Draft version April 5, 2024

Typeset using LATEX twocolumn style in AASTeX631

Identifying Quasars from the DESI Bright Galaxy Survey

S. Juneau ,1 R. Canning,2 D. M. Alexander ,3, 4 R. Pucha ,5, 6 V. A. Fawcett ,7 A. D. Myers,8 J. Moustakas ,9

O. Ruiz-Macias,4 S. Cole ,4 Z. Pan ,10 J. Aguilar,11 S. Ahlen ,12 S. Alam ,13 S. Bailey ,11 D. Brooks,14
E. Chaussidon ,11 C. Circosta,14 T. Claybaugh,11 K. Dawson,5 A. de la Macorra ,15 Arjun Dey ,1 P. Doel,14
K. Fanning ,16, 17 J. E. Forero-Romero ,18, 19 E. Gaztañaga,20, 2, 21 S. Gontcho A Gontcho ,11 G. Gutierrez,22 C. Hahn ,23
K. Honscheid,24, 25, 26 R. Kehoe,27 T. Kisner ,11 A. Kremin ,11 A. Lambert,11 M. Landriau ,11 L. Le Guillou ,28
arXiv:2404.03621v1 [astro-ph.GA] 4 Apr 2024

M. Manera ,29, 30 P. Martini ,24, 31, 26 A. Meisner ,1 R. Miquel,32, 30 A. Muñoz-Gutiérrez,15 J. Nie ,33
N. Palanque-Delabrouille ,34, 11 W. J. Percival ,35, 36, 37 C. Poppett,11, 38, 39 F. Prada ,40 C. Ravoux ,41, 34, 42 M. Rezaie ,43
G. Rossi,44 E. Sanchez ,45 E. F. Schlafly ,46 D. Schlegel,11 M. Schubnell,47, 48 H. Seo ,49 J. Silber ,11 M. Siudek ,21
D. Sprayberry,1 G. Tarlé ,48 Z. Zhou ,33 and H. Zou 33
1 NSF NOIRLab, 950 N. Cherry Ave., Tucson, AZ 85719, USA
2 Institute of Cosmology and Gravitation, University of Portsmouth, Dennis Sciama Building, Portsmouth, PO1 3FX, UK
3 Centre for Extragalactic Astronomy, Department of Physics, Durham University, South Road, Durham, DH1 3LE, UK
4 Institute for Computational Cosmology, Department of Physics, Durham University, South Road, Durham DH1 3LE, UK
5 Department of Physics and Astronomy, The University of Utah, 115 South 1400 East, Salt Lake City, UT 84112, USA
6 Steward Observatory, University of Arizona, 933 N, Cherry Ave, Tucson, AZ 85721, USA
7 School of Mathematics, Statistics and Physics, Newcastle University, Newcastle, UK
8 Department of Physics & Astronomy, University of Wyoming, 1000 E. University, Dept. 3905, Laramie, WY 82071, USA
9 Department of Physics and Astronomy, Siena College, 515 Loudon Road, Loudonville, NY 12211, USA
10 Kavli Institute for Astronomy and Astrophysics at Peking University, PKU, 5 Yiheyuan Road, Haidian District, Beijing 100871, P.R. China
11 Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720, USA
12 Physics Dept., Boston University, 590 Commonwealth Avenue, Boston, MA 02215, USA
13 Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai 400005, India
14 Department of Physics & Astronomy, University College London, Gower Street, London, WC1E 6BT, UK
15 Instituto de Fı́sica, Universidad Nacional Autónoma de México, Cd. de México C.P. 04510, México
16 Kavli Institute for Particle Astrophysics and Cosmology, Stanford University, Menlo Park, CA 94305, USA
17 SLAC National Accelerator Laboratory, Menlo Park, CA 94305, USA
18 Departamento de Fı́sica, Universidad de los Andes, Cra. 1 No. 18A-10, Edificio Ip, CP 111711, Bogotá, Colombia
19 Observatorio Astronómico, Universidad de los Andes, Cra. 1 No. 18A-10, Edificio H, CP 111711 Bogotá, Colombia
20 Institut d’Estudis Espacials de Catalunya (IEEC), 08034 Barcelona, Spain
21 Institute of Space Sciences, ICE-CSIC, Campus UAB, Carrer de Can Magrans s/n, 08913 Bellaterra, Barcelona, Spain
22 Fermi National Accelerator Laboratory, PO Box 500, Batavia, IL 60510, USA
23 Department of Astrophysical Sciences, Princeton University, Princeton NJ 08544, USA
24 Center for Cosmology and AstroParticle Physics, The Ohio State University, 191 West Woodruff Avenue, Columbus, OH 43210, USA
25 Department of Physics, The Ohio State University, 191 West Woodruff Avenue, Columbus, OH 43210, USA
26 The Ohio State University, Columbus, 43210 OH, USA
27 Department of Physics, Southern Methodist University, 3215 Daniel Avenue, Dallas, TX 75275, USA
28 Sorbonne Université, CNRS/IN2P3, Laboratoire de Physique Nucléaire et de Hautes Energies (LPNHE), FR-75005 Paris, France
29 Departament de Fı́sica, Serra Húnter, Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain
30 Institut de Fı́sica d’Altes Energies (IFAE), The Barcelona Institute of Science and Technology, Campus UAB, 08193 Bellaterra Barcelona, Spain
31 Department of Astronomy, The Ohio State University, 4055 McPherson Laboratory, 140 W 18th Avenue, Columbus, OH 43210, USA
32 Institució Catalana de Recerca i Estudis Avançats, Passeig de Lluı́s Companys, 23, 08010 Barcelona, Spain
33 National Astronomical Observatories, Chinese Academy of Sciences, A20 Datun Rd., Chaoyang District, Beijing, 100012, P.R. China
34 IRFU, CEA, Université Paris-Saclay, F-91191 Gif-sur-Yvette, France
35 Department of Physics and Astronomy, University of Waterloo, 200 University Ave W, Waterloo, ON N2L 3G1, Canada
36 Perimeter Institute for Theoretical Physics, 31 Caroline St. North, Waterloo, ON N2L 2Y5, Canada
37 Waterloo Centre for Astrophysics, University of Waterloo, 200 University Ave W, Waterloo, ON N2L 3G1, Canada

Corresponding author: Stéphanie Juneau

stephanie.juneau@noirlab.edu
2 S. Juneau et al.

38 Space Sciences Laboratory, University of California, Berkeley, 7 Gauss Way, Berkeley, CA 94720, USA
39 University of California, Berkeley, 110 Sproul Hall #5800 Berkeley, CA 94720, USA
40 Instituto de Astrofı́sica de Andalucı́a (CSIC), Glorieta de la Astronomı́a, s/n, E-18008 Granada, Spain
41 Aix Marseille Univ, CNRS/IN2P3, CPPM, Marseille, France
42 Université Clermont-Auvergne, CNRS, LPCA, 63000 Clermont-Ferrand, France
43 Department of Physics, Kansas State University, 116 Cardwell Hall, Manhattan, KS 66506, USA
44 Department of Physics and Astronomy, Sejong University, Seoul, 143-747, Korea
45 CIEMAT, Avenida Complutense 40, E-28040 Madrid, Spain
46 Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA
47 Department of Physics, University of Michigan, Ann Arbor, MI 48109, USA
48 University of Michigan, Ann Arbor, MI 48109, USA
49 Department of Physics & Astronomy, Ohio University, Athens, OH 45701, USA

ABSTRACT
The Dark Energy Spectroscopic Instrument (DESI) cosmology survey includes a Bright Galaxy Survey (BGS)
which will yield spectra for over ten million bright galaxies (𝑟 < 20.2 AB mag). The resulting sample will
be valuable for both cosmological and astrophysical studies. However, the star/galaxy separation criterion
implemented in the nominal BGS target selection algorithm excludes quasar host galaxies in addition to bona
fide stars. While this excluded population is comparatively rare (∼ 3 − 4 per square degrees), it may hold
interesting clues regarding galaxy and quasar physics. Therefore, we present a target selection strategy that
was implemented to recover these missing active galactic nuclei (AGN) from the BGS sample. The design of
the selection criteria was both motivated and confirmed using spectroscopy. The resulting BGS-AGN sample
is uniformly distributed over the entire DESI footprint. According to DESI survey validation data, the sample
comprises 93% quasi-stellar objects (QSOs), 3% narrow-line AGN or blazars with a galaxy contamination
rate of 2% and a stellar contamination rate of 2%. Peaking around redshift 𝑧 = 0.5, the BGS-AGN sample
is intermediary between quasars from the rest of the BGS and those from the DESI QSO sample in terms of
redshifts and AGN luminosities. The stacked spectrum is nearly identical to that of the DESI QSO targets,
confirming that the sample is dominated by quasars. We highlight interesting small populations reaching 𝑧 > 2
which are either faint quasars with nearby projected companions or very bright quasars with strong absorption
features including the Ly𝛼 forest, metal absorbers and/or broad absorption lines.

Keywords: Active galactic nuclei (16) — Quasars (1319) — Spectroscopy(1558) — Redshift surveys (1378) —
Galaxy spectroscopy (2171)

1. INTRODUCTION quasi-stellar objects (QSOs) as there might be distinct modes

Obtaining a complete census of active galactic nuclei of black hole activity (e.g., Heckman & Best 2014).
(AGN) is of fundamental importance for our understanding Observationally, one method to identify AGN is searching
of black hole growth as well as for a complete picture of for spectral signatures of AGN-ionized gas such as broad
galaxy formation and evolution (Madau & Dickinson 2014; emission lines (> 1000 km s−1 ) arising from high-velocity
Somerville & Davé 2015). The presence of an AGN can have gas near the black hole from the so-called broad line region
significant impacts on its host galaxy. For example, AGN (BLR), and highly-ionized but lower velocity gas from the
feedback is thought to regulate star formation by heating up narrow-line region (NLR) with typical line widths in the range
and/or expelling gas from the host galaxy (Fabian 2012; Har- 200 − 400 km s−1 (Osterbrock & Ferland 2006). However, the
rison et al. 2018). Furthermore, AGN play a crucial role in broad lines may be missing if the central region is obscured
the cosmological evolution of black holes as the likely main by intervening dusty material along the line of sight, whereas
drivers of the growth of supermassive black holes across cos- the NLR can be extended on scales of hundreds of parsecs
mic time (Marconi et al. 2004; Alexander & Hickox 2012). to ∼ 10 kiloparsecs (e.g., Bennert et al. 2002; Hainline et al.
As a result, understanding cosmic history requires to study 2014). AGN and quasars with both broad and narrow lines
a representative galaxy population including active phases in their spectra are classified as Type 1 whereas those with
of black hole accretion ranging from moderate AGN lumi- only narrow lines are classified as Type 2 (Khachikian &
nosities to the most luminous instances known as quasars or Weedman 1974). In addition to broad lines, the spectra of
 DESI BGS AGN Target Selection 3

