CANDELS MULTIWAVELENGTH CATALOGS: SOURCE IDENTIFICATION AND PHOTOMETRY IN THE CANDELS UKIDSS ULTRA-DEEP SURVEY FIELD

The CANDELS UDS field was covered by HST using a mosaic grid of tiles that was observed over two epochs. During each epoch, the field was imaged in one orbit (∼2000 s) with the Wide Field Camera 3 (WFC3) split into F125W (1/3 orbit) and F160W (2/3 orbit) together with parallel exposures using the Advanced Camera for Surveys (ACS) in F606W and F814W.

The WFC3 mosaics are composed of a grid of 4 × 11 tiles (see Figure 3; top) at spacing intervals that maximize the coverage without introducing gaps between tiles, resulting in a final rectangular field of view of ∼223 × 9'. The long axis is at a position angle of −90°. Exposures were oriented so that the ACS parallels are offset along the long axis of the mosaic, producing a similar-sized mosaic overlapping most of the WFC3 mosaic, except at its edges where some tiles are only covered by WFC3 or by ACS. We refer to Grogin et al. (2011, Figure 14) and Koekemoer et al. (2011, Figures 17–22) for details on the HST data set, mosaics, and data reduction. The final UDS HST mosaics (e.g., drizzled science images and inverse variance weight images) are publicly accessible via the STScI archive.²⁵

Zoom In Zoom Out Reset image size

The CANDELS UDS field is covered by a large number of ground-based data. Deep near-infrared data of an area of 0.8 deg² (including the CANDELS field) were obtained in J, H, and K as part of the UKIDSS (Lawrence et al. 2007) using the Wide Field Camera (WFCAM) on the UKIRT telescope. The current public data release (UKIDSS DR8) reaches median depths of J = 24.9, H = 24.2, K = 24.6 (5σ). Intermediate data releases including images are available from the WFCAM Science Archive.²⁶ O. Almaini et al. (in preparation) provide details on the data.

The full UKIDSS field of view was also imaged by the Canada–France–Hawaii Telescope (CFHT) in the u-band with MegaCam (PIs: O. Almaini, S. Foucaud; O. Almaini et al., in preparation).

The CANDELS UDS field was also observed in other near-infrared bands as part of the High Acuity Wide field K-band Imager (HAWK-I) UDS and GOODS-S survey (HUGS; VLT large program ID 186.A-0898, PI: A. Fontana; A. Fontana et al., in preparation) using HAWK-I on the VLT. About 95% of the CANDELS UDS was covered by three HAWK-I pointings in both the Y and K_s bands. All three pointings were imaged in Y and K_s for ∼8 hr and ∼13 hr, respectively. We refer to A. Fontana et al. (in preparation) for more details on the HUGS data.

A large set of optical imaging observations in the UDS field were taken with the Suprime-Cam on the Subaru Telescope as part of the Subaru/XMM-Newton Deep Survey (SXDS) in B, V, R_c, i', and z'-band. These data reach a 3σ limit (2 arcsec diameter aperture) magnitude of B = 28.4, V = 27.8, R = 27.7, i = 27.7, and z = 26.6 (Furusawa et al. 2008). Mosaics (registered to the CANDELS astrometry) are available online²⁷ (see also Cirasuolo et al. 2010).

Apart from the shallow (4 × 30 s) IRAC coverage obtained by the Spitzer Wide-Area Infrared Extragalactic Survey (SWIRE; Lonsdale et al. 2003), the UDS was surveyed as part of a Spitzer cycle 4 Legacy Program: the Spitzer UKIDSS Ultra Deep Survey (hereafter SpUDS; PI: J. Dunlop). SpUDS covers an area of about 1 deg² (see Figure 2) in the four IRAC channels (3.6, 4.5, 5.8, and 8.0 μm) and reaches a depth of 24.7 mag (1σ) at 4.5 μm. Images (and SExtractor catalogs) for SpUDS can be found online.²⁸

A subregion of SpUDS (0.40 deg²; see Figure 2)—that also contains the full CANDELS UDS field—has recently been observed as part of the Spitzer Extended Deep Survey (SEDS hereafter; PI: G. Fazio) during the Spitzer Warm Mission²⁹ at 3.6 and 4.5 μm. A final UDS mosaic incorporating all coextensive exposures from SEDS, SWIRE and SpUDS was generated by the SEDS team so as to reach the desired 12 hr total integration time/pixel within the SEDS footprint (Ashby et al. 2013). SEDS reaches a point-source sensitivity of 26 AB mag (3σ) at both 3.6 and 4.5 μm.

