THE GHOSTS SURVEY. I. HUBBLE SPACE TELESCOPE ADVANCED CAMERA FOR SURVEYS DATA

The favored Λ-Cold Dark Matter (ΛCDM) cosmology has met with much success in reproducing the large-scale structure of the universe (see, e.g., Springel et al. 2005; Percival et al. 2007; Komatsu et al. 2010; Reid et al. 2010). However, the paradigm has yet to be conclusively tested on small scales. Indeed, cosmological studies on the scope of individual galaxies may provide the most stringent tests of the theory.

For example, the Sloan Digital Sky Survey (SDSS; York et al. 2000; Adelman-McCarthy et al. 2008) has now more than doubled the number of known dwarf satellite galaxies around our Galaxy (Willman et al. 2005; Zucker et al. 2006; Belokurov et al. 2007a; Koposov et al. 2008; Tollerud et al. 2008). However, the dwarf-galaxy number counts still strongly disagree with the dark matter sub-halo populations predicted by ΛCDM (Klypin et al. 1999; Moore et al. 1999; Diemand et al. 2008; Springel et al. 2008). This "missing satellite" problem has sparked much debate as it places strong constraints on the nature of dark matter (e.g., Zentner & Bullock 2003; Strigari et al. 2007; Polisensky & Ricotti 2011) as well as the behavior of baryons in dark-matter potentials (e.g., Bullock et al. 2000; Ricotti & Gnedin 2005; Okamoto et al. 2010).

ΛCDM models of hierarchical formation also predict that the ongoing accretion and disruption of many of these satellite galaxies should be observable today in our own stellar halo (e.g., Bullock et al. 2001; Bullock & Johnston 2005; Font et al. 2006). Following the initial discovery of the tidally disrupted Sagittarius Dwarf galaxy (Ibata et al. 1994), increasing evidence for these remnants has indeed been found. In particular, the SDSS has now detected significant substructure in our stellar halo both spatially (Belokurov et al. 2006; Bell et al. 2008) and in metallicity space (Belokurov et al. 2007b; Carollo et al. 2007; Ivezić et al. 2008). However, these observations still leave much room for interpretation due to both limited sky coverage and a vantage point internal to the Galaxy. For instance, whether the halo is built entirely from disrupted satellites with no smooth underlying component is still unclear (Bell et al. 2008). The problem is further complicated by recent theoretical work, which has suggested that a non-trivial amount of the stellar mass in the halo could have been contributed throughout cosmic history by the ejection of stars originally formed in the host galaxy disk (Zolotov et al. 2009; Purcell et al. 2010).

In M31, these issues are being addressed through a comprehensive effort to map the entire visible stellar halo. The inner 55 kpc was first explored in the Isaac Newton Telescope Wide-Field Camera Panoramic Survey of M31, which revealed the massive infalling giant southern stream (Ibata et al. 2001; Ferguson et al. 2002). Subsequent surveys with Megacam on the Canada–France–Hawaii Telescope covered a quadrant of the halo out to ∼150 kpc with an extension to M33 at ∼200 kpc (Martin et al. 2006; Ibata et al. 2007). The remaining quadrants are currently being mapped by the Pan-Andromeda Archaeological Survey (PAndAS; McConnachie et al. 2009). These deep surveys find a rich network of filaments out to some 120 kpc. Some of this substructure has been targeted with deep Hubble Space Telescope Advanced Camera for Surveys (HST/ACS) pointings, revealing a complex composition of stellar populations, which, in contrast to our Galaxy, is indicative of substantial mixing in the M31 system (Brown et al. 2006; Richardson et al. 2008).

In the hierarchical formation paradigm of ΛCDM, stellar halos are predominantly built up from the early accretion and disruption of low-metallicity dwarf galaxies. Once they have entered the host system, these dwarfs undergo little or no active star formation. Galaxy halos therefore provide an unfettered view of early star formation in the universe as well as the merger history of the host system (e.g., Searle & Zinn 1978; Johnston et al. 1996; Helmi & White 1999; Bullock et al. 2001; Helmi 2008). Ongoing studies of the Milky Way and M31 stellar halos will thus provide some insight on new high-resolution simulations that explore the assembly of ∼L* galaxies (e.g., Cooper et al. 2010). However, with a stochastic process like hierarchical formation, our Galaxy and M31 may be atypical of the larger population. For instance, Hammer et al. (2007) have empirically argued that the accretion history of the Milky Way has been unusually quiescent in comparison to similar systems in the local universe. Thus, to further assess our understanding of galaxy formation we have undertaken the GHOSTS (Galaxy Halos, Outer disks, Substructure, Thick disks, and Star clusters) survey of 14 nearby disk galaxies.

This survey represents the largest study of resolved stellar populations in the outer envelopes of nearby disk galaxies to date. By resolving individual stars with space-borne observatories, we are able to address a number of issues that cannot be otherwise studied. Notably, integrated light observations are limited in sensitivity principally due to the sky brightness, but also flat-field non-uniformities, Milky Way cirrus emission, bright star masking, and scattered light corrections (de Jong 2008). More generally, the integrated colors are subject to degeneracies between age, metallicity, and extinction, thus preventing discrimination of the various mechanisms believed to be responsible for the formation and evolution of these systems (de Jong et al. 2007). Alternately, for galaxies beyond the Local Group, current ground-based observations of resolved stellar populations lack the required resolution to disentangle stars in regions of high density and to reject contamination from unresolved background galaxies. Although these contaminants do not prevent the detection of substantial substructure, e.g., tidal streams (Martínez-Delgado et al. 2008, 2009), they do severely compromise the measurement of faint features like the extended stellar halo.