Type 1 quasars typically feature a prominent blue power-law bridge the gap between the rest of the BGS galaxy sample
component attributed to the accretion disk (Elvis et al. 1994). and the QSO sample. Lastly, we summarize our main find-
The wide area cosmological survey conducted with the ings in Section 7. We use AB magnitudes throughout this
Dark Energy Spectroscopic Instrument (DESI Collaboration manuscript.
et al. 2022) will obtain spectra for approximately 40 million
galaxies and quasars (DESI Collaboration et al. 2016). While 2. OBSERVATIONS
DESI is primarily a Stage IV cosmology experiment (Levi 2.1. Tractor photometric catalogs
et al. 2013), the wealth of spectroscopic data is bound to The DESI survey target selection is based on photometric
generate a large number of astrophysical discoveries. The catalogs from the ninth data release (DR9) of the Legacy
selection of DESI targets is based on photometric properties Surveys1 (LS; see overview by Dey et al. 2019). The total
and was carefully optimized for each tracer. The dark-time LS footprint surpasses 14,000 square degrees and is divided
tracers consist of luminous red galaxies (Zhou et al. 2023, into a northern portion covered by the Beijing-Arizona Sky
LRGs), emission-line galaxies (Raichoor et al. 2023, ELGs), Survey (BASS; Zou et al. 2017) in the 𝑔 and 𝑟 bands and
and quasars (Chaussidon et al. 2023, QSOs). by the Mayall 𝑧-band Legacy Survey (MzLS) in the 𝑧 band.
At the low redshift end, the Bright Galaxy Survey (BGS) The southern portion is covered by the Dark Energy Camera
will achieve a high spectroscopic sampling for galaxies span- Legacy Survey (DECaLS) in all 𝑔, 𝑟, 𝑧 bands. The photometry
ning 0 < 𝑧 < 0.6 over a ∼14,000 square degree footprint is extracted by running the Tractor pipeline (Lang et al. 2016)
(Hahn et al. 2023). The BGS consists of a magnitude-limited on a detection image made by coadding those three broad
component (BGS Bright, 𝑟 < 19.5), augmented by a compo- bands, and then fixing the shape and size to measure fluxes
nent that reaches fainter magnitudes (BGS Faint, 𝑧 ≲ 22.2) and their uncertainties in each filter. For a nominal galaxy
but that further includes color-dependent fiber magnitude cuts exponential profile with a half-light radius of 0.45′′ , DECaLS
to ensure a high redshift success rate (> 95%). Together, the reaches median 5𝜎 detection limits of 𝑔 = 23.72, 𝑟 = 23.27,
BGS Bright and BGS Faint samples (BGS-BF) will produce and 𝑧 = 22.22 AB magnitudes. Similarly, the BASS/MzLS
redshifts for > 10 million galaxies during the course of the surveys reach 𝑔 = 23.48, 𝑟 = 22.87, and 𝑧 = 22.29 AB
DESI survey. magnitudes.
During early Survey Validation (SV; DESI Collaboration The ground-based optical and near-infrared photometry
et al. 2023a), we discovered that the star-galaxy separation was augmented by forced photometry in the mid-infrared
step of the BGS target selection removes quasars which spanning 3 − 22 microns. The latter is based on images from
present photometric signatures of a dominant point source the Wide-field Infrared Survey Explorer (WISE) satellite sur-
and, as a result, appear stellar-like rather than galaxy-like. vey (Wright et al. 2010) and its NEOWISE extension (Mainzer
While some of those quasars are eventually selected as DESI et al. 2014) that were reprocessed by Meisner et al. (2021) to
QSOs (Chaussidon et al. 2023), the latter favor 𝑧 > 1 targets. create the unWISE maps in the 𝑊1 and 𝑊2 channels (also
Thus, there remained an intriguing sub-population of bright see Meisner et al. 2018; Schlafly et al. 2019). The redder
AGN that risk being unaccounted by failing the BGS selection 𝑊3 and 𝑊4 photometry is extracted from the original WISE
criteria due to their star/galaxy cut but also failing the QSO maps (AllWISE) because the NEOWISE extension and un-
target selection criteria due to being at lower redshifts and/or WISE maps only exist for the first two channels. Lastly, the
possibly having a non-PSF morphology. We thus developed LS are matched to Gaia DR2 photometry in the broad 𝐺 band
a new sample selection algorithm to recover these missed (hereafter 𝐺 Gaia ; Gaia Collaboration et al. 2018).
luminous AGN and create the BGS-AGN sample, which is In particular, Gaia-detected sources are fixed to a point
complementary to BGS-BF by design. spread function (PSF) morphological type when running the
In this paper, we describe how we solved this problem by Tractor pipeline if:
defining a set of color and magnitude cuts trained on a spec-
troscopically confirmed sample from the Sloan Digital Sky (𝐺 Gaia ≤ 18 & AEN < 100.5 )|(𝐺 Gaia ≤ 13) (1)
Survey (SDSS), and applied to the DESI One-percent survey
portion of SV. The photometric and spectroscopic observa- where AEN is the ASTROMETRIC EXCESS NOISE from
tions are introduced in Section 2, followed by the primary Gaia DR2. While this condition is built-in when generating
BGS target selection and dedicated BGS-AGN target selec- the photometric catalog, it is not explicitly used when de-
tion in Sections 3 and 4, respectively. Results include the veloping the BGS target selection (Hahn et al. 2023) nor the
target selection validation from the One-percent survey and BGS-AGN criteria described below in Section 4 so we discuss
high signal-to-noise stacked spectra (Section 5). Lastly, we its impact on the BGS-AGN sample separately (Section 4.3).
present physical properties of the BGS-AGN sample in Sec-
tion 6, which demonstrate that BGS-AGN targets essentially 1 https://www.legacysurvey.org/dr9/description/
4 S. Juneau et al.

2.2. DESI Targeting We use the SDSS pipeline classification provided (CLASS)
to split the SDSS spectroscopic sample into STAR (795,683),
GALAXY (2,612,295) and QSO (861,906). In the case of
DESI targets comprise both dark time and bright time trac- stars and galaxies, we further require ZWARNING=0 to en-
ers as well as secondary targets. A detailed description of sure a reliable spectrum. This SDSS-LS test sample has the
DESI targeting is provided by Myers et al. (2023). In this advantage to include a large number of objects but is limited
work, we focus on BGS and QSO targets. The DESI QSO to brighter magnitudes compared to the DESI survey depth.
target class is designed to favor 𝑧 > 1 QSOs and is intended By cross-matching to the LS photometry, we can apply the
to be observed during dark time (Chaussidon et al. 2023). By same criteria as the DESI targeting to select subsamples of in-
contrast, the BGS is intended to be observed during bright terest to predict the DESI targeting outcome before validating
time (Hahn et al. 2023). However, some targets are assigned with the actual DESI SV observations.
to both the bright and dark time programs for data calibra- 2.3.2. DESI SV catalogs
tion purposes. In addition, a given object may be targeted by
more than one target class if it meets the respective selection The BGS-AGN target selection was included starting with
criteria of two or more classes. We describe how the target the One-Percent survey portion of SV, which was the third
information is captured and quantify the overlap between for phase (SV3; described by Myers et al. 2023, in their Sec-
objects that fulfill more than one targeting classes. tion 2.3). We use the SV3 redshift catalog as the validation
The DESI targeting information is encoded using bit- dataset (Section 5). The redshift catalogs produced by the
masks.2 The primary targeting column is called Redrock pipeline (Fuji version; Bailey et al., in preparation)
DESI TARGET for the main survey, and SV3 DESI TARGET include best-fit redshift, error estimates, quality flags and
for the SV3 survey (similarly for SV1, and SV2). The BGS the spectral classification based on the best-fitting template
is further divided into Bright, Faint and AGN components (SPECTYPE=STAR, GALAXY, QSO). Restricting to BGS-
and this targeting information is encoded as BGS TARGET for BF targets, the healpix redshift catalog from the SV3 survey
the main survey and SV3 BGS TARGET for the SV3 survey. includes a total of 274,890 entries with a valid fiber status4 for
and we compile the information most directly relevant to this 259,181 unique BGS-BF targets. This total includes 154,974
work in Table 1. unique BGS Bright targets and 104,207 BGS Faint targets.
By definition, there is no overlap between BGS-BF and We augment the default pipeline redshift catalogs with cata-
BGS-AGN. However, there is a small overlap of these classes logs from the QSO afterburner QuasarNet (Busca & Balland
with the QSO targets. Looking at the number of unique 2018) and the broad Mg II finder pipelines (see Section 6.2
targets with a good fiber status3 in SV3, we obtain N(BGS- of Chaussidon et al. 2023, for details). We compare the spec-
BF)=259,181 with only 92 also targeted as QSO (0.035%). tral classification from both classifiers and the redshift value
For BGS-AGN, there are 519 unique targets targets with 72 from QuasarNet.5 Alexander et al. (2023) found that using
also targeted as QSO (14%). We compare the physical prop- a modified pipeline with these additional classifiers identi-
erties of the BGS-BF, BGS-AGN, and the QSO samples in fied 94% of visually inspected quasars relative to 86% with
Section 6. the standard Redrock pipeline alone. We therefore examine
the results from both approaches in this work as well (Sec-
2.3. Spectroscopic catalogs tions 5.1.2, 5.1.3).
In addition to redshift and spectral type information, we
We use spectroscopic catalogs from SDSS DR16 (Ahu-
gather emission line fluxes resulting from spectral fitting with
mada et al. 2020) and from DESI SV (DESI Collaboration
version 2.5.0 of the FastSpecFit6 package (J. Moustakas
et al. 2023a,b) to motivate and validate the target selection
et al., in prep.; Moustakas 2023). FastSpecFit models
criteria. We further use spectra from the DESI SV campaign
the continuum using a stellar population component and a
for visual inspection, to select representative examples and to
smooth continuum component where the latter accounts for
create stacked spectra.
potential non-stellar contributions (such as an AGN accretion
2.3.1. SDSS DR16 disk power-law) and for possible effects from flux calibration
or other instrumental errors. The FastSpecFit package
In the case of SDSS DR16, we cross-matched the main red- additionally fits emission lines using Gaussian profiles with
shift table (SpecObj) to the LS DR9 photometry using a 1′′
matching radius, resulting in a sample of 4,531,048 objects.
4 We apply COADD FIBERSTATUS=0.
5 The MgII classifier uses the Redrock redshift as an input and does not
2 https://desidatamodel.readthedocs.io/en/latest/bitmasks.html#target-masks compute an independent redshift estimate.
3 COADD FIBERSTATUS=0 6 https://fastspecfit.readthedocs.io/en/latest/
 DESI BGS AGN Target Selection 5

Table 1. DESI samples and targeting summary.

Sample name SV3 DESI TARGET SV3 BGS TARGET Reference

Bit Bit name Bit Bit name

BGS-BF 60 BGS ANY 0 BGS FAINT Hahn et al. (2023)
1 BGS BRIGHT
BGS-AGN 60 BGS ANY 2 BGS WISE This work
QSO 2 QSO — — Chaussidon et al. (2023)

either narrow lines or a combination of narrow and broad ple spectra for a single object, we select the best spectrum
lines for the allowed transitions such as the Balmer lines using a SPEC PRIMARY flag determined by the redshift
of Hydrogen and Helium. In this work, we use rest-frame warning (ZWARN) and ranked by the value TSNR2 QSO,
optical lines such as H𝛽, [O iii]𝜆5007 (hereafter [O iii]) to which a proxy for the signal-to-noise ratio (SNR) for QSO
assess the ionization properties via the [O iii]/H𝛽 ratio, and spectra (Guy et al. 2023). In SV3, there are 782 coad-
estimate the AGN luminosity via the [O iii] luminosity. We ded spectra for BGS-AGN targets with no fiber warning
correct the latter for dust attenuation via the Balmer decrement (COADD FIBERSTATUS=0). We visually examined all
H𝛼/H𝛽 ratio assuming a reference intrinsic ratio of 3.0 as spectra but we report statistics for 519 unique objects in the
expected for AGN ionized gas (Osterbrock & Ferland 2006). remainder of this Paper to avoid double counting.
We further extend toward the rest-frame UV by searching
for broad Mg ii𝜆𝜆2796,2803 doublet emission, which is fit 3. DESI BGS TARGET SELECTION AND MISSING
separately from the optical lines by FastSpecFit. We use AGN
version 3.2 of the EDR value-added catalog7, except for some As detailed by Hahn et al. (2023), the BGS target selection
edge cases that we needed to refit as described in Appendix A. process involved magnitude and fiber-magnitude limits, star-
2.4. DESI SV spectra galaxy separation, spatial masking around bright stars and
other data quality cuts. Generally speaking, BGS-Bright is
We use the spectra from SV3 for visual inspection of all
nearly magnitude limited at 𝑟 < 19.5, while BGS-Faint ex-
BGS-AGN targets (Section 5.1.1) and to produce represen-
tends to fainter magnitudes but has additional color cuts to
tative examples as well as stacked spectra to demonstrate
maximize redshift success rate of those fainter targets. The
average spectral properties. The spectra were reduced by the
most relevant criterion for AGN and quasar subpopulations
Redrock pipeline. We used the Everest internal reduction
is the star-galaxy separation step because it is challenging to
for visual inspection, and the more recent Fuji reduction for
remove stars without also removing point-source dominated
the rest of the analysis in this paper including the final Re-
quasars and bright AGN.
drock classification results, afterburner classifier results and
the example spectra and stacked spectra. The Fuji version has
3.1. Star-galaxy separation
now been publicly released as the DESI Early Data Release
(EDR; DESI Collaboration et al. 2023b). The spectra shown The star-galaxy separation criterion was based on the com-
in this paper can be retrieved via the main DESI data access parison between the Gaia 𝐺 magnitude extracted assuming a
page8 or via the Spectra Access and Retrievable Catalog Lab narrow PSF profile (appropriate for stars) and the total Tractor
(SPARCL, Juneau et al. 2024).9 𝑟 magnitude accounting for the full light profile which can be
To maximize the spectral quality, we employ spectra that extended for galaxies. Stars and objects otherwise dominated
have been coadded per location on the sky (HEALPix coadds by a PSF-like morphology such as QSOs are expected to have
Górski et al. 2005) though we note that the spectra are coad- a small difference 𝐺 Gaia −𝑟 < 0.6 while resolved galaxies will
ded separately for the dark and bright programs. While BGS typically have an excess in the profile integrated 𝑟 magnitude
is designed to be conducted primarily during bright time, and tend to have 𝐺 Gaia − 𝑟 > 0.6 (Ruiz-Macias et al. 2020,
some targets are also observed during dark time for calibra- 2021). This is supported by the distributions of 𝐺 Gaia − 𝑟
tion and quality assurance purposes. When there are multi- values for the spectroscopically classified SDSS sample with
LS DR9 photometry from Section 2.3.1 as shown in Fig-
ure 1. This star-galaxy separation criterion is only applicable
7 https://fastspecfit.readthedocs.io/en/latest/fuji.html to objects with a Gaia detection.
8 https://data.desi.lbl.gov/doc/releases/edr/ We note that the majority (88.5%) of SDSS galaxies are
9 https://astrosparcl.datalab.noirlab.edu/ not detected in 𝐺 Gaia and so are not shown in Figure 1 and
6 S. Juneau et al.