Additional IRAC observations of parts of each SEDS field are now underway as part of the SEDS-CANDELS program (PI: G. Fazio). The S-CANDELS observations of the UDS (PID 80218), when completed, will cover ∼150 arcmin² within the CANDELS area and reach a maximum depth of about 50 hr (four times the existing depth).

A number of spectroscopic observations were conducted in the UDS field. They often targeted specific types of objects: passively evolving galaxies (Yamada et al. 2005), radio sources (Simpson et al. 2006, 2012; Vardoulaki et al. 2008; M. Akiyama et al., in preparation; H. Pearce et al., in preparation), QSOs (Smail et al. 2008), Lyα emitters (Ouchi et al. 2008), galaxy cluster members (Geach et al. 2007; van Breukelen et al. 2007; Papovich et al. 2010; Tanaka et al. 2010; Finoguenov et al. 2010).

Extensive spectroscopic campaigns have recently increased the number of known redshifts in the UDS field. A spectroscopic campaign using the Magellan/IMACS multi-object spectrograph (PIs: M. Cooper & B. Weiner) provides reliable spectroscopic redshifts for 272 sources within the CANDELS UDS field of view (M. Cooper et al., in preparation³⁰). The UDSz ongoing ESO Large Programme (PI: O. Almaini) is being carried out with the VLT/VIMOS and VLT/FORS2 spectrographs and is obtaining spectra for ∼3500 sources in the full UDS field. Most of these spectroscopic redshifts are still proprietary, however, and will be added to the CANDELS catalog as soon as they become publicly released. At the time of publication, only 210 sources in the catalog have a non-proprietary spectroscopic redshift.

The field of view covered by the F160W data is about 201.7 arcmin². Exposures are shorter at the eastern end of the mosaic (to accommodate F350LP exposures during the same orbit, see Koekemoer et al. 2011, Figure 17). We estimate the limiting magnitudes as a function of position (from the rms map). They are computed for each pixel rescaled to an area of 1 arcsec² at a 1σ level. Figure 3 shows the distribution of the limiting magnitudes across the F160W image and its associated histogram. The peaks in the distribution are found at 27.75, 28.06, and 28.30 and we therefore clearly identify three main areas at different depths: mag  <  27.90 (the ∼1/3 shallower eastern tiles), 27.90 < mag  <  28.26 (the 2/3 deeper Western tiles and tile overlaps in the shallow region), and mag >28.26 (tile overlaps in the deeper region), respectively, corresponding to regions of about 58.7, 136.9, and 5.9 arcmin². The present catalog also contains for each source an estimation of the limiting magnitude at the position where the source is detected (see Section 3.3 for details).

3.2.1. Modifications of the SExtractor Software

3.2.2. SExtractor Cold and Hot Detection Modes

Traditional wide field surveys were usually not very deep. The source extraction was therefore focused on deblending bright, extended sources, avoiding, for example, the separation of galaxy sub-structures into multiple objects. For deeper surveys, source extraction was usually tuned to push the detection in order to pick up faint and small galaxies. In contrast, recent surveys, and in particular CANDELS, reach unprecedented depth over wide areas. The scientific goals of such surveys extend from the study of nearby sources to the discovery of the farthest and faintest galaxies in the early universe. It has been infeasible to find a single setup for the extraction of sources. We therefore need to adapt the detection methodology. Recent works have adopted a rather simple strategy to deal with this issue, using a two-step approach (see, e.g., Gray et al. 2009, and references therein). First, SExtractor is run in a so-called "cold" mode that efficiently extracts and deblends the brightest sources. It is then run in a so-called "hot" mode optimized for the detection of faint objects. The parameters were further adjusted to obtain a consistent photometry between the cold and hot modes. The adopted SExtractor cold and hot mode parameters are provided in Appendix A.

Figure 4 illustrates the differences between the cold and hot source extraction modes on the F160W image in the vicinity of several extended, clumpy galaxies. The cold mode detects a clumpy galaxy as a single object, merging all the clumps together into a single source, while the hot mode tends to separate clumps into individual objects. Nevertheless, the hot mode detects the faintest objects of the image that were missed by the cold mode. By making use of these two complementary detection modes, we aim to produce the most reliable and complete source catalog possible. SExtractor was run twice to create the cold and hot mode catalogs and associated segmentation maps. The cold and hot mode catalogs contain 27,167 and 37,715 sources, respectively.

Zoom In Zoom Out Reset image size

3.2.3. Cold + Hot Combination Routine

The cold and hot catalogs were merged. The adopted cold + hot combination routine was adapted from GALAPAGOS, a software designed for source extraction, light-profile modeling and catalog compilation of large astronomical surveys (see Barden et al. 2012 and the GALAPAGOS manual for details³²).