By resolving individual stars with the HST, we can reliably distinguish very low surface brightness structures while constraining their ages and metallicities. Such data allow us to address a number of issues. For example, using the abundant red giant branch (RGB) stars as tracers of the faint underlying population, we are able to study the size and shape of the stellar halo. These structural parameters are intrinsically connected to a number of critical phases of galaxy formation, including the small-scale behavior of the primordial power spectrum of density fluctuations; the suppression of star formation in small dark-matter potentials due to reionization, where an earlier reionization epoch leads to a fainter and more concentrated halo; and how deep the baryons of accreted satellites sit in their host dark-matter potentials (Bullock & Johnston 2005; Bekki & Chiba 2005; Abadi et al. 2006). Additionally, the triaxiality of the stellar halo depends on the shape of the dark-matter halo, which is still debated. Dissipationless (dark-matter-only) simulations, and simulations that model the formation of low-mass galaxies, typically predict strongly flattened halos (Jing & Suto 2002; Bailin & Steinmetz 2005; Allgood et al. 2006; Kazantzidis et al. 2010). Conversely, simulations that model the formation of more massive galaxies find that the processes of galaxy formation back-react on the dark matter to produce more spherical halo shapes (Abadi et al. 2006; Kazantzidis et al. 2010).

The stellar halos of galaxies also constrain models accounting for the dearth of low-mass satellites. Scenarios where star formation in smaller halos may be suppressed by reionization or baryon blowout (e.g., Bullock et al. 2000) predict that the ratio of the stellar-halo luminosity to the total luminosity of the system varies strongly with the dark-matter halo mass: from large intra-cluster light fractions for galaxy clusters, to smaller halo light fractions around our L_* targets, to perhaps nearly non-existent stellar halos for our low-mass targets (Purcell et al. 2007). By covering a range of galaxy masses and morphological types, we will fully assess these predictions.

Studying resolved stellar populations also allows us to compare and contrast the metallicity of any companion dwarf galaxies and globular clusters with that of the host galaxy. Furthermore, the metallicity distribution of the host system can be studied both as a function of galaxy type and as a function of position within the galaxy. As substructure is more readily identified in the outer regions (Johnston et al. 2008), we can thus place limits on the presence of a uniform low-metallicity halo beneath the remnants of recent accretion events.

Ongoing accretion is often manifest as distinct spatial structure such as tidal streams. Studying resolved stellar populations allows us to detect these features at very faint limits. Additionally, by resolving ages and metallicities in the thin and thick disks we can study the impact of such accretion events on the host galaxy disk. This includes phenomena such as disk heating, disk truncations, and disk flaring (e.g., van der Kruit & Searle 1981; Kazantzidis et al. 2008; Purcell et al. 2010).

Future GHOSTS papers will explore all these features in depth. In particular, forthcoming papers will discuss the luminosity and density distributions of the stellar halo, bulge, and thick disk; stellar-halo substructure; stellar-halo color distributions; disk truncations; star clusters; and substructure in the M83 outer disk. This paper however focuses on a detailed description of the survey and data reduction procedures. Accordingly, we present in Section 2 the sample and our observations. Section 3 provides a discussion of the data reduction and our assessment of photometric errors. The survey completeness, the effects of contaminants, and the results of artificial star tests are presented in Section 4. Our artificial star tests are also used to define the selection criteria for the stars in the survey. The color–magnitude diagrams (CMDs) presented in Section 5 are then used to measure the tip of the red giant branch (TRGB) distances listed in Section 6. The final data products are collected in Section 7 and we summarize the survey in Section 8.

The GHOSTS survey began with our HST  Cycle 12 SNAPshot program 9765 to study the dust structure of nearby edge-on galaxies (Seth et al. 2005a, 2005b). The quality of the observations allowed us to investigate the properties of both the thin and thick disks of the galaxies as well as the innermost volume of their stellar halos (Seth et al. 2005b). In HST  Cycles 14 (SNAP program 10523) and 15 (General Observer program 10889), we extended these observations farther out into the stellar halos and expanded the sample to cover a range of galaxy masses, luminosities, inclinations, and morphological types.

Our Cycle 14 sample consists of all undisturbed disk galaxies in the list of Karachentsev et al. (2003c) with distances between 1 and 5 Mpc with low galactic foreground extinction (A_V < 0.3 mag) and rotation velocities in the HyperLeda database (Paturel et al. 2003) greater than 90 km s⁻¹. Given the short exposure times of HST SNAPshot observations, this distance limit allows us to reach a signal-to-noise ratio (S/N) >10 at one magnitude below the TRGB.

For Cycle 15 we targeted edge-on galaxies (i > 87°) in order to study halos uncontaminated by disk stars and to avoid uncertainties due to projection effects when investigating halo structure. This edge-on configuration also allows us to study thick disks and associated globular clusters. As the likelihood of observing disk galaxies at such high inclinations is low, they are typically observed at distances greater than the limits of the Karachentsev et al. (2003c) catalog and so require longer exposures. We therefore selected all undisturbed edge-on disk galaxies with rotation velocities larger than 80 km s⁻¹, A_V < 0.5 mag, and distance moduli less than 30.75 (∼14.1 Mpc). We also required at least one reliable distance estimate for each target galaxy, besides a Hubble flow distance (e.g., based on Cepheids, surface brightness fluctuations (SBFs), or accurate Tully–Fisher calibrations). In spite of these requirements, our subsequent observations showed that two of our target systems resided beyond our distance limit. Specifically, the Tully et al. (2009) Extragalactic Distance Database recorded a distance modulus (μ) of 30.64 ± 0.35 for NGC 5907, while SBF measurements of NGC 7814 have consistently measured μ < 30.75 (Tonry et al. 2001; Jensen et al. 2003). The measurements presented in Section 6 place these systems at μ = 31.13 ± 0.10 and 30.80 ± 0.10, respectively.

Data from our Cycle 12 survey of dust lanes in edge-on galaxies were added to the GHOSTS survey for those galaxies with V_rot > 80 km s⁻¹ that were close enough to resolve stars one magnitude below the TRGB. For each of these galaxies observed during Cycles 12, 14, and 15, we also searched the Multimission Archive at STScI¹² (MAST) for additional archived fields with similar exposures in both the F606W and F814W filters. Note that due to a guide-star acquisition failure, the World Coordinate System (WCS) for the archive field NGC 5236-Field 11 is incorrect in the original MAST repository. We have thus updated the WCS using the IRAF commands ccmap and ccsetwcs to fix the drizzled image to stars listed in the USNO-A2.0 Catalog (Monet 1998). The resulting astrometry is good to ∼04. Table 1 lists the key properties of the 14 galaxies in the GHOSTS survey and the number of suitable ACS/WFC fields available for each.