line removes 99% of SDSS stars (578,093/584,165), 10% of

SDSS galaxies (149,415/1,506,924), and 91% of SDSS QSOs
(261,297/287,257). We note that criteria in 𝑟 magnitudes re-
sult in strong diagonal features and namely the BGS-Faint
limit is around 𝑟 ∼ 20.22 (the solid blue line corresponds to
𝑟 = 20.3 for visual reference given the bin size).

3.3. Missing QSOs due to non-PSF morphology

One could expect that a significant fraction of the QSOs
that are cut from the BGS sample (in red on the left-hand
panel of Figure 2) may be recovered as DESI QSO targets.
We check this by cross-referencing with potential DESI QSO
targets based on the QSO targeting algorithms (Chaussidon
et al. 2023; Myers et al. 2023) applied to the LS DR9 pho-
tometry. We find that 246,885 out of the 261,297 (94.5%)
SDSS spectra with a QSO class that were cut due to the
Figure 1. Distribution of morphology-sensitive 𝐺 Gaia −𝑟 magnitude
difference for SDSS galaxies with spectral classification of STAR 𝐺 Gaia − 𝑟 criterion are potential QSO targets. Among the
(light purple), GALAXY (blue) or QSO (open red histogram). Given remaining 14,412 missing SDSS QSOs that are not potential
the much larger number of stars and their narrow distribution, we DESI QSO targets, a majority (56%) have a resolved (i.e., non-
only show one fifth of that subsample. The vertical green dashed PSF) morphology according to the Tractor MORPHTYPE.10
line is the adopted value to reject stars from the BGS target selection. As detailed by Chaussidon et al. (2023), a PSF morphology
Galaxies are either undetected in 𝐺 Gaia or tend to have 𝐺 Gaia − 𝑟 > is required for the nominal selection of DESI QSO targets
0.6.
as well as other photometric criteria tailored to preferentially
select cosmological tracer QSOs at 𝑧 > 1, which are expected
the criterion in not applicable in such cases. Galaxies with to appear point-like in imaging.
𝐺 Gaia − 𝑟 > 0.6 account for 10%, and only 1.5% are below However, the subpopulation of 8,115 quasars with an ex-
the threshold. In contrast, most (89%) SDSS stars have a tended morphology could be physically interesting as possibly
𝐺 Gaia detection and they display a very narrow distribution consisting of quasars with a resolved host galaxy component,
in 𝐺 Gaia − 𝑟. Among all 795,683 stars, 88% are below the in which case we would expect them to lie at lower red-
0.6 threshold while only 0.8% are above and the remaining shifts and/or probe moderate luminosity AGN. We indeed
11% lack a 𝐺 Gaia detection. The QSO distribution is slightly find a preferentially lower redshift (median of 𝑧 = 0.46) for
broader than that of stars but the majority of QSOs with a non-PSF morphological types compared to their counterparts
𝐺 Gaia detection would be cut out by the threshold. However, with a PSF morphological type (median of 𝑧 = 1.57). This
a significant fraction of QSOs lack a 𝐺 Gaia detection (43.5%; sets our expectations for the BGS AGN sample relative to the
not shown), which leaves 54.3% of the SDSS QSO subsample bulk of the DESI QSO target sample, which we will verify in
below the threshold and 2.2% above. Some of the cases Section 6.1.
undetected by Gaia will be selected as part of the BGS-BF Given that not all the missing QSOs will be recovered by
and/or QSO targets. the DESI QSO target selection, we need to develop criteria to
3.2. SDSS sample with DESI BGS selection criteria distinguish between stars and QSOs that are independent from
We apply the BGS target selection criteria (Hahn et al. 𝐺 Gaia − 𝑟 given the strong overlap of this quantity between
2023) with the exception of the 𝐺 Gaia − 𝑟 > 0.6 cut to stars and QSOs. We use combinations of optical and infrared
the SDSS-LS test sample with spectroscopic classification colors to achieve this goal. The BGS-AGN selection that we
(Section 2.3.1) to investigate the impact of that criterion introduce in this work (Section 4) would target 7,179 out of
on the three spectral types. Figure 2 shows the result- the 8,115 missing QSOs with a non-PSF morphology (88%).
ing bivariate distributions of 𝐺 Gaia − 𝑟 as a function of Overall, it would target 10,671 (4.1%) of the SDSS QSOs cut
𝐺 Gaia . For each spectral type (class in SDSS nomencla- with 𝐺 Gaia − 𝑟< 0.6. When taken together with the potential
ture), we compare objects that meet the BGS Bright or BGS DESI QSO targets, we obtain a 97.3% potential recovery rate.
Faint criteria (BGS-BF; in grey) with those that meet all
the other criteria except for the star-galaxy separation and 10 The morphological type is called TYPE in the original Tractor catalogs but
are therefore cut from BGS-BF (in red). We find that the was renamed MORPHTYPE in the DESI catalogs to distinguish it from the
𝐺 Gaia − 𝑟 > 0.6 criterion shown with the horizontal green spectral type SPECTYPE.
 DESI BGS AGN Target Selection 7

Figure 2. SDSS-LS cross-matched sample on the 𝐺 Gaia − 𝑟 vs. 𝐺 Gaia plane divided according to the SDSS spectral class: (left) QSOs, (middle)
Stars, (right) Galaxies. This figure illustrates some of the selection criteria of the BGS-BF sample (grey color map) relative to the complement
sample that is cut (red color map) based on 𝐺 Gaia − 𝑟 > 0.6 (green horizontal line). For reference, the solid blue line corresponds to 𝑟 = 20.3
chosen to be slightly fainter than the formal BGS faint limits (𝑟 < 20.175 and 𝑟 < 20.22) to minimize overlap with the contours. The number
density contours are spaced in increments of 1 dex with the outermost contour corresponding to 10 objects per bin. Numbers are labeled and
show that the 𝐺 Gaia − 𝑟 criterion removes the majority of stars (99%), a modest percentage of galaxies (10%), but also unfortunately removes a
significant percentage of QSOs (91%).

4. BGS AGN TARGET SELECTION Figure 3 demonstrates the first three cuts for the SDSS-LS
The BGS AGN target selection relied on both precursor DR9 sample after applying the selection criteria for BGS-BF
SDSS data and on early DESI SV1 data and was therefore (grey) and for the new BGS-AGN sample introduced here and
implemented starting with the SV3 phase of survey validation in Section 4.2 (red). Namely, the (𝑧 − 𝑊1) − (𝑔 − 𝑟) cut is
and carried over the main DESI survey (Myers et al. 2023). shown directly on the horizontal axis (vertical dashed line),
the 𝑊1 − 𝑊2 cut is made along the vertical axis (horizontal
dashed line) while the last cut yields the diagonal dashed
4.1. Main AGN criteria
line creating the top-right selection box for the BGS-AGN
We identify AGN hosts primarily via the presence of a sample. We can see that the QSOs that happen to be selected
hot dust component, which we infer by comparing the WISE as part of BGS-BF form a sequence in this parameter space
infrared colors to optical colors, and by requiring a red 𝑊1 − and that BGS-AGN QSOs appear to follow and extend the
𝑊2 color (e.g., Stern et al. 2012; Wu et al. 2012). We further same sequence away from the bulk of galaxies. For this
select for the presence of a strong point source component via SDSS-LS DR9 sample, applying the BGS-AGN selection
the comparison of Gaia and Tractor photometry. yields a sample comprising 97.4% QSOs, 0.3% stars and
Relative to the BGS design paper (Hahn et al. 2023), we 2.3% galaxies.
adopt a similar approach in terms of applying magnitude,
color and quality cuts on the photometry but we employ the
opposite strategy to preferentially select objects that may be 4.2. Quality criteria
point-source dominated according to 𝐺 Gaia − 𝑟. We use pre- We apply data quality criteria to ensure adequate imaging
dictions from the SDSS-LS DR9 sample together with SV1 coverage, valid fluxes and the absence of masked pixels due
spectra that were visually inspected to define the following to, e.g., proximity to a bright source in the 𝑊2 band.
BGS AGN criteria: Specifically, we require at least one observation in all of
𝑔, 𝑟, 𝑧 bands, with physical values of flux and inverse variance
(ivar):
(𝑧 − 𝑊2) − (𝑔 − 𝑟) > −0.5 (2)
(𝑧 − 𝑊1) − (𝑔 − 𝑟) > −0.7 (3) nobs𝑖 > 0 for 𝑖 = 𝑔, 𝑟, 𝑧 (6)
(𝑊1 − 𝑊2) > −0.2 (4) flux𝑖 > 0 for 𝑖 = 𝑔, 𝑟, 𝑧 (7)
(𝐺 Gaia − 𝑟) < 0.6 (5) ivar𝑖 > 0 for 𝑖 = 𝑔, 𝑟, 𝑧. (8)
8 S. Juneau et al.

Figure 3. 𝑊1 − 𝑊2 color as a function of (𝑧 − 𝑊1) − (𝑔 − 𝑟) from the LS DR9 photometry for each SDSS spectral class: QSO (left-hand
panel), STAR (middle panel) and GALAXY (right-hand panel). The BGS-BF sample and BGS-AGN sample are shown in grey and red colors,
respectively. BGS-AGN with spectral class of STAR or GALAXY are plotted with individual symbols rather than histograms and contours
due to the small number of data points. The blue dashed lines delineate the three color-based BGS-AGN selection criteria from Section 4.
The number density contours are spaced in increments of 1 dex (0.5 dex) for the BGS-BF (BGS-AGN) sample, with the outermost contour
corresponding to 10 objects per bin. Notably, the BGS-AGN sample extends the QSO population toward more QSO-like colors and away from
galaxy-like colors.