The combined cold + hot catalog includes all the sources detected in the cold mode + the sources detected in the hot mode at positions where no cold source was detected. In short, for each source in the hot mode, the routine checks whether it falls within the Kron ellipse of a cold mode detected source. If it does, the source is ignored. The final merged catalog therefore contains the full cold mode catalog followed by the "hot mode only" sources (with new updated identification indices).

For example, in Figure 4, the routine will consider the different clumps of the extended galaxies as part of the same object—because they are all detected within the same Kron ellipse in the cold mode—and therefore appear as one unique object (one line) in the merged catalog (whose SExtractor characteristics are reproduced from the cold mode catalog). Sources that are only detected in the hot mode and isolated (i.e., not within a Kron ellipse of a cold mode source) are simply added to the catalog with the SExtractor measurements of the hot mode catalog.

Figure 5 shows a region of the F160W image with the SExtractor segmentation maps of the cold, hot, and cold + hot modes. In the final combined segmentation map, the isophotal areas for sources detected in the cold mode are directly reproduced from the cold mode segmentation map. The isophotal areas for sources only detected in the hot mode are then added with their new identification number from the cold + hot merged catalog.

Zoom In Zoom Out Reset image size

The final combined F160W-selected catalog contains 35,932 sources. Figure 6 shows the source number counts (per bin of 0.5 mag; also listed in Table 2) with the contributions of the cold and hot modes overlaid. The flux densities and uncertainties in the F160W-band reported in the multiwavelength catalog correspond to the SExtractor outputs FLUX_BEST and FLUXERR_BEST, converted into μJy using the Koekemoer et al. (2011) zero points.

Zoom In Zoom Out Reset image size

Table 1. UDS Data Set

Instrument	Filter	Central	FWHM	Limiting	Surveya
		Wavelength		Magnitude
		(nm)	(arcsec)	(5σ, 1 FWHM radius, AB)
CFHT/MegaCam	u	386	0.86	27.68	(1)
Subaru/Suprime-Cam	B	450	0.82	28.38	(2)
	V	548	0.82	28.01	(2)
	Rc	650	0.80	27.78	(2)
	i'	768	0.82	27.69	(2)
	z'	889	0.81	26.67	(2)
HST/ACS	F606W	598	0.10	28.49	(3)
	F814W	791	0.10	28.53	(3)
HST/WFC3	F125W	1250	0.20	27.35	(3)
	F160W	1539	0.20	27.45	(3)
VLT/HAWK-Ib	Y	1019	0.42/0.52/0.49	27.05/26.73/26.69	(4)
	Ks	2147	0.36/0.42/0.37	26.16/25.92/25.98	(4)
UKIRT/WFCAM	J	1251	0.76	25.63	(1)
	H	1636	0.80	24.76	(1)
	K	2206	0.70	25.39	(1)
Spitzer/IRAC	3.6 μm	3562	∼1.9	24.72	(5)
	4.5 μm	4512	∼1.9	24.61	(5)
	5.8 μm	5686	2.08	22.30	(6)
	8.0 μm	7936	2.20	22.26	(6)

Notes. ^a(1) UKIDSS: O. Almaini et al., in preparation; (2) SXDS: Furusawa et al. 2008; (3) CANDELS: Koekemoer et al. 2011; (4) HUGS: A. Fontana et al., in preparation (5) SEDS: Ashby et al. 2013. (6) SpUDS. ^bFWHM and limiting magnitudes are provided for the three HAWK-I pointings following the scheme Pointing1/Pointing2/Pointing3 (i.e., Central/West/East; see A. Fontana et al., in preparation).

Download table as:  ASCIITypeset image

Table 2. Number Counts in the CANDELS UDS F160W Image

Mag	N mag−1 deg−2a
15.75	250 ± 94
16.25	428 ± 124
16.75	571 ± 143
17.25	749 ± 164
17.75	1071 ± 195
18.25	785 ± 167
18.75	1784 ± 252
19.25	2855 ± 319
19.75	4176 ± 386
20.25	7209 ± 507
20.75	9101 ± 570
21.25	13740 ± 700
21.75	20664 ± 859
22.25	26946 ± 981
22.75	36974 ± 1149
23.25	53070 ± 1376
23.75	73449 ± 1619
24.25	99003 ± 1880
24.75	122950 ± 2095
25.25	154357 ± 2347
25.75	184443 ± 2566
26.25	216207 ± 2778

Note. ^aUncertainties on the number counts are Poissonian.

Download table as:  ASCIITypeset image

The completeness limit of the UDS F160W catalog was estimated by comparing with deeper data in the GOODS-South field. The central part of this field was observed in 10 epochs (Grogin et al. 2011), i.e., five times the depth of UDS. We ran SExtractor (cold + hot) on the ten-epoch GOODS-S combined F160W mosaic and derived number counts similar to those of the UDS. By comparing number counts in GOODS-S Deep with UDS, we estimate that the 90% (50%) limit of completeness of the UDS F160W data is 26.70 (27.05).

