The DESI Survey Validation: Results from Visual Inspection of the Quasar Survey Spectra

TL;DR

<3-5 sentence high-level summary> DESI’s QSO visual inspection (VI) validates spectroscopic pipelines and QSO target-selection by constructing robust truth tables from deep-field SV spectra and testing afterburner-based identifications on sparse VI data. The study demonstrates that the standard Redrock pipeline reliably identifies most QSOs but misses a non-negligible fraction, which can be recovered by afterburners to yield a higher overall QSO fraction with preserved good-redshift purity. At the main DESI depth (~1000 s), both the standard and modified pipelines meet good-redshift purity, velocity-precision, and accuracy requirements, with the modified pipeline achieving substantially higher QSO recovery (~94% vs ~86%) and better redshift performance for challenging, missed QSOs. The VI analysis also reveals a diverse QSO population, including host-galaxy dominated, dust-reddened, BAL, and narrow-line QSOs, as well as significant non-QSO galaxy contamination and a handful of strong-lensing-like multiple-source spectra, underscoring the richness and scientific potential of the DESI QSO sample.>

Abstract

A key component of the Dark Energy Spectroscopic Instrument (DESI) survey validation (SV) is a detailed visual inspection (VI) of the optical spectroscopic data to quantify key survey metrics. In this paper we present results from VI of the quasar survey using deep coadded SV spectra. We show that the majority (~70%) of the main-survey targets are spectroscopically confirmed as quasars, with ~16% galaxies, ~6% stars, and ~8% low-quality spectra lacking reliable features. A non-negligible fraction of the quasars are misidentified by the standard spectroscopic pipeline but we show that the majority can be recovered using post-pipeline "afterburner" quasar-identification approaches. We combine these "afterburners" with our standard pipeline to create a modified pipeline to improve the overall quasar yield. At the depth of the main DESI survey both pipelines achieve a good-redshift purity (reliable redshifts measured within 3000 km/s) of ~99%; however, the modified pipeline recovers ~94% of the visually inspected quasars, as compared to ~86% from the standard pipeline. We demonstrate that both pipelines achieve an median redshift precision and accuracy of ~100 km/s and ~70 km/s, respectively. We constructed composite spectra to investigate why some quasars are missed by the standard spectroscopic pipeline and find that they are more host-galaxy dominated (i.e., distant analogs of "Seyfert galaxies") and/or dust reddened than the standard-pipeline quasars. We also show example spectra to demonstrate the overall diversity of the DESI quasar sample and provide strong-lensing candidates where two targets contribute to a single spectrum.

Figures (15)

• Figure 1: Example DESI spectra illustrating the different VI quality classes (left) and thumbnail images ($18^{\prime\prime}\times18^{\prime\prime}$) centered on each target (right). Salient information for each target is plotted at the top of each spectrum including the VI quality flag. For each target both an unsmoothed (grey) and smoothed (black) spectrum are plotted along with the associated error spectrum (orange) and the overlapping wavelength ranges for the three different spectral arms (pale pink shaded regions). The most prominent emission lines identified in the two high-quality spectra (with VI quality flags of 3 and 4) are highlighted using blue vertical dashed lines as well as the potential identification of a single emission line ([ o ii] at $z$ = 1.261) for the VI quality flag 2 target. We note that these visually inspected spectra were obtained with 10x longer exposures than those used in the main DESI survey.
• Figure 2: Venn diagrams showing (top) the overlap between the missed QSO sample, the Mg II absorption systems, and the targets with non-repeatable redshifts and (bottom) the overlap in the selection of missed QSOs between the three afterburner selections across the sparse VI fields.
• Figure 3: Distribution of (left) VI quality flags and (right) high-quality VI redshifts for the deep-field VI (top) and missed QSOs in the sparse VI (bottom). Targets meeting the SV1 and main selection are indicated in pale and dark grey, respectively. The dashed line indicates the threshold required for a high-quality redshift (VI $\ge2.5$).
• Figure 4: Pie chart showing the distribution of high-quality optical spectral classifications and low-quality spectra from the deep-field VI for targets meeting the SV1 (top) and main (bottom) target selections.
• Figure 5: $r$-band fiber magnitude versus redshift for (left) the deep-field VI (see section 2.3.1) and (right) the missed QSOs from the sparse VI (see section 2.3.2); the redshift is taken from the VI for the high-quality spectra and from Redrock for the low-quality spectra. The target classifications are based on the visually inspected optical spectral type for the high-quality spectra (VI $\ge$ 2.5): QSO (blue circle), galaxy (green square), star (orange star). The low-quality (VI $<2.5$) identifications are plotted as black triangles. The top panels show the redshift distributions for the high-quality QSOs (blue) and the galaxies (green).