Given the importance of the WISE bands for a reliable AGN Unlike the BGS selection, we further require an entry in
identification, we place comparatively stronger constraints on the Gaia DR2 catalog (𝐺 Gaia ≠ 0) and impose the following
the 𝑊1 and 𝑊2 bands: Gaia magnitude cut:

S/N𝑖 > 10 for 𝑖 = 𝑊1, 𝑊2 (9) 𝐺 Gaia > 16. (14)

WISEM2 = 0 (10) Lastly, we apply two more quality flags from the photo-
metric Tractor catalog (MASKBITS) to avoid objects near
where WISEM2 indicates whether the object touches a pixel bright stars (bit “BRIGHT”) or in areas identified as part of a
in the WISEMASK W2 bright star mask11 as recorded in the globular cluster (bit “CLUSTER”).
Tractor photometric catalog (MASKBITS).12
Next, we apply magnitude cuts similar − but not identical 4.3. Criterion inherited from the photometry
− to the rest of the BGS target selection (Hahn et al. 2023): We re-examine the bivariate distribution of 𝐺 Gaia − 𝑟 as a
function of 𝐺 Gaia after applying all the BGS-BF and BGS-
𝑟 fibtot > 15 (11)
AGN criteria to the SDSS-LS sample including the quality
𝑟 fiber < 22 (12) cuts (Figure 4). For objects fainter than 𝐺 Gaia > 18 we can
see a continuation of the contours of a small population of
QSOs selected by BGS-Faint (close to the limiting magni-
16 < 𝑟 < 17.5 if PSF





 tudes around 𝑟 ∼ 20.22). At the bright end (𝑟 < 17.5,
16 < 𝑟 < 19.5 if not PSF (13) diagonal blue dashed line), the selection includes PSF mor-
phology quasars, some of which may be bright due to their



 19.5 ≤ 𝑟 < 20.3
 if not PSF & (𝑟 fiber < 21.5) lower redshifts while others might overlap with intrinsically
luminous quasars which will also be selected as QSO targets.
where 𝑟, 𝑟 fiber and 𝑟 fibtot are respectively the 𝑟-band magni-
The small gap between 𝐺 Gaia < 18 and 𝑟 > 17.5 is ex-
tude, the 𝑟-band fiber magnitude, and the total 𝑟-band fiber
plained by a criterion inherited from the Tractor photometry,
magnitude corrected for aperture loss.
which forces a morphological type of ‘PSF’ for objects with
a bright Gaia magnitude: 𝐺 Gaia < 18 (dotted blue line) while
11 WISE masks are documented here: our criteria only allowed PSF morphology for objects brighter
https://catalog.unwise.me/files/unwise bitmask writeup-03Dec2018.pdf than 𝑧 < 17.5 (dashed blue line). This photometric criterion
12 https://www.legacysurvey.org/dr9/bitmasks/#maskbits was not initially considered during the sample selection. It
 DESI BGS AGN Target Selection 9

In this section, we describe how we used the DESI SV3

observations to validate the BGS AGN sample selection.
Namely, we check the spectral type (§5.1) and redshift accu-
racy (§5.2) of the BGS AGN spectra compared to results from
visual inspection. We further compare the average spectral
shape to that of the average DESI QSO (§5.3) and confirm the
presence of AGN ionization using an emission line diagnostic
diagram (§5.4).
5.1. Spectral classification
We assess the BGS AGN selection by checking the spectral
classification from three different methods: (1) visual inspec-
tion; (2) default classification from the Redrock pipeline; (3)
classification from the two QSO classifiers that are run in
post-processing: QuasarNet (QN) and the MgII classifier.
5.1.1. Visual inspection

Figure 4. SDSS cross-matched sample on the plane defined by We conducted a visual inspection campaign to examine
𝐺 Gaia −𝑟 as a function of 𝐺 Gaia , illustrating some of the selection cri- each BGS-AGN spectrum and determine the spectral type
teria of the BGS-BF sample (grey color map) and BGS-AGN sample (STAR, GALAXY or QSO), the redshift and the quality of
(red color map). These include a cut based on 𝐺 Gaia − 𝑟= 0.6 (green the spectra (from 0 being unusable to 4 being the best qual-
horizontal line), and cuts based on 𝑟 band magnitudes (dashed and ity). We followed a very similar method as for the DESI QSO
solid blue lines values of 𝑟 = 17.5 and 20.3, respectively). Further- targets (Alexander et al. 2023). Each spectrum is indepen-
more, the LS DR9 Tractor photometry catalog includes a threshold dently examined by two people, and a third person reviews
at 𝐺 Gaia =18 (dotted lines) in the sense that objects brighter than this
value are assigned a PSF morphological type. The number density
all the results to merge them, resolving conflicts if they arise.
contours are spaced in increments of 0.5 dex with the outermost Because the BGS-AGN targets tend to be bright by selection,
contour corresponding to 10 objects per bin. most of the spectra have a high signal-to-noise ratio and their
redshift and spectral class can readily be identified visually.
impacts the subset of objects with 16 < 𝐺 Gaia < 18 and As a result, 516 out of 519 visual classifications were assigned
with 𝑟 > 17.5, which are currently excluded from the sample a high quality (≥ 2.5) translating into a high confidence in
(small triangular region in Figure 4). the visual inspection results. The three cases with low quality
were all found to be stellar contaminants (Appendix B).
4.4. Sky coverage of BGS AGN targets In addition to the three main spectral types (STAR,
GALAXY, QSO), visual inspectors were asked to note if the
After applying the full set of BGS AGN selection criteria, spectrum appeared to be a narrow-line AGN (Type 2 AGN)
we compute the target number density, which is applicable for based on the presence of a clear [Ne iv]𝜆3425 line or based on
both the SV3 and the main DESI surveys over the photometric a visual assessment of the typical strong emission lines (e.g.,
footprint. As displayed in Figure 5 we find a uniform sky elevated [O iii]/H𝛽 ratio). For the 519 unique BGS-AGN
distribution with a mean around 3−4 targets per square degree targets in SV3, the initial visual spectral classification yield
and with 90% of the values in the range 1−7 targets per square 485 QSOs (93.4%), 12 Type 2 AGN candidates (2.3%), 13
degree. Importantly, there is no enhanced target density close galaxies (2.5%) and 9 stars (1.7%).
to the Galactic plane (solid black line), which could occur if However, we consider the visual Type 2 classification ten-
there were significant stellar contamination. tative so we examine quantitative emission line ratio clas-
These modest values indicate that BGS AGN targets are sification in Section 5.4, display individual cases in Ap-
rare relative to other target classes such as the full BGS targets pendix A, and summarize the outcome here as well as in
(∼ 80 deg−2 ) and QSO targets (∼ 320 deg−2 ) (Myers et al. Table 2. Based on emission line measurements, we relabel
2023). While not numerous, they still contribute to increasing one Type 2 to GALAXY due to the emission lines being very
the completeness of bright galaxies observed by DESI, and narrow (𝜎 <100 km s−1 ) and typical of low-mass star-forming
can be of particular interest for galaxy and black hole studies. galaxies. Conversely, one spectrum labeled as GALAXY was
Namely, they are part of the DESI AGN Summary Catalog changed to Type 2 due to having strong [O iii]/H𝛽 together
(Canning et al., in preparation). with a greater line width and tentative [Ne iv]𝜆3425 and Mg ii
emission. Furthermore, one spectrum assigned as GALAXY
5. SURVEY VALIDATION RESULTS that was flagged as “Type 1?” indicating tentative broad lines
10 S. Juneau et al.

Figure 5. Sky density map of BGS AGN targets. As indicated by the colorbar, the target number density is fairly uniform across the entire
photometric footprint with 90% of the values in the range 1 − 7 targets per square degree. The solid black line (red dotted line) delineates the
Galactic plane (Ecliptic plane).

Figure 6. Spectral Classification of the BGS-AGN SV3 sample according to the default Redrock pipeline (left), the QN and MgII afterburner
classifiers (center), and from visual inspection (right). While Redrock misses several cases, they are recovered using the afterburners, which
yield results very similar to visual inspection.

has now been confirmed as a QSO with a more recent DESI The Redrock pipeline assigns a SPECTYPE of STAR,
spectrum (Figure 21). GALAXY or QSO based on the best-fit model template. In
Three of the 13 spectra assigned a GALAXY spectral type this work, we report the results from the Fuji spectral reduc-
are consistent with blazars with weak or no emission lines, a tion used to produce the DESI EDR (DESI Collaboration et al.
blue continuum, and a clear detection in the Very Large Array 2023b). When comparing Redrock with visually classified
Sky Survey (VLASS; Lacy et al. 2020, shown in Figure 25). spectra from the first phase of SV (SV1) for DESI QSO tar-
Lastly, three of the Type 2 may instead be narrow-line Seyfert gets, Alexander et al. (2023) found that the Redrock pipeline
1 (NLSy1), which are intermediate in terms of line widths tends to correctly identify stars but occasionally misidentifies
(Figure 22). If we consider those as “narrow-line” AGN for QSOs as galaxies such as cases with both a significant stellar
the purpose of the final spectral classification, we obtain: 487 continuum and AGN/QSO emission lines. This is consistent
QSOs (93.8%), 14 narrow-line AGN or blazar candidates with the main findings for our sample of BGS AGN spec-
(2.7%), 9 galaxies (1.7%) and 9 stars (1.7%). Considering tra. Redrock assigns a very similar percentage of stars (1.9%
them as QSO would slightly increase our QSO success rate versus 1.7%) as the visually inspection results but assigns
to 94.4%. a lower QSO percentage (75.5% compared to 93.4%), and
conversely a higher percentage of galaxies (22.5%). While
5.1.2. Redrock pipeline Redrock missed ∼ 19% of the quasars, the redshift was not
 DESI BGS AGN Target Selection 11

necessarily incorrect for such a high fraction. We evaluate classified as either Type 2 (N=5; Figure 24) or as NLSy1
the redshift success rate separately in Section 5.2. (TARGETID=39633413522588456; Figure 22). These cases
were all found to have signatures of AGN and therefore this
5.1.3. Afterburner QSO classifiers does not represent a major disagreement in the physical nature
of those objects. Conversely, there were five visually iden-
Similarly to the QSO target identification procedure out-
tified QSOs not classified as such by the afterburners. One
lined by Chaussidon et al. (2023), we use the results from
of the five was assigned an incorrect redshift from Redrock
two QSO classifiers that were ran in post-processing after the
likely due to prominent broad absorption lines (BALs): TAR-
Redrock pipeline. The QN classifier (Busca & Balland 2018;
GETID=39627939041510701 (Figure 16). The remaining
Farr et al. 2020) is a machine-learning approach that assigns a
four spectra are shown in Figure 17 and have relatively lower
probability of spectral emission line features being consistent
signal-to-noise ratios and/or redder continua relative to nor-
with a QSO giving a confidence from 0 to 1, as well as de-
mal blue quasars. Overall, the reliable QSO recovery by the
termining a redshift for the spectrum. In this work, we apply
combined QN and MgII classifiers supports their application
the same criterion as for the QSO targets, which is to require
to the BGS AGN target class for the main DESI survey.
that at least one of the main expected quasar emission lines
considered (Ly𝛼, C iv, C iii], Mg ii, H𝛼, H𝛽) is assigned a
5.2. Redshift accuracy
confidence over 0.95 (C LINE BEST>0.95).13
The second classifier is based on the presence of a broad In addition to the spectral classification, a main objective of
Mg ii line. As described by Chaussidon et al. (2023), the MgII visual inspection is assessing the redshift. The prospect14
Classifier algorithm fits a Gaussian within a 250 Å window visual inspection tool presents a few redshift solutions from
centered at the expected position of Mg ii given the Redrock the Redrock pipeline as a starting point with an overlaid best-
redshift. There are three criteria to accept the presence of a fit model and the option to interactively adjust the redshift
broad Mg ii line: (i) the improvement of 𝜒2 must be better with both a coarse and a fine slider. We consider redshifts
than 16, (ii) the width of the Gaussian greater than 10 Å, and with a visual quality flag greater or equal to 2.5 to be reli-
(iii) the amplitude-over-noise of the Gaussian be greater than able, and use them as the reference truth to compare with the
three. If these conditions are met, the spectrum is reclassified values determined from the Redrock pipeline and QN clas-
as a QSO. sifier. While in marginal quality cases the VI redshift could
Using the combined QN and MgII classifiers yields very itself be wrongly assigned, we note that the majority of the
similar numbers and percentages as visual inspection: namely BGS-AGN quasar spectra have a very high quality above 3.5
488 QSOs (94%), 21 galaxies (4%) and 10 stars (1.9%). There (480/485=99%) with only 1% having been assigned a mod-
is no distinction for potential Type 2 AGN with neither the erately good quality flag (2.5 − 3.5).
standard Redrock pipeline nor this modified QSO classifier The MgII classifier can change the spectral type but it uses
pipeline so the 21 “galaxies” include some narrow-line AGN the Redrock redshift value. However, QuarsarNet computes
candidates as determined by the visual classification (Sec- likely redshifts based on its own models and probability cal-
tion 5.1.1). culations. If the QN redshift differs from the Redrock redshift
We show the proportion of each spectral type according by more than 0.05 and the QSO identification has a high con-
to each method in Figure 6. In conclusion, either visual fidence (C LINE BEST>0.95), the Redrock pipeline is run
classification or using the afterburner QSO classifiers improve again with only QSO templates and using the QN redshift
the completeness of QSO classification by 19-20% among as a prior to measure a new redshift for the quasar spectrum
the BGS AGN sample. We note that the spectral type of all (Chaussidon et al. 2023). For QSO targets, accurate red-
BGS-AGN objects that were identified as QSOs by the default shifts are defined as within 3000 km s−1 of the true redshift
Redrock pipeline were confirmed during the visual inspection (compared to 1000 km s−1 for galaxies; Lan et al. 2023).
campaign. We discuss the small remaining contamination by We define the redshift accuracy by:
stars and galaxies in Appendix B.
|𝑧 𝑅𝑅 − 𝑧𝑉 𝐼 |
In detail, while the numbers of QSOs found by visual 𝑑𝑧 = (15)
inspection and by the modified QSO classifier pipeline are (1 + 𝑧 𝑉 𝐼 )
similar, there are some small discrepancies in classification. where 𝑧 𝑉 𝐼 is the redshift from visual inspection and 𝑧 𝑅𝑅
Namely, there are six afterburner QSOs that were visually is the redshift from Redrock. Below, we consider the Re-
drock redshifts before and after applying the QN classifier to
13 An original threshold value of 0.5 was reported by Chaussidon et al. (2023) compare their relative accuracy.
but we subsequently revised it to 0.95 based on calculating classification
precision and completeness as a function of C LINE BEST values (Canning
et al., in prep.) 14 https://desi-prospect.readthedocs.io/en/latest/index.html
12 S. Juneau et al.