• 1.  
  F160W limiting magnitude: for each source in the catalog, we derive a limiting magnitude indicating the depth of the F160W image in the region where the source falls. This limiting magnitude is derived from the square root of the average of the rms squared in the SExtractor segmentation map of each source scaled to an area of 1 arcsec². In practice, the limiting sensitivity in HST images depends on the source size, and faint galaxies can be significantly smaller than 1 arcsec² and thus be detected fainter than this fiducial magnitude limit. The knowledge of the depth fluctuation of the detection image is fundamental for any future volume-sensitive statistics (e.g., luminosity functions, etc.).
• 2.  
  Flag: not all objects detected by SExtractor are real. For example, SExtractor typically detects spurious sources in the spikes of very bright stars. Some sources also fall at the edges of the detection image where the photometry is not optimal. A flag column is therefore included in the catalog which indicates star spikes and bright halos as well as large artifacts and noisy edges (see Appendix B for details).

TFIT requires a careful preparation of the low-resolution images.

• 1.  
  Astrometry. The low-resolution images should also have a compliant astrometry with the high-resolution images. The pixel scale of the low-resolution image should be equal to (or an integer multiple of) the pixel scale of the high-resolution image. All ground-based images were resampled to the F160W pixel scale of 006 and aligned to the CANDELS HST data astrometry using swarp (Bertin 2010). We keep the original pixel scale (06) of the Spitzer data (SEDS, SpUDS). The astrometry of SEDS (Ashby et al. 2013) is compliant with the astrometry of the HST images so no realignment was needed. This was not the case for the SpUDS data that were therefore reprojected to the HST astrometry also using swarp.
• 2.  
  Background subtraction. The low-resolution images first need to be background-subtracted. For each image, a first rough background approximation was determined by smoothing the image on large scales using a large-annulus ring-median filter and then subtracted from the image. The image was then smoothed to the corresponding image PSF scale and sources were masked. To account for the wings of the sources, the source masks were enlarged by convolving with a Gaussian kernel. Several iterations were performed, starting from the brightest sources with high signal-to-noise (S/N) ratio down to sources with a S/N ∼5 above the mean background. The non-masked pixels were used to create a "noise" map, which was interpolated to determine the background in the masked pixels.

We tested the background subtraction routine by sampling the noise in the background-subtracted low-resolution images. We randomly added 1000 artificial sources in the F160W image (privileging regions without sources) and derived the photometry of their non-existent counterparts in the low-resolution images. We expect the S/N (flux density over uncertainty) to be a Gaussian, centered to zero with σ = 1. The background subtraction technique works well for ground-based data.

For IRAC data, however, both the mean S/N and the TFIT residual image (see Section 5.4) are found to be slightly negative (∼10⁻⁴ MJy sr⁻¹), i.e., the routine is over-subtracting the background. Although this quantity is negligible for bright objects, it becomes dominant for faint sources. We therefore implemented an additional background correction for the IRAC data. We ran TFIT on the background-subtracted images, subtracted the TFIT output collage—an image of the sources as modeled by TFIT—and repeated the background subtraction procedure. We then repeated the test with artificial sources and found a Gaussian-like distribution of the S/N and mean residuals <10⁻⁶ MJy sr⁻¹). TFIT was run a fourth time on the final image.

One of the key processes within TFIT is the smoothing of the high resolution detection image to the PSF of the lower resolution measurement image. Such a smoothing process is performed by applying a convolution kernel. The derivation of the convolution kernel is not done by TFIT and must be provided beforehand by the users.

Several techniques have been developed to derive an accurate convolution kernel, in particular those by Alard & Lupton (1998) and Alard (2000). In the case of the UDS visible and near-infrared ground-based data, we adopt a slightly simpler method based on analysis in the Fourier space, similar to the convolution kernel derivation technique presented in Aniano et al. (2011). Such a method can be implemented with standard astronomical tools for image analysis and has been successfully used for the GOODS-MUSIC (Grazian et al. 2006; Santini et al. 2009) and GOODS-ERS database (Santini et al. 2012). In brief, we first derive the PSFs of the detection and measurement image, PSF₁ and PSF₂, respectively, e.g., by summing up stars in the field. These two PSFs are then normalized. The convolution kernel K is, by definition,

$$\begin{equation} {\rm PSF}_2(x,y)=K(x,y) \otimes {\rm PSF}_1(x,y). \end{equation} \tag{ 1 }$$