Table 1. Details of the 14 Disk Galaxies Selected for the GHOSTS Survey

Name	α (J2000)	δ (J2000)	b	μ	D	AV	Incl.	Vrot	Morphological	Number of ACS Fields
							Angle		Type	GHOSTS	Archival
	(h m s)	(d m s)	(°)	(mag)	(Mpc)			(km s−1)
(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
NGC 7814d	00 03 14.89	+16 08 43.5	−45.17	31.1	17.2	0.15	706	231	SA(s)ab	5	0
NGC 0253e	00 47 33.12	−25 17 17.6	−87.96	27.5	3.2	0.06	780	194	SAB(s)c	5	5
NGC 0891ab	02 22 33.41	+42 20 56.9	−17.41	30.0	10.2	0.28	880	212	SA(s)b	2	3
NGC 2403de	07 36 51.40	+65 36 09.2	29.19	27.7	3.6	0.13	600	122	SAB(s)cd	7	0
NGC 3031 (M 81)bcde	09 55 33.17	+69 03 55.1	40.90	27.8	3.7	0.27	590	225	SA(s)ab	6	6
NGC 4244ade	12 17 29.66	+37 48 25.6	77.16	28.3	4.7	0.07	880	89	SA(s)cd	8	0
NGC 4565ad	12 36 20.78	+25 59 15.6	86.44	30.6	13.3	0.05	900	245	SA(s)b	4	0
NGC 4631cde	12 42 08.01	+32 32 29.4	84.22	29.1	6.7	0.06	850	139	SB(s)d	2	0
NGC 4736 (M 94)de	12 50 53.06	+41 07 13.7	76.01	28.5	5.0	0.06	350	167	(R)SA(r)ab	3	0
NGC 5023de	13 12 12.60	+44 02 28.4	72.58	29.8	9.5	0.06	900	80	Scd	1	0
NGC 5236 (M 83)be	13 37 00.95	−29 51 55.5	31.97	28.3	4.7	0.22	463	164	SAB(s)c	4	8
NGC 5907bd	15 15 53.77	+56 19 43.6	51.09	31.1	16.5	0.04	870	227	SA(s)c	1	0
IC 5052ae	20 52 01.63	−69 11 35.9	−35.80	29.5	8.1	0.17	900	80	SBd	2	0
NGC 7793ae	23 57 49.83	−32 35 27.7	−77.17	28.2	4.4	0.07	530	96	SA(s)d	4	0

Notes. Key to columns: (1) NGC/IC and Messier identifier; (2) and (3) right ascension and declination; (4) Galactic latitude in degrees; (5) and (6) approximate distance modulus, and distance to the galaxy in Mpc, as listed in the NASA/IPAC Extragalactic Database (NED; http://nedwww.ipac.caltech.edu) as of 2011 February 1; (7) V-band Galactic extinction from Schlegel et al. (1998); (8) inclination angle of the galaxy, as listed in HyperLeda (Paturel et al. 2003, http://leda.univ-lyon1.fr); (9) rotational velocity in km s⁻¹ (HyperLeda); (10) morphological type (NED); (11) number of HST  (ACS WFC) fields taken for each galaxy as part of the GHOSTS survey; (12) number of HST  (ACS WFC) fields of sufficient depth and with F606W and F814W observations available in the HST  archive (MAST). ^aThis system has a known disk truncation. ^bThis system is known to have significant stellar tidal features. ^cH i studies have revealed this galaxy to be part of an interacting system. ^dThis system was imaged by SDSS. ^eThis system was imaged with Spitzer/IRAC by the Spitzer Infrared Nearby Galaxies (SINGS; Kennicutt et al. 2003) or Local Volume Legacy (LVL; Dale et al. 2009) surveys.

Download table as:  ASCIITypeset image

2.1. HST/ACS Observations

The ACS/WFC data were downloaded from the Hubble Data Archive as *_flt FITS files. These observations have already been passed through the calacs package, which performs the initial data reduction as part of the opus pipe line (see the HST  Data Handbook for ACS; Gonzaga et al. 2011). More specifically, the data have been bias-subtracted, passed through an initial cosmic-ray identification routine, corrected for the dark image, and finally flat-fielded. Bad pixels have also been flagged, however we used a series of IRAF routines (namely ccdmask) to pick out any additional bad columns not previously identified. We then passed the images through a further cleaning step, masked out extended sources, and finally performed the photometry.

3.1. Drizzling

3.2. Extended Source Identification

3.3. Photometry

We used the ACS module of dolphot (Dolphin 2000, a modified version of HSTphot) to identify stars and for photometric measurements. This code initially uses a library of precomputed Tiny Tim point-spread functions (PSFs) to center and measure the magnitude of each star in both filters. The code was run on the flat-fielded CCD images (*_flt), which have been masked for bad pixels (e.g., hot pixels, cosmic rays, bad columns, etc.) as described earlier. We configured dolphot with parameters similar to those used in the ACS Nearby Galaxy Survey Treasury program (Dalcanton et al. 2009, hereafter ANGST).

dolphot initially aligns the dithered exposures using bright stars in common between the frames; generally, the alignment is good to ∼001 precision. The code then scans for stars by searching for peaks in image brightness after subtracting the background sky, chosen here as the region bound between the photometry aperture and the PSF. For completeness, a second pass is made with the previous identifications subtracted from the image. dolphot then iteratively solves for the magnitude of each detected star by accounting for the photometry of neighboring stars at each step. During this process the flux is computed by simply scaling the PSF. As the photometry of faint stars is dependent on the shape of the assumed PSF, dolphot modifies the reference PSFs by minimizing the residuals of bright, isolated stars. This step is computed only after a reasonable convergence of all stellar magnitudes is achieved. Overall, ANGST found that these deviations affect the final photometry on the order of 0.01 mag.

Aperture corrections to the PSF magnitudes were calculated and applied by dolphot using isolated stars. Typically, these were found to be <0.05 mag. The code also implements the latest corrections for charge transfer efficiency (CTE) loss using the CTE analytical expressions given in the Instrument Science Report ACS-ISR 2009-01 (Chiaberge et al. 2009). For this survey, CTE corrections typically increase from zero at the chip edge to 0.08 mag at the chip center.

