UV-TO-FIR ANALYSIS OF SPITZER/IRAC SOURCES IN THE EXTENDED GROTH STRIP. I. MULTI-WAVELENGTH PHOTOMETRY AND SPECTRAL ENERGY DISTRIBUTIONS

The EGS region has been observed in the X-ray (0.5–10 keV) using Chandra/ACIS during two observation cycles in 2002 and 2005. The AEGIS-X survey covers an area of 0.67 deg² in eight pointings (∼17' × 17' each), completely overlapping with the region covered by IRAC. The nominal exposure time of the frames is 200 ks per pixel, reaching limiting fluxes of 5.3 × 10⁻¹⁷ and 3.8 × 10⁻¹⁶ erg cm² s⁻¹ in the soft (0.5–2 keV) and hard (2–10 keV) bands, respectively. In this work, we use the data reduction and point source catalogs published in Laird et al. (2009; see also Nandra et al. 2005 for a first version of the catalogs). The two-band (soft and hard) merged catalog comprises 1325 sources with <1.5% spurious detections. The authors identified optical and NIR counterparts for 1013 and 830 sources, respectively, from the CFHTLS optical catalog (see Section 2.3.1), and the Spitzer/IRAC catalog of Barmby et al. (2008). The cross-match is based on maximum likelihood method with a search radius of 2'' (more than a factor of 1.5 the rms of their astrometric accuracy).

The GALEX (Martin et al. 2005) observed the EGS over three consecutive years providing deep UV data in two channels (FUV at 153 nm, and NUV at 231 nm) on a 1.13 deg² circular area around α = 14^h20^m, δ = +52°47'. The total exposure times for the composite stacks are 58 ks and 120 ks in the FUV and NUV filters, respectively. The approximate limiting magnitude in both bands is ∼25.1 mag.

2.3.1. CFHTLS/CFHT12K

2.3.2. MMT/Megacam

2.3.3. HST/ACS

2.3.4. Subaru Suprime-Cam

2.4.1. HST/NICMOS

2.4.2. Subaru MOIRCS

2.4.3. Palomar and Calar Alto Imaging

2.5.1. Spitzer/IRAC

2.5.2. Spitzer/MIPS 24 μm and 70 μm

A radio survey at 1.4 GHz (20 cm) of the northern half of the EGS (∼50% of the IRAC mosaic) was conducted with the Very Large Array (VLA) in its B configuration during 2003–2005. The AEGIS20 survey covers 0.73 deg² down to 130 μJy beam⁻¹ including a smaller region of 0.04 deg² with a 50 μJy detection limit (5σ). The data reduction and the source catalog, comprising 1123 sources, were presented in Ivison et al. (2007).

EGS has also been the target of an exhaustive and unique spectroscopic follow-up. As one of the DEEP2 fields, optical multi-object spectroscopy has been carried out from 2003 to 2005 with the Deep Imaging Multi-Object Spectrograph (DEIMOS; Faber et al. 2003) on the Keck II telescope. The observations cover the spectral range 640 nm < λ < 910 nm with a resolution of 0.14 nm. The DEEP2 DR3¹⁰ contains spectroscopic redshifts for 13,867 sources at 0 < z < 1.4 with a median redshift z = 0.75. The targets were selected from the CFHT12k-BRI images (within 1.31 deg²) in the magnitude range 18.5 ⩽ R ⩽ 24.1. The spectroscopic redshifts for around 70% of the sources are labeled with a high-quality flag (values of 3 and 4, meaning >95% success rate). Lower quality flags are considered unreliable and will be excluded from our analysis here and in Paper II.

We complemented the DEEP2 spectroscopy of z < 1.4 galaxies with redshifts for z ∼ 3 sources from the Lyman Break Galaxy (LBG) survey of Steidel et al. (2003). This survey covers a total area of 0.38 deg² divided in several fields, one of them centered in the EGS. The observations in EGS consist of a single 15' × 15' mask centered at α = 14^h17^m43^s, δ = 52°28'48'' observed with the Low Resolution Imaging Spectrometer (LRIS; Oke et al. 1995) on Keck. The spectra cover the 400–700 nm range with a median resolution of 0.75 nm. The targets were pre-selected based on the LBG color–color criteria (Steidel et al. 1996) including only candidates brighter than R = 25.5. The EGS catalog contains a total of 334 LBG candidates in the surveyed area. Out of them, 193 are spectroscopically confirmed to be at z ∼ 3. Unfortunately, the overlap with the IRAC frame is not complete (and some of the galaxies are extremely faint in the IRAC bands), and we were only able to identify 243 (72% of the spectroscopic sample) LBGs in our 3.6 + 4.5 μm selected catalog (we give more details on these sources in Paper II).

The source detection in the IRAC data was carried out separately in the 3.6 μm and 4.5 μm images using SExtractor (Bertin & Arnouts 1996). The complementary detection in the slightly shallower 4.5 μm band helps to alleviate the source confusion problems arising from the point-spread function (PSF) size and the remarkable depth of the IRAC data. Both catalogs were cross-matched using a 1''  search radius to remove repeated sources. This produces a master IRAC-selected catalog containing the sources detected in any of the two channels. Eventually, most sources are simultaneously detected in both channels.

The average survey depth is remarkably homogeneous across the strip, t_exp ∼ 10 ks, with the exception of two small areas with lower exposure at the top (δ>53525) and bottom (δ < 52025) of the mosaic. We took into account the lower exposure times near the edges of the images by defining two different areas: a shallower region with exposure time shorter than 3800 s (N(frames) ⩽ 20), and a deeper region covering the majority of the strip. The detection was carried out with different SExtractor parameters in each region, using a more conservative configuration for the shallower region. Then, we used this (more restrictive) catalog to purge some low significance detections in the other catalog within an overlapping area between them (N(frames) = 18–25). The purged catalog restricted to the area with N(frames) > 20 constitutes our master photometric catalog, and it covers an area of 0.50 deg².