The derivation of the exact shape of the convolution kernel is done in the Fourier space, i.e., if ℵ = FT(K), ℘₁ = FT(P₁), and ℘₂ = FT(P₂), the Fourier transform of the kernel is given by

$$\begin{equation} {\aleph }=\frac{{\wp }_2}{{\wp }_1}. \end{equation} \tag{ 2 }$$

A low passband filter (LPBF)³⁴ was applied in the Fourier domain to remove the effects of noise and suppress the high frequency fluctuations. A residual was computed from the two PSFs in order to check for the validity of the derived filter. Finally, ℵ was transformed back to the pixel space and normalized to unity (to preserve the flux of the objects) following:

$$\begin{equation} K(x,y)={\rm FT}^{-1}({\aleph }\cdot {\rm LPBF}). \end{equation} \tag{ 3 }$$

If the high and low-resolution images have significantly different resolutions—such as HST and Spitzer—the PSF of the low-resolution images can be used directly as the convolution kernel. For IRAC, we obtained a set of 5 × spatially oversampled PSFs (W. Hoffman 2010, private communication; see also the IRAC Data Handbook, Appendix C). There are 25 such oversampled PSFs for each channel, representing a 5 × 5 sampling across the detector to map PSF variation. We averaged together these 25 location-specific PSFs to construct a single, location-averaged, PSF for each channel. The SEDS and SpUDS images are mosaics of a series of Astronomical Observation Requests (AOR; 107 and 26 respectively) that were executed with different PAs. The model PSF is rotated by the PAs for each AOR. The average of all rotated PSFs is then smoothed by a boxcar kernel, a mandatory step because the model PSF is sharper than the true IRAC PSF. Tests on the SEDS and SpUDS data showed that the optimal smoothing is obtained using a boxcar size of 23 pixels (006 pixel⁻¹). Finally, the smoothed PSF are circularized by putting the pixels outside a diameter of 25'' to zero and normalizing the total "flux" in the template to unity.

In order to derive photometry, TFIT uses as the source template only pixels within a segmented region. In past works using TFIT (e.g., Laidler et al. 2007), this region was defined by the SExtractor segmentation map. This implies that only pixels within the HST isophotal area were used to construct templates for TFIT. However, past studies (e.g., De Santis et al. 2007) have shown that SExtractor tends to underestimate the size of the source isophotal areas especially for faint and/or small sources and that therefore part of the source flux was missed. Unfortunately, it is challenging to estimate exactly how much of the flux is missed because (1) we do not know how much flux is actually outside the segmented region for each source, and (2) a portion of this flux will be considered as background in the preparation of the low-resolution images for TFIT. The size of the isophotal area depends on the threshold adopted in the detection image. However, for faint sources, a significant fraction of their flux probably falls below the isophotal detection threshold. This isophotal area is fed as an input parameter to TFIT which, as a consequence, tends to miss a significant fraction of the flux and underestimate the total magnitude. We run a series of tests to quantify how SExtractor underestimates the isophotal area and implement a "dilation" correction to compensate for this underestimation. This issue mostly concerns sources that are either faint or have a small isophotal area, two populations that are overlapping as size strongly correlates with brightness. We will interchangeably refer in the following sections to faint/bright or small/large sources.

5.3.1. SExtractor Estimation of Source Isophotal Area

The following tests were done on the F160W band since it is the catalog detection band and the reference high-resolution image for our TFIT run.

As mentioned in Section 3.2.3, the central part of the GOODS-South field was observed in 10 epochs, five times the UDS depth. We ran SExtractor ("cold + hot" then merged) on the full GOODS-S F160W image ("10Epoch") and also on an image made only from the first two epochs ("2Epoch"). Figure 8 shows the ratio of ISOAREAF_IMAGE parameter between the "10Epoch" and "2Epoch" images. SExtractor derives larger source isophotal areas (ISOAREAF_IMAGE, the number of pixels above the adopted threshold) in the deeper image. As expected from previous studies, the isophotal area is smaller in the shallower image for faint sources. Although SExtractor nicely recovers the total ISOAREAF_IMAGE for bright sources (ISOAREAF_IMAGE >1000 pixels), it gradually loses part of the source for fainter objects with an underestimation of at least a factor of 4 for sources with ISOAREAF_IMAGE =60 pixels.

