Target Selection and Validation of DESI Luminous Red Galaxies

TL;DR

This work presents the DESI luminous red galaxy (LRG) target selection from the DESI Legacy Imaging Surveys, detailing region-specific cuts that yield a high-density, high-bias sample intended for precise BAO and RSD measurements. It demonstrates robustness to imaging systematics through linear-regression corrections and carefully designed veto masks, and it validates spectroscopic performance using Survey Validation and early Main Survey data, achieving a high redshift success with minimal stellar contamination. A simple, calibrated model links redshift failure probability to source brightness and spectroscopic depth, enabling robust corrections for clustering analyses. The paper also provides ancillary resources, including SV1/SV3 selections and stellar masses estimated via Random Forests, to support future DESI analyses and related surveys.

Abstract

The Dark Energy Spectroscopic Instrument (DESI) is carrying out a 5-year survey that aims to measure the redshifts of tens of millions of galaxies and quasars, including 8 million luminous red galaxies (LRGs) in the redshift range of $0.4<z<{\sim}\,1.0$. Here we present the selection of the DESI LRG sample and assess its spectroscopic performance using data from Survey Validation (SV) and the first 2 months of the Main Survey. The DESI LRG sample, selected using $g$, $r$, $z$, and $W1$ photometry from the DESI Legacy Imaging Surveys, is highly robust against imaging systematics. The sample has a target density of 605 deg$^{-2}$ and a comoving number density of $5\times10^{-4}\ h^3\mathrm{Mpc}^{-3}$ in $0.4<z<0.8$; this is a significantly higher density than previous LRG surveys (such as SDSS, BOSS and eBOSS) while also extending to $z \sim 1$. After applying a bright star veto mask developed for the sample, $98.9\%$ of the observed LRG targets yield confident redshifts (with a catastrophic failure rate of $0.2\%$ in the confident redshifts), and only $0.5\%$ of the LRG targets are stellar contamination. The LRG redshift efficiency varies with source brightness and effective exposure time, and we present a simple model that accurately characterizes this dependence. In the appendices, we describe the extended LRG samples observed during SV.

Figures (22)

• Figure 1: The redshift distribution of the DESI LRG sample and its comparison with LRG samples from earlier surveys. The y-axis is the number of objects in each redshift bin (of width $\Delta z=0.05$) per deg$^2$. The survey area and the total number of LRGs that have been or will be observed in each survey are listed in the legend. The dashed curve corresponds to the redshift distribution of a hypothetical sample with a constant comoving density of $5\times10^{-4}\ h^3\mathrm{Mpc}^{-3}$, which is approximately the DESI LRG target density in the redshift range of $0.4<z<0.8$; the area under the curve is proportional to the enclosed comoving volume. We describe how we obtain the redshifts for DESI LRGs in §\ref{['sec:spectro_data']}.
• Figure 2: The footprint of the DESI Legacy Imaging Surveys DR9, with the colors representing the surface density (in deg$^{-2}$) of the LRG targets (after applying the LRG veto masks). DESI will only observe regions above DEC$>$-20, and the DESI footprint also avoids regions close to the edge of the imaging footprint. See the actual DESI footprint in schlafly_survey_ops. This density map is computed with a HEALPix resolution of NSIDE=256, and we only plot pixels that are $>20\%$ occupied by the imaging survey footprint. The curve that separates the two regions is the Galactic plane.
• Figure 3: Selection cuts for the LRG targets in the South footprint. The points are color-coded by their redshifts measured by DESI. The upper left panel shows the stellar rejection cut, with gray points representing stars (which are plotted to show the stellar locus and are not LRG targets). The upper right panel shows the cut that removes lower-redshift and bluer galaxies. The lower left panel shows the sliding color-magnitude cut that serves as the luminosity cut and also shapes the redshift distribution; the "knee" at $W1={\sim}\,19$ introduces more galaxies at higher redshift. The lower right panel shows the magnitude limit in $z$-band fiber magnitude that ensures enough S/N for DESI observations.
• Figure 4: Stellar mass completeness of the DESI LRG sample as a function of stellar mass and photometric redshift. The dashed curve shows the fraction of galaxies above a given stellar mass that has been selected as a DESI LRG compared to a magnitude-limited sample. The blue histogram shows the distribution of stellar masses of a magnitude-limited sample of galaxies (having the same magnitude limit as the DESI LRG sample) whereas the black histogram denotes the subset of galaxies that have been selected as DESI LRGs. The stellar masses were obtained using a random forest-based algorithm (described in appendix \ref{['app:rf_masses']}) and the photometric redshifts are from zhou_clustering_2021. As spectroscopic redshifts are not available for the magnitude-limited sample, we use photometric redshifts for this demonstration. The figure uses objects from both the North and the South where valid photometry in $g$, $r$, $z$, $\mathrm{W}_{1}$ and $\mathrm{W}_{2}$ is available. The selected sample is highly complete for the most massive galaxies (i.e. $\log_{10}(\mathrm{M}_{\star}[\mathrm{M}_{\odot}])>11.5$) in the redshift range of 0.4 to 1.0. The completeness decreases significantly for redshifts lower than 0.4 but the decrease is less steep for redshifts above 1.0.
• Figure 5: Density of LRG targets in bins of imaging/foreground systematics values in the three imaging regions. (DECaLS and DES are both observed with DECam and have the same selection cuts and linear regression coefficients, but DES is significantly deeper and we plot it as a separate region to illustrate the difference.) The error bars represent "the error of the mean" assuming Gaussian distribution. The histograms show the distribution of each systematics property for each imaging region. "Galaxy depth" is based on 'the 'GALDEPTH" value in the LS DR9 catalog and it is the $5\sigma$ detection magnitude of an ELG-like galaxy and it assumes zero Galactic extinction; to account for Galactic extinction, we add an $E(B-V)$ term to obtain the imaging depth relevant for extragalactic sources. "PSF size" is the PSF FWHM and measures the seeing. The trends are computed using a HEALPix density map with NSIDE=512 by averaging over the pixels in bins of imaging/foreground properties.