The final output lists the position of each star relative to the drizzled image, together with the measured magnitudes and a slew of diagnostic data for the combined image, as well as for the individual exposures. These include the χ² for the fit to the PSF; an S/N value; sharpness, which can be used to discriminate detections that are either too sharp, such as cosmic rays and hot pixels, or too broad, i.e., background galaxies; roundness, which is used to detect extended or diffracted objects; and crowding, which indicates the change in brightness if nearby stars were not simultaneously fit. Also, the output for each detection are the more general parameters of object type and error flag. The former describes the general shape of the detection, while the latter indicates any issues with the fit, including the presence of bad or saturated pixels. Magnitudes are calculated in ST Instrumental VEGA magnitudes as well as Johnson–Cousins equivalent values using the transformations described in Sirianni et al. (2005). The code also used the latest zero points reported on the STScI Web site.¹⁴ These are separately quoted for data taken before and after the ACS anomaly in 2006 June, which led to an increase in operating temperature after the switch to side-2 electronics.

dolphot was executed with a number of user-defined parameters that primarily affect the star fitting, sky fitting, and aperture correction routines. These parameters were investigated in depth by ANGST and for crowded regions we have adopted their values. Most notably, setting Fitsky = 3 simultaneously fits the sky and the star PSF as a two-parameter fit within a reasonably large aperture of radius (rAper =) 10 pixels. However, to ensure the code runs reliably in crowded regions with these settings, the Force1 flag needs to be set, such that all detections are treated as stars, regardless of the morphology of the object. For our sparser fields, we found these settings resulted in too many contaminants relative to the true number of stars (see Section 4.2). We therefore re-ran dolphot on all fields with Force1 set to off in order to fit only the most likely candidates. Correspondingly, to ensure the most robust fitting in the sparser regions, albeit with a greater noise level in the crowded fields, we used a smaller photometry aperture (rAper = 4 pixels) and set Fitsky = 2, which fits the sky outside the photometry aperture but within the PSF. We thus produced catalogs using both prescriptions for all fields. The two methods produced similar results in regions of intermediate density that are neither sparse nor crowded.

To test the reliability of the dolphot output, we present in Figure 1 a comparison of the photometry carried out on Fields 01 and 02 of NGC 5236 using the parameters for sparse fields. These fields overlap by approximately 18%, with some 2500 RGB stars in common. For coincident detections, we plot the difference in magnitudes reported by dolphot for both filters up to the ∼50% completeness limit. Also plotted are binned means, with error bars indicating the 1σ spread in the scatter. These are to be compared with the shaded regions that denote the mean expected error, computed for each pair by combining in quadrature the dolphot error for both measurements.

Zoom In Zoom Out Reset image size

In each case the observed distribution is consistent with the dolphot errors, although a small offsets of −0.04 mag is found in the F814W filter. This offset is similar to the average CTE correction applied by dolphot to each star. However, when comparing corrections between chips, as we have done here, Chiaberge et al. (2009) find that CTE corrections may overcorrect, leading to an apparent offset of up to ∼0.05 mag. This is similar to the error arising from not correcting for CTE, but in the opposite sense. Hence this may be the cause of the offset noted here. Alternatively, the offset may be due to variances in the focus due to thermal heating of the observatory (known as "breathing"), although dolphot does attempt to account for this. Finally, the difference may be caused by aperture corrections, as the stars chosen to measure these residuals will be different in each field. Regardless, we note that although the offset is statistically significant, the magnitude of the difference is substantially less than the typical uncertainty reported by dolphot. We thus find the dolphot code to be photometrically reliable and accurate. Furthermore, we use the reported errors as our final total uncertainty on the measured magnitudes. This approach is shared by the ANGST survey who, through a series of photometric tests, found that any systematics in the dolphot code are dwarfed by the final errors assigned by the code. At the TRGB of our closest galaxy, NGC 2403, the average dolphot error is only 0.04 mag. This degrades to 0.22 mag at the faint limit of the survey, defined here as the 50% completeness limit, as averaged over all fields. The mean dolphot error for all stars in all fields at S/N >10 is 0.06 mag.

By equating the density of observed stars to surface brightness, we can reach extremely faint surface-brightness levels. The steps involved in this process are outlined in Figure 2. Here, a triangular region of the stellar halo of NGC 4565 denotes the area used to measure the surface brightness along the minor axis of the system. The number of stars in radial bins in this region are counted and divided by the area covered by the observation (corrected for the area lost to bright foreground stars and resolved background galaxies). In this case, RGB stars, selected by color and magnitude from the CMD, are used. These 8–12 Gyr old stars are the most readily observed and are responsible for the majority of light in these older outer regions. Then, using a least-squares algorithm, the logarithm of the stellar density is linearly offset to match a surface brightness profile extracted from an all-galaxy image of the inner region. The stellar densities for each GHOSTS field in a galaxy are offset by the same amount to ensure consistency. We find that an SDSS z-band image provides a good fit to an RGB-dominated population. However, SDSS i-band and Spitzer/IRAC 3.6μ or 4.5μ observations also provide reasonable fits with no significant differences. Galaxies in the sample that are covered by SDSS or observed with Spitzer/IRAC either through the Spitzer Infrared Nearby Galaxies (SINGS; Kennicutt et al. 2003) or Local Volume Legacy (LVL; Dale et al. 2009) surveys are identified in Table 1.

Zoom In Zoom Out Reset image size

As shown in Figure 2, there is generally good agreement between the density of RGB stars and the surface brightness as measured by SDSS. Note that the I-band magnitude plotted here is a combination of the SDSS i_AB and z_AB bands as described on the SDSS Web site.¹⁵ There can, however, be some disagreement in regions of high surface brightness, like the thin disk and bulge, where it can be difficult to resolve individual stars due to stellar crowding. This occurs when the PSFs of the stars overlap. We can statistically correct for this effect by populating the image with artificial stars and measuring the fraction that we successfully recover. This enables the correlation of the internal structure of the galaxy with that of the stellar halo (see Section 4.1).

Figure 2 also illustrates the great depths which resolved stellar populations enable us to reach (typically V ∼ 30 mag arcsec⁻² for this survey). This limit is set by sampling statistics and the accuracy of contaminant subtraction. As we reach out into the halo, the detection rate of individual RGB stars becomes low enough that Poisson errors dwarf the signal. Ever deeper observations will reach further down the luminosity function yielding many more stars and even fainter limiting surface brightnesses. However, contamination from faint foreground stars and background unresolved galaxies often becomes the dominant source of uncertainty at such low detection limits. This is discussed further in Section 4.2.