Zoom In Zoom Out Reset image size

As a test of how isophotal area affects photometry, we smoothed the F160W image to the 04 resolution of the VLT/HAWK-I data by convolving the image by the corresponding kernel. Using this same kernel, we ran TFIT, intending to recover the input photometry of the original (unsmoothed) F160W image. Figure 9 shows the difference between the output of TFIT and the input F160W magnitude versus input magnitude (top) and SExtractor isophotal area (bottom; i.e., the ISOAREAF_IMAGE parameter). Figure 9 shows that the TFIT output is in good agreement with the F160W input values for bright sources (mag_F160W < 20 and ISOAREAF_IMAGE >1000 pixels) but worsens rapidly with faintness (and smaller isophotal area) with an average offset of −0.24 for 26 < mag_F160W < 27 (−0.26 for sources with ISOAREAF_IMAGE <60 pixels). As expected, SExtractor tends to underestimate the isophotal area of sources and therefore TFIT fails to recover the expected input photometry for faint objects. Smoothing the F160W data to match even lower-resolution data, e.g., 08 (Subaru/Suprime-Cam) or Spitzer/IRAC gives trends similar to those of Figure 9, although the discrepancies between input and TFIT output are less dramatic with an average offset of −0.1 for 26 < F160W < 27.

Zoom In Zoom Out Reset image size

5.3.2. Dilation Correction

In order to deal with this particular issue, we adopt a correction technique similar to the one implemented within the ConvPhot software. ConvPhot automatically dilates the area of every source on the segmentation map generated by SExtractor. Sources with an area above a minimum threshold (m_(AREA)) are dilated by a constant factor of 4 (i.e., doubling the area). Sources smaller than m_(AREA) are dilated to reach this minimum threshold. The dilation correction is done using a publicly available routine called dilate (see de Santis et al. 2007). dilate expands the segmentation map by a fixed factor but prevents the merging between close sources.

We refine this dilation technique by modifying the dilate routine to adapt it to the present data set. As observed in Figure 9 (as well as in Figure 8), TFIT recovers well the photometry of sources with ISOAREAF_IMAGE  > 1000 pixels with a difference between the input and TFIT output magnitude of less than <0.01. We therefore do not apply any dilation for these sources. The largest discrepancies are obtained for sources with an isophotal area smaller than 60 pixels (with an underestimation of a factor of four of the isophotal area in the case of GOODS-South "2Epoch"/"10Epoch" comparison). We therefore adopt a constant dilation factor of 4 for sources with ISOAREAF_IMAGE <60 pixels. As observed in Figure 8, the loss of isophotal area by SExtractor can be parameterized by an hyperbola. We therefore opt for a smooth transition for sources with 60 < ISOAREAF_IMAGE <1000. To summarize, we adopt the following dilation factor corrections:

• 1.  
  If ISOAREAF_IMAGE > 1000: Dilated_area = area.
• 2.  
  If 60 < ISOAREAF_IMAGE < 1000: Dilated_area =−0.0004 × area² + 1.25 × area +166.
• 3.  
  If ISOAREAF_IMAGE <60: Dilated_area =4 × area.

Figure 8 shows the adopted dilation correction. Figure 10 shows its effect on the segmentation map. It also shows the final discrepancies between the input F160W photometry and the TFIT output after dilation. The adopted dilation correction permits us to recover, with more accuracy, the F160W input photometry.

Zoom In Zoom Out Reset image size

This dilated segmentation map (and updated input source catalog) is adopted for all the TFIT runs on the low-resolution data (ground-based and Spitzer/IRAC data).

TFIT incorporates a procedure that quantifies any small geometric distortions and shifts that could persist even after a careful alignment of the high- and low-resolution images and creates a series of shifted kernels based on these distortions (as part of the "registration dance" stage; see Laidler et al. 2007, for details). We ran TFIT a second time with these kernels in order to improve the alignment of the model with the data. The shifts are typically less than 01 for ground based data and less than 03 for the Spitzer images. For background-subtraction purposes described in Section 5.1, TFIT was run a third time on the IRAC data.

TFIT produces a residual image derived by subtracting the model image—a collage of the source template scaled by the TFIT flux measurement—from the low-resolution image. The residual image permits us to visualize any TFIT problem such as a non-optimal convolution kernel. Objects that are detected in a low-resolution image but are not in the input catalog derived from the high-resolution data also remain unchanged on this residual image. Residual images of the TFIT second pass are shown in Figure 11 for the B, Y and SEDS 3.6 μm.

Zoom In Zoom Out Reset image size

We finally convert the TFIT output into μJy using the low-resolution image zero points.

• 1.  
  J and H bands. Figure 12 shows the comparison between the HST/WFC3 (F160W, and F125W) and UKIDSS DR8 (J and H) photometry. The WFC3 and UKIDSS WFCAM filters are slightly different (see Figure 1). In order to account for the discrepancy between filters, the UKIDSS flux densities are converted to the WFC3 photometric system using the color corrections adopted by Koekemoer et al. (2011): F125W = J + 0.05(J − H) and F160W = H + 0.25(J − H). In Figure 12, we only plot sources with S/N > 10 and F160W CLASS_STAR <0.98 to avoid bright (possibly saturated) stars. The agreement is good and shows no systematic offset (i.e., no zero-point issue) or trends.
• 2.  
  K bands. Similarly, we compare the HAWK-I K_s and UKIDSS K photometry (see Figure 12). No correction is applied because the filters are consistent and correction would be negligible. We also find a good agreement between the two bands with no specific discrepancy.

