellanic Clouds and the Sagittarius Dwarf Galaxy at the large-diameter end. After this cut, the largest angular-diameter galaxies which remain are NGC0224=M31 and NGC0598=M33 with \(D(25)\) diameters of 178 and 62 arcmin, respectively.

   Among smaller systems, we implement the \(D(25)<20\) arcsec cut because we find that the fraction of spurious sources (or sources with incorrect diameters) in Hyperleda increases rapidly below this diameter; moreover, galaxies smaller than this size are modeled reasonably well as part of the standard Tractor pipeline (see Tractor implementation details).

2. We remove approximately 3800 galaxies with no magnitude estimate in Hyperleda (as selected by our query), galaxies which we find to be largely spurious based on visual inspection.

3. We remove an additional roughly 6500 spurious sources (or galaxies with significantly overestimated diameters) based on visual inspection.

4. Finally, we reject approximately 1700 galaxies whose primary galaxy identifier (in Hyperleda) is from either SDSS or 2MASS and whose central coordinates place it inside the elliptical aperture of another (non-SDSS and non-2MASS) galaxy with diameter greater than 0.5 arcmin. Based on visual inspection, we find that many of these sources are due to shredding or are spurious sources with grossly over-estimated diameters.

In addition, we visually inspect all galaxies in the sample with \(D(25)>0.75\) arcmin, including all the NGC/IC galaxies, and assess their published elliptical geometry and coordinates. Where necessary, we update the diameter, position angle, minor-to-major axis ratio, and, in some cases, central coordinates "by hand", as indicated in the BYHAND column described in the SGA-2020.fits catalog.

We note that the NASA/IPAC Extragalactic Database (NED) proved invaluable for these cross-checks.

Supplemental Catalogs

To improve the completeness of the Hyperleda catalog, we supplement the sample with several additional catalogs:

1. We add the sample of Local Group Dwarf Galaxies from McConnachie (2012), making sure to account for any systems already in the Hyperleda catalog. Using visual inspection, we determine that approximately half these systems are insufficiently resolved to be part of the SGA-2020 (e.g., Ursa Minor), and so we remove them from the sample.

2. Next, we identify the sample of galaxies in the RC3 and OpenNGC catalogs which are missing from the Hyperleda sample. Surprisingly, many of these systems are large and have high average surface brightness.

3. Finally, we use the DR8 photometric catalogs to identify additional large-diameter galaxies. This supplemental catalog consists of two subsamples:

   1. First, after applying a variety of catalog-level quality cuts (and extensive visual inspection), we identify all objects in DR8 with half-light radii \(r(50)>14\) arcsec based on their Tractor model fits;

   2. And second, we construct a candidate sample of compact galaxies which would otherwise be forced to be point sources in DR9 based on their Gaia catalog properties (see this notebook for details).

Final Parent Sample

The final parent sample contains 531,677 galaxies approximately limited to \(D(25)>20\) arcsec, spanning a wide range of magnitude and mean surface brightness. Of these, 383,620 have grz imaging from DR9 and end up in the final SGA-2020 catalog (see Custom Mosaics & Ellipse-Fitting).

Group Catalog

Galaxies which are close to one another (in apparent, angular coordinates) must be analyzed jointly. Consequently, we build a simple group catalog from the Final Parent Sample using a friends-of-friends algorithm and a 10 arcmin linking length, taking care to ensure that galaxies which overlap (within two times their circularized \(D(25)\) diameter) are assigned to the same group.

Using this procedure, we identify 512,825 unique groups, of which roughly 93% have just one member. Among the remaining groups, approximately 15,000 have two members, 1585 groups have 3-5 members, 51 have 6-10 members, and just four groups have more than 10 galaxies, including the center of the Coma Cluster.

Custom Mosaics

We run the DR9 pipeline on a "custom brick" based on the estimated center and diameter of the galaxy group (using GROUP_RA, GROUP_DEC, and GROUP_DIAMETER defined in SGA-2020.fits). Specifically, we generate mosaics according to the following criteria:

• For groups with GROUP_DIAMETER\(\,<14\) arcmin we use a mosaic diameter of \(3\, \times\) GROUP_DIAMETER;

• For groups with \(14\,<\) GROUP_DIAMETER\(\,<30\) arcmin we use a mosaic diameter of \(2\, \times\) GROUP_DIAMETER;

• And for groups with GROUP_DIAMETER\(\,>30\) arcmin (which only affects NGC0598_GROUP) we use a mosaic diameter of \(1.4\, \times\) GROUP_DIAMETER.

In all cases, for the grz imaging we adopt a fixed pixel scale of 0.262 arcsec/pixel and for the unWISE mosaics we use 2.75 arcsec/pixel.