Finally, we quantify the percentage of quasars with good blue). In both cases, the stacked spectrum of the missed
redshift accuracy as the percentage with good VI quality QSOs are redder in terms of spectral slope and with slightly
(≥ 2.5) and a difference 𝑑𝑧 ≤ 0.010 (which corresponds more pronounced stellar absorption features (e.g., between
to 3000 km s−1 ). Among 485 visually identified quasars, 3500-4000 Å). This difference is larger for the QSO sample
we find that 99.4% have a good redshift accuracy after us- (light blue) and more moderate for the BGS AGN missed
ing the QN priors, compared to 95.3% based on the original sample (orange). Altogether, this confirms that, on average,
Redrock redshifts. Similarly, we find that 99.6% of the 488 the BGS AGN sample largely comprises typical QSOs. How-
quasars identified with the combined Redrock+MgII+QN cri- ever, individual objects can still significantly deviate from
teria have a good redshift accuracy, which is also a 4% im- these median spectra.
provement relative to the original Redrock redshifts (95.5%).
The Redrock algorithm only identified 392 quasars (∼ 20% 5.4. Emission-line classification
incomplete relative to the combined and VI methods) but Optical emission line ratios are commonly used to identify
99.0% of that sample have a good redshift accuracy. Based the primary source of ionization in galaxies. The well-known
on these results, we conclude that the most complete and ac- BPT diagnostic diagram (Baldwin et al. 1981) combines the
curate redshift selection of quasars for the BGS AGN sample [O iii]/H𝛽 and [N ii]/H𝛼 ratios, which are sensitive to a combi-
can be obtained by combining Redrock with the MgII and QN nation of ionization parameter and gas-phase metallicity and
classifiers and that using QN priors can improve the fraction can be enhanced when gas is ionized by non-stellar sources
of targets with an accurate redshift by 4% to reach > 99%. such as AGN and/or shocks (e.g., see review by Kewley et al.
2019). Those line ratios are further selected to be closely
5.3. Stacked spectra spaced in wavelength and therefore have little to no sensitiv-
ity to dust obscuration. When using optical spectra, the BPT
To examine the typical spectral features of BGS AGN tar-
is only applicable at low redshifts given that H𝛼 and [N ii]
gets, we create stacked spectra following the same procedure
shift to the near infrared starting at redshift around 𝑧 > 0.45.
as Alexander et al. (2023). Namely, the spectra are shifted
A few alternative diagnostic diagrams have been suggested
to the rest-frame, aligned on a common wavelength grid, and
by keeping the [O iii]/H𝛽 ratio which is accessible in the
normalized at a rest-frame wavelength of 3000 Å. They are
optical range up to redshift 𝑧 ∼ 1 but replacing the redder H𝛼
then stacked by taking the median at each spectral bin.
and [N ii] lines. These include substituting [N ii]/H𝛼 with
In the top panel of Figure 7, we show the median spectrum
absolute 𝐻-band magnitude (Weiner et al. 2007), the rest-
for the bulk of the sample, which comprises 395 visually iden-
frame 𝑈 − 𝐵 color as part of the Color-Excitation diagram
tified QSOs at 𝑧 < 0.95. This stacked spectrum resembles a
(CEx; Yan et al. 2011), the stellar mass as part of the Mass-
typical broad-line (Type 1) QSO spectrum with a blue contin-
Excitation diagram (MEx; Juneau et al. 2011, 2014) or the
uum attributed to the accretion disk and several broad permit-
[O iii] emission line width as part of the Kinematic-Excitation
ted lines such as Mg ii, H𝛾, H𝛽 and H𝛼 (e.g., Vanden Berk
diagram (KEx; Zhang & Hao 2018).
et al. 2001). In the bottom panel, the stacked spectrum was
In this work, we use both the original KEx diagram (Zhang
built by using the 90 QSOs at 𝑧 ≥ 0.95, for which we obtain a
& Hao 2018) and a modified KEx diagram that we adapt to
spectral range probing further into the rest-frame UV. The re-
depend on the maximum line width (𝜎max ) of H𝛽, H𝛼 and
sulting stack is again typical and strongly resembles the lower
Mg ii instead of the originally proposed [O iii] line width.
redshift counterpart except with a slightly bluer continuum.
The motivation for this change is to visually and quantitatively
The bluer wavelength coverage allows us to probe additional
distinguish narrow-line (Type 2) and broad-line (Type 1) AGN
broad lines such as Ly𝛼, C iv and C iii]. Together, these
as part of a single diagram.15 In all cases, the [O iii]/H𝛽 refers
two stacked spectra represent 93% of the BGS-AGN sam-
to only the narrow component of the H𝛽 line, while [O iii]
ple. We examine spectra for other classes such as narrow-line
is assumed to be a single component in the FastSpecFit
(Type 2) AGN, galaxies and stars separately in Section 5.4
fitting algorithm (Section 2.3.2). The maximum line width
and in Appendix B.
corresponds to:
Next, we compare directly with the stacked spectra from the
visually inspected DESI QSO sample to further confirm that 𝜎max = max{𝜎H𝛽 (n), 𝜎H𝛽 (b), 𝜎H𝛼 (b), 𝜎Mg ii }, (16)
the BGS AGN sample indeed consists of bona fide quasars.
Similarly to Alexander et al. (2023), we created new stacks to where 𝜎H𝛽 (n) and 𝜎H𝛽 (b) are respectively the narrow and
separate QSOs identified with Redrock from those missed by broad H𝛽 components, 𝜎H𝛼 (b) is the broad H𝛼 component,
the default Redrock pipeline that were however recovered vi- and 𝜎Mg ii is the Mg ii line width. In all cases, a value of line
sually. As Figure 8 demonstrates, the QSO spectra identified
by Redrock are very similar between the BGS AGN sample
presented here (red) and the main QSO survey of DESI (dark 15 also see Mullaney et al. (2013), where the authors used the H𝛼 width
 DESI BGS AGN Target Selection 13

Figure 7. Stacked spectra of BGS-AGN targets visually identified as QSOs split between the bulk of the sample at low redshift (𝑧 < 0.95;
𝑁 = 395) and the remaining higher redshift QSOs (𝑧 ≥ 0.95; 𝑁 = 90). In each panel, the light line shows the stacked spectrum, and the darker
line is a smoothed version with a Gaussian kernel (𝜎 = 2 pixels). Vertical dashed lines mark the expected location of emission lines, as labeled
at the top.

Figure 8. Stacked spectra of BGS-AGN QSOs identified by Redrock (red) or missed by Redrock but identified by VI (orange) compared to the
average spectra of DESI QSO targets identified by Redrock (dark blue) and those missed by Redrock but identified from VI (light blue). In each
case, the light line shows the stacked spectrum, and the darker line is a smoothed version with a Gaussian kernel (𝜎 = 2 pixels). Vertical dashed
lines mark the expected location of emission lines, as labeled at the top. Short vertical dotted lines mark the location of absorption lines.
14 S. Juneau et al.

Figure 9. (Top) Distribution of [O iii] line widths for the BGS-BF Figure 10. (Top) Distribution of the maximum line widths (𝜎max ;
(grey) and BGS-AGN (red) subsamples with H𝛽 and [O iii] detec- Equation 16) for the BGS-BF (grey) and BGS-AGN (red) subsamples
tions. Overall the BGS-AGN sample is characterized by broader with H𝛽 and [O iii] detections. Overall the BGS-AGN sample is
lines compared to the BGS-BF sample. (Bottom) Original KEx characterized by broader lines compared to the BGS-BF sample.
AGN diagnostic diagram consisting in the [O iii]/H𝛽 line ratio ver- (Bottom) Modified KEx AGN diagnostic diagram consisting in the
sus the line width for the [O iii] line, for the BGS-BF sample (shaded [O iii]/H𝛽 line ratio versus 𝜎max , shown for the BGS-BF sample
gray) and the BGS-AGN sample shown with symbols coded to their (shaded gray) and the BGS-AGN sample split per spectral type
final spectral classification as labeled. We note two QSOs for which classification as labeled and as described in Figure 9. The diagonal
the visually identified broad lines were not detected by the automated dashed line is the demarcation of the original KEx diagram.
spectral fitting due to their low signal-to-noise spectra (open red cir-
cles). The diagonal dashed line is the demarcation of the original
ponent only) as a function of 𝜎[O iii] . The dashed line is the
KEx diagram.
empirical division proposed by Zhang & Hao (2018) to dis-
tinguish between star-forming galaxies to the lower left and
width is only considered if the line flux reaches S/N> 3. The AGN host galaxies to the upper right. The underlying grey
minimum requirement for inclusion on the KEx diagram is bivariate distribution corresponds to the BGS-BF sample and
the availability of narrow H𝛽 and [O iii] emission lines, for we find that only 13.8% is located above and to the right of
which we also apply a S/N> 3 detection threshold. Among the dividing line in the AGN side. The color symbols show
the 519 unique BGS-AGN targets, there are 416 that cover the BGS-AGN sample and in contrast to the BGS-SF sample,
the necessary spectral range and 390/416 (94%) with a line we find that the majority (96.2%) are on the AGN side.
detection for both [O iii] and H𝛽. The color and shape of the symbols used for the BGS-
First, we show the original KEx diagnostic diagram in Fig- AGN sample are based on a revised spectral classification that
ure 9. The top panel displays a comparison of the [O iii] line involved examining all individual spectra and images as well
width distribution between the BGS-BF sample (grey) and as the emission line fitting results for all non-QSO spectral
the BGS-AGN sample (red). One can see that the BGS-AGN types as well as for all cases with conflicting results from
sample is overall biased toward broader lines (𝜎 >100 km various classification methods (see Appendix A and B for
s−1 ) but there is no clear dichotomy in either line width distri- details and notes on individual objects). The final categories
bution. The bottom panel shows the KEx diagram consisting are as follows:
of the [O iii]/H𝛽 line flux ratio (based on the narrow H𝛽 com-
 DESI BGS AGN Target Selection 15

• QSO: visually identified QSOs from the presence of line flux ratio from the narrow lines. The underlying distri-
broad lines in the spectrum including both normal blue bution of BGS-BF galaxies (shaded grey) is characterized by
quasars and red quasars with reddened continua (red a dominant population of star-forming galaxies with a faint
circles). For plotting, we use open circles to mark two plume of Type 2 AGN (with 𝜎max <500 km s−1 ; 2.7 dex) and
cases for which the visually identified broad lines are a yet fainter plume of Type 1 AGN with 𝜎max > 1000 km
not detected by the automated fitting routines; s−1 (3 dex; gray shaded distribution under the colored sym-
bols). The BGS-AGN sample is explicitly divided according
• NLSy1 (narrow-line Seyfert 1; filled triangles): visu- to the spectral classification listed above and with identi-
ally flagged as Type 2 candidates but found to have line cal plotting symbols and demarcation line as in Figure 9.
widths in the NLSy1 range (400 <FWHM< 2000 km The modified KEx diagram allows us to compare the rela-
s−1 , which is 170 < 𝜎 < 850 km s−1 ) with [O iii]/H𝛽< tive importance of star forming galaxies, narrow-line AGN
3; and broad-line AGN at a glance, showing a more striking
• Sy2: Type 2 from the visual inspection campaign ex- difference between the BGS-AGN sample and the BGS-BF
cept for two cases relabeled based on their [O iii]/H𝛽 samples compared to the original KEx diagram. That said,
ratio and 𝜎max (one Sy2 relabeled as galaxy, and one the original still had an excellent success rate at identifying
galaxy relabeled as Sy2); AGN signatures.