4.1. Artificial Star Tests

Artificial star tests were carried out to assess the completeness in recovering all the stars in a given field using the dolphot package. Artificial stars were added to each field image and passed through dolphot one at a time in parallel batches. The properties of these "fake" stars, i.e., color and brightness, are chosen to mimic the observed stars, but extended to fainter magnitudes. We vary the number of artificial stars generated based on surface brightness, populating the higher surface brightness regions of an observation with more artificial stars. Typically, we computed the photometry for 500,000 to 2,000,000 artificial stars per field. The probability of recovery was then assessed as a function of the stellar magnitude in each filter as well as the local surface brightness. We can thus correct our stellar densities for the effects of crowding, where individual stars may no longer be resolved. This correction is shown in Figure 3 where the radial light profile along the major axis of NGC 4244 is plotted both before and after this correction. At the center of the galaxy, where the stars appear merged, a significant deficit in the brightness is seen. However, after correction, the surface brightness, as inferred from the stellar density, precisely matches the observed SDSS i-band photometry.

Zoom In Zoom Out Reset image size

Overall, the dolphot recovery of the input parameters for each individual star is very accurate. At F606W =23 mag, the 1σ deviation in magnitude is 0.01 mag as averaged over all the artificial stars tests for all our fields. The reliability decreases toward fainter magnitudes, as expected. However, even at F606W =26 mag, the deviation is still only 0.1 mag, well within the typical errors assigned to each detection by dolphot.

4.2. Contaminants

Using the Besançon model (Robin et al. 2003) for the distribution of Galactic stars, we expect only a few Milky Way foreground star per arcmin², or approximately ∼50 per ACS field (for our lowest galactic-latitude pointing). In contrast, the typical galaxy density at 28 mag, as calculated by Ellis & Bland-Hawthorn (2007) for galaxies unresolvable beyond a 005 PSF, is 300 arcmin⁻². Thus, the principal contaminants in our images are unresolved background galaxies.

To assess the impact of both foreground stars and unresolved sources, we have repeated our analysis on a number of deep ACS/WFC exposures from the HST archive that are far removed from any nearby galaxies and are thus free from any resolvable stars outside our own galaxy (see Table 3). As with our target galaxies, these fields also avoid any significant galaxy clusters at z < 1. These observations are thus ideal for empirically assessing the number of background galaxy contaminants and their contribution to different parts of the CMD.

Table 3. Empty Fields

Field Source	PID	No. of Fields	b	Exposures (s)
		Processed	(°)	F606W	F814W
CT344.......................	9498	1	−86.8	8 × 510	8 × 510
P9468.......................	9468	2	+48.6	3 × 720	3 × 720
P10195.....................	10195	1	+48.6	3 × 720	3 × 720
P10196.....................	10196	1	+48.6	3 × 720	3 × 720
Extended Groth Strip	10134	3	+59.6	4 × 565	4 × 525

Notes. Field source names are those given in MAST, while PID is the HST proposal number under which the observations were executed. Number of fields processed indicates the number of fields used in our analysis of contaminants, not the number of fields available from each program.

Download table as:  ASCIITypeset image

Panel (a) of Figure 4 plots the CMD of false stellar detections reported by dolphot at an S/N greater than 3.5 for the deep CT344 field. Because we expect almost no foreground stars in this field, we can assume that all of the "stellar" detections are in fact unresolved background galaxies. Clearly, for this deep exposure, there are many contaminants. As our outermost fields may contain only hundreds of real stars, we need to define a number of selection criteria to ensure that as many spurious detections as possible are removed while passing the majority of real stars. We thus initially identify a number of cuts using the dolphot diagnostic output described in Section 3.3.

Zoom In Zoom Out Reset image size

To optimize these selection criteria, we pass them to a routine that searches for improved solutions over a limited parameter space using a downhill simplex method. This was achieved by first populating all the "empty" archival fields listed in Table 3 with approximately 350,000 artificial stars. The distribution of these artificial stars was chosen to approximate the magnitude and color of the real detections in NGC 2403; they thus act as proxies for the true stars we wish to recover in our observations. The code then varied the parameters of the selection criteria to minimize the number of spurious detections in the "empty" fields, while maximizing the number of artificial stars recovered (see the Appendix for further details). The final selection criteria adopted for selecting stars in sparse fields are

$$\begin{equation} {-}0.06<\textsc {sharpness}_{\rm {F606W}}+\textsc {sharpness}_{\rm {F814W}}<1.30 \end{equation} \tag{ 1 }$$

$$\begin{equation} \textsc {crowding}_{\rm {F606W}}+\textsc {crowding}_{\rm {F814W}}<0.16 \end{equation} \tag{ 2 }$$

$$\begin{equation} \textsc {S/N}_{\rm {F606W|F814W}}>5.0 \end{equation} \tag{ 3 }$$

where each parameter is a diagnostic of the stellar photometry output from dolphot. Additionally, to select only the most likely candidates, we use the detections for which dolphot reports an object of type 1 (i.e., a clean point source) and an error flag of 2 or less (not enough bad or saturated pixels to compromise the fit). We use the same selection criteria for crowded fields with the exception of the crowding cut, which we remove completely in uncrowded fields (see the Appendix).

After culling the data, some unresolved background galaxies inevitably remain (as shown for the CT344 archival field in panel (b) of Figure 4). However, the number of remaining contaminants is significantly fewer than the number of stars in the sparsest GHOSTS fields (e.g., panel (d) of Figure 4). To further quantify the effects of these contaminants, the surface density of false detections across all the "empty" fields in Table 3 is plotted as a function of F814W magnitude in Figure 5. Exposure time is found to have little effect on the distribution, other than varying the depth of the CMDs. Hence, at densities greater than 2.5 × 10⁻⁴ detections per mag per arcsec², the trend across all the contaminant fields is a clear power law of log (dN/dF814W) =0.65 × F814W-19.0. This decreases to 0.62 × F814W-19.1 after culling the data with the crowded field selection criteria, and further flattens to 0.43 × F814W-14.4 after using the sparse field selection criteria.