We study the (optical to near-infrared) colors of stars in the catalog. We first isolate stars in the multiwavelength catalog by selecting sources that have CLASS_STAR ⩾0.98. We build a library of synthetic models of stars from the Bruzual–Persson–Gunn–Stryker Atlas of stars (Gunn & Stryker 1983) that we convolve with the response curves of the different filters. We then compare their colors to the colors of the stars in the catalog and derive a series of color–color diagrams; Figure 13 shows four of these diagrams. The agreement is excellent.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We also look at the distribution of the sources in a BzK color–color diagram (introduced by Daddi et al. 2004) which preferentially isolates z > 1.4 galaxies. Figure 14 shows the BzK diagram derived from the catalog (using the K-band photometry from the UKIDSS DR8 data). Small corrections were applied to account for the differences in filters between UDS and Daddi et al. (2004).³⁵ Sources with CLASS_STAR >0.98 (orange dots) are, as expected, preferentially found in the stellar locus of such diagrams (Daddi et al. 2004). All (but one) sources with spectroscopic redshift 1.4 < z_spec < 2.5 (red squares) are sBzK-selected galaxies (within error bars), i.e., have BzK ⩾ −0.2 (where BzK = (z − K) − (B − z)) and therefore lie in the typical locus of star-forming galaxies at 1.4 < z < 2.5. The only source not sBzK-selected is found in the locus of pBzK-selected galaxies (BzK < −0.2 and z − K > 2.5), i.e., has colors consistent with a passive galaxy at 1.4 < z < 2.5.

Zoom In Zoom Out Reset image size

As mentioned earlier, we also derive TFIT photometry for the SpUDS-only 3.6 and 4.5 μm images. Figure 15 shows the comparison between the TFIT photometry for the SEDS and SpUDS IRAC data. The agreement is good with no systematic offset and no suspicious trend with magnitude—i.e., the difference is fairly symmetric at a given magnitude—suggesting that there is no zero-point offset and no systematic background subtraction issue that could bias the faint sources either in the SEDS or the SpUDS mosaics.

Zoom In Zoom Out Reset image size

We also compare our TFIT SEDS photometry to the (StarFinder) SEDS catalog (Ashby et al. 2013). We made use of an early distribution of their catalog to the CANDELS group (SEDS team 2012, private communication) and cross-matched their source list with our F160W-selected catalog. In the SEDS data, confusion limit is an issue and it is therefore important to use a small aperture in order to avoid flux contamination from the wings of nearby sources. We therefore compare the TFIT SEDS photometry to their aperture-corrected StarFinder photometry derived in a 24 diameter aperture. The agreement is good. The discrepancy increases at faint magnitudes since StarFinder does not deblend faint sources. Their photometry may be contaminated by neighboring objects and over estimated hence the positive offset. At the bright end (mag <20), stars have magnitudes consistent in both catalogs. Bright galaxies however show a larger discrepancy. This was expected when considering magnitudes derived in such a small aperture. The match between TFIT SEDS and StarFinder is excellent for bright galaxies (<0.1 mag) when using the StarFinder 6'' diameter aperture-corrected photometry.

Figure 16 shows the source number counts derived from the present catalog in the four Spitzer/IRAC bands (SEDS UDS in red and SpUDS-only in blue). Counts are derived for different cuts in signal-to-noise (S/N =3, 4, 5) since IRAC photometry at lower S/N becomes rapidly unreliable and strongly biased by background contamination issues. We therefore strongly advise users to regard the IRAC flux density estimates for sources with S/N <3 with caution. Published source number counts from Fazio et al 2004 in the Boötes field and from Ashby et al. (2013) derived from the full SEDS survey (M. Ashby 2010, private communication) are overplotted for information. We also derive IRAC number counts from GOODS-MUSIC.³⁶ We precise that the GOODS IRAC 5.8 and 8.0 μm data (shown here as the open points) are much deeper than the SpUDS data.

Zoom In Zoom Out Reset image size

Our IRAC source number counts are in good agreement with previously published number counts. As mentioned earlier, the SpUDS data for 3.6 and 4.5 μm are included in the SEDS mosaic, although here we measure the photometry independently from a different TFIT run on the SpUDS images alone. The number counts are in perfect agreement up to the completeness limit of the SpUDS data, i.e., about 1.5 mag brighter than the SEDS UDS data. The agreement is also excellent with both the GOODS-MUSIC and the full SEDS survey (that also includes the present SEDS UDS data). As expected, the full-SEDS number counts (derived from the SEDS StarFinder aperture photometry catalog) fall below our TFIT SEDS UDS counts (starting 0.5 mag fainter) since we were able to detect and deblend more efficiently the faintest sources thanks to our reference F160W-selected source catalog coupled with TFIT photometry.

