CANDELS MULTI-WAVELENGTH CATALOGS: SOURCE IDENTIFICATION AND PHOTOMETRY IN THE CANDELS COSMOS SURVEY FIELD

The CANDELS COSMOS field was observed by the WFC3 in F125W and F160W (J₁₂₅ and H₁₆₀) and in parallel by the ACS in the F606W and F814W filters (V₆₀₆ and i₈₁₄). The WFC3 observations covered a rectangular grid of 4 × 11 tiles ($\sim 8\buildrel{\,\prime}\over{.} 6\times 23\buildrel{\,\prime}\over{.} 8$) running north to south, allowing for maximum contiguous coverage in the near-infrared. The field was observed over two epochs with each tile observed for one orbit in each epoch. The one-orbit observations were divided into two exposures in F125W ($\sim 1/3$ orbit depth) and F160W ($\sim 2/3$ orbit depth) along with parallel ACS observations in the F606W and F814W (Grogin et al. 2011; Koekemoer et al. 2011). The exposures in each orbit were dithered using a small-scale pattern providing half-pixel subsampling of the PSF and also ensuring that the hot pixels and persistences were moved around. The output is a calibrated and astrometry-corrected mosaics of all the exposures in the four individual HST bands. The astrometry is based on the CFHT/MegaCam i* imaging supplemented by deep Subaru/Suprime-Cam ${i}^{+}$ imaging with absolute astrometry registered to the VLA 20 cm survey of the COSMOS field (Schinnerer et al. 2007; see also Koekemoer et al. 2007). This is also the adopted reference grid by the COSMOS team (Capak et al. 2007). All the ground-based and Spitzer data (described in the next Section) were aligned to the HST data astrometry using SWARP. For the present work we used the V0.5 release of the HST/ACS and WFC3 data available from the CANDELS website.³⁶ The observation depth, effective wavelength, and the PSF information for each of the HST filters are summarized in Table 1.

Table 1.  Summary of the CANDELS COSMOS Broadband Data

Instrument	Filtera	Effective	PSF	5σ limiting depthc	Version	References
		Wavelengthb	FWHM
		(Å)	(arcsec)	(AB Magnitude)
CFHT/MegaPrime	ua	3817	0.93	27.31	July 2009	Gwyn (2012)
	ga	4860	1.08	27.69	⋯	⋯
	ra	6220	0.84	27.18	⋯	⋯
	ia	7606	0.85	27.23	⋯	⋯
	za	8816	0.84	26.18	⋯	⋯
Subaru/Suprime-Cam	B	4448	0.95	27.98	V2	Taniguchi et al. (2007)
	${g}^{+}$	4761	1.58	26.78	⋯	⋯
	V	5470	1.33	26.86	⋯	⋯
	${r}^{+}$	6276	1.05	27.18	⋯	⋯
	${i}^{+}$	7671	0.95	26.95	⋯	⋯
	${z}^{+}$	9096	1.15	25.55	⋯	⋯
HST/ACS	F606W	5919	0.10	28.34	V0.5	Koekemoer et al. (2011)
	F814W	8060	0.10	27.72	⋯	⋯
HST/WFC3	F125W	12486	0.14	27.72	⋯	⋯
	F160W	15369	0.17	27.56	⋯	⋯
VISTA/VIRCAM	Y	10210	1.17	25.47	DR1	McCracken et al. (2012)
	J	12524	1.07	25.26	⋯	⋯
	H	16431	1.00	24.87	⋯	⋯
	Ks	21521	0.98	24.83	⋯	⋯
Spitzer/IRAC	3.6 μm	35569	1.80	24.41	V1.2	Ashby et al. (2013)
	4.5 μm	45020	1.86	24.40	⋯	⋯
	5.8 μm	57450	2.13	21.28	V2	Sanders et al. (2007)
	8.0 μm	79158	2.29	21.20	⋯	⋯

Notes.

^aFilters adopted from http://cosmos.astro.caltech.edu/page/filterset. ^bCalculated as ${\lambda }_{\mathrm{eff}}=\sqrt{(\int S(\lambda )\lambda d\lambda )/(\int S(\lambda ){\lambda }^{-1}d\lambda )}$ with $S(\lambda )$ the filter response function (Tokunaga & Vacca 2005). ^cThe 5σ limiting magnitude calculated within a circular aperture with a radius ${r}_{\mathrm{ap}}={FWHM}$ of the PSF in each filter.

Download table as:  ASCIITypeset image

The full COSMOS field was targeted by the Canada–France–Hawaii Telescope (CFHT) 3.6 m telescope MegaPrime instrument in the u*, g*, r*, i*, and z* optical bands as part of the CFHT Legacy Survey³⁷ with COSMOS being in the second Deep field. The MegaCam/MegaPrime camera used for the observations has a pixel scale of 0.187 arcsec pixel⁻¹ (Boulade et al. 2003). The final images are from MegaPipe³⁸ (Gwyn 2008) with zero-point adjustments. The image processing and stacking of the data is further described by Gwyn (2012). In this work we used the CFHTLS D2 mosaics 2009.³⁹

The COSMOS field was also observed by the Subaru/Suprime-Cam in the B, ${g}^{+}$, V, ${r}^{+}$, ${i}^{+}$ and ${z}^{+}$ broadband filters. The Suprime-Cam has a field of view of $34^{\prime} \times 27^{\prime}$ with a pixel scale of 0.202 arcsec pixel⁻¹. The data were processed using the IMCAT package.⁴⁰ The individual frames were combined, flat-fielded, and photometry and astrometry calibrated. We refer the reader to Capak et al. (2007) and Taniguchi et al. (2007, 2015) for further description of the observations and data processing. There is also further observations of the COSMOS field by Subaru/Suprime-Cam in twelve intermediate bands along with narrowband data across two filters (Capak et al. 2007; Taniguchi et al. 2007, 2015) covering the wavelength range of $\sim 4000\mbox{--}8500\,\mathring{\rm A}$. These observations were processed similarly to the broadband optical data discussed above (Taniguchi et al. 2015). Although slightly shallower than the optical broadband observations, these filters have higher resolving power than the former (with $R=\lambda /{\rm{\Delta }}\lambda \sim 23;$ Taniguchi et al. 2015), equivalent to low-resolution spectroscopy in the optical. The resolving power is even higher for the two narrowband filters ($R\sim 50\mbox{--}100;$ Taniguchi et al. 2015) although with smaller wavelength coverage. This provides a unique data set for studying emission-line galaxies and high-redshift systems such as Lyα emitters (Shimasaku et al. 2006; Iwata et al. 2009; Koyama et al. 2014). Table 2 summarizes the Subaru intermediate and narrowband observations. We used the Subaru V2 mosaics for broadband and NB816 observations and V1 mosaics for intermediate and NB711 data available from the IRSA⁴¹ in this work.