Unlike in DR9, we use a couple different options when calling the legacypipe photometric pipeline:

1. We invoke the --fit-on-coadds option, which triggers the following specialized behavior:

   • After reading the individual, sky-subtracted CCD images and rejecting outlier pixels, the inverse variance pixel weights are rescaled to prevent Tractor from fitting the central part of the (typically large, high-surface brightness) galaxy at the expense of the outer envelope.

   • The source detection and model fitting steps are carried out on the coadded images using the average, inverse-variance weighted pixelized PSF in each bandpass.

   • Objects detected within the elliptical mask of each SGA large galaxy are not forced to be point sources.

2. We increase the threshold for detecting and deblending sources by specifying --saddle-fraction 0.2 (the default value is 0.1) and --saddle-min 4.0 (versus the default 2.0).

   The saddle-fraction parameter controls the fractional peak height for identifying new sources around existing sources, and the saddle-min parameter is the minimum required saddle point depth (in units of the standard deviation of pixel values above the noise) from existing sources down to new sources.

   We find these options necessary in order to prevent excessive shredding and overfitting of the "resolved" galactic structure in individual galaxies (e.g., HII regions).

Ellipse-Fitting

We measure the multi-band surface brightness profiles of each galaxy in the SGA using the ellipse-fitting tools in the astropy-affiliated package photutils. Once again, we analyze each galaxy group independently and use MPI parallelization to process the full sample.

Specifically, we carry out the following steps for each galaxy group:

1. We begin by reading the GROUP_NAME-largegalaxy-tractor.fits and GROUP_NAME-largegalaxy-sample.fits catalogs for each group (see the Images and Catalogs section) and reject the following sources from the subsequent ellipse-fitting step, if any:

   • objects missing from the Tractor catalog (i.e., they were dropped during Tractor modeling);

   • objects with negative r-band flux or objects fit by Tractor as type PSF;

   • galaxies fit as Tractor type REX which have a measured half-light major-axis length shape_r \(<5\) arcsec;

   • galaxies fit as Tractor types EXP, DEV, or SER which have a measured half-light major-axis length shape_r \(<2\) arcsec;

   The first two criteria identify spurious sources in the initial parent catalog or objects with grossly over-estimated diameters, and all these objects already have been removed from the SGA-2020.fits catalog.

   The second two criteria identify galaxies which are too small to benefit from ellipse-fitting, i.e., they are well-fit by the standard photometric pipeline and have been deemed to not require special handling. These sources will likely be removed from future versions of the SGA.

2. Next, we read the grz images and corresponding inverse variance and model images. Here and throughout our analysis we use the r-band image as the "reference band." We also read the GROUP_NAME-largegalaxy-maskbits.fits image (see Images and Catalogs) but only retain the BRIGHT, MEDIUM, CLUSTER, ALLMASK_G, ALLMASK_R, and ALLMASK_Z bits (defined on the DR9 bitmasks page). Hereafter, we refer to this mask as the starmask.

   With these data in hand, we carry out the following steps:

   1. First, we build a residual_mask which accounts for statistically significant differences between the data and the Tractor models. In detail, we flag all pixels which deviate by more than \(5\sigma\) (in any bandpass) from the absolute value of the Gaussian-smoothed residual image, which we construct by subtracting the model image from the data and smoothing with a 2-pixel Gaussian kernel. This step obviously masks all sources including the galaxy of interest, but we restore those pixels in the next step. In addition, we iteratively dilate the mask two times and we also mask pixels along the border of the mosaic with a border equal to 2% the size of the mosaic.

   2. Next, we iterate on each galaxy in the group from brightest to faintest based on its r-band flux (from Tractor). For each galaxy, we construct the model image from all the Tractor sources in the field except the galaxy of interest, and subtract this model image from the data.

      We then measure the mean elliptical geometry of the galaxy based on the second moment of the light distribution using a modified version of Michele Cappellari's mge.find_galaxy algorithm (hereafter, the ellipse moments). When computing the ellipse moments, we only use pixels with surface brightness \(>27\ \mathrm{mag\ arcsec}^{-2}\) and we median-filter the image with a 3-pixel boxcar to smooth out any small-scale galactic structure.

      Finally, we combine the residual_mask with the starmask (using Boolean logic), but unmask pixels belonging to the galaxy based on the ellipse moments geometry, but using 1.5 times the estimated semi-major axis of the galaxy.