• Blazar: visually classified as galaxy candidates but with

an unusual combination of weak to no emission lines,
blue continuum and prominent radio (VLASS 3 GHz) 6. PROPERTIES OF THE BGS AGN SAMPLE
detection; Now that we have checked that the majority of the BGS
AGN sample consists of quasars, we study a few properties
• Galaxy: visually classified as galaxy (except for one such as their redshift distributions, infrared colors, and AGN
relabeled case from Type 2) with either no emission luminosities. To put these properties in a broader context, we
lines or narrow emission lines consistent with stellar consider the BGS AGN sample relative to what we expected
photoionization according to the KEx diagrams. from the precursor SDSS-LS sample, and in relation to the
The KEx diagram classification is mostly consistent with the BGS-BF and to QSO samples from DESI SV3.
spectral types listed above16 with a few discrepancies at the 6.1. Redshift distribution N(z)
level of 2% (8/375) QSO or NLSy1 in the star-forming galaxy
side (filled red circles and green triangles). We tabulate the We compute the redshift distributions per unit area ex-
total number of objects in each spectral category in Table 2 pressed as 𝑁 (𝑧) in redshift slices of 0.1 as shown in Fig-
noting that the subsets with [O iii] and H𝛽 emission lines ure 11. The left-hand panel shows the expected distributions
include: 370/485 QSO, 2/2 QSO (NL), 3/3 NLSy1, 8/8 Sy2, for the SDSS-LS sample that is spectroscopically classified
2/3 blazars and 5/8 galaxies. as QSO and which meets the BGS-BF selection (filled grey),
By design due to being based on a forbidden line width, BGS-AGN selection (filled red) in comparison with the QSOs
the original KEx diagram does not distinguish between the that were removed from BGS due the 𝐺 Gaia − 𝑟< 0.6 cut but
narrow line (Type 2) and broad line QSO objects. For this could be potential DESI QSO targets (orange histogram).
purpose, we next investigate using the maximum line width The latter is scaled down by a factor of 5 to more easily com-
introduced above (Equation 16) as a substitute for 𝜎[O iii] . pare the shape of each distribution. This panel suggests that
We compare the 𝜎max distributions for the BGS-BF sample QSOs from the BGS-BF sample are mostly at low redshifts
(in gray) and the BGS-AGN sample (in red) in the top panel (𝑧 < 0.5), the DESI QSO targets are mostly at high redshifts
of Figure 10. We find that the BGS-AGN sample is strongly (𝑧 > 1) while the BGS-AGN sample somewhat bridges the
biased toward broad lines. Even among just the Type 1 AGN gap between the two with a peak around 𝑧 ∼ 0.5.
from both samples (log(𝜎max ) > 2.6), there is still a tendency The right-hand panel shows a version of the same figure
for BGS-AGN targets to have broader lines relative to the but with the DESI SV3 data for the corresponding samples
BGS-BF sample, which may indicate more massive black (same color scheme). While the trends noted above remain
holes. applicable, we note that the BGS-BF redshift distribution is
In the bottom panel of Figure 10, we show the modified KEx even more strongly limited to 𝑧 < 0.5 and reaches a much
diagram with the maximum line width versus the [O iii]/H𝛽 higher number density peaking around 7 per square degree
(shown scaled down by a factor 3). For the DESI QSO targets,
we split them between the subset that would meet the BGS
16 This is by definition for two objects out of 390 for which we used the line magnitude limits (solid orange histogram) to be directly com-
width and [O iii]/H𝛽 values to relabel parable with the subsample shown for the SDSS-LS sample.
16 S. Juneau et al.

Table 2. Spectral type classification

Spectral type N Definition AGN? Figure

QSO 485 Broad lines among Ly𝛼, C iv, Mg ii, H𝛽, H𝛼, with either a typical blue or an Yes Fig. 7
attenuated red continuum
QSO (NL) 2 Visually a QSO but the broad lines are not detected with FastSpecFit Yes Fig. 16, 20
NLSy1 3 Visually flagged as Type 2 but with moderate broad lines (𝜎max <850 km s −1 ) Yes Fig. 22
and [O iii]/H𝛽< 3
Sy2 8 AGN-like values or [O iii]/H𝛽 and 𝜎max (possibly visually flagged as Type 2) Yes Fig. 23, 24
Blazar 3 Blue continuum with weak or no emission lines and a radio counterpart in VLASS Yes Fig. 25
Galaxy 9 Stellar continuum without or with emission lines consistent with stellar photoionization No Fig. 29-31
Star 9 Best-fit by a stellar template with Redrock and either confirmed visually (N=6) or No Fig. 26-28
assigned a low confidence (N=3)
Total 519

Figure 11. Redshift distributions as predicted from the SDSS QSOs (left-hand panel), and as observed with DESI SV3 (right-hand panel) for
the QSO spectral type. In both panels, the population that fulfills the selection criteria for the BGS-BF sample is shown in gray while the
BGS-AGN sample is shown in red. On the right-hand side panel, the distribution of QSOs from the BGS-BF sample was scaled down by a
factor of three. We attribute the much larger number density to the difference in magnitude limits for SDSS and DESI. This figure truncates
the tail of the distribution at 𝑧 = 2.5 to focus on the bulk of the sample. Nine additional BGS-AGN QSOs at 2.5 < 𝑧 < 3.3 are included in
Figure 13, which shows the complete sample of 519 unique BGS AGN targets in SV3. To compare with higher redshift QSOs, we add the
redshift distribution for objects selected as DESI QSO targets in orange histograms. The solid orange histograms are QSO targets that meet the
magnitude limits of the BGS sample. On the right-hand side, we further show QSO targets fainter than BGS with a dashed orange histogram
scaled by a factor of 20 due to the high target density.

Those bright QSO targets show a very similar distribution the SDSS predictions were accurate in the sense that the BGS-
shape as the SDSS-LS sample, with the solid line orange his- AGN quasars peak around 𝑧 ∼ 0.5 making them intermediary
tograms peaking around 𝑧 ∼ 1.3 in both panels. As a further between the lower redshift BGS-BF quasars (𝑧 < 0.5) and the
comparison, we show DESI QSO targets that are fainter than higher redshift DESI QSO targets (𝑧 > 0.5).
the BGS limits with a dashed line histogram. As expected, the
6.2. AGN luminosity vs. redshift
redshift distribution of the faint QSO targets peak at a higher
redshift (around 𝑧 ∼ 1.8), and reaches a much higher target In this section, we estimate the AGN bolometric luminos-
density (the histogram was scaled down by a factor of 20). ity based on the [O iii] luminosity, which is appropriate for
Overall, modulo some quantitative normalization differences, low-redshift massive galaxies because their [O iii] flux is dom-
inated by the narrow line region gas when an AGN is present
 DESI BGS AGN Target Selection 17

Figure 12. (Left) Uncorrected [O iii] luminosity for the BGS-BF (gray) and BGS-AGN (red symbols) subsamples that have [O iii] and
H𝛽 measurements and were classified as AGN from the modified KEx diagnostic diagram (Figure 10). The dashed line corresponds to the
[O iii] luminosity for a fixed line flux of 4 × 10 −17 erg s −1 cm −2 . (Right) Similar to the left panel but showing the dust-corrected bolometric
luminosities of the AGN as a function of redshift. The horizontal dotted line marks a common threshold used to define the quasar regime with
𝐿 𝑏𝑜𝑙 > 1044 erg s −1 .

(Kauffmann et al. 2003). We follow the approach of Lamas- obscuration and this reduces the apparent offset between the
tra et al. (2009) and first apply a dust obscuration correction BGS-BF and BGS-AGN samples in their AGN bolometric
based on the H𝛼/H𝛽 Balmer Decrement assuming an intrin- luminosities (right-hand panel). The BGS-AGN bolometric
sic dust-free ratio of 3.0 (Osterbrock & Ferland 2006) and luminosity values seem to be nearly consistent with those for
the dust attenuation prescription from Bassani et al. (1999). the BGS-BF quasars at a given redshift. But because they
When the Balmer Decrement is not available, we use the typ- are located preferentially at higher redshifts, they also exhibit
ical value of 4.0 found for the BGS sample. In the small higher luminosities on average, reaching well into the quasar
number of cases with a measured ratio below 3, we assume regime at 𝐿 bol > 1044 erg s−1 (horizontal dotted line).
that the dust obscuration is negligible and do not attempt to
apply a correction. 6.3. Infrared 𝑊1 − 𝑊2 color trend
We restrict this part of the analysis to the subsample of The WISE 𝑊1 − 𝑊2 color has been suggested as a cri-
galaxies that pass the AGN criterion from the modified KEx terion to identify AGN by, e.g., Stern et al. (2012) and
diagram (above the dashed line from Figure 10). This choice Wu et al. (2012). The former study selects objects red-
is motivated to further increase the likelihood that the [O iii] der than (𝑊1 − 𝑊2)Vega ≥ 0.8, which corresponds to
luminosity is indeed dominated by the NLR gas rather than (𝑊1 − 𝑊2) 𝐴𝐵 ≥ 0.16. The latter employs a slightly
by ionized gas from star-forming regions within the 1.5- lower cut of (𝑊1 − 𝑊2)Vega ≥ 0.57, which corresponds
arcsecond fiber aperture. We show both the uncorrected to (𝑊1 − 𝑊2) 𝐴𝐵 ≥ −0.07. In this section, we compare
[O iii] luminosities (left panel) and the dust-corrected bolo- the trend of the WISE 𝑊1 − 𝑊2 color as a function of
metric luminosities (right panel) as a function of redshift in redshift for the BGS AGN sample with that of the SV3
Figure 12. There is a general increase in luminosity with sample divided into galaxies and QSOs based on the Re-
redshift due in part to the sensitivity limit for emission line drock pipeline spectral type. The comparison samples are
flux measurements (with S/N>3), which modulates the lower required to have 𝑆/𝑁 > 10 in both WISE bands, good qual-
envelope of the distribution. For reference, the dashed line ity Tractor photometry (MASKBITS=0) and reliable spectra
shows the luminosity corresponding to a constant line flux of (COADD FIBERSTATUS=0), yielding 336,067 galaxies at
4 × 10−17 erg s−1 cm−2 , which highlights that the BGS-AGN 0.025 < 𝑧 < 1.6 and 27,999 QSOs at 0.025 < 𝑧 < 3.5.
sample mostly comprises well detected emission lines. Inter- As Figure 13 illustrates, there is a different trend and locus
estingly, we note that the BGS-AGN lower envelope is higher for the bulk of galaxies (top panel; in blue) and QSOs (bottom
indicating instead a trend toward higher [O iii] luminosities panel; in orange). The BGS AGN sample (colored and black
compared to BGS-BF galaxies at the same redshifts. How- symbols) tends to follow the rest of the QSOs. However, there
ever, the BGS-AGN sample has a slightly lower average dust are galaxies scattered over the full range of WISE colors (blue
18 S. Juneau et al.