The ground-based near-infrared observations are from the VISTA (Emerson & Sutherland 2010) 4.1 m telescope VIRCAM large-format array camera (Dalton et al. 2006) in the Y, J, H, and K_s bands with mean pixel scale of 0.34 arcsec pixel⁻¹. The complete contiguous $\sim 1.5\,{\deg }^{2}$ of UltraVISTA observations were done using a stripes pattern with $\sim 0.7\,{\deg }^{2}$ of the field observed with longer exposure (in four stripes) separating the observations into deep and ultra-deep regions (McCracken et al. 2012) with the CANDELS HST observations inside one of the ultra-deep stripes. The data were pre-processed at CASU,⁴² which includes dark subtraction, flat fielding, gain normalization, and initial sky subtraction (Irwin et al. 2004; McCracken et al. 2012). The data were further processed at TERAPIX using an iterative sky-background removal technique and resampled to a pixel scale of 0.15 arcsec pixel⁻¹. McCracken et al. (2012) give more details on the data processing. We used the final DR1 mosaics of the UltraVISTA data for the CANDELS multi-wavelength catalog.⁴³

The COSMOS field also has been observed by the NOAO Extremely Wide-Field Infrared Imager (NEWFIRM) on the Mayall 4 m telescope as part of the NEWFIRM Medium Band Survey⁴⁴ (NMBS; Whitaker et al. 2011). The NEWFRIM observations in the COSMOS cover an area of $27\buildrel{\,\prime}\over{.} 6\times 27\buildrel{\,\prime}\over{.} 6$ encompassing the CANDELS HST observations. The observations are over five medium-band filters of J₁, J₂, J₃, H₁, and H₂ covering the wavelength of $1\mbox{--}1.8\,\mu {\rm{m}}$ and the K filter centered at 2.2 μm. The three medium-band J₁, J₂, and J₃ filters are a single broadband J filter split into three and the two medium-band H₁ and H₂ filters combine into a single broadband H (van Dokkum et al. 2009; Whitaker et al. 2011). The final mosaic has a pixel scale of 0.3 arcsec pixel⁻¹. Whitaker et al. (2011) further discuss the data processing. Here we used the first data release of the NMBS data (DR1) of the COSMOS field available from the NOAO science archive.⁴⁵

The COSMOS field was observed by the Spitzer Space Telescope IRAC instrument (Fazio et al. 2004) at 3.6 μm, 4.5 μm, 5.8 μm, and 8.0 μm as part of the S-COSMOS (Sanders et al. 2007). The 3.6 μm and 4.5 μm bands have much deeper data from the Spitzer Extended Deep Survey⁴⁶ (SEDS; Ashby et al. 2013). The SEDS observations cover a strip of ${10}^{\prime }\times 1^\circ$ oriented north–south coinciding with the deep VISTA data mentioned above and incorporate previous 3.6 μm and 4.5 μm data from the S-COSMOS providing a uniform depth of 26 mag (3σ) for all observations (Ashby et al. 2013). The 5σ limiting magnitude and FWHM size of the PSF in each IRAC band are reported in Table 1. In this work we used the SEDS first data release (V1.2) (Ashby et al. 2013) available from https://www.cfa.harvard.edu/SEDS/data.html for photometry measurements in the 3.6 μm and 4.5 μm bands and the S-COSMOS Spitzer/IRAC 5.8 μm and 8.0 μm GO2 (V2) data (Sanders et al. 2007) available from http://irsa.ipac.caltech.edu/data/SPITZER/S-COSMOS/. Figure 2 shows the sky coverage of the different ground-based and space data in the COSMOS CANDELS.

Zoom In Zoom Out Reset image size

We performed photometry on the high-resolution (ACS + WFC3) data using SExtractor software version 2.8.6 (Bertin & Arnouts 1996) in dual mode with the WFC3 F160W as the detection band consistently with the multi-wavelength catalogs in the other four CANDELS fields (Galametz et al. 2013; Guo et al. 2013). SExtractor software was modified in several ways to enhance the sky measurement, add a new cleaning procedure, and fix isophotal-corrected magnitude calculations as discussed by Galametz et al. (2013). In order to measure the photometry in the visible bands we PSF-matched the ACS and WFC3 images and extracted the photometry from the matched images in dual mode.

As shown in the photometric studies of the CANDELS UDS and CANDELS GOODS-S fields (Galametz et al. 2013; Guo et al. 2013), it is impossible to identify all galaxies using a single set of parameters (signifying the area, signal to noise, background, etc.) for the extraction. The challenge is to detect the brightest targets while avoiding blending and also detect the faintest objects without introducing spurious sources into the catalog. To this end we used two sets of SExtractor input parameters. One set of parameters is aimed at bright source detection with a focus on deblending extended sources (cold mode), and a second set on faint galaxies (hot mode). The two catalogs generated by the hot and cold parameters were then combined following a routine adopted from GALAPAGOS⁴⁷ (Barden et al. 2012). The combined catalog includes all the sources from the cold-mode catalog plus sources in the hot-mode catalog that do not exist in the cold mode as identified by the Kron ellipse of a cold mode detected source as discussed by Galametz et al. (2013). Figure 3 shows the 1σ detection limit of the combined catalog computed over a circular aperture with a radius of 1 arcsec along with the cumulative distribution of the detection area and the exposure time distribution in the F160W band. Figure 4 shows the magnitude distribution of the sources in the hot, cold, and combined catalogs along with the comparison of the F160W combined counts with the CANDELS UDS (Galametz et al. 2013) and 3D-HST (Skelton et al. 2014). Table 3 gives the number counts in magnitude bins for the combined catalog along with the associated Poissonian uncertainties.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 3.  HST/WFC3 F160W Number Counts from the Combined Hot+Cold Catalog

WFC3 F160Wa	N(${\deg }^{-2}\,{\mathrm{mag}}^{-1}$)	Poisson Uncertainty
15.25	70	50
15.75	387	117
16.25	563	141
16.75	633	149
17.25	1196	205
17.75	1653	241
18.25	2638	305
18.75	3166	334
19.25	4960	418
19.75	6965	495
20.25	9990	593
20.75	13789	696
21.25	17377	782
21.75	27086	976
22.25	33735	1089
22.75	44956	1257
23.25	62720	1485
23.75	82841	1707
24.25	106200	1933
24.75	133426	2166
25.25	156045	2343
25.75	178769	2508
26.25	201810	2664

Note.