Additionally, we look at the distribution of sources in IRAC color–color diagrams known to be an efficient tool to isolate AGN. Figure 17 shows the well known mid-infrared AGN selection wedges in the [3.6]–[4.5] versus [5.8]–[8.0] color–color diagram first introduced by Stern et al. (2005) and the log(S_8.0/S_4.5) versus log(S_5.8/S_3.6) color–color diagram first introduced by Lacy et al. (2004) and recently revised by Donley et al. (2012). We plot only sources with S/N >3 in all four IRAC bands. As expected, stars (i.e., sources with CLASS_STAR >0.98; orange dots) have colors consistent with zero in the Vega system. The "Stern" selection wedge is more appropriate for bright sources and may fail quickly at the depth we are probing in this catalog (Donley et al. 2012). Similarly, the "Lacy" wedge is greatly contaminated by faint non-AGN sources and one should think to restrict the AGN selection to the less contaminated "Donley" selection wedge.

Zoom In Zoom Out Reset image size

X-ray point sources (blue dots; Ueda et al. 2008) are found to lie preferentially within the AGN selection wedges. Indeed, 61% are also mid-infrared-selected according to Stern et al. (2005) AGN selection (78% for Lacy et al. 2004), a result somehow surprising when comparing to past studies such as Gorjian et al. (2008) which find that only 28% of the X-ray detected sources in the Boötes field were also mid-infrared selected. The investigation of such discrepancy is however beyond the scope of this paper. As expected from past studies (Stern et al. 2000), the overlap of mid-infrared selected AGN with radio sources is smaller with only 33% (58%) of the (>100 μJy) radio-sources (green dots; Simpson et al. 2006) falling within the "Stern" ("Lacy") AGN wedge.

Intense work is currently being done within the CANDELS team to derive robust photometric redshifts and stellar mass estimates for all sources in the CANDELS multiwavelength catalogs. T. Dahlen et al. (in preparation) summarize the CANDELS team efforts to (1) compare photometric redshift and mass estimates from different codes (13 in total), (2) assess how well codes manage to recover the redshifts of objects with known spectroscopic redshift, and (3) determine how well codes reproduce stellar masses of simulated galaxies. The final goal is to converge all of them into one unique and optimal photometric redshift and stellar mass estimates recipe that is to be adopted for all CANDELS multiwavelength catalogs.

Photometric redshifts and stellar masses derived from the present UDS multiwavelength catalog will be presented in an upcoming paper. In this section, we make use of one of the 13 codes presented in T. Dahlen et al. (in preparation), namely zphot (Giallongo et al. 1998) as a first test of the photometry of the present catalog. zphot is a χ²-minimization procedure that finds the best-fitting template to the observed colors of a source out of a spectral library of galaxies. The spectral library was built using PEGASE 2.0 models (Fioc & Rocca-Volmerange 1997). We refer to Grazian et al. (2006) and references therein for further details on zphot.

We first run zphot on sources with available spectroscopic redshifts, derive "flux corrections," i.e., small shifts to be applied to the photometry in order to better match the galaxy templates and then run zphot on the full catalog.

Figure 18 shows the distribution of photometric redshifts for all sources with flag =0 in the UDS catalog. A series of peaks in redshifts is observed, suggesting the CANDELS UDS field contains a number of galaxy groups/clusters. We refer to A. Galametz et al. (in preparation) for a more detailed study of galaxy overdensities in the UDS. We will just note that two structures were already known in the CANDELS UDS field: a galaxy cluster at z ∼ 0.65 (Geach et al. 2007; see also Figure 7) and one of the highest redshift galaxy clusters known to date at z = 1.62 (Papovich et al. 2010, 2012; Tanaka et al. 2010), which are both clearly visible in the photometric redshift distribution.

Zoom In Zoom Out Reset image size

A comparison between photometric and spectroscopic redshifts is shown in Figure 19. The quality of the photometric redshifts is excellent. We further study the distribution of the relative difference z_spec − z_phot / (1 + z_spec). Its central peak can be cleanly fit by a tight Gaussian with an average scatter of 0.022. However, a large majority of the sources with a spectroscopic redshift in the UDS catalog are AGNs and the reliability derived from these sources may not reflect that of the full catalog, especially for normal faint galaxies.

Zoom In Zoom Out Reset image size