After the detection of sources, we removed spurious sources in the wings of bright stars (where the PSF shows bright knots). For that purpose, first we made a preliminary detection of star-like sources based on the IRAC color–color criteria of Eisenhardt et al. (2004; see Section 5.4). Then, we eliminated detections within the typical distance where the contamination from the star is significant (typically ∼9'' for the typical star magnitudes and depth of the EGS observations). After masking the regions around stars, the total area covered by the catalog is 0.48 deg².

Aperture photometry for all 3.6 μm + 4.5 μm detected sources was measured with our own dedicated software (which takes into account pixel fractions appropriately; see PG08) in all four IRAC images, previously registered to the same World Coordinate System (WCS). The flux is measured in the four IRAC bands simultaneously. If a source is undetected in the shallower bands (i.e., [5.8] and [8.0]) we still measure an upper limit flux as three times the rms of the sky. The majority of the IRAC sources are unresolved in the 3.6 μm image (FWHM ∼ 2''). However, most of them are not point-like, but slightly extended. Consequently, PSF fitting is not effective and photometry is best measured with small circular apertures (PG08; Barmby et al. 2008; Wuyts et al. 2008; Ilbert et al. 2009). The flux measurement in all bands was carried out at the positions specified in the IRAC master catalog. We used a 2''  radius aperture and applied aperture corrections for each band derived from the PSF growth curves. The values of the correction are [0.32 ± 0.03,0.36 ± 0.02,0.53 ± 0.02, 0.65 ± 0.02] mag at [3.6,4.5,5.8,8.0] μm. The errors account for the typical WCS alignment uncertainties. For a small number of extended sources (∼2% of the total catalog, and 75% of them presenting [3.6] < 22.3), the 2'' aperture tend to underestimate the total magnitudes (by more than a 10%). These sources are typically bright nearby galaxies whose Kron (1980) radius is larger than ∼45. The flux measurement for these sources was performed in larger apertures enclosing the full object and applying the extended source aperture corrections given in the Spitzer/IRAC cookbook.

The uncertainties in the IRAC photometry were computed taking into account the contributions from the sky emission, the readout noise, the photon counting statistics, the uncertainties in the aperture corrections, and a 2% uncertainty from the zero-point absolute calibration (Reach et al. 2005). We did not assume the uncertainties resulting from SExtractor flux measurement. Instead, we used a more realistic method to determine the background noise that takes into account the effects of pixel-to-pixel signal correlation. This procedure has also been applied to measure the photometric uncertainties in all the bands and is outlined in Section 4.3. Nevertheless, a straightforward comparison to the SExtractor errors indicates that the noise correlation does not introduce a significant contribution to the flux uncertainties at bright magnitudes, leading to a median increment ⩽0.02 mag up to [3.6,4.5] ∼ 23.75.

A detailed comparison of our IRAC catalog of the EGS survey to the one published by Barmby et al. (2008) is presented in Section 3.3, including a discussion on the source confusion levels.

We estimated the completeness of the IRAC catalog by analyzing the recovery of simulated sources added in the mosaicked images. The simulations were carried out in the central regions of the mosaic where the coverage is uniform (t_exp ⩾ 10 ks). Artificial sources spanning a wide range of sizes (from 1'' to 6'') and brightnesses ([3.6] = 16–25 mag) were created on the IRAC images at random locations. The number of simulated sources was chosen to be representative of the Poisson uncertainty in the observed number densities. The source detection and photometry was performed again in the simulated images keeping the same SExtractor parameters as in the original frames. The success rate recovering the simulated sources allows us to estimate the completeness level as a function of magnitude. Figure 2 summarizes the completeness analysis in the four channels. For simplicity, we show results only for point-like sources. The completeness for extended sources is typically ∼10% lower at faint magnitudes, however, these sources represent a very small fraction of the catalog at these magnitudes. Figure 2 also shows the source density as function of the magnitude in each band. We find that the catalog is 85% complete for point sources with [3.6,4.5] = 23.75 mag and 75% complete at [3.6,4.5] = 24.75 mag. In the two other channels, the detection efficiency is significantly lower, with 85% completeness at [5.8,8.0] = 22.25, and the completeness dropping rapidly beyond that magnitude. Note that the forced photometric measurement in these bands provides a significant number of <5σ (vertical dashed lines in Figure 2) detections that would be missed otherwise.

Zoom In Zoom Out Reset image size

The lower panels of Figure 2 show the photometric uncertainties as a function of magnitude in the four IRAC bands. The red and green lines indicate the median and the level enclosing 90% of the distribution, respectively. For the region with the deepest coverage, we estimated a 3σ limiting magnitude of [3.6,4.5] ∼ 24.75 and [5.8,8.0] ∼ 22.90 from the median value of the sky rms in our default photometric apertures (2'' radius).

The 3.6+4.5 μm catalog contains 70,048 and 99,618 sources with [3.6] < 23.75 and [3.6] < 24.75, respectively. The median magnitude of the sample up to [3.6] < 24.75 is [3.6] = 23.14, and 75% of the sources present [3.6] < 23.96. Note that these numbers correspond to the IRAC 3.6+4.5 μm catalog before applying the deblending technique discussed in Section 5. Consequently, the number of sources quoted above is lower than in the final catalog (see Table 2).

Here we compare our IRAC photometric catalog to that published by Barmby et al. (2008, hereafter BAR08). BAR08 used the same data set (except for the GO flanking regions) and obtained final mosaics in all four IRAC bands with a very similar reduction to ours. Concerning the source detection, our method is slightly different from BAR08 since we detect galaxies in both the 3.6 and 4.5 μm channel instead of only using the bluer band. Our dual detection technique helps to alleviate the source confusion problems arising from the PSF size and the remarkable depth of the IRAC data. In addition, we measure fluxes in the four channels simultaneously, obtaining upper limit values for undetected sources in the shallower bands ([5.8] and [8.0]). Note also that we have increased the resolution of our catalog by deconvolving blended sources using higher resolution information from ground-based observations (Section 4.1). However, for the sake of clarity, we compare here the BAR08 catalog with ours before carrying out the deblending procedure.