^aBin center magnitude.

Download table as:  ASCIITypeset image

At this stage we also assigned a photometry flag to every object in the catalog. The flagging system is the same as that adopted by the other CANDELS fields (Galametz et al. 2013; Guo et al. 2013) and discussed in detail by Galametz et al. (2013). We use a zero for a good photometry in the flagging and assigned a value of one for bright stars and spikes associated with those stars as photometry for objects contaminated by this would be unreliable. A photometric flag of two is associated with the edges of the image as measured from the F160W rms maps.

In order to measure the photometry in the ground-based and Spitzer bands, we used the Template FITting method (TFIT; Laidler et al. 2007) similarly to the other CANDELS multi-wavelength catalogs (Galametz et al. 2013; Guo et al. 2013, M. Stefanon et al. 2017, in preparation). TFIT is a robust algorithm for measuring photometry in mixed-resolution data sets. Sources that are well separated in the high-resolution image (HST) could be blended in the low-resolution image (ground-based or Spitzer). TFIT uses position and light profiles from the high-resolution image to calculate templates that are used to measure the photometry in the low-resolution image. It does that by smoothing the high-resolution image to match the PSF of the low-resolution image using a convolution kernel (Galametz et al. 2013; Guo et al. 2013). Fluxes in the low-resolution image are then measured using these templates while fitting the sources simultaneously.

TFIT requires some pre-processing of low-resolution image in terms of orientation and pixel scale. The individual steps taken are as follows.

Background Subtraction: The low-resolution images must be background subtracted before running TFIT. We used a background subtraction routine with several iterations that included a first estimate through smoothing the image on large scales followed by PSF smoothing and source masking that led to a noise map that was interpolated to determine the background (Galametz et al. 2013).

Image Scale: TFIT requires the low-resolution image pixel scale to be an integer multiple of the high-resolution detection image (0.06 arcsec pixel⁻¹ for the F160W) and that both images have the same orientation. Furthermore, the astrometry of the low-resolution images must be consistent with the high-resolution observations. We used SWARP to resample low-resolution data sets to the next larger pixel scale that is an integer multiple of the WFC3 mosaic and also used it for astrometry and image alignment.

Point-spread Function and Kernel: The point-spread function of both the high-resolution and low-resolution images are needed for the TFIT pre-processing. We constructed the PSF by stacking isolated and unsaturated stars in each band using custom IDL routines. We also constructed a kernel to convolve the high-resolution templates to the low-resolution ones. The kernel was constructed using a Fourier space analysis technique similar to Galametz et al. (2013), which takes the ratio of the Fourier transform of each PSF. This gives the Fourier transform of the kernel, which is then transformed back into normal space generating the kernel. As discussed by Galametz et al. (2013) a low passband filter is applied in the Fourier space to cancel the high-frequency fluctuations and remove the effect of noise. For Spitzer/IRAC, which has the largest difference in resolution from HST, one could use the PSFs directly as the convolution kernel (Galametz et al. 2013). We generated a model PSF by averaging a set of oversampled PSFs that measure PSF variations across the detector. The final PSF is a boxcar kernel smoothed and flux normalized model, which also incorporates all the PAs associated with different Astronomical Observation Requests (AORs). We refer the reader to Galametz et al. (2013) and Guo et al. (2013) for more detail.

Dilation Correction for High Resolution Images: TFIT uses the area of the Galaxy identified from the high-resolution HST bands to measure the photometry. These are pixels defined by the high-resolution segmentation maps and are fed into TFIT in the form of the isophotal areas of the high-resolution image. However, as demonstrated by Galametz et al. (2013) and previously by De Santis et al. (2007), SExtractor usually underestimates the isophotal area of faint or small galaxies, and this leads to an underestimate of the flux for such systems. Galametz et al. (2013) performed extensive tests to quantify and correct for this effect, the so-called "dilation correction," by refining and applying the public dilate code (De Santis et al. 2007). These simulations showed that the correction factor is negligible for objects with large isophotal areas and that it is largest for objects with area <60 pixels. The original isophotal area size from SExtractor hence defines the dilation factor applied. We used the same criteria to correct our high-resolution SExtractor segmentation maps as outlined by Galametz et al. (2013) and refer the reader to this work for further details. Even after this correction, the total flux measurement is always bound by uncertainties incorporated into the above assumptions and this constitutes one of the limitations of photometry estimation.

We ran TFIT in two stages. During the first step TFIT measured any remaining misalignment in the form of distortion or mis-registration between the high-resolution and low-resolution images in the form of shifted kernels (Laidler et al. 2007; Galametz et al. 2013; Guo et al. 2013). In the second step, TFIT used the kernels measured from the first step to correct the misalignment and construct a difference residual map of the low-resolution bands. Figure 5 shows the TFIT residual maps in the low-resolution visible, near-infrared, and infrared bands along with the high-resolution F160W detection band. The residual maps in the visible and near-infrared are close to zero and only show residuals in the center of very bright objects. However, as argued by Guo et al. (2013), the residual maps are qualitative representations of the TFIT photometry measurements, and we later show in our data quality checks that the photometry is properly measured for these bright objects. Tables 1 and 2 summarize the PSF FWHMs used for the high- and low-resolution images.

Zoom In Zoom Out Reset image size

The final F160W SExtractor catalog has 38671 sources over the 216 arcmin² area of the CANDELS COSMOS field with WFC3 F160W observations. TFIT keeps the original F160W SExtractor ID and coordinates for each object in the combined hot+cold catalog and therefore we only need to combine corresponding entries from the high-resolution and low-resolution catalogs. Figure 6 shows the 5σ limiting depth for all the filters in the catalog as computed and tabulated in Tables 1 and 2. We further derive and report a weight for each target in the catalog calculated from the F160W rms maps at the SExtractor positions for each object as described in Guo et al. (2013).