shaded bivariate distribution in the top panel) overlapping However, most DESI targets are not detected in the compar-
with the bulk of the QSOs. This overlap implies that a small atively much less sensitive 𝑊3 map so it was not used for
fraction of inactive galaxies exhibit AGN-like infrared colors, target selection. Nonetheless, future work could investigate
which could naturally explain the small galaxy contamination the infrared properties for the subset of objects that do have
rate found in the BGS-AGN sample (∼ 2% after correcting a 𝑊3 detection, which we could expect to be relevant at the
for narrow-line AGN and blazars as described in Section 5.4 low and moderate redshifts of the BGS-AGN targets.
and Appendix A).
An alternative explanation could be that some AGN host 6.4. Quasars at 𝑧 > 2
galaxies possess IR signatures of AGN heated dust (red High redshift quasars can be used as probes of the in-
𝑊1 − 𝑊2 colors) but lack optical QSO signatures due to, tervening structures along the line of sight to these bright
e.g., dust obscuration. This is consistent with at least the background sources. In particular, the DESI QSO target
Type 2 AGN (black triangles), which are characterized by class includes QSOs at 𝑧 > 2.1 that will be observed at higher
spectra with only narrow lines and lacking a blue power-law signal-to-noise in order to study the Ly𝛼 Forest (Schlafly et al.
continuum, overlapping with the QSO in terms of their in- 2023; Ramı́rez-Pérez et al. 2023). While the redshift distribu-
frared WISE colors but usually classified as a galaxy spectral tion of the BGS-AGN sample peaks at 𝑧 ∼ 0.5, there is a small
type from Redrock (although five of the eight were classi- tail reaching 𝑧 > 2, which we examine in this section. The 23
fied as QSO by QN or MgII; Appendix A.5). Yet another quasars can be divided into two strongly distinct categories.
explanation could be that the WISE photometry of some On the one hand, 11 of them are bright (𝑟 < 17.5) and were
objects is affected by nearby blended sources. Among the assigned a PSF morphological type. The spectra of those tar-
small number of spectra visually identified as galaxies with gets (Figure 14) are all blue with prominent broad emission
no obvious AGN contribution (dark blue star symbols), we lines, and further display Ly𝛼 Forest absorption lines as well
note that 5/9 are located on the edge of the distribution to- as metal absorbers such as Mg ii absorbers (e.g., Napolitano
ward low redshift and blue WISE colors, which leaves only et al. 2023). Two of them have obvious broad absorption lines
a small number of four galaxies (at 𝑧 > 0.25) with AGN- (BALs; Filbert et al. 2023).
like infrared colors. We show the spectra and color images On the other hand, the remaining 12 quasars at 𝑧 > 2 are
of all galaxy type in Appendix B and two out of the four faint and were not assigned a PSF morphological type. Their
galaxies at 𝑧 > 0.25 appear to have an additional red compo- spectra and images are displayed in Figure 15 in order of
nent on their images (TARGETID=39632961917682386 and increasing redshift. Strikingly, nearly all the images indicate
39632951826190082 in Figure 30). close angular proximity or even a possible overlap with other
By selection, BGS AGN targets are redder than 𝑊1 −𝑊2 > sources or components. We attribute the departure from PSF
−0.2 (Section 4). For reference, we draw the Stern et al. morphology to this apparent presence of companions at small
(2012) criterion with a dotted black line. At redshift 𝑧 < 1, angular separation (most likely in projection). Nine of the
we find that nearly half of BGS AGN with visually identified 12 are best-fit with a Sersic profile while two are fit with
QSO or Type 2 AGN signatures have colors bluer than the a DeVaucouleurs profile and one with a round exponential
Stern criterion. In contrast, at 𝑧 > 1 the bulk of QSOs and of profile.17
BGS AGN targets have redder colors. We also compare with Despite this trend, the spectra are all confidently classified
the 𝑊1 − 𝑊2 criteria from Wu et al. (2012) (green dashed as QSO and display the usual broad emission lines (e.g., Ly𝛼,
line) and from Wang et al. (2016) (black dot-dashed line; C iii], C iv, Mg ii). Some of them further show a Ly𝛼
𝑊1 − 𝑊2 > −0.14 AB). The latter was used together with forest, metal absorbers and/or BALs. We conclude that they
other cuts to search for high-redshift (𝑧 > 5) QSOs (e.g., are genuine quasars and that the potential chance alignment
Yang et al. 2016, 2023). They are similar to the threshold with other sources mean that they risk being excluded from
of 𝑊1 − 𝑊2 > −0.2 used in this work. In all cases, we can traditional QSO target selection algorithms.
see that a single 𝑊1 − 𝑊2 criterion would not be sufficient
to disentangle QSOs from galaxies given their overlap at 𝑧 < 7. CONCLUSIONS
0.5 and possibly also at 𝑧 > 1 but at such redshifts, DESI This manuscript presents the motivation and selection cri-
primarily selects ELGs which tend to have faint continuum teria for the BGS AGN target class of the DESI survey. This
and therefore usually not strongly detected in the WISE bands target class was introduced to supplement the BGS bright
(Raichoor et al. 2023). and faint surveys, which were missing AGN hosts that ap-
A commonly used strategy is to augment the WISE pho-
tometry with the longer wavelength channel 𝑊3 (12 𝜇m) and
use the 𝑊1 − 𝑊2 color together with the 𝑊2 − 𝑊3 color (e.g., 17 The LS DR9 morphological types are defined here:
https://www.legacysurvey.org/dr9/description/#morphological-
Jarrett et al. 2011; Mateos et al. 2012; Hviding et al. 2022).
classification
 DESI BGS AGN Target Selection 19

Figure 13. Observed 𝑊1 − 𝑊2 color as a function of redshift for the BGS AGN sample (colored symbols) compared to SV3 galaxies (top panel)
and SV3 QSOs (bottom panel). The BGS AGN symbols are placed at the location of the VI redshift and are coded to indicate the final spectral
classification: stars in light blue star symbols, galaxies with no signatures of AGN in dark blue star symbols, blazars in filled blue circles,
galaxies with Type 2 AGN signatures in black open triangles, NLSy1 in filled black triangles, and QSOs in red filled circles. The majority of
the BGS AGN sample comprises QSOs at 0.1 < 𝑧 < 1 and have WISE colors consistent with QSOs from the overall SV3 sample. Horizontal
lines show color cuts from the literature: Stern et al. (2012, dotted black line), Wu et al. (2012, green dashed line) and Wang et al. (2016, black
dash-dotted line).

pear point-like according to the Gaia and Tractor photometry QuasarNet and MgII afterburner classifiers raises this
comparison. The main conclusions are as follows: fraction to 94%. In all cases, the stellar contamination
remains around 2%, indicating that the missed quasars
• To select the BGS AGN while keeping the stellar con-
were instead assigned to a galaxy spectral type;
tamination low, we employ optical and infrared color
cuts, namely infrared colors typically expected from
AGN-heated dust. The DESI SV3 sample comprises • Stacking the spectra of BGS AGN results in a typical
782 spectra of 519 unique BGS AGN target; QSO spectrum. Additionally, splitting the subset iden-
tified (missed) by Redrock result in stacked spectra that
• The resulting target number density of 3 − 4 per square
are typical of (redder than) regular blue quasars;
degree is modest but uniform across the full DESI foot-
print, as expected;
• Applying an emission-line diagnostic diagram modi-
• Visual inspection of the BGS AGN spectra revealed
fied from the KEx diagram shows that BGS AGN tar-
a high success rate in terms of quasar classification
gets lacking broad lines (Type 1 AGN) were typically
(94%) with small percentages of narrow-line AGN or
found to exhibit emission line properties consistent with
Blazars (3%). galaxy contaminants (2%) and stellar
narrow-line (Type 2) AGN, confirming that the majority
contaminants (2%);
host an active black hole. Furthermore, this new ver-
• Using the default Redrock pipeline along only identi- sion of the KEx diagram (Section 5.4) could become a
fied 76% of the targets as QSO, whereas adding the useful tool for other AGN and galaxy studies;
20 S. Juneau et al.

Figure 14. Observed frame DESI spectra of 𝑧 > 2 QSOs with a Tractor PSF morphology (left-hand panels). The observed spectrum is in dark
gray; the error spectrum is shown in yellow, and is comparatively negligible in the case of these bright QSOs. The high average signal-to-noise
ratio allows us to display the original spectra without any smoothing thus maintaining narrow absorption features. Color images come from LS
DR9 (middle panels) and either HSC DR2 if available or otherwise LS DR10 (right-hand panels). The 1.5 arcsec DESI fiber aperture is marked
with a circle on the LS DR9 images. QSOs are displayed in order of increasing redshifts from top to bottom.
DESI BGS AGN Target Selection 21

Figure 14. Continued.

22 S. Juneau et al.

Figure 15. Observed frame DESI spectra of 𝑧 > 2 QSOs with a Tractor non-PSF morphology (left-hand panels). The observed spectrum is in
gray, a Gaussian-smoothed spectrum in black and the error spectrum is in yellow. Color images come from LS DR9 (middle panels) and either
HSC DR2 if available or otherwise LS DR10 (right-hand panels). The 1.5 arcsec DESI fiber aperture is marked with a circle on the LS DR9
images. QSOs are displayed in order of increasing redshifts from top to bottom.
DESI BGS AGN Target Selection 23

Figure 15. Continued.

24 S. Juneau et al.

• We computed the redshift distribution of the BGS AGN with the National Science Foundation. DMA thanks the
targets classified as QSOs and found it to be peak Science Technology Facilities Council (STFC) for support
around 𝑧 ∼ 0.5, which is intermediate between BGS from the Durham consolidated grant (ST/T000244/1). RP ac-
QSOs from the bright and faint survey (𝑧 < 0.5), and knowledges support from the University of Arizona and Astro
that of the QSO targets 𝑧 > 0.5. The redshift ranges are Data Lab (NSF NOIRLab, AURA). VAF acknowledges fund-
similar to our expectations from the SDSS test sample ing from an United Kingdom Research and Innovation grant
(Fig. 11); (code: MR/V022830/1). ADM was supported by the U.S. De-
partment of Energy, Office of Science, Office of High Energy
• There is an interesting high-redshift tail of 𝑧 > 2
Physics, under Award Number DE-SC0019022. JM grate-
quasars (Ly𝛼 QSOs), which are included either be-
fully acknowledges funding support for this work from the
cause they are among the brightest (a PSF morphology
U.S. Department of Energy, Office of Science, Office of High
is allowed only at 𝑟 < 17.5 for BGS-AGN targets) or
Energy Physics under Award Number DE-SC0020086. SMC
because the had a non-PSF morphology due to close
acknowledges the support of STFC grant ST/X001075/1.
projected neighbors (e.g., foreground star, galaxy, or
This material is based upon work supported by the U.S. De-
possible background lensed galaxies). We feature other
partment of Energy (DOE), Office of Science, Office of High-
rare AGN subcategories and conflicting cases in Ap-
Energy Physics, under Contract No. DE–AC02–05CH11231,
pendix A.
and by the National Energy Research Scientific Comput-
Looking forward, the first year data of the main DESI sur- ing Center, a DOE Office of Science User Facility under
vey (Year 1 observations obtained until June 2022) already the same contract. Additional support for DESI was pro-
yield 18,302 good quality spectra of 15,523 unique BGS- vided by the U.S. National Science Foundation (NSF), Di-
AGN targets. This corresponds to a sample size increase by vision of Astronomical Sciences under Contract No. AST-
a factor 30 relative to the 519 unique BGS-AGN targets from 0950945 to the NSF’s National Optical-Infrared Astronomy
DESI SV3 presented in this work. They have a very similar Research Laboratory; the Science and Technology Facili-
redshift distribution as shown in Figure 11 with a marked ties Council of the United Kingdom; the Gordon and Betty
peak at 𝑧 ∼ 0.5 and a tail reaching 𝑧 > 2 − 2.5 so we expect Moore Foundation; the Heising-Simons Foundation; the
much better statistics of the same types of interesting ob- French Alternative Energies and Atomic Energy Commission
jects that bridge the gap between the lowest redshift quasars (CEA); the National Council of Science and Technology of
and the bulk of the DESI (or even SDSS) QSO targets. The Mexico (CONACYT); the Ministry of Science and Innova-
DESI Year 1 data will be published as part of Data Release 1 tion of Spain (MICINN), and by the DESI Member Insti-
similarly to the currently public DESI EDR.18 tutions: https://www.desi.lbl.gov/collaborating-institutions.
Any opinions, findings, and conclusions or recommendations
DATA AVAILABILITY
expressed in this material are those of the author(s) and do
The Data Release 9 of the DESI Legacy Imaging Surveys not necessarily reflect the views of the U. S. National Science
is available at https://www.legacysurvey.org/dr9/. Foundation, the U. S. Department of Energy, or any of the
Documentation of DESI data access is maintained at https: listed funding agencies.
//data.desi.lbl.gov/doc/access/. DESI EDR data can also be The authors are honored to be permitted to conduct scien-
accessed via searchable catalog databases at the Astro Data tific research on Iolkam Du’ag (Kitt Peak), a mountain with
Lab (Fitzpatrick et al. 2014; Nikutta et al. 2020) at https: particular significance to the Tohono O’odham Nation.
//datalab.noirlab.edu and via a spectral database SPARCL The DESI Legacy Imaging Surveys consist of three indi-
(Juneau et al. 2024) at https://astrosparcl.datalab.noirlab.edu/, vidual and complementary projects: the Dark Energy Cam-
both of which were used in this work. era Legacy Survey (DECaLS), the Beijing-Arizona Sky Sur-
Besides the images and spectra available above, the data vey (BASS), and the Mayall z-band Legacy Survey (MzLS).
shown in the figures are available at https://doi.org/10.xxxxx/ DECaLS, BASS and MzLS together include data obtained,
zenodo.xxxxxxxx (URL will be updated upon manuscript respectively, at the Blanco telescope, Cerro Tololo Inter-
acceptance). American Observatory, NSF’s NOIRLab; the Bok telescope,
Steward Observatory, University of Arizona; and the May-
ACKNOWLEDGEMENTS
all telescope, Kitt Peak National Observatory, NOIRLab.
SJ’s research is supported by the U.S. NSF NOIRLab, NOIRLab is operated by the Association of Universities for
which is operated by the Association of Universities for Re- Research in Astronomy (AURA) under a cooperative agree-
search in Astronomy (AURA) under a cooperative agreement ment with the National Science Foundation. Pipeline pro-
cessing and analyses of the data were supported by NOIRLab
18 https://data.desi.lbl.gov/doc/releases/ and the Lawrence Berkeley National Laboratory (LBNL).
 DESI BGS AGN Target Selection 25