Zoom In Zoom Out Reset image size

These remaining spurious detections will limit the equivalent surface brightness that can be reached using resolved star counts. Specifically, the overall uncertainty on the measured stellar density will be a combination of the random Poisson error on the expected count rate of contaminants and halo stars, as well as a systematic uncertainty on the level of contamination. The systematic uncertainty on the expected mean level of contamination will remain the same for all binning prescriptions, but both Poisson noise terms will increase for decreasing bin sizes.

To illustrate the impact of these errors on surface brightness measurements, we calculate the S/N for detections in NGC 4565 (Figure 2). We restrict our analysis to contaminants with the same color as the RGB stars, assuming a distance modulus for NGC 4565 of 30.4. By integrating the contaminant luminosity function, we predict a total of 64.1 ± 8 (random) ± 7.7 (systematic) contaminants in a typical bin size of 80 arcsec⁻². The systematic uncertainty is principally due to the uncertainty in the slope of the number of contaminants per magnitude. After culling by the sparse-field selection criteria, the expected number of contaminants decreases to 2.7 ± 1.6 (random) ± 1 (systematic). The overall noise before culling will thus be $7.7+\sqrt{8^2+N}$, where N is the number of actual halo stars detected. Hence for an S/N greater than 10, we require 256 stars (noise =17.9 random +7.7 systematic). For NGC 4565, this equates to an equivalent surface brightness limit of I = 28.2 mag arcsec⁻² (V ∼ 29.2 mag arcsec⁻², converted using the typical RGB color measured in this survey). This limit improves substantially after culling, after which we only need 121 stars to reach the same S/N (noise =11.1 random +1 systematic). This number of stars equates to I = 29.0 mag arcsec⁻² (V ∼ 30.0 mag arcsec⁻²), almost a magnitude fainter.

For ground-based observations, seeing limits the resolvable PSF to ∼04. Thus, the majority of contaminants removed by these selection culls will be indistinguishable from stars in ground-based images. This limits these observations to much higher equivalent surface brightnesses (e.g., Mouhcine et al. 2010; Bailin et al. 2011). Using the previous example, such observations would require over 960 stars in the same size bin to reach the same S/N, as the contaminants themselves now comprise the majority of noise. This equates to a surface brightness limit of I ∼ 27 mag arcsec⁻², over 2 mag brighter.

As well as reaching low equivalent surface brightnesses, a key advantage to using resolved stellar populations over integrated light studies is the ability to differentiate the components of a galaxy by age and metallicity. Figure 6 highlights the distinct features observed in our CMDs, shown here for one of the outer fields of NGC 2403. The ages of these distinct regions depend on the star formation history (SFH) of the system. However, we can assign approximate age ranges by assuming a constant rate of formation (e.g., Figure 4 of Seth et al. 2005b). These ages, together with a description of the stellar populations that contribute to each region, are as follows.

• Main Sequence (MS): <300 Myr—these bluer stars are actively fusing hydrogen to helium in their cores. Their position on the sequence is primarily dictated by their mass which also determines their time on the sequence. Brighter stars toward the tip of the sequence are typically less than 100 Myr old.
• Red and Blue Helium-burning branches (RHeB and BHeB): 25–600 Myr—these massive, intermediate-age stars burn helium in their cores. During this phase, some of the red stars may quickly evolve to the blue sequence before eventually returning to the right of the CMD leading to blue loops. The rapid migration between states leads to the two distinct populations seen here. The specific color of the sequences, and so their separation, is metallicity dependent.
• Asymptotic Giant Branch (AGB): 1–3 Gyr—although the bulk of the helium-shell burning stars in this short-lived phase are 1–3 Gyr old, they will generally be at least 0.3 Gyr in age and possibly as old as 10 Gyr. However, the stars forming the observed tail to redder colors are mainly carbon stars, which are at most 3–5 Gyr old. Part of the AGB sequence overlaps with the RGB. However, the brighter AGB stars observed in the region identified here are most likely thermally pulsing AGB stars. In this state, the stars cycle between explosive burning of a helium shell and fusion of a hydrogen shell.
• Red Giant Branch (RGB): 8–12 Gyr—this region defines the oldest stars in the CMD and is principally composed of RGB stars undergoing shell hydrogen burning. Typically the RGB sequence is dominated by stars formed in the last significant starburst event, however RGB stars are always older than 1 Gyr. Metallicity strongly influences the position of the RGB stars in the CMD with higher metallicities leading to redder stars at a given magnitude. However, all RGB sequences with [Fe/H] <−0.7 reach approximately the same maximum magnitude at the TRGB.

Zoom In Zoom Out Reset image size

The younger populations are most evident in the fields that lie on the galaxy disks, while the halo fields (and late-type bulge systems) predominantly show older stars. The color and slope of the RGB is sensitive to metallicity, as illustrated by the colored lines in panel (a) of Figure 6. Specifically, as the metallicity increases the RGB branch becomes redder and fainter.

The color of the RGB stars just below the well-defined TRGB can thus be used to estimate metallicity (e.g., Da Costa & Armandroff 1990; Bellazzini et al. 2001). There is some degeneracy with age using this technique. Worthey (1994) finds that stellar population models which differ according to Δage/ΔZ ∼ 3/2 are nearly indistinguishable. Using isochrones from the Dartmouth Stellar Evolution Database (Dotter et al. 2008), decreasing the age of a [Fe/H] = -1.0 RGB population from 10 to 3.33 Gyr shifts the color at half a magnitude below the TRGB by −0.11 mag. In comparison, decreasing the metallicity of the same 10 Gyr population by Δ[Fe/H] = log 2 shifts the color at the same magnitude by −0.16 mag. However, as this effect is not linear with age, the inferred metallicities can be underestimated when a range of stellar ages are present. Specifically, younger stars, which evolve quicker in color and form the bulk of the RGB track, are bluer than models based on globular clusters or 10 Gyr isochrones suggest. Hence metallicities inferred from colors should be treated as lower limits. For more accurate metallicities, a full SFH analysis is recommended. Despite these effects, our limited coverage of the top few magnitudes of the CMD still allows for better age/metallicity resolution than is currently possible with integrated light studies.