Zoom In Zoom Out Reset image size

The color of stars changes as a function of their spectral type, which in turn depends on the mass (and hence temperature) among other parameters (Kurucz 1979; Vandenberg 1985; Baraffe et al. 1998). Using this, we could compare the measured color of the point-like objects in our catalog against predictions of the colors of stars derived from stellar physics. The predicted colors of stars were computed from the stellar library of BaSeL (Lejeune et al. 1997; Westera et al. 2002) to measure the model stellar colors. We present the comparisons on color–color plots using several observed filters. To measure the predicted photometry from the templates, we integrated the model stellar SED over the wavelength range of each filter taking into account the filter response functions. This provides the predicted colors of point sources. Figure 7 shows our TFIT measured colors for the point-like objects compared to the colors from the BaSeL stellar library. The color trend of our point-like objects agrees with the general distribution of colors predicted by the stellar models. This further confirms our measured photometry, specifically for the brighter sources, and shows no systematic bias in the photometry. The scatter at the redder color is mostly associated with the fainter sources in the catalog and also due to the intrinsic scatter of colors inherent to the library because of the degeneracies among the different populations of stars.

Zoom In Zoom Out Reset image size

A validation check of the infrared colors of objects in the catalog was done by using the Spitzer/IRAC TFIT measured photometry of galaxies to identify luminous active galactic nuclei (AGNs). Mid-infrared observations of galaxies have been used extensively to identify and study AGN host galaxies (e.g., Laurent et al. 2000; Farrah et al. 2007; Petric et al. 2011; Yan et al. 2013; Lacy et al. 2015). Several recent studies have used wide and deep observations in the mid-infrared by Spitzer/IRAC to successfully identify large samples of AGNs using flux ratios (Lacy et al. 2004, 2007; Stern et al. 2005, 2012; Donley et al. 2012; Messias et al. 2012). These color selections are based on the power-law behavior of the mid-infrared continuum of luminous AGNs caused by heated dust, producing a thermal continuum (Neugebauer et al. 1979; Ivezić et al. 2002; Donley et al. 2012; Messias et al. 2012).

Figure 8 shows the TFIT-measured IRAC color distributions of galaxies in our catalog that have a ${\rm{S}}/{\rm{N}}\gt 5$ in all four channels. According to these selections the red IRAC colors are expected to be dominated by emission from AGN-heated dust, as also predicted by previous studies of SDSS and radio-selected quasars (Lacy et al. 2004), whereas any stellar components usually shifts the ${S}_{5.8}/{S}_{3.6}$ ratio to bluer colors. The Donley et al. (2012) criteria are more conservative in selecting AGNs by removing star-forming and quiescent galaxies identified through other selection methods from optical and near-infrared observations (such as the BzK and LBG selections; Madau et al. 1996; Giavalisco 2002). Figure 8 also shows the IRAC color distribution of the X-ray detected sources in COSMOS (Cappelluti et al. 2007) along with the IRAC color distribution of point sources. The majority of the X-ray detected AGNs have IRAC color distributions consistent with the selections by Lacy et al. (2004) and Donley et al. (2012). As discussed by Donley et al. (2012), not all the X-ray luminous and IRAC detected sources are identified by AGN color criteria. In fact, Donley et al. (2012) argue that the X-ray-detected QSOs that fall outside the color criteria seem to be more heavily obscured with lower luminosity AGNs such that the host galaxy contributes more to the optical-near-IR flux. The stellar sources and the general population, however, follow different IRAC color distributions as demonstrated by Lacy et al. (2004) and Donley et al. (2012). We also present a check of stellar colors using IRAC 3.6 μm and 4.5 μm bands in Section 6.2 and Figure 22, where our photometry is found to be consistent with synthetic colors.

Zoom In Zoom Out Reset image size

We also checked the TFIT photometry by comparing it with the public photometry available from the 3D-HST survey⁴⁸ (Skelton et al. 2014). The 3D-HST photometry was measured by SExtractor on PSF-matched combined HST/WFC3 images in three bands (F125W, F140W, and F160W) as the detection (Skelton et al. 2014). The HST images were reduced similarly to CANDELS using the same pixel scale and tangent point (Skelton et al. 2014). Photometry on the low-resolution bands was performed using the MOPHONGO code (Labbé et al. 2005, 2006; Wuyts et al. 2007; Labbé et al. 2013), which takes into account the variations in the PSF size across different filters and in particular the source confusion problem in the low-resolution images by using a combined PSF-matched WFC3 images as a high-resolution image prior for photometry estimation (Labbé et al. 2005; Skelton et al. 2014). The 3D-HST adjusted the AUTO fluxes by an aperture correction derived from growth curves and furthermore performed Galactic extinction corrections. The Galactic extinction correction was measured at the center of each filter and was based on Finkbeiner et al. (1999). These corrections are relatively small ($\lesssim 0.07;$ Table 5 in Skelton et al. 2014). We took this into account when comparing our fluxes with that of the 3D-HST. All fluxes in the 3D-HST catalogs were converted to AB magnitudes using a zero-point of 25 (Skelton et al. 2014).

Figure 9 shows the comparison between TFIT measured photometry and the public photometry from the 3D-HST. For the source matching we used a radius equal to the FWHM size of the PSF in the F160W ($\sim 0.17\mathrm{arcsec}$). We find good agreement between the measured fluxes in our catalog and that of the 3D-HST. The offset is generally $\lesssim 0.1$ mag. There is, however, a magnitude-dependent trend when comparing the CANDELS and 3D-HST photometry. This is related to the difference in the photometry extraction between the CANDELS (TFIT) and the 3D-HST (aperture photometry with fixed apertures for each band; Tables 4–8 of Skelton et al. 2014). Figure 10 shows the comparison between CANDELS COSMOS TFIT measured $(\mathrm{Band}-{\rm{F}}160{\rm{W}})$ color and the corresponding colors of sources measured from the 3D-HST catalog (Skelton et al. 2014) as a function of the F160W magnitude. For both catalogs the F160W band is taken as the reference band for measuring the color. While comparing the ${Spitzer}-F160W$ color for the different fields as a function of photometric redshift, we noticed a deviation of ∼0.6 mag between the colors in the COSMOS and GOODS-S fields for the red objects in the Spitzer 8.0 μm band. There is an offset (though smaller) in similar color between the 3D-HST COSMOS and GOODS-S fields. Looking at the variation of this color difference between the CANDELS COSMOS and 3D-HST as a function of the F160W, we notice that most of the difference is associated with objected fainter than ∼22 mag, which is similar to our 5σ detection limit.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We provided the CANDELS COSMOS photometric catalog to individual teams in the collaboration. The teams were asked to estimate photometric redshifts to galaxies using a calibrating sample containing spectroscopic redshifts. The spectroscopic redshifts were taken from the zCOSMOS compilation (Lilly et al. 2007). Only redshifts for sources in the CANDELS COSMOS area with clear emission-line features were used with uncertain redshifts left out. The spectroscopic redshifts used here are all in public domain. However, for training purposes, a larger sample of unpublished spectroscopic redshifts were used from zCOSMOS (M. Salvato 2016, private communication). Measurements from different teams are in good agreement and also agreed with an independent sample of 448 high-quality spectroscopic redshifts not used to calibrate the photometric redshift methods. A total of six individuals participated. Methods included various fitting codes using minimum ${\chi }^{2}$ and MCMC along with varying Star Formation Histories (SFH). The details of these methods are outlined in Table 7 in the Appendix. Using simulations, we showed that one could determine redshifts to an accuracy of 0.025 in ${\rm{\Delta }}z=({z}_{\mathrm{phot}}-{z}_{\mathrm{spec}})/(1+{z}_{\mathrm{spec}})$. For each galaxy, we derived the median redshift from different methods and considered that as the redshift estimate for that galaxy. The confidence intervals were measured by combining the intervals from different methods, following the procedure used by Dahlen et al. (2013). Taking the median of the photometric redshifts does not necessarily produce a better measurement as this depends heavily on the codes and templates used in calculating the redshift and the corresponding scatter when compared to the spectroscopic redshifts. Comparing with an independent sample of spectroscopic redshifts not used for photometric redshift training, we find that while taking the median reduces the outlier fractions compared to some of the individual codes, marginally lower outlier fractions are obtained using only a subset of the codes (those of Wuyts, Gruetzbach, and Salvato for this field). Given the relatively small number of spectroscopic redshifts available for this test (262), we do not consider the difference significant and choose to report the median of all the codes as our recommended best photometric redshift. Figure 11 shows a direct comparison between the photometric and spectroscopic redshifts for galaxies in the CANDELS COSMOS field, only using high quality spectroscopic redshifts. The training was done using a larger sample of spectroscopic redshifts in CANDELS COSMOS area from the zCOSMOS survey. These do not overlap with the comparison sample here, which is a smaller subsample that is in public domain. Figure 12 shows the spectroscopic redshift distributions of the training and comparison samples. The high-quality spectroscopic redshifts used for the training of the photometric redshifts are available over the range $0\lt z\lt 1$ and hence photometric redshifts reported are most reliable for that redshift range. The values of the redshift difference, as defined in Dahlen et al. (2013), along with the outlier fractions are listed in Table 4. As mentioned above, reliable spectroscopic redshifts are available out to $z\sim 1$ and the numbers listed in Table 4 are only representative for galaxies in this range.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 4.  Spectroscopic Redshift Comparison Table

Survey	OLFa	${\sigma }_{F}$ b	${\sigma }_{\mathrm{NMAD}}$ c	${\sigma }_{O}$ d	Number of Galaxiese
CANDELS COSMOS	0.008	0.035	0.011	0.016	506
COSMOS Team	0.05	0.071	0.008	0.017	504
3D-HST	0.012	0.045	0.008	0.015	499

Notes.

^aDefined as fraction of objects with $| {\rm{\Delta }}z| /(1+{z}_{\mathrm{spec}})\gt 0.15$ where ${\rm{\Delta }}z=({z}_{\mathrm{phot}}-{z}_{\mathrm{spec}})/(1+{z}_{\mathrm{spec}})$ (Dahlen et al. 2013). ^b ${\sigma }_{F}\equiv \mathrm{rms}({\rm{\Delta }}z/(1+{z}_{\mathrm{CANDELS}}))$. ^c ${\sigma }_{\mathrm{NMAD}}\equiv 1.48\times \mathrm{median}(| {\rm{\Delta }}z| /(1+{z}_{\mathrm{CANDELS}}))$. ^d ${\sigma }_{O}\equiv \mathrm{rms}({\rm{\Delta }}z/(1+{z}_{\mathrm{CANDELS}}))$ after removing the outliers. ^ewith reliable spectroscopic redshift used for the comparison. This is for objects with $0\lt z\lt 1$ and hence the reported numbers are valid for this redshift range where high quality spectroscopic redshifts are available.

Download table as:  ASCIITypeset image

We cross-compared the CANDELS photometric redshift catalog with those from COSMOS (Ilbert et al. 2013) and 3D-HST (Skelton et al. 2014). Figure 11 shows the comparison between the photometric and spectroscopic redshifts in the COSMOS and 3D-HST catalogs. The rms values are listed in Tableapjsaa53b1t4 4 and show similar trends as CANDELS COSMOS. We compare photometric redshifts between the CANDELS COSMOS and those from the COSMOS (Ilbert et al. 2013) and 3D-HST (Skelton et al. 2014) catalogs, in Figure 13. The two photometric redshift comparison plots present the full scatter (${\sigma }_{F}\equiv \mathrm{rms}({\rm{\Delta }}z/(1+{z}_{\mathrm{CANDELS}}))$), the normalized median absolute deviation (${\sigma }_{\mathrm{NMAD}}\equiv 1.48\times \mathrm{median}(| {\rm{\Delta }}z| /(1+{z}_{\mathrm{CANDELS}}))$) and the scatter after excluding outliers (${\sigma }_{O}\equiv \mathrm{rms}({\rm{\Delta }}z/(1+{z}_{\mathrm{CANDELS}}))$) as well as the outlier fraction (defined as fraction of objects with $| {\rm{\Delta }}z| /(1+{z}_{\mathrm{CANDELS}})\gt 0.15$). There is consistency in the photometric redshift measurements for the majority of galaxies. The spread seen in Figure 13, is due to log-binning of the histograms and does not indicate any large inconsistency. Table 5 shows the photometric offsets measured from SED fitting for different bands. The offsets are measured through the SED fitting, simultaneously with photometric redshifts. Different independent SED fitting methods estimated the offsets and they agreed fairly well. The magnitude offsets were then used in the photometry to estimate the final photometric redshifts. The photometric redshifts have the correction for the photometric offsets, but the photometry presented here is not corrected for the offset.