BAR08 measured aperture photometry with SExtractor. The publicly available catalog¹¹ includes MAG_AUTO and MAG_ISO measurements, jointly with aperture magnitudes for several radii corrected to total magnitudes with empirical PSF corrections. We compare our photometry to the magnitudes measured in the 3.5 pixel aperture (21) by BAR08. Their aperture corrections agree with our measurements for the 4'' diameter aperture within the errors (due to alignment uncertainties).

BAR08 and our catalog are cross-correlated in the region of highest exposure (t_exp>4 ks) using a 1'' radius. The comparison of the WCS between the two mosaics is in very good agreement, with an rms of <005. We find 41,514 and 52,130 sources in common up to [3.6] < 23.75 and [3.6] < 24.75, respectively.

Attending to the density of sources per unit area, we find that our catalog includes 11% ± 5% and 18% ± 7% more sources than the catalog published by BAR08 at [3.6] < 23.75 mag and [3.6] < 24.75 mag, respectively. The uncertainties in these measurements were estimated by comparing number counts in 0.5 mag bins, including Monte Carlo simulations on the photometric errors.

The different source densities are a consequence of their more conservative SExtractor detection threshold. We have carefully chosen the SExtractor parameters differentiating between the regions with high/low coverage (Section 3.1), trying to push down the detection limits as much as possible without degrading the reliability of the entire catalog. As a consequence, our catalog recovers a larger number of sources at fainter magnitudes, and the completeness of our catalog is larger than BAR08 for the same magnitude. For example, they quoted a ⩽50% completeness for point-like sources at [3.6] = 23.75 mag, compared to our estimated 85%. Unfortunately, lowering the detection threshold inevitably increases the number of spurious detections. However, in the context of a merged multi-band photometric catalog, spurious detections can be efficiently identified as sources only detected by IRAC (cf. N(band) < 5), and we will show that the reliability is >97%, with false detections located almost uniquely around very bright sources (see Section 5.3).

Figure 3 shows the comparison of the photometric magnitudes in the four bands for both catalogs. In both cases, the flux was measured on circular apertures and corrected to total magnitudes using aperture corrections. We have corrected the comparison by a constant value of −0.05 mag in [3.6] and [4.5] and by −0.04 mag and −0.03 mag in [5.8] and [8.0], respectively. Such small offsets are attributed to slight differences in the data reduction (final absolute calibration, frame stacking, registering, and mosaicking) and in the aperture corrections. Despite the small offsets, the overall results in the four IRAC bands are in good agreement. Note that the average photometric error in our catalog for each magnitude bin (green bars) encloses 1σ of the values around the median difference (red and cyan lines).

Zoom In Zoom Out Reset image size

The lower panel of each plot in Figure 3 shows the comparison of the photometric errors in BAR08 and in our catalog. Our quoted photometric uncertainties tend to differ from BAR08, specially at faint magnitudes. In contrast with that paper, we have considered zero-point and WCS uncertainties, resulting in slightly larger uncertainties (∼0.05 mag) in our catalog for bright sources up to [3.6][ 4.5] ∼ 21 mag and [5.8][ 8.0] ∼ 19 mag. At fainter magnitudes, the photometric uncertainties increase with magnitude at a faster rate in the catalog of BAR08. The cause for this difference is the procedure to measure the background noise. Similarly to BAR08, we estimated this value from the sky variance measured in circular apertures at different locations of the images that are empty of sources (Section 4.3). However, the definition of an empty region depends on the limits of source detection. Therefore, given our higher detection fraction, our sky regions would contain, in principle, lower signal pixels effectively decreasing the rms. In addition, our photometric procedure estimates uncertainties on a source-by-source basis studying the background around each object in a independent way, while BAR08 relied on SExtractor photometric errors and applied a correction to them based on the average properties of the mosaic. Nevertheless, our estimates of the photometric errors are consistent with the observed scatter of the comparison between our photometry and that measured by BAR08.

The Rainbow code starts from a primary selection catalog (in our case, IRAC selected) and obtains merged photometry and spectroscopy in other bands. The first step is to identify the counterparts of the IRAC sources in the other bands, where SExtractor catalogs have been built following typical procedures. These catalogs are cross-correlated to the 3.6+4.5 μm positions using a 2'' search radius. An exception to this rule are the MIPS, radio, and X-ray catalogs. For the MIPS and radio bands, we used a 25 and 3'' radius, respectively, recognizing possible alignment and center estimation problems of the order of one pixel. For the latter, instead of using the WCS of the X-ray sources, we cross-matched to the positions of the IRAC counterparts (given in Laird et al. 2009) using a 1'' radius. These authors used as reference the IRAC catalog of BAR08, which covers a slightly lower area than ours. Thus, for sources outside of the BAR08 mosaic we cross-matched to the X-ray coordinates using a 2'' radius (see Section 5.2 for more details).

The IRAC sources are identified in this way with objects in all the other catalogs. One of the optical catalogs (typically the deepest; in the EGS, the Subaru R-band data) is used as a reference to narrow the following cross-correlations and alleviate confusion problems present in the IRAC images. As the cross-correlation to the catalog of spectroscopic redshifts is done to the coordinates of the counterpart in the reference (optical) band, it is possible to choose a more reliable 075 search radius for this catalog.

Before the cross-matching procedure is carried out between a pair of images, these are re-aligned locally within a 4' × 4' square region using the positions of several sources (typically more than 20) as reference. The mean rms between the central coordinates of matched sources in optical/NIR images is typically ⩽01 and ⩽02 between the IRAC and the ground-based images. This procedure allows to overcome small misalignment problems between the frames and assures a reliable identification of counterparts and an accurate positioning of the photometric aperture in all bands. It also allows to obtain reliable photometry (in the appropriate aperture) even if the source is very faint and/or undetected in an individual image.

The combination of the remarkable depth and the ∼2'' FHWM of the IRAC observations inevitably leads to issues of source confusion, specially around crowded environments. However, based on the (ground-based) reference image, it is possible to deblend IRAC sources which have not been separated by SExtractor in the original IRAC images and lie at least 1'' away (half the FWHM, chosen as our resolution criterion). HST images reveal that the multiplicity is larger, but the deconvolution of sources separated by less than 1''  is very uncertain. When multiple counterparts are found in the optical/NIR images during the cross-matching, the IRAC photometry is recomputed following a deconvolution method similar to that used in Grazian et al. (2006), Wuyts et al. (2008), Williams et al. (2009), or Wang et al. (2010).

In our case, first, the coordinates of the photometric aperture are re-positioned to that of the optical/NIR counterparts. Then, the PSF of the higher resolution image is convolved to the IRAC PSF, and the flux of each source is scaled to match that of the real IRAC sources measured in 09 apertures (after re-centering to the positions of the optical counterparts). Finally, total magnitudes are computed applying an aperture correction of [1.30 ± 0.07,1.02 ± 0.08,1.2 ± 0.10,1.44 ± 0.14] mag in the [3.6,4.5,5.8,8.0] μm bands, respectively. Figure 4 illustrates the deconvolution procedure using an HST/ACS image as reference. The red and green apertures depict the standard 2'' aperture (for isolated sources) and the 09 aperture, respectively. Pixel-by-pixel variations in the residual from subtracting the model PSF do not exceed a 5% within the 09 and 2'' apertures. We also checked that the average rms (∼3%) is well within the photometric error of the sources. The analysis of an average PSF, derived from observed sources across the image, indicates that for the typical separation between blended sources, ∼22 (>18 for 75% of them), the flux contamination from the nearby neighbor does not exceed a 10% for sources with a flux ratio around 1:2–3. Approximately 75% of the blended sources present flux ratios lower than 1:3.5.

Zoom In Zoom Out Reset image size

After applying the deblending method, our IRAC-selected catalog contains 76,936 (113,023) sources to [3.6] < 23.75 (24.75). This means that we were able to deblend 8% of the sources in the original IRAC catalog built with SExtractor (presented in Section 3.1), and 16% of the final catalog of 76,936 sources were deblended (typically, each blended sources was a combination of two sources). We find no significant difference in the brightness distribution of the blended sources compared to the resolved sources.

In the following sections, and in Paper II, we will analyze the SEDs and physical properties of the IRAC sources, concentrating in the sample with [3.6] < 23.75 mag, which count with more accurate IRAC photometry (S/N ≳ 8). Therefore, this will be the working sample for the rest of the paper unless explicitly stated otherwise. Nonetheless, all the procedures discussed in the following are also applied to the sources down to [3.6] < 24.75 mag. Although these objects are not included in the accompanying catalog (presented in Section 6, restricted to the most reliable detections), it is possible to retrieve their data through our online database (see Section 6.3).

The photometry is carried out in the same (Kron) elliptical aperture in all bands. The parameters of that aperture are obtained from a reference image (the same as for the cross-matching procedure) whose resolution is representative of the entire data set. Normally, this reference image is a ground-based optical/NIR frame with a PSF of approximately 1''–15, which is easy to translate to other ground-based images avoiding aperture issues.

The bands are sorted according to depth to facilitate the cross-match to the optical bands. The typical aperture band for the majority of the sources (83% of the sample) is the Subaru R band, followed by the CFHTLS i' band (4%), MMT −i,−z (3% each) and MOIRC-K (2%). For the remaining 5% of the sources, the aperture is computed from other bands (CFHTLS-r, i₈₁₄, V₅₀₅) that account for less than 1% of the total. The Subaru imaging was preferred to the CFHTLS as primary aperture band due to the uniform coverage of the whole IRAC mosaic (the CFHTLS frames cover only ∼60% of the IRAC survey). In order to ensure that the aperture radius is always large enough to enclose the full PSF profile in all ground-based images, we established a minimum aperture radius equal to the worst value of the seeing (15). Once the best photometric aperture is defined, if no counterpart is found in one band but data exist at that position, the flux is measured within the same aperture. Note that the local WCS re-centering of the images allows an accurate positioning of that aperture. In this way, we recover fluxes for very faint sources whose detection was missed by SExtractor. When the counter-part source is too faint to be detected, the sky-rms value is stored to be used as an upper limit in the SED analysis.

Some bands are deliberately excluded from the merged photometric measurement described above due to a significant difference of the resolution compared with the reference band or because images are not available. For all these bands, the photometry is obtained differently and later incorporated in the merged catalog during the cross-correlation procedure. For the four IRAC channels, the optical/NIR Kron apertures are typically not large enough to enclose the entire PSF. Therefore, we keep the values measured in Section 3.1 with aperture photometry. Given the comparatively large PSFs of the MIPS (24 μm, 70 μm) and GALEX (FUV, NUV) images, the flux measurement in these bands was carried with IRAF-DAOPHOT and SExtractor, respectively (see PG05; PG08). Figure 5 illustrates the different resolutions and aperture sizes involved in the photometric measurement (see also the captions in Figures 9–12).

Zoom In Zoom Out Reset image size

The HST -ACS bands were included in the general photometric procedure. Although the much higher resolution may lead to multiple cross identifications even when compared with the optical ground based images, the high spatial resolution (after the local re-alignment described above) allows a reliable cross-matching within 015 (rms of the local WCS solution for HST images). As shown in Figure 5, apertures were placed correctly even when multiple counterparts are identified, and the photometry includes the fluxes from all of them. We have conducted an additional test to check the accuracy of the ACS photometry measured with this method. We compared the observed photometry to synthetic magnitudes derived from SED templates fitted to the multi-band photometry (except ACS) of spectroscopically confirmed galaxies. We find a very small offset (⩽0.02 mag) and a scatter consistent with the typical photometric errors in the bands (see Section 3.1.3 of Paper II for more details).

The NIR source catalog of the Palomar/WIRC survey (Bundy et al. 2006) is also included in the merged photometric catalog, although no images are available to match apertures. In this case, we use the SExtractor MAG_AUTO value of the closest neighbor in the J and K bands.

The uncertainties in the photometry are derived simultaneously to the process of flux measurement in each individual band. As mentioned in Section 3.1, the photometric errors obtained using SExtractor often underestimate the true background noise due to signal correlation on adjacent pixels (Labbé et al. 2003; Gawiser et al. 2006). In order to properly account for this effect, we estimated the flux uncertainties in three different ways, as described in PG08 (Appendix A). First, we measured the background noise in a circular corona of 5'' width around each source. This procedure is similar to that used by SExtractor and provides uncertainties σ_AP ∝ N^1/2_pixels. In addition, we measured the average sky signal on non-connected artificial apertures built with random pixels around each source, with the same size as the photometric aperture, and containing only pure sky pixels (rejecting pixels >5σ the rms obtained with the first method). Finally, we also estimated the background noise following the method by Labbé et al. (2003). The flux measured on several photometric apertures around the source, identical to the one employed for the photometry, is fitted to a Gaussian function to yield the rms background fluctuation. The sky background is set to the resulting average value of the three methods, and the final photometric uncertainty is set to the largest estimate, typically, the one measured with the second method.

Table 3 shows the fraction of the entire IRAC catalog detected in the different bands compiled for this work. We give specific values for different sub-samples divided as a function of brightness in the selection band, detection level (S/N), and location in the IRAC mosaic. The majority of sources are detected in the four IRAC channels. However, while most sources (>90%) have reliable detections at 3.6 and 4.5 μm, only half of the sample is detected with S/N > 5 at the two longer wavelengths. Note that the photometry at 5.8 and 8.0 μm was not performed independently (see Section 3.1).

The fraction of IRAC sources detected in the optical bands is also elevated (∼95%) for the majority of the bands. However, cutting the measurements at S/N > 5, the efficiency decreases to ∼85% for the deepest CFHTLS, MMT, Subaru and ACS bands, and to 50%–60% for the u and z bands (in both CFHTLS and MMT data) and the shallower CFHT BRI bands. For the GALEX data, the detection is typically lower than 10% and 25% in the FUV and NUV bands, respectively. Around 12% and 6% of the sources are matched to high-quality spectroscopic redshift estimates in the main and flanking regions, respectively. The lower efficiency in the flanking regions is caused by the inhomogeneous DEEP2 coverage of the IRAC mosaic.

The NIR coverage of the strip surveyed by IRAC is also discontinuous. The WIRC catalog provides the most uniform coverage, including 40% and ∼100% of the area in the main region in the J and K bands, respectively. The fraction of IRAC sources detected in each band is very similar, ∼50%–60% (∼20% with S/N > 5). The deepest NIR observations are those taken in CAHA-J and Subaru-MOIRCS-K_s bands. The latter covers ∼1/4 of the main region up to 1 mag deeper than the WIRC-K data, presenting a much higher detection fraction, ∼90% (70% with S/N > 5). The CAHA-J data cover 40% of the main region recovering ∼30% (S/N > 5) of the IRAC detections, a slightly larger fraction than WIRC-J.

We note that for most bands, the fraction of sources detected with S/N < 5 can be significantly higher than the overall value. In the shallowest bands, this is caused by the higher photometric uncertainties (e.g., the WIRC J and K bands). However, for most bands this is the result of the enhanced detection achieved with the forced photometric measurement. For each IRAC source undetected in a given image, but having a counterpart in any other band, we still measure the flux in the same aperture at the position of the counterpart. With this method, we increase the detection efficiency at fainter magnitudes recovering sources that would be missed otherwise. However, given the extreme faintness of these sources, some of the measurements produce low significance detections (S/N < 2). As a precaution, to preserve the overall quality of the SED, we do not include these values (photometric uncertainties >0.4 mag) in the SED fitting procedure, nor in the estimate of the physical parameters in Paper II. Nevertheless, we keep these values as they can be useful as upper limits. Another estimate of the limiting magnitude in each band is obtained from the value of the sky rms in failed forced measurements (i.e., those for which the measured flux in the aperture is negative).

Figure 6 illustrates the characteristics of the different flux measurements in four bands probing representative wavelengths in the UV-to-NIR range. The central panel of each plot shows a color–magnitude diagram in [3.6] versus u*Ri' and K, respectively. The gray-scale density map depicts the distribution of standard (non-forced) photometric measurements, typically detected with S/N > 5 (magenta line). Forced detections for which we obtain a valid flux or a sky-rms value are shown as green and blue dots, respectively. The cyan markers are sources detected in IRAC only (for which we assign arbitrary magnitudes in each band). For the deep R and i' bands, which present the highest detection efficiency, forced measurements account for less than 5% of the flux measurements and present a median S/N ≲ 2 (black line). In the u* band, the source density for a similar limiting magnitude is much lower, and consequently the fraction of forced detections is higher, around 20%.

Zoom In Zoom Out Reset image size

The upper panels in each quadrant of Figure 6 show the fraction of undetected sources in each band. In the R and i' bands, these sources make up for <4% of the total sample. Most of them are IRAC-only sources (plotted in cyan), whereas the rest are at least marginally detected in one other (typically red) band, but the forced measurement fails for the particular band shown in the plot (blue). The fraction of undetected sources in the u* band (and other shallower optical bands, e.g., CHFT12K BRI) increases to 10%, being in this case dominated by failed forced measurements at the positions of R or i' detections. These kinds of measurement are also predominant for undetected sources in the K band.

Note that some of the IRAC-only sources are relatively bright, 21 < [3.6] < 22. Most of them are detected in deep K-band observations. However, there are a few galaxies detected only in the IRAC bands (typically the faintest ones). The nature of these interesting sources, potential candidates to massive high-redshift galaxies, will be explored in a future paper.

Figure 6 also shows the color distribution of IRAC sources in bins of [3.6] mag (right panels in each quadrant) and spectroscopically confirmed galaxies in several redshift bins (red ellipses and dots in the central panels). The general trend with color in the R and i' bands is that faint IRAC sources are on average bluer than brighter ones. The median colors for the faintest [3.6] bin, 22.50 < [3.6] < 23.75, are R − [3.6] = 1.6 and i' − [3.6] = 1.4. These colors are similar to those of galaxies at 1 < z < 1.5. This is consistent with the fact that the median redshift for the IRAC 3.6+4.5 μm magnitude limited sample ([3.6] < 23.75) is z ∼ 1 (see Paper II and PG08). Note also that most of the forced detections in the R band would qualify as IRAC extremely red objects (by the criteria of Yan et al. 2004, R − [3.6] > 4.0) that target dusty starbursts or passively evolving galaxies at z>1.5.

In general, all galaxies tend to become redder in optical−IRAC colors with increasing redshifts. Indeed, for a typical galaxy SED, the observed optical bands shift into the (fainter) UV whereas the [3.6] mag becomes brighter as it approaches the stellar bump at 1.6 μm rest frame (for z ≲ 1.2). Therefore, it is not surprising that the u* − [3.6] median color is redder (u* − [3.6] = 2.1, typical of a z ∼ 1 galaxy) than the median color involving the Ri'K bands. Note also that the observed UV progressively shifts into the Lyman break, producing the large fraction of z>2.5 (red dots in the top left quadrant of Figure 6) galaxies that are u-dropouts. Finally, we find that the K − [3.6] color presents an almost constant value as a function of [3.6] mag for a given redshift range, becoming progressively redder as we move to higher redshifts: K − [3.6] < 0 at z < 0.5 and K −  [3.6]>0 at z>1. Again, this is consistent with the fact that both K and [3.6] transit through the peak of the stellar bump for z < 1, changing their relative positions (see, e.g., Huang et al. 2004). An interesting consequence of the red K − [3.6] colors of high-z galaxies is that IRAC 3.6 μm observations are equivalent, in terms of source densities, to a K-selected sample down to slightly deeper limiting magnitudes.

Figure 7 shows the density map (central panel) and histograms (right panel) of the number of photometric bands with measured fluxes, N(band), in the main (left) and flanking regions (right). Typically, the SED of a galaxy in the main region has a median of 19 photometric data points. The average spectral coverage is ∼8 bands larger than in the flanking regions, mainly due to the lack of data from the CFHTLS and NIR surveys. Galaxies with an available spectroscopic redshift (R ≲ 24 mag; green dots) typically present the highest band coverage (∼22 and ∼14 bands in the main and flanking regions, respectively), since they are relatively bright in the optical. Interestingly, the faintest IRAC sources, [3.6] > 23.75 mag (red histogram), present a relatively high band coverage, N(band) = 17 in the main region. In most of the cases, these are low significance detections in the deepest optical/NIR bands where the forced photometric measurement (see Section 4.2) is able to recover a flux (see, e.g., Figure 9). On the opposite side, there is a non-negligible number of IRAC-faint but optically bright sources with more than 19 photometric data points. These constitute a population of blue dwarf galaxies at intermediate redshift easily detected in the optical but with faint IRAC counterparts (i.e., not very massive).

Zoom In Zoom Out Reset image size

The upper panel of Figure 7 shows the [3.6] magnitude distribution for the full sample (black) and for sources with poor spectral coverage (N(band) < 5; blue histogram). The latter defines a clearly isolated group in the density contours (central panel) and in the histograms (right panel) of Figure 7. These sources represent less than 3%(4%) of the sample up to [3.6] < 23.75 (24.75). Most of them are clear IRAC-only detections (as the ones discussed in Figure 6) relatively bright in [3.6,4.5]. However, some of the faintest sources can be affected by contamination of spurious sources (we give more details on these sources in Section 5.3).

Table 3 shows the fraction of IRAC sources in our sample ([3.6] < 23.75) detected in the X-ray, FIR, and radio surveys within the overlapping area. About 30% (20% with S/N > 5 detections) of the IRAC sources are detected at 24 μm (S/N = 5 reached at ∼60 μJy), 10% (2% with S/N > 5) at 70 μm (S/N = 5 reached at ∼3.5 mJy), 1% in the X-rays, and ⩽1% at 20 cm. For the radio and X-ray surveys, we cross-matched our sample to the catalogs of Ivison et al. (2007) and Laird et al. (2009), respectively. As explained in Section 4, the cross-match to the radio catalog was performed using a 3'' radius, whereas for the X-ray catalog we followed a two-step identification. First, we divided the X-ray sample attending to the availability of pre-identified IRAC counterparts (Laird et al. used the IRAC catalog of BAR08 as reference). For the galaxies identified in IRAC, we used a 1'' cross-matching radius, whereas for the other we used a more conservative 2'' radius. With this procedure, we found 848 out of the 882 galaxies with IRAC counterparts in their catalog (815 among the 830 with high reliability flag; see Laird et al. 2009 for more details). The remaining galaxies were either outside of the N(frame) > 20 region or too close to a bright star. In addition, we were able to identify 175 additional X-ray sources in the area of our mosaic not covered by the data of BAR08, i.e., we recover a total of 1023 out of 1325 X-ray sources from the catalog of Laird et al. For the MIPS 24 μm and 70 μm and radio surveys, we identify 10758, 868, and 590 (out of 1122) sources, respectively. Noticeably, about 20% and 54% of the sources detected in MIPS 24 μm and 70 μm present a multiple IRAC counterparts (typically 2–3) and ∼6% in X-ray and Radio.

All of these surveys cover an area larger than the IRAC observations, but none of them fully cover the IRAC strip (although the X-ray and MIPS 24 μm data cover >90% of the area). Table 4 shows the total area covered by the X-ray, MIPS, and radio surveys, the fraction that area in common with the IRAC mosaic, and the number of sources with an IRAC counterpart in our catalog. All the sources in the different surveys located within the area covered by IRAC are detected in our catalog (with the exception of a few objects too close to bright stars).

Note that despite the larger area of the VLA observations, this survey covers only ∼40% of the region surveyed by IRAC. In fact the data are limited only to the upper region of the IRAC mosaic (δ>5310; see Figure 1).

In this section, we analyze in more detail the sources with a poor spectral coverage to asses the reliability of the IRAC catalog as a function of the magnitude in the [3.6] band.

First, we test the reliability with the widely used method of comparing the number of detections at faint levels in the original and a negative image. This test reveals that only ∼1% of the sources up to [3.6] < 23.75 mag are spurious.

However, this procedure is conceived to detect faint spurious sources based on the assumption of a symmetric noise, whereas most of the spurious detections are concentrated around bright regions as a consequence of PSF and saturation artifacts and an excessive deblending. Therefore, we chose to follow a different approach to analyze the reliability of the catalog based on the fact that faint [3.6, 4.5] detections lacking a measurable counterpart in other bands are potential candidates for spurious detections.

As discussed in Section 5, we find a nearly isolated group of sources detected in N(band) ⩽ 4. Most of these sources are IRAC-only detections (cyan histogram of Figure 6) that, in the absence of low significance optical detections to validate them, could be artifacts of the source extraction procedure.

The visual inspection of galaxies with N(band) ⩽ 4 reveals a clear dichotomy. We find that these sources are either strongly clustered around bright stars and extended sources (typically low-z galaxies), or uniformly distributed across the image. Most of the sources in the first group are easily identified as spurious, whereas for the isolated sources, their high S/N (∼5–10) and the visual inspection seem to favor that they are real. We dealt with the reliability of sources with N(band) < 4 by considering spurious all detections in a 20''–30'' radius region around the brightest objects ([3.6] = 16–15 mag). Up to [3.6] ⩽ 23.75, approximately half of the N(band) ⩽ 4 sources are spurious by this criteria, accounting for less than 2% of the total sample. Nevertheless, even for the more conservative scenario, the sum of all N(band) ⩽ 4 detections constitutes less than ∼4% of the sample in the main region. Finally, we also check the reliability of these sources by studying how many of them are simultaneously detected at both 3.6 and 4.5 μm. Unfortunately, we find that some real detections are missed at 4.5 μm due to the slightly worse image quality, while some of the spurious sources are detected in both images in regions where the density of artifacts is larger (i.e., around bright stars).

Summarizing, we conclude that the overall reliability of the catalog is very high (>97%) and that the contamination by spurious sources is strictly restricted to the surrounding areas of very bright sources. For the sake of completeness, we do not remove the sources with N(band) ⩽ 4 from the sample. Instead, we include in the data catalog (Table 6) both the number of bands in which the source was detected (N(band)) and a flag indicating if the source is located in the vicinity of a bright object (see Section 6.2).

Following PG08, we used eight different photometric and morphological criteria to identify stars in the merged photometric catalog: (1) the average of SExtractor star/galaxy separation parameter (stellarity) for all bands where the source is detected; (2) when available, the stellarity parameter and FWHM of the source in the HST images are also considered as an independent, more reliable, criterion; (3, 4) the IRAC-based color–magnitude criteria of Eisenhardt et al. (2004) and Barmby et al. (2004); (5) the concentration parameter (i.e., the difference of the [3.6] mag measured in a 3'' radius aperture and the SExtractor MagAuto); (6) the color–color and (7) color–magnitude criteria of Eisenhardt et al. (2004; this time based on optical and NIR bands); and (8) the BzK criteria of Daddi et al. (2004). In spite of using multiple stellarity criteria, the heterogeneous band coverage makes difficult the simultaneous application of the eight criteria (e.g., the HST data cover only ∼40% of the full area). For this reason, we chose the bulk of the criteria to be based on IRAC or optical colors, which are available for the majority of the sources independently of the region in which they are located. We verify that at least five criteria can be estimated for 86% of the sources with [3.6] < 23.75.

A source was identified as a star if three or more of the stellarity criteria are satisfied. Based on this method, we found 2913 stars. Among them, we are able to identify 69% of the spectroscopically confirmed stars. The rest of them satisfy at least one or two of the stellarity criteria. Figure 8 illustrates the efficiency of the IRAC morphological criteria by Eisenhardt et al. (2004) and the BzK criteria of Daddi et al. (2004). Red dots indicate sources with N(criteria) ⩾ 3. Clearly, the BzK criterion is more efficient identifying stars over a wider magnitude range. Note that, as discussed in PG08, an IRAC-selected catalog down to [3.6] < 23.75 only includes a minor fraction of stars (less than 5% at all magnitudes), the majority of them at bright magnitudes (∼40% at [3.6] < 18). At the faintest magnitudes, the overall dimming of the objects could lead us to identify some stars as galaxies due to the lack of applicable criteria. However, we find that both the galactic and stellar number counts in the IRAC bands are in good agreement with those presented in Fazio et al. (2004) down to our limiting magnitude. From the relative distribution of galaxies and stars, we also find that the fraction of stars makes up for only 3% of all sources at magnitudes fainter than the median of the sample ([3.6] = 22.4).

Zoom In Zoom Out Reset image size

These are the fields included in Table 5.

• 1.  
  Object. Unique object identifier starting with irac000001. Objects labeled with an underscore plus a number (e.g., irac000356_1) are those identified as a single source in the IRAC catalog built with SExtractor, but deblended during the photometric measurement carried out with the Rainbow software (see Section 4.1). Note that, although the catalog contains 76,936 elements, the identifiers do not follow the sequence irac000001 to irac076185. This is because the catalog is extracted from a larger reference set by imposing coordinate and magnitude constraints. The table is sorted according to this unique identifier.
• 2.  
  α, δ. J2000.0 right ascension and declination in degrees.
• 3.  
  zspec. Spectroscopic redshift determination drawn from the DEEP2 spectroscopic survey or the catalog of LBGs of Steidel et al. (2003).
• 4.  
  qflag. Spectroscopic redshift quality flag from 1 to 4. Sources with qflag > 3 have a redshift reliability larger than 80%.
• 5.  
  FUV, NUV,u',.... Effective wavelengths (in nanometers), magnitudes, and uncertainties (in the AB system) for each of the 30 photometric bands compiled for this paper. The bandpasses are GALEX FUV and NUV, CFHTLS u*g'r'i'z', MMT u'giz, CFHT12k BRI, ACS V₆₀₆ and i₈₁₄, Subaru R, NICMOS J₁₁₀, H₁₆₀, MOIRCS K_s, WIRC JK, CAHA-JK_s IRAC 3.6-8.0, and MIPS 24 μm and 70 μm. The bands are sorted according to the effective wavelength of the filter. We refer to Section 3 for details on the photometric measurement and error calculation. The magnitudes do not include zero-point corrections (see Paper II, Section 3.1.3). A value of the magnitude equal to −99.0 with error equal to 0.0 indicates a not valid photometric measurement. A negative error indicates that the source was undetected by SExtractor but a positive flux was obtained when forcing the measurement at that position using the appropriate aperture (see Section 4.2). An error equal to 0.0 indicates that the source was undetected and the forced measurement returned a negative flux. In this case, the value of the magnitude is an upper limit equal to the sky rms (1σ) in the photometric aperture (see Section 4.2).

These are the fields included in Table 6.

• 1.  
  Object. Unique object identifier (the same as in the photometric catalog in Table 5).
• 2.  
  α, δ. J2000.0 right ascension and declination in degrees.
• 3.  
  N(bands,detect). Number of UV-to-NIR in which the source is detected.
• 4.  
  N(bands,forced). Number of UV-to-NIR in which the source is a priori undetected, but the forced photometry is able to recover a valid flux.
• 5.  
  Flag. Quality flag indicating that the source is located in the vicinity of a bright object. Sources detected only in the IRAC bands (N(band) < 5) and close to a bright ([3.6] > 16) source are likely to be spurious (see Section 5.3). The values of the flag indicate: (5) source within 70''–100''of the brightness saturated stars in the field, (4) source within 30'' of a [3.6] < 15 source, (3) source within 20'' of a 15 < [3.6] < 16 source, (2) source within 15'' of a 16 < [3.6] < 17 source, (1) source within 10'' of a 17 < [3.6] < 18 source, and (0) source unflagged.
• 6.  
  Stellarity. Total number of stellarity criteria satisfied (see Section 5.4). A source is classified as a star if it satisfies three or more criteria. (Section 5.4).
• 7.  
  Region. Region of the field in which the source is located (Section 5): A value of 1 or 0 indicates that the source is in the main or flanking region, respectively. The main region is defined as the area of the IRAC mosaic within 52.16 < δ < 53.20 and α>214.04, the flanking regions are those containing the remaining area.

The photometric catalog presented here and the inferred stellar parameters discussed in Paper II are obtained using the Rainbow software (see PG05; PG08). The program includes different sub-routines for each task, and the output of each step serves as the input for the next. After the data processing, both the input (images, spectra, templates) and the resulting catalogs are stored in a database with individual sources as building blocks. On doing so, we achieve several goals: (1) each object is fully characterized with all the available information; (2) the data can be easily sorted and retrieved according to several criteria; (3) the data are homogeneously stored, which allows a straightforward combination with data from other Rainbow fields and projects (such as those in PG05 or PG08).

In order to provide worldwide access to the data stored in our database, we have developed a publicly available Web interface, dubbed Rainbow Navigator.¹² Here we briefly describe the main features of the utility. A detailed description of its capabilities can be found at the Web site.

Rainbow Navigator is essentially a user-friendly Web interface to a database containing all the data products resulting from the process of creating and analyzing the IRAC-selected catalog presented in these two papers, from the initial source detection to the estimate of the stellar parameters.

As many other astronomical query interfaces (e.g., NED, SIMBAD), Rainbow Navigator allows to retrieve the information for single sources searching for the source name or by coordinates. It also allows to create subsets of the complete catalog based on multiple constraints over the multi-band photometry, the redshifts or the stellar parameters. In addition, we have incorporated a cross-matching tool that allows to compare catalogs uploaded by the user to the IRAC-selected sample stored in the database, returning the sources in common. Rainbow Navigator also has an on-the-fly utility to create sky maps of a selected area, including point-and-click access to the individual sources.

Furthermore, an interesting feature that we do not include in the public catalog for the sake of simplicity is the possibility of retrieving observed and rest-frame synthetic magnitudes over a predefined grid of 52 different filters covering the whole spectral range from the UV to the radio wavelengths. These values are computed by convolving the best-fitting template (see Paper II) for each source with the appropriate filter transmission curve.

Each source of the catalog has its own data sheets that provide all the available information including not only the photometry, the redshifts, or the stellar parameters but also detailed information of the stellarity, synthetic magnitudes, and the multiplicity in other bands, jointly with the unique identification and coordinates of the counterparts in each given band. The tool also provides postage stamps of the source in each of the available bands that can be modified on-the-fly or combined to create simple false color images. Furthermore, each page includes a figure depicting the full SED showing the fit of the optical and IR data jointly with the best-fitting templates.

Figures 9–12 show examples of the multi-band postage stamps, the UV-to-FIR SED and fitting templates, and the clickable map utility for a few galaxies at different redshifts. The postage stamps also illustrate the different photometric measurements in each band (aperture matched and circular apertures) and the forced detections for very faint sources (see, e.g., the source in Figure 11).

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

Rainbow Navigator has been conceived to serve as a permanent repository for future versions of the catalogs containing improvements over the previous results (the present version is data release 1), and also to similar data products in other cosmological fields (such as GOODS-N and GOODS-S, presented in PG08). Currently, it provides public access to the IRAC-selected catalog in the EGS presented in this paper, and also a similar release of the data described in PG08 for a small piece of the central region of the GOODS-S region.