Figures 7–20 plot the field locations and CMDs for all the GHOSTS fields, including the reprocessed archival fields, which can vary significantly in exposure depth. Hess density diagrams are used when the number of stars is greater than 30 stars in a 0.1 × 0.1 mag bin in F814W and color, and are plotted with contours every 100, 200, 350, 500, 750, 1000, and 1500 stars per 0.01 mag². Photometric errors, as reported by dolphot and averaged over the stars in the given magnitude bins, are indicated by ellipses. For each CMD the best-fit TRGB for the entire galaxy (described in Section 6) is marked by a red dot-dashed line, while a solid red line denotes the TRGB measurement for that specific field. Arrows in the upper right of each CMD indicate the direction of the reddening vector, and 20% completeness limits, as determined from the artificial star tests, are indicated with blue dashed lines.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Stars evolving along the RGB sequence undergo a helium flash when the electron-degenerate helium core reaches a certain pressure and temperature. At this point, a rapid runaway fusion of helium is triggered that lasts until the degeneracy is removed (e.g., see Dearborn et al. 2006). As this process occurs at a constant critical mass, the maximum luminosity of the RGB sequence is theoretically well defined, particularly in the I band where the bolometric luminosity peaks (Lee et al. 1993; Salaris & Cassisi 1997; Salaris et al. 2002). For low-metallicity stars ([Fe/H] < -0.7), this limiting magnitude is known to be independent of age and only weakly dependent on metallicity. However, as the color of the stars at the TRGB can be used to measure their metallicity, this degeneracy can generally be accounted for. Although note that the precise color of the TRGB is partially dependent on the age and alpha abundance of the underlying population. Using the Da Costa & Armandroff (1990) sample of [Fe/H] < -0.7 globular clusters, Bellazzini et al. (2001) found that

$$\begin{equation} (V-I)_{\rm {TRGB}}=0.581[{\rm {Fe/H}}]^2+2.472[{\rm {Fe/H}}]+4.013 \end{equation} \tag{ 4 }$$

where the metallicity is calculated on the Zinn & West (1984) scale. By inverting this relation and converting it to the Carretta & Gratton (1997) scale, Rizzi et al. (2007) have calculated the metallicity of HST-observed stars in several nearby galaxies. They subsequently used these values to infer the absolute magnitude of the horizontal branch (HB) stars, which is dependent on metallicity, for each system. Thus, by measuring the apparent magnitude of the TRGB relative to the HB stars, they were able to calibrate the following TRGB absolute magnitudes as functions of color of the TRGB:

$$\begin{equation} M_{I}=-4.05+0.217[(V-I)-1.6] \end{equation} \tag{ 5 }$$

or

$$\begin{equation} M_{\rm {F814W}}=-4.06+0.20[({\rm {F606W}}-{\rm {F814W}})-1.23]. \end{equation} \tag{ 6 }$$

By propagating the errors through this procedure, Rizzi et al. (2007) place an uncertainty of 0.02 mag on these calibrations.

To measure the apparent magnitude of the TRGB, we followed the same method employed by Seth et al. (2005a). Specifically, we used the logarithmic edge-detection algorithm defined in Méndez et al. (2002) on stars selected with a color cut to remove both MS and redder AGB stars. This cut was typically set at 0.3 < F606W-F814W < 1.6 but for fields with significant helium-burning sequences the range was shifted to redder colors (>1.0 mag). These intermediate-age stars are typically confined to the galaxy disks, but for a few systems, notably NGC 2403 and NGC 5236 (see Figures 10 and 17), they can be far removed from the galaxy center.

Generally, we only report TRGB measurements for fields that do not cover the main galaxy disk. This further minimizes contamination from younger stars and reduces the effects of internal extinction and crowding. However, for systems with only one field, or for galaxies that are sufficiently distant to easily discriminate the disk and halo populations in the same field, we still measured the TRGB. For these fields, we excluded the stars in regions of high surface brightness by using the galaxy masks produced in Section 3.2. Additionally, we only included fields with enough stars for convergence in the detection. Typically, this required at least 10 stars within the color cut and within 0.25 mag of the TRGB.

Although we measure colors of the RGB both at the TRGB and at M_I = −3.5 (for comparison with Seth et al. 2005a), we do not infer metallicities. For significantly metal-poor stars, such as halo stars, the metallicities calculated from RGB colors may carry substantial uncertainties. We thus leave detailed metallicity calculations using analyses of full SFHs to future work.

Random errors in the TRGB detections were assessed by reprocessing 500 Monte Carlo resamples of the stellar magnitudes with replacement. At the TRGB magnitude, these random errors dominate the stellar photometric errors reported by dolphot. Hence an uncertainty in the TRGB magnitude was assigned by fitting a Gaussian to the distribution of the resamples. Further details of each TRGB detection and measurement are available on the GHOSTS Web site.¹⁶

Table 4 presents the results of the best TRGB detection for the measured fields together with the RGB color measurements and the associated distance modulus using the absolute magnitude defined in Equation (6). The error quoted on the distance modulus incorporates both the error in the TRGB detection and the error in the measurement of the color at the TRGB. Overall, some variation in the TRGB magnitude is seen between fields of the same galaxy. This could be due to shifts in the age and metallicity of the populations or increased reddening in the fields lying closer to the host galaxy. However, these variations are typically within the errors calculated from the Monte Carlo resamples.

A few notable features are evident in the TRGB detections. First, although stars are observable below the TRGB in NGC 0891-Field 03 and NGC 4565-Field 04 as required by the survey constraints, the observations are too shallow for a reliable TRGB detection after applying the selection criteria. Both these fields were observed in the Cycle 12 SNAPshot program, which was principally designed to study the dust lanes across the bright central regions of these systems. Second, the TRGBs in Fields 04 and 05 of NGC 5236 are slightly brighter than the detection in the inner field. This can be attributed to the stellar stream that passes through these fields, which also produces a distinct AGB population not seen in the other halo fields. The brighter TRGB suggests that the stream may lie slightly in the foreground of the host galaxy. This feature will be discussed further in a future paper. Lastly, although NGC 5236-Field 09 lies on the bright spiral galaxy MCG-05-32-042, the system is too distant to resolve individual stars (although globular clusters in this system may be mistaken for NGC 5236 stars). However, NGC 5236-Field 06 covers the dwarf galaxy UGCA 365, which lies in the vicinity of NGC 5236 and so accounts for the bulk of the stars in this field. Using the distance modulus in Table 4, we infer a distance of 5.2 ± 0.1 Mpc for this system.