Zoom In Zoom Out Reset image size

Table 5.  SED Fitting Measured Photometric Offsets

Filter	Median Offseta
	(AB Mag)
CFHT-u*	0.041
CFHT-g*	−0.044
CFHT-r*	−0.005
CFHT-i*	−0.025
CFHT-z*	0.000
Subaru-B	−0.004
Subaru-${r}^{+}$	−0.071
Subaru-${i}^{+}$	−0.074
Subaru-${z}^{+}$	−0.150
ACS-F606W	0.072
ACS-F814W	0.019
WFC3 F125W	0.104
WFC3 F160W	0.091
UVISTA-Y	−0.027
UVISTA-J	0.046
UVISTA-H	0.067
UVISTA-Ks	−0.031
IRAC-3.6 μm	−0.061
IRAC-4.5 μm	−0.176
IRAC-5.8 μm	−0.069
IRAC-8.0 μm	−0.697
NEWFIRM-J1	−0.061
NEWFIRM-J2	−0.027
NEWFIRM-J3	0.000
NEWFIRM-H1	−0.025
NEWFIRM-H2	−0.060
NEWFIRM-K	−0.073

Note.

^aPositive offset: measured flux fainter than expected from template.

Download table as:  ASCIITypeset image

One method for estimating the photometric redshift uncertainties that is specially useful for fainter flux limits where not enough spectroscopic redshift information is available is the pair statistics estimates (Quadri & Williams 2010; Dahlen et al. 2013; Huang et al. 2013; Hsu et al. 2014). The method, as outlined in Dahlen et al. (2013), relies on the fact that spatially close pairs (defined as objects with separation less than 15 arcsec) have a high probability of being associated with each other and therefore being at similar redshifts (Dahlen et al. 2013). This close-pair association would show up as excess power at small separations when compared to the distribution based on random galaxies (Dahlen et al. 2013). Figure 14 shows the random and close-pair photometric redshift difference distribution along with the excess distribution of the close pairs in photometric redshift after subtracting the photometric redshift distribution of the random galaxies. By fitting a Gaussian function to this excess distribution we measured an uncertainty of 0.017 ($\sqrt{2}\times {\sigma }_{\mathrm{Gaussian}}$) in the photometric redshift distribution.

Zoom In Zoom Out Reset image size

Mobasher et al. (2015) studied stellar mass measurement using CANDELS data. With extensive simulations, they explored different sources of uncertainty in stellar mass measurements and estimated the error budget associated with them. The stellar masses were measured from different methods independently and were compared with their expected (input) mass. All methods produced stellar masses in good agreement. We use the median mass (among all the methods) as the reported CANDELS estimate. As we discussed earlier, taking the median does not necessarily produce a more robust mass estimate because the results of individual codes used are not instances of the same stochastic process.

The TFIT multi-waveband photometric catalog for the CANDELS COSMOS field was provided to the CANDELS teams outlined in Tableapjsaa53b1t8 8 and they measured the stellar masses through separate SED fitting techniques. For all the independent measurements, redshifts were fixed to their median values (as described in the previous section). Given that the photometric redshifts are calibrated with a spectroscopic sample with $0\lt z\lt 1$, the stellar mass estimates are also well calibrated and most robust within this redshift range, as discussed in the previous section, although we report stellar mass estimates out to $z\sim 5$ in this catalog. Some of the methods included nebular emission when fitting the SEDs as outlined in Table 8. A total of eight entries were received from different teams. We measured the median stellar mass between different methods. We used the Hodges–Lehmann⁴⁹ method to estimate the median stellar mass which accounts for the small number of entries when measuring the median value.

Figure 15 presents a direct comparison between different methods used to measure stellar masses. Here, we plot estimates from each method against the median from all the rest of the methods. There is good internal consistency between different methods and the tails seen in Figure 15 help identify where the outliers are in each method compared to the rest. These outliers will not bias our measurements as we report the median of all methods for our final stellar mass measurements as demonstrated in Mobasher et al. (2015).

Zoom In Zoom Out Reset image size

The median stellar mass estimates from the CANDELS are compared to those from the COSMOS (Ilbert et al. 2013) and 3D-HST catalogs (Skelton et al. 2014) in Figure 16. In these comparisons, we measure the scatter similarly to the photometric redshift where ${\sigma }_{F}\equiv \mathrm{rms}(\mathrm{log}({M}_{\mathrm{CANDELS}})-\mathrm{log}({M}_{\mathrm{other}}))$ and ${\sigma }_{O}\equiv \mathrm{rms}(\mathrm{log}({M}_{\mathrm{CANDELS}})-\mathrm{log}({M}_{\mathrm{other}}))$ after removing the outliers where outlier fraction is defined as the fraction of objects with ${\rm{\Delta }}\mathrm{log}(M)\equiv (\mathrm{log}({M}_{\mathrm{CANDELS}})-\mathrm{log}({M}_{\mathrm{other}})\gt 0.5$ (Mobasher et al. 2015), as shown on Figure 16. The CANDELS stellar mass measurements are consistent with both COSMOS and 3D-HST stellar masses. The stellar mass offsets between CANDELS median measurement and 3D-HST and COSMOS are also plotted as a function of F160W magnitude in Figure 17. As expected, the larger discrepancies between the stellar mass estimates occur at fainter magnitudes. The rms values of the stellar mass estimate offsets as defined above are presented in Table 6 as a function of H-band magnitude, confirming a good agreement at brighter magnitudes and an overall consistency for the majority of galaxies.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Table 6.  Variations and Outlier Fractions in Mass Measurements

	Magnitude Cut (AB)	OLFa	${\sigma }_{{\rm{F}}}$	${\sigma }_{\mathrm{NMAD}}$	${\sigma }_{{\rm{O}}}$
COSMOS
	$21\lt H\lt 22$	0.032	0.307	0.124	0.149
	$22\lt H\lt 23$	0.048	0.309	0.138	0.153
	$23\lt H\lt 24$	0.077	0.384	0.179	0.175
	$24\lt H\lt 25$	0.172	0.549	0.251	0.204
3D-HST
	$21\lt H\lt 22$	0.049	0.597	0.133	0.131
	$22\lt H\lt 23$	0.054	0.463	0.138	0.144
	$23\lt H\lt 24$	0.089	0.563	0.148	0.148
	$24\lt H\lt 25$	0.138	0.687	0.201	0.179

Note.

^aDefined as $| {\rm{\Delta }}\mathrm{log}(M)| \gt 0.5$ (Mobasher et al. 2015).

Download table as:  ASCIITypeset image

When measuring the stellar mass, we fit the SEDs by fixing redshifts to the median of the photometric redshifts from the CANDELS COSMOS team, as discussed in Section 5.1. Therefore, if the redshifts for the same galaxies are different in those from 3D-HST and COSMOS teams, the effect would propagate to the estimated stellar masses. As a result, the observed offsets between the stellar mass values in Figures 16 and 17 could partly be explained by the discrepancy between the redshifts. To explore this, we studied the residual diagrams between redshifts and stellar masses for the three measurements. For this we look at the ratio of the stellar masses as measured by different methods ($\mathrm{log}({M}_{\mathrm{CANDELS}})-\mathrm{log}({M}_{\mathrm{other}})$) as a function of the redshift difference $({z}_{\mathrm{CANDELS}}-{z}_{\mathrm{other}})/(1+{z}_{\mathrm{CANDELS}})$ in Figure 18 showing a 0.25 dex scatter in stellar mass for galaxies with similar redshifts. This is in agreement with results of Mobasher et al. (2015), who measured the combined error budget in stellar mass values due to different parameters. Therefore, the distribution in residual mass here is consistent with the expected uncertainties in the stellar mass measurements (i.e., the vertical scatter). The galaxies with deviant redshifts also have deviant stellar mass estimates, partly explaining the observed scatter between the CANDELS and 3D-HST and CANDELS and COSMOS stellar masses.

Zoom In Zoom Out Reset image size

Figure 19 compares the difference between stellar mass measurements with and without correction for nebular emission lines. This shows a small scatter (0.25 dex) in the stellar mass, consistent with Mobasher et al. (2015), but no significant offset over the whole population over the large redshift range.

Zoom In Zoom Out Reset image size

We use the method introduced by Pozzetti et al. (2010) to estimate the stellar mass completeness limit of the general population of galaxies (see also Ilbert et al. 2013 and Darvish et al. 2015a). Given the magnitude limit of the sample (${H}_{\mathrm{lim}}=27.56;$ $5\sigma$ limiting magnitude in the F160W detection band), we assigned a limiting stellar mass (${M}_{\mathrm{lim}}$) to each galaxy. ${M}_{\mathrm{lim}}$ is the stellar mass that a galaxy would have at its estimated redshift, if its apparent magnitude was the same as the magnitude limit of our sample (${H}_{\mathrm{lim}}=27.56$). This was evaluated by $\mathrm{log}({M}_{\mathrm{lim}}/{M}_{\odot })=\mathrm{log}(M/{M}_{\odot })+0.4(H-{H}_{\mathrm{lim}})$, where M is the estimated stellar mass of the Galaxy with its apparent magnitude H. This results in a distribution of ${M}_{\mathrm{lim}}$ values at any given redshift. The 90% (70%) stellar mass completeness limit at each redshift is therefore equivalent to the mass with 90% (70%) of the galaxies having their ${M}_{\mathrm{lim}}$ value below the stellar mass completeness limit. In general, the stellar mass completeness limit depends on the M/L ratio and is higher for quiescent and dusty galaxies. The stellar mass estimate is also sensitive to the presence of the Balmer break in the SED of galaxies. At $z\gtrsim 3$ the WFC3 H-band will no longer be probing the Balmer break and a redder filter (such as K_s-band or 3.6 μm) is more suited at estimating the stellar mass at these redshifts (Ilbert et al. 2013). Using the UltraVISTA K_s band limiting magnitude reported in Table 1, we found that while the completeness estimates change at $z\gtrsim 3$, the deviations are at the level of 0.1–0.2 dex, which is within the stellar mass uncertainties. Figure 20 shows the distribution of stellar mass as a function of redshift, along with the estimated 90% and 70% completeness limits.

Zoom In Zoom Out Reset image size

One of the main methods of identifying high-redshift galaxies is by targeting the pronounced Lyman break at rest-frame 912 Å that exist in the SED of these galaxies (e.g., Pettini et al. 2002; Shapley et al. 2003; Bolton & Haehnelt 2013; Faucher-Giguère et al. 2015; Williams et al. 2015). The break is caused by the absorption of the UV light from hot and young stars by the neutral Hydrogen (Madau 1995; Madau et al. 1996). Because of this break in the SED, these galaxies would be undetected in the bluer bands and appear in the redder filters (the so-called dropout technique or LBG selection; Madau et al. 1996; Steidel et al. 1999; Giavalisco 2002; Stark et al. 2009). This technique has been used extensively over the past few years in conjunction with deep multi-waveband data and spectroscopic observations to identify and study star-forming galaxies all the way to the cosmic dawn and epoch of reionization (e.g., Yan & Windhorst 2004; Capak et al. 2011; Stark et al. 2011; Steidel et al. 2011; Oesch et al. 2013, 2015; Treu et al. 2013; Bouwens et al. 2014).

Here we use the following LBG color selection from Bouwens et al. (2015) to identify candidates at $\langle z\rangle \sim 4$ and $\langle z\rangle \sim 5$ respectively:

$$\begin{eqnarray}&&({Subaru}(B)-{Subaru}(V))\gt 1,\end{eqnarray} \tag{ 1a }$$

$$\begin{eqnarray}&&(\mathrm{CFHT}({i}^{* })-\mathrm{WFC}3({\rm{F}}125{\rm{W}}))\lt 1,\end{eqnarray} \tag{ 1b }$$

$$\begin{eqnarray}&&({Subaru}(B)-{Subaru}(V))\gt 1.6\\ &&\quad \times (\mathrm{CFHT}({i}^{* })-\mathrm{WFC}3({\rm{F}}125{\rm{W}}))+1.\end{eqnarray} \tag{ 1c }$$

for the B-dropout and

$$\begin{eqnarray}&&({Subaru}(V)-\mathrm{CFHT}({i}^{* }))\gt 1.2,\end{eqnarray} \tag{ 2a }$$

$$\begin{eqnarray}&&(\mathrm{CFHT}({z}^{* })-\mathrm{WFC}3({\rm{F}}160{\rm{W}}))\lt 1.3,\end{eqnarray} \tag{ 2b }$$

$$\begin{eqnarray}&&({Subaru}(V)-\mathrm{CFHT}({i}^{* }))\gt 0.8\\ &&\quad \times (\mathrm{CFHT}({z}^{* })-\mathrm{WFC}3({\rm{F}}160{\rm{W}}))+1.2\end{eqnarray} \tag{ 2c }$$

for the V-dropout. Figure 21 shows the color–color diagrams for the B-dropout and V-dropout selections and the corresponding photometry used along with the photometric redshift distribution of the selected candidates, as measured in the previous section. The LBG selected sources have photometric redshifts consistent with the selections. There are five spectroscopically confirmed sources (three in the B-dropout and two in the V-dropout) among the selected candidates with colors consistent with the corresponding criteria.

Zoom In Zoom Out Reset image size

The presence of passively evolving galaxies at $z\sim 2$ has been established for some time now (Daddi et al. 2005; Kriek et al. 2006, 2008, 2009; Papovich et al. 2006, 2015; Ilbert et al. 2013; Muzzin et al. 2013). These galaxies are argued to be the progenitors of the most massive systems that form the red sequence at low redshifts (Cassata et al. 2013; Barro et al. 2014; Williams et al. 2014). In the past few years there have been several studies that predict the presence of these systems at $z\sim 3$ (Nayyeri et al. 2014; Wang et al. 2016) and some predictions as high as $z\sim 5\mbox{--}6$ (Mobasher et al. 2005; Wiklind et al. 2008). Identification of these objects requires deep observations in infrared wavelengths where they have the peak of their emission due to colder and redder stars (Wuyts et al. 2007; Nayyeri et al. 2014). The CANDELS COSMOS deep HST/WFC3 near infrared observations and the Spitzer/IRAC TFIT photometry measurements are crucial for separating red galaxies from blends of fainter sources and lower redshift dusty star-forming galaxies. The multi-band data in the CANDELS COSMOS, especially the wealth of data available on both sides of the Balmer/4000 Å break help constrain the SEDs of old systems and separate them from the red dusty star-forming galaxies.

One well-known color–color diagram is the BzK color plot that is used to identify star-forming and quiescent galaxies at $z\sim 1\mbox{--}2$ (Daddi et al. 2004). The BzK method uses $(z-K)$ versus $(B-z)$ colors of galaxies to separate the two populations at high redshift (Daddi et al. 2004, 2007). This is mostly based on the presence of the Balmer/4000 Å break in the SED of older galaxies that is observed in the $(z-K)$ color at $z\sim 1\mbox{--}2$. The color-selection diagram is such that objects with a ${BzK}\equiv (z-K)-(B-z)\gt -0.2$ are identified as actively star-forming galaxies at $z\sim 1\mbox{--}2$ while objects with a ${BzK}\lt -0.2$ and $(z-K)\gt 2.5$ are identified as passively evolving systems at similar redshifts. Variations of the BzK diagram can be used to extend this method to higher redshifts (Daddi et al. 2004; Guo et al. 2012). The so-called VJL and iHM diagrams use the $(J-3.6\,\mu {\rm{m}})$ versus $(V-J)$ and $(H-4.5\,\mu {\rm{m}})$ versus $(i-H)$ colors to identify star-forming and quiescent galaxies at $z\sim 2.5$ and 3.5 respectively. It is based on the same principle as the BzK diagram with the $(J-3.6\,\mu {\rm{m}})$ and $(H-4.5\,\mu {\rm{m}})$ colors probing the 4000 Å break at higher redshifts.

We use the BzK, VJL, and iHM selections (Daddi et al. 2004; Guo et al. 2012) to identify passively evolving systems in the CANDELS COSMOS at $z\gt 1$. Figure 22 shows the BzK color–color plot along with the higher redshift VJL and iHM plots. The star-forming and quiescent galaxies in each plot are identified in the blue and red regions, respectively. The inset in each plot shows the photometric redshift distribution. As expected, these sources are very red in both the $(B-z)$ and $(z-K)$ colors and the corresponding ones in the VJL and iHM diagrams. This is indicative of the pronounced Balmer/4000 Å break in the SED of these galaxies and the general old stellar population. Figure 22 inset shows the redshift distribution of the quiescent galaxies in red. The BzK-identified passive systems have a mean photometric redshift of z = 1.61 and a distribution that is consistent with the population being at $z\gt 1$. Furthermore this population has a mean stellar mass of ${10}^{10.97}\,{M}_{\odot }$. The photometric redshift and stellar mass distributions and mean values along with the colors of the passive BzK galaxies are consistent with the quiescent galaxy selection at high redshift. One of the passive galaxies identified in the CANDELS COSMOS area has been spectroscopically confirmed at z = 1.265. This galaxy has a stellar mass of ${10}^{10.98}\,{M}_{\odot }$ and photometric redshift of ${z}_{\mathrm{phot}}=1.29$ consistent with the spectroscopic redshift and with the Galaxy being massive and old at $z\gt 1$. The VJL-selected passive systems have mean photometric redshift of z = 2.33 and mean stellar mass of ${10}^{11.03}\,{M}_{\odot }$. The photometric redshift and mass distributions are consistent with passive old galaxy selection at $z\sim 2\mbox{--}3$. The iHM redshift distribution shows a bi-modality with a large fraction of lower redshift galaxies in the selection. Guo et al. (2012) discussed the 50% contaminant fraction in the iHM color selection. We have 16 spectroscopically confirmed star-forming galaxies in the BzK plot at $z\sim 1.5$ and two at $z\sim 2.5$ in the VJL. All these sources have colors consistent with the expected values from the selection criteria. There is only one source in the spectroscopic sample with ${z}_{\mathrm{spec}}=1.564$ with inconsistent $(B-z)$ and $(z-K)$ colors compared to what is expected for the sBzK sample. This source has ${z}_{\mathrm{phot}}=0.682$ and is one of the outliers in Figure 11.

Zoom In Zoom Out Reset image size

The BzK color diagram is well calibrated using large and highly complete spectroscopic samples of galaxies (Daddi et al. 2004) and furthermore it predicts the location of stars on the $(z-K)$ versus $(B-z)$ plane (through the criterion $(z-K)\lt 0.3(B-z)-0.5;$ Daddi et al. 2004). Guo et al. (2012) recently extended this analysis to higher redshifts using redder filters and in particular the Spitzer 3.6 μm and 4.5 μm observations. This allows us to examine the colors of the stars measured in our catalog against predictions made by the BzK, VJL, and iHM color–color diagrams. Figure 22 shows the BzK colors of point sources from our catalog, identified from the SExtractor CLASS_STAR parameter, along with the predicted locus from Daddi et al. (2004). We further show the $(J-3.6\,\mu {\rm{m}})$ versus $(V-J)$ and $(H-4.5\,\mu {\rm{m}})$ versus $(i-H)$ colors of stars in our catalog along with the corresponding synthetic colors of stars from BaSeL stellar library (Lejeune et al. 1997; Westera et al. 2002) similar to Guo et al. (2012). We see from Figure 22 that the measured near-infrared and specifically infrared Spitzer colors of the stars in our catalog are consistent with the predictions from stellar models.