   3. The preceding algorithm fails in fields containing more than one galaxy if the central coordinates of one of the galaxies is masked by a previous (brighter) system. (We consider a source to be impacted if any pixels in a 5-pixel diameter box centered on the Tractor position of the galaxy are masked.) In this case, we iteratively shrink the elliptical mask of any of the previous galaxies until the central position of the current galaxy is unmasked.

      Note that this algorithm is not perfect, particularly in crowded fields (e.g., the center of the Coma Cluster), but will be improved in future versions of the SGA.

   4. Another occasional failure mode is if the flux-weighted position of the galaxy based on the ellipse moments differs by the Tractor position by more than 10 pixels, which can happen in crowded fields and near bright stars and unmasked image artifacts. In this case we revert to using the Tractor coordinates and model geometry and record this occurence in the largeshift bit (see the Bitmasks page).

   5. Finally, we convert the data images and variance images to surface brightness in units of \(\mathrm{nanomaggies\ arcsec}^{-2}\) and \(\mathrm{nanomaggies}^2\ \mathrm{arcsec}^{-4}\), respectively.

3. With the images and individual masks for each galaxy in hand, we can now measure the multi-band surface-brightness profiles of each galaxy. We assume a fixed elliptical geometry based on the ellipse moments previously measured, and robustly determine the surface brightness along the elliptical path from the central pixel to two times the estimated semi-major axis of the galaxy (based on the ellipse moments), in 1-pixel (0.262 arcsec) intervals.

   In detail, we measure the surface brightness (and the uncertainty) using nclip=3, sclip=3, and integrmode=median, i.e., two sigma-clipping iterations, a \(3\sigma\) clipping threshold, and median area integration, respectively, as documented in the photutils.isophote.Ellipse.fit_image method.

   From the r-band surface brightness profile, we also robustly measure the size of the galaxy at the following surface brightness thresholds: 22, 22.5, 23, 23.5, 24, 24.5, 25, 25.5, and \(26\ \mathrm{mag\ arcsec}^{-2}\) . We perform this measurement by fitting a linear model to the surface brightness profile converted to \(\mathrm{mag\ arcsec}^{-2}\) vs \(r^{0.25}\) (which would be a straight line for a de Vaucouleurs galaxy profile), but only consider measurements which are within \(\pm1\ \mathrm{mag\ arcsec}^{-2}\) of the desired surface brightness threshold. To estimate the uncertainty in this radius we generate Monte Carlo realizations of the surface brightness profile and use the standard deviation of the resulting distribution of radii.

   Finally, we also measure the curve-of-growth in each bandpass using the tools in photutils.aperture. Briefly, we integrate the image and variance image in each bandpass using elliptical apertures from the center of the galaxy to two times its estimated semi-major axis (based on the ellipse moments, again, in 1-pixel or 0.262 arcsec intervals).

   We fit the resulting curve-of-growth, \(m(r)\) using the following empirical model:

   \begin{equation*} m(r) = m_{tot} + m_{0} \log_{e}\left[1 + \alpha_{1} \left(\frac{r}{r_{0}}\right)^{-\alpha_{2}} \right] \end{equation*}

   where \(m_{tot}\), \(m_{0}\), \(\alpha_{1}\), \(\alpha_{2}\), and \(r_{0}\) are constant parameters of the model and r is the semi-major axis in arcsec. In our analysis we take the radius scale factor \(r_{0}=10\) arcsec to be fixed. Note that in the limit \(r\rightarrow\infty\), \(m_{tot}\) is the total, integrated magnitude.

   With this model, the half-light semi-major axis, \(r_{50}\), can be inferred from the best-fitting model parameters:

   \begin{equation*} r_{50} = r_{0} \left\{ \frac{1}{\alpha_{1}} \left[ \exp\left( -\frac{\log_{10}(0.5)}{0.4 m_{0}} \right) -1 \right] \right\}^{-1/\alpha_{2}} \end{equation*}

   Finally, we package all the measurements, one per galaxy, into an astropy.QTable table (including units on all the quantities), and write out the results documented in the Data Products section.

SGA-2020.fits

Group Files

For each galaxy group in the SGA-2020 (i.e., each row in SGA-2020.fits) we produce the set of files described in the Images and Catalogs table and the Custom Mosaics & Ellipse-Fitting documentation section.

These files are organized into the directory structure RASLICE/GROUP_NAME, where GROUP_NAME is the name of the galaxy group and RASLICE (000-359) is the one-degree wide slice of the sky that the object belongs to. Specifically, in Python:

RASLICE = '{:06d}'.format(int(GROUP_RA*1000))[:3]

ELLIPSEBIT

The following bits largely pertain to galaxies with DR9 imaging; they indicate why a given object in the SGA-2020.fits catalog was not ellipse-fit.