Table 5 presents an overall distance modulus for each galaxy together with distance moduli taken from the literature. These include measurements inferred from Cepheid variables, SBFs, planetary nebula luminosity functions (PNLFs), and TRGB measurements. For systems with detections in more than one field, we compute the mean distance modulus. However, for NGC 5236 we do not include Fields 04 and 05, as they are dominated by a significant stellar stream as previously mentioned, nor do we include Field 06 as this covers the dwarf galaxy UGCA 365. For the overall uncertainty, the error for each field is combined in quadrature, and for systems with three or more measurements, this error is also combined in quadrature with an estimate of the standard deviation of the results. Overall, measurements from the literature agree favorably with the final galaxy TRGB distances measured here.

Using the artificial star tests, we find that the median depth at 50% completeness across all the fields in this survey is 2.7 mag below the TRGB. However, only 67 of the 76 GHOSTS fields reach at least 20% completeness at depths of 1.5 mag below the TRGB.

The data from the GHOSTS survey will be distributed as a "High-Level Science Product" on the MAST Web site.¹⁶ For each field, including those taken from the archive, the following will be released.

• 1.  
  *.phot_sparse.gz: the raw photometry files output by dolphot using the parameters for sparsely populated fields (which are also listed in *sparse.param). Descriptions of the columns are given in the dolphot manual.¹⁷
• 2.  
  *.phot_crowded.gz: similarly, the raw photometry files output by dolphot but using the parameters for crowded fields (which are also listed in *crowded.param).
• 3.  
  *.stars.fits: the final filtered catalog of stars, listed separately for photometry using either the sparse or crowded field parameters together with the corresponding selection criteria from Section 4.2. The columns are X, Y (relative to the F814W drizzled image), RA, DEC, MAG1_ACS, MAG1_JC, MAG1_ERR,  CHI1,  SNR1, SHARP1,  ROUND1,  CROWD1,  FLAG1,   MAG2_ACS, MAG2_JC,  MAG2_ERR,  CHI2,  SNR2,  SHARP2,  ROUND2, CROWD2, FLAG2, where "1" refers to the ACS filter listed under the header keyword "FILTER1" (typically F606W) and "2" refers to the "FILTER2" keyword (typically F814W). The *_ACS column lists the instrumental magnitudes recorded in VEGAMAG, and the *_JC column lists the corresponding Johnson–Cousins magnitudes, which are V for F606W and I for F814W. Subsequent columns are the dolphot parameters summarized in Section 3.3 with "FLAG" indicating the error flag. The header lists the zero-point corrections applied, together with additional data such as completeness at F606W–F814W = 0.75 (as computed from the artificial star tests) and the detected TRGB magnitude.
• 4.  
  *.drz.fits: the final cleaned drizzled images, both in the F606W and F814W filters. The F814W image is used for coordinate references.
• 5.  
  *.gal.fits: where applicable, the galaxy mask, which was used to exclude spurious detections of extended objects coincident with the host galaxy disk, and to exclude high-extinction regions when calculating the TRGB distance.
• 6.  
  *.seg.fits: the segmentation mask, which identifies the extended objects in the drizzled image
• 7.  
  *.se: the SExtractor catalog from which the above mask is constructed.
• 8.  
  *.sky.fits: the background sky image computed by running the calcsky routine included in the dolphot package on the drizzled F814W image. This file is used for sky brightness in the artificial star tests.
• 9.  
  *.artificial.fits: the three-dimensional probability distribution of recovering a star, computed separately for the sparse and crowded field culls, as a function of magnitude in the F814W filter, F814W–F606W color, and background sky brightness. Bins without sufficient coverage in the artificial star tests to compute a probability are set to zero.

Additionally, more diagnostic plots that display the results of the culls are presented on the GHOSTS Web site, together with further information regarding the TRGB detection for individual fields.

We have presented an overview of the GHOSTS survey and described the reduction of the ACS/WFC F606W and F814W imaging in 14 nearby disk galaxies. Across the survey, we have reliably measured the photometry of some 3.1 million stars in 76 ACS fields. For each of these fields we have also generated a few million artificial stars, which reveal the median depth of the survey at 50% completeness to be 2.7 mag below the TRGB. The mean photometric error for all stars with an S/N >10 is ∼0.06 mag. By repeating our analysis on fields from the MAST archive that are free of resolvable stars, we find the principal contaminant in the survey to be unresolved background galaxies. In sparse fields, these sources lead to a larger error than can be attributed to the small number statistics of the RGB tracers. Hence, when equating RGB stellar densities to surface brightness, the background galaxies limit the measurable brightness. However, by specifying various selection criteria, which have subsequently been optimized by an automated routine, approximately 95% of the contaminants can be removed. For NGC 4565, this corresponds to an equivalent surface brightness limit of I ∼ 29 mag arcsec⁻² (V  ∼  30 mag arcsec⁻²). Without the selection criteria, this limit is almost a magnitude brighter, and by simulating atmospheric seeing conditions, we expect the same limit for ground-based observations to be some two magnitudes brighter. We have used this culled photometry to measure the TRGB distances to all the galaxies in the survey. The uncertainties on these measurements ranges from only ∼2% to 7% of the total value.

This survey represents the largest sample of resolved stellar populations in the outskirts of disk galaxies to date. The combination of our observations with high-resolution N-body simulations will undoubtedly yield significant advances in the study of disk galaxy assembly and evolution.

Support for Proposal numbers 9765, 10523, and 10889 was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Incorporated, under NASA contract NAS5-26555. Additionally, some of the data presented in this paper were obtained from the Multimission Archive at the Space Telescope Science Institute (MAST). Support for MAST for non-HST data is provided by the NASA Office of Space Science via grant NNX09AF08G and by other grants and contracts. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. We acknowledge the usage of the HyperLeda database (http://leda.univ-lyon1.fr). This research has also made use of NASA's Astrophysics Data System Bibliographic Services. S.C. acknowledges the support of a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada.

Facility: HST (ACS) - Hubble Space Telescope satellite