Figure 16. Rest-frame DESI spectrum of a LoBAL quasar. The unsmoothed spectrum is in gray with a smoothed spectrum overplotted in
black. The error spectrum is shown in yellow. Color images come from LS DR9 (middle panels) and either HSC DR2 or LS DR10 (right-hand
panels). The 1.5 arcsec DESI fiber aperture is marked with a circle on the LS DR9 images.

Legacy Surveys also uses data products from the Near-Earth Energy. The complete acknowledgments can be found at
Object Wide-field Infrared Survey Explorer (NEOWISE), https://www.legacysurvey.org/acknowledgment/
a project of the Jet Propulsion Laboratory/California In- This research uses services and data provided by the SPec-
stitute of Technology, funded by the National Aeronautics tra Analysis and Retrievable Catalog Lab (SPARCL) and the
and Space Administration. Legacy Surveys was supported Astro Data Lab, which are both part of the Community Sci-
by: the Director, Office of Science, Office of High Energy ence and Data Center (CSDC) program at NSF NOIRLab.
Physics of the U.S. Department of Energy; the National En-
ergy Research Scientific Computing Center, a DOE Office
of Science User Facility; the U.S. National Science Founda- Facilities: Mayall (DESI), Mayall (Mosaic-3), Blanco
tion, Division of Astronomical Sciences; the National As- (DECam), Bok (90Prime), Astro Data Lab, Sloan, WISE,
tronomical Observatories of China, the Chinese Academy of NEOWISE, Gaia
Sciences and the Chinese National Natural Science Foun- Software: astropy (Astropy Collaboration et al. 2013,
dation. LBNL is managed by the Regents of the Univer- 2018, 2022), desispec (Guy et al. 2023), desitarget (Myers
sity of California under contract to the U.S. Department of et al. 2023) FastSpecFit (Moustakas 2023), Redrock (Bai-
ley et al. 2023), SPARCL (Juneau et al. 2024)

APPENDIX

A. REFINED CLASSIFICATION FROM EMISSION LINES

As described in Section 5.1, the spectra were classified based on different methods including a campaign of visual inspection
(Section 5.1.1) and applying emission line diagnostic diagrams (Section 5.4). In this Appendix, we present additional details that
led to the adopted spectral classification and display individual spectra and images to illustrate each category. In particular, we
examine the spectra of all cases that were subject to a conflicting classification between various methods, and we highlight rare
AGN subpopulations such as red quasars (Section A.3), NLSy1 (Section A.4) and blazars (Section A.6).
A.1. QSO missed by the modified QSO classifier pipeline
In Section 5.1.3, we describe the classification results from the modified QSO pipeline that combines Redrock with the
afterburner QN and MgII classifiers. There were five conflicting cases not classified as QSO by the pipeline but classified as
such visually. One striking case is a 𝑧 ∼ 1.51 QSO with prominent low-ionization BALs in the spectrum as shown in Figure 16.
The original redshift as determined by Redrock was inaccurate likely due to the BAL features. Even though it is challenging to
determine a precise redshift for such a spectrum, the visual inspection classification is secure in terms of assigning a QSO spectral
type.
As shown in Figure 17 the other four cases are a combination of faint spectra with low signal-to-noise, comparatively weak and/or
narrow broad lines. The object in the top row (TARGETID=39633145330403453) has a maximum line width of 𝜎max =362 km s−1 ,
which is in the NLSy1 regime (𝜎 = 170−850 km s−1 ) but it has [O iii]/H𝛽=5.3, which is higher than the formal definition for NLSy1
([O iii]/H𝛽< 3). The second and third row show cases with weak broad line and we note that TARGETID=3962789086993068
may be a red QSO (shown in the third row;) also see Section A.3).
Lastly, we note that TARGETID=39627776260576636 (bottom row) has tentative broad Mg ii and H𝛽 lines in its spectrum
with otherwise a low signal to noise ratio. This target has a higher quality spectrum with more exposure time in DESI DR1
26 S. Juneau et al.

Figure 17. Rest-frame DESI spectra of visually identified QSOs that were missed by the afterburner pipeline. The top two examples may be
NLSy1 given that their broad lines appear fairly narrow. The spectrum in the third row is reddened in addition to having relatively faint (low
signal-to-noise) features. The spectrum in the bottom row has a blue continuum with faint broad Mg ii and H𝛽 lines which are apparent when
interactively examining the spectrum but have a low significance in the unsmoothed spectrum. In all rows, the unsmoothed spectrum is in gray
with a smoothed spectrum overplotted in black. The error spectrum is shown in yellow. Color images come from LS DR9 (middle panels) and
either HSC DR2 or LS DR10 (right-hand panels). The 1.5 arcsec DESI fiber aperture is marked with a circle on the LS DR9 images.

(in preparation), which shows the lines more clearly and that spectrum is assigned a QSO spectral type by the default Redrock
pipeline. Therefore, this gives us confidence in the visual classification. Overall, we keep the QSO visual classification for all
five targets shown in this Section but we understand that these cases are challenging to identify with an automated pipeline due
to BALs or faint signal. Still, these missed QSOs are only at the 1% level for the BGS-AGN sample (5/519) so we also conclude
that the modified pipeline performs adequately.

A.2. Broad line identification

A clear signature of typical QSOs is the presence of broad permitted lines in their spectra and those were the main features
identified visually. In the rest-frame optical, the main broad features are the Balmer series of Hydrogen. In the rest-frame UV,
the main features are Mg ii, C iii], C iv, and Ly𝛼. In this section, we discuss the quantitative identification of broad lines from
automated spectral fitting, and our motivation for combining both UV and optical lines when adopting the maximum line width
𝜎max from the Mg ii, H𝛽, and H𝛼 lines in Section 5.4.
 DESI BGS AGN Target Selection 27

Figure 18. Maximum line width as a function of redshift. The top panel is the result when considering only the Balmer lines H𝛼 and H𝛽, while
the bottom panel shows the result after adding the MgII doublet. In both panels, the points are color-coded with the original VI classification as
labeled. Furthermore, the plotting symbols are either filled or open depending on whether a broad line was detected (Type 1) or not (Type 2),
respectively.

In the FastSpecFit emission line fitting algorithm, the H𝛽 and H𝛼 Balmer line profiles are tied together in terms of their
width and position. However, the presence of a broad component is comparatively more challenging to detect for H𝛽 due to being
> 3 times fainter line than H𝛼. Consequently, when relying on the Balmer lines alone, there is a marked decrease of cases with
detected broad lines starting at 𝑧 > 0.45 where H𝛼 is redshifted outside of the spectral range (Figure 18a). At 𝑧 < 0.45 where both
H𝛼 and H𝛽 are available, most visually identified QSOs are also assigned a broad line (solid red circles). However, at 𝑧 > 0.45,
there are many visually classified QSOs that lack a significant broad Balmer component (open red circles).
Luckily, the Mg ii doublet becomes available within the DESI spectral range at 𝑧 > 0.3 and is characterized by an easily
detectable broad profile in Type 1 AGN. Therefore, combining the line width information from H𝛼, H𝛽, and Mg ii allow us to
classify most of the BGS-AGN samples with emission lines as broad-line QSOs (solid red circles in Figure 18b).
Among spectra visually classified as QSO and with H𝛽 and [O iii] lines, there were originally 14 for which the emission line
measurements from FastSpecFit failed to detect a broad line with S/N>3 in any of H𝛽, H𝛼, Mg ii. Inspecting the emission-line
fitting results of those 14 spectra revealed the following:

• One case failed due to the extreme broad line being outside the normal prior range (top panel of Figure 19). The solution
was to refit the spectrum with a new prior, which converged with 𝜎max =5051.5 km s−1 (and 𝜎[O iii] =188.5 km s−1 ). In
detail, the broad line profile deviates from a single Gaussian but a more complex characterization is beyond the scope of
this work.
28 S. Juneau et al.

Figure 19. Rest-frame DESI spectra for which we obtain a new fit with FastSpecFit for the following reasons: (top) very broad H𝛼 line
which required different priors. Noticeably, the H𝛽 broad line is much weaker and harder to visually disentangle from the continuum; (bottom)
originally had an erroneous redshift but here we can see obvious Mg ii emission as well as a broad H𝛽. The unsmoothed spectrum is in gray
with a smoothed spectrum overplotted in black. The error spectrum is shown in yellow. Color images come from LS DR9 (middle panels) and
either HSC DR2 or LS DR10 (right-hand panels). The 1.5 arcsec DESI fiber aperture is marked with a circle on the LS DR9 images.

Figure 20. Rest-frame DESI spectrum for a low signal-to-noise case which fails our S/N>3 criterion due to FastSpecFit finding S/N=2.8 for
the broad H𝛼 component. The unsmoothed spectrum is in gray with a smoothed spectrum overplotted in black. The error spectrum is shown in
yellow. Color images come from LS DR9 (middle panel) and LS DR10 (right-hand panel). The 1.5 arcsec DESI fiber aperture is marked with
a circle on the LS DR9 image.

• Two cases had an inaccurate redshift from the Redrock pipeline. The solution was to refit them with the corrected redshifts,
and we show the case within our redshift range of interest in the bottom panel of Figure 19.

• Nine cases have broad lines with low signal-to-noise and thus are sensitive to small variations of the smooth continuum
component of the FastSpecFit model. The solution was to refit these nine without a smooth continuum correction, which
successfully recovered at least one of broad lines for those objects.

• The last two were not recovered when fitting without a smooth continuum contribution. One has a broad H𝛼 line that could
be measured at a significance of S/N=2.8 (𝜎max =2555.6 km s−1 ), which fails the formal requirement S/N>3. The spectrum
is shown in Figure 20. