The scientific goals of the DESpec spectroscopic survey lead to a survey definition that acquires at least 1000 galaxy redshifts per square degree up to z~1.5 over at least the DES footprint of 5000 square degrees. The area will be extended up to 15,000 square degrees by later using LSST photometry for spectroscopic target selection. The DES and LSST photometry (together with VISTA JHK photometry) yields not only fluxes, colors, and photometric redshifts but also galaxy image shapes and surface brightnesses. All of this information can be exploited to select a sample of galaxies that satisfy the joint requirements of large redshift range, adequate volume sampling, and control over any bias introduced due to sample selection or redshift failures. In practice, we expect to use galaxy flux, color (and photo-z), and surface-brightness to optimize the redshift distribution and galaxy types of the survey. We plan to target a mix of emission-line galaxies (ELGs) – which predominate at high redshift and which yield efficient redshift estimates to z~1.7 based upon their prominent emission lines – and luminous red galaxies (LRGs), which have brighter continuum spectra and higher clustering amplitude and offer good redshift success rates up to z~1. The target selection will ultimately be chosen to optimize the science yield; at this stage, we present some examples of survey selection to demonstrate the feasibility of the survey, to set the scale, and to derive constraints upon the instrument design. A key question is whether we need to maximize high-redshift galaxies (z>1) in contrast to a better sampling of the whole galaxy population at intermediate redshifts (z~0.5-1.0); the latter covers less volume but optimally complements weak lensing science from DES (see Fig.2.2). The combination of DES and DESpec can be used to measure galaxy biasing and recover the full modes of the 3d matter distribution, as opposed to just BAO or RSD information. The DESpec sample optimization is being worked out in detail, but we next show some examples that illustrate the potential of combining DES with DESpec.
3.A Simulated Galaxy Catalogs
We have created simulated catalogs of galaxies to quantify the yield of particular spectroscopic selection schemes, to predict the number of potential targets, and to compute the sensitivities and thereby the rate of progress of the survey. The simulations are built directly from the observed COSMOS catalog of Capak et al. (2009) and Ilbert et al. (2009). We refer to this simulation as the COSMOS Mock Catalog (CMC) (Jouvel et al. 2009).
The COSMOS photometric-redshift catalog (Ilbert et al. 2009) was computed with 30 bands over 2 deg2 (GALEX for the UV bands, Subaru for the optical (U to z), and CFHT, UKIRT, and Spitzer for the NIR bands). It achieves very good photo-z accuracy and low catastrophic redshift rates due to careful calibration with the spectroscopic samples zCOSMOS and MIPS. The CMC is restricted to the area fully covered by HST/ACS imaging, 1.24 deg2 after removal of masked areas. There are a total of 538,000 simulated galaxies for i < 26.5 in the mock catalog, leading to a density of roughly 120 gal/arcmin2. AGN, stars, and X-ray sources were removed from the input COSMOS catalog.
A photo-z and a best-fit template spectrum (including possible additional extinction) are assigned to each galaxy of the COSMOS catalog. We first integrate the best-fit template through the instrument filter transmission curves to produce simulated magnitudes in the instrument filter set. We then apply random errors to the simulated magnitudes based upon a simple magnitude-error relation in each filter. The simulated mix of galaxy populations is then, by construction, representative of a real galaxy survey, and additional quantities measured in COSMOS (such as galaxy size, UV luminosity, morphology, stellar masses, correlation in position) can be easily propagated to the simulated catalog. The COSMOS mock catalog is limited to the range of magnitude space where the COSMOS imaging is complete (iAB ~ 26.2 for a 5 sigma detection, see Capak et al. (2007) and Capak et al. (2009)).
We assign emission-line fluxes to each galaxy of the CMC by modeling the emission-line fluxes (Ly alpha, [OII], H beta, [OIII], and H alpha) using the Kennicutt (1998) calibration. We first estimate the star-formation rate (SFR) from the dust-corrected UV rest-frame luminosity already measured for each COSMOS galaxy. The SFR can then be translated to an [OII] emission-line flux using another calibration from Kennicutt (1998). The relation found between the [OII] fluxes and the UV luminosity is in good agreement with the VVDS data and is valid for different galaxy populations. For the other emission lines, we adopt intrinsic, unextincted flux ratios of [OIII]/[OII] = 0.36; H beta/[OII] = 0.28; H alpha/[OII] = 1.77 and Ly alpha/[OII] = 2 (McCall et al. 1985; Moustakas and Kennicutt 2006; Mouhcine et al. 2005; Kennicutt 1998).
The CMC does an excellent job of reproducing the counts and color distributions of galaxies observed in bands from 0.4 to 2.3 microns, for example comparing to the GOODS (Giavalisco et al. 2004) and UDF (Coe et al. 2006) surveys. The CMC also provides an excellent match to the redshift-magnitude and redshift-color distributions for I<24 galaxies in the VVDS spectroscopic redshift survey (Le Fèvre et al. 2005). For more detail about the CMC and its validation, see Jouvel et al. (2009).
3.B Spectroscopic Sensitivity
To compute the sensitivities (exposure time required to obtain a certain S/N for a particular simulated galaxy spectrum, and redshift success rates), we start with input spectra from the DES galaxy catalog simulations (provided by R. Wechsler & M. Busha) or from the COSMOS Mock Catalog (CMC, Jouvel et al. 2009), both of which model the detailed properties of the galaxy population at magnitudes and redshifts appropriate to DESpec. The DES catalog simulations provide galaxy spectra based upon the template models generated by the Kcorrect package (Blanton & Roweis 2007), while as just described the CMC provides spectra based on the COSMOS data, providing in particular detailed distributions of line ﬂuxes for emission-line galaxies. We account for the transmission as a function of wavelength through the atmosphere.
For this initial study, we have assumed that the DESpec system has 50% of the net throughput of DECam, accounting for losses in the fibers, grating, and other optics. We have initiated a more realistic estimate of the throughput of DESpec, and in the R&D phase we will use the reference design to obtain a reliable throughput curve as a function of wavelength. We also account for the fraction of galaxy light entering a ﬁber, using either an analytic Gaussian model and the galaxy size distribution from the CMC or the detailed DES galaxy image simulations. The dominant source of noise for the DESpec spectra comes from the night sky, which we model using the high-S/N, high-resolution spectral atlas of Hanuschik (2003) plus the extensive spectroscopic archives from the SDSS and BOSS surveys. We take care to use data with resolution similar to or better than DESpec, in order to properly sample the much lower sky continuum levels in between the forest of atmospheric airglow emission lines in the red.
Figure 3.1: The signal-to-noise ratio in 2.5 Å bins as a function of wavelength for an input spectrum with mAB = 22 and an exposure of 30 minutes. The calculation assumes the DESpec throughput and sky noise as described in the text.
Figure 3.2: Same as Figure 3.1, except that the signal is an unresolved emission line with flux
The above components are implemented in a simulation package written in IDL, which generates ﬂux-calibrated, one-dimensional extracted spectra with the signal and noise properties and wavelength resolution appropriate to DESpec. The simulated DESpec spectra are then used to calculate S/N and redshift success rates as functions of different galaxy properties, in particular galaxy magnitude, redshift, and emission-line strength. For emission-line galaxies, we make measurements of the S/N of the emission-line ﬂux, especially of the [OII] 3727 line that will be the most useful for redshift determination at z>0.5. We calculate the spectroscopic redshift success rate following Jouvel et al. (2009), i.e., we require at least two lines to be detected at more than 5 sigma in the wavelength range of the spectrographs, or in the case of the OII line, we require that it is detected at greater than 5 sigma and that it is detected in its corresponding DES imaging band at 24th magnitude. In addition, we use the standard redshift measurement technique of cross-correlation with template spectra, as implemented in the IRAF external package rvsao (Kurtz & Mink 1998), to compare measured vs. true redshifts and to estimate the fraction of successful redshift measurements vs. galaxy properties. This will be useful for absorption-line galaxies, for which redshift measurement will be based upon spectral features spread over a range of wavelengths rather than on individual strong emission lines. We plot two examples of DESpec signal-to-noise calculations, one for a hypothetical object that has a continuum level with constant AB mag = 22 (Fig. 3.1) and one for an emission line with ﬂux = 1.0×10−17 erg/s/cm2 (Fig. 3.2). In both cases we assume a 2-arcsec-diameter ﬁber, 70% of the light entering the ﬁber, a 30-minute exposure time, airmass 1.3, and the sky from Hanuschik (2003) as appropriate for dark time. We plot the S/N in ﬁxed bins of 2.5 Å. In the following we adopt a flux of 3.0×10−17 erg/s/cm2 (three times larger than the flux illustrated in Fig. 3.2) and S/N = 5 for detection. For the emission line we assume the case of a narrow line for which the ﬂux is entirely contained within a single resolution element. As noted below in Section 4, the results are not very sensitive to the assumed fiber diameter.
3.C Luminous Red Galaxy Selection
For galaxy clustering measurements for baryon acoustic oscillations (BAO) and redshift-space distortions (RSD), LRGs are convenient because they occupy dense regions and are thus strongly biased relative to the dark matter, b~2, so the clustering amplitude is high. With our simulations we have estimated the completeness rate for objects in the CMC catalogue. We plot in Fig. 3.3 a preliminary example of the redshift completeness for LRGs, showing the magnitude, as a function of redshift, where the DESpec redshift success rate is expected to be 90%. (The 90% completeness magnitude gets fainter with increasing redshift, because strong absorption lines enter the DESpec spectral range with increasing z, which more than offsets the lower flux levels.) Also shown is the apparent magnitude vs. redshift relation for an LRG with luminosity of 3L* (Eisenstein et al. 2001), showing that the survey is expected to be about 90% efficient for bright (>3L*) LRG redshift measurement out to z = 1.3 in a 30 minute exposure, modulo target selection efficiency, fiber collisions, etc.
We have considered two different selection algorithms for LRG's. The first is based on optical and near-infrared data from the DES+VHS datasets. The second one is based on a full selection from DES bands combined with near-infrared bands from VHS as well as WISE data. We have made two LRG selections for the following reasons: LRG's at z~0.5 usually inhabit massive dark-matter halos and hence are tracers of the dark matter field that produces the lensing measured by the DES sample. We therefore can extract extra information via cross-correlation techniques described earlier. The two selections are a proof-of-concept showing that these galaxies can be selected in large numbers at the appropriate redshifts both with a simple color cut and also much more efficiently with more complicated techniques (which might yield better results in terms of finding larger cross-correlations but might be harder to understand in terms of selection effects). We describe the selections below including the number density of galaxies in each selection.
Using DES+VHS imaging, LRGs can be selected to yield a relatively flat redshift distribution over the range 0.5 < z < 1 by using selection cuts in r-z vs. z-H color space. We plot these color-color diagrams in Figure 3.4. These cuts supply more than the needed density of LRG targets, so the targets can be randomly sampled. The galaxies yielded by Cut I amount to 770 galaxies per square degree at magnitudes in the z band brighter than 21. For Cuts II and III, we have ~60 objects per square degree at magnitudes brighter than 22 in the z band, but they are at significantly higher redshifts0.5. Should give this number here in the text as well if that's the case. -->. These cuts will also select some non-LRG objects.
The second selection criterion is based on a neural network method for selecting galaxies. Here we assume that we may select LRG's from the full multi-wavelength dataset including DES, VHS and WISE. In this selection, we classify galaxies as LRG's or not based on the spectrum from which they were created in the mock catalogue. We use a subset of around 10,000 galaxies to train the neural network to select galaxies that have spectra similar to those of LRGs based solely on the information provided by the colors. This selection method yields 1241 LRG targets per square degree at i< 22 with a photometric redshift larger than 0.5. We can see from Fig. 3.5 the distribution of selected galaxies vs. redshift from the mock catalog for these two selection methods. Also included in Fig. 3.5 for comparison is the number of galaxies that could be selected via PTF and WISE photometry (cf. BigBOSS).
Figure 3.3: The red points and curve indicate the iAB magnitude vs. redshift where the redshift success rate for an LRG spectrum is 90%, based on our spectroscopic simulations. Black curve shows the magnitude vs. redshift relation for an LRG with L=3L* (Eisenstein et al. 2001), showing that DESpec should be 90% successful for bright LRG redshift measurements out to z = 1.3 in a 30 minute exposure.
Figure 3.4: Illustration of targeting efficiency of z-H vs. r-z color-color cuts I, II, and III for galaxies with z-mag <22: for redshift z<0.5 (blue), 0.51.1 (gold, red and purple).
Figure 3.5: Predicted redshift distribution for three methods of selecting luminous red galaxies based on the Cosmos Mock Catalog.