Highly Efficient Selection of High-Redshift Emission-Line Galaxies for future DESI-like surveys with Deep Multi-band Imaging

Abstract

Emission-line galaxies (ELGs) are an important tracer of baryon acoustic oscillations (BAO) and large-scale structure (LSS) at $z > 1$. In this work, we investigate the feasibility of using deep wide-area multi-band imaging (e.g., from the Rubin Observatory) to efficiently select high redshift ELGs. Using Hyper Supreme-Cam $grizy$ photometry and COSMOS2020 many-band photometric redshifts, we designed simple color cuts guided by a probabilistic random forest classifier to select galaxies at $z = 1.1$--$1.6$. We then empirically tested and refined these color cuts using two samples of galaxies with deep spectroscopy and broad color coverage obtained with the Dark Energy Spectroscopic Instrument (DESI). Compared to DESI ELGs at $z = 1.1$--$1.6$, we achieve a higher redshift measurement success rate (89% versus 69%), a much higher correct redshift range success rate (84% versus 34%), and a far higher net surface density yield (1372 $\mathrm{deg^{-2}}$ versus 660 $\mathrm{deg^{-2}}$). Combining our sample with current DESI ELGs would increase the net ELG number density by a factor of $\sim3$, moving it out of the shot-noise limited regime and reducing the uncertainties on the BAO scale parameter at $z = 1.1$--$1.6$ by a factor of $\sim 2$.

Figures (14)

• Figure 1: $1\sigma$ uncertainty on the BAO scale parameter $\alpha$ as a function of the factor by which the ELG number density is increased compared to the expected final DESI sample (solid black), for measures of both the transverse scale ($\alpha_{\perp}$) and the line of sight scale ($\alpha_{\parallel}$). The biggest gains are found for the $1.3 < z < 1.5$ redshift bin (blue curve), where expanding the ELG sample at these redshifts by $\sim3\times$ (indicated by the dashed black line) would yield a factor of $\sim2\times$ improvement on $\alpha_{\perp}$ and $\alpha_{\parallel}$. Enlarging the sample by more than this factor would yield only marginal additional improvements.
• Figure 2: Color-color diagrams of the DESI spec-truth sample using $g-r/r-z$ plotted using imaging from either deep (left) or wide (right) depth HSC imaging, with galaxies color-coded according to their spec-z. This sample was obtained using HSC deep photometry. We show the color cuts for the standard DESI ELG sample (dashed black lines) and the spec-truth sample (solid black). Although the wide photometry is noisier, the selection limits are far enough from where ELGs reside in color space to ensure that the sample still retains all ELG-like objects down to the desired depths.
• Figure 3: [O II] flux S/N vs. the $\Delta\chi^2$ between the best-fit and second best-fit redshift solutions for the spec-truth sample (points color coded by spec-z). We consider objects above the solid black diagonal line (the DESI ELG reliable-redshift criteria; Raichoor_2023) or rightward of the dashed black line ($\Delta\chi^2$) to have reliable redshifts (i.e., not in the shaded region of this diagram).
• Figure 4: The effective exposure time distribution for the spec-truth sample. Only those objects whose exposure times were within the range bounded by the solid black vertical lines ( $700 < t < 1400$ seconds) were used for calculating redshift measurement success rates, whereas all objects with $t > 700$ seconds were included when calculating the fraction of targets in the desired redshift range ($1.1 < z < 1.6$). The metrics used are described in more detail in Section \ref{['sec:optimization']}.
• Figure 5: Target density ($\Sigma_\mathrm{target}$), redshift range success rate ($f_\mathrm{z\,= \, 1.1\, -\, 1.6}^{700 \,<\,t\, <\,1400}$), and the net surface density yield ($\Sigma_\mathrm{yield}^{z\,= \, 1.1\, -\, 1.6}$) for different $g$-fiber limiting magnitudes (black solid lines). The $g$-fiber limiting magnitude optimized to yield a net surface density of 1372 redshifts per square degree is indicated by the dashed purple lines in each panel. The DESI ELG sample is shown by the gray stars while the results of our sample are shown by the blue stars. Despite both samples having similar target densities and $g$-fiber limiting magnitudes, the HSC-based selection presented here is much more efficient at yielding redshifts in the desired range, as can be seen by the much higher values of redshift range success rate (middle panel) and net surface density yield (right panel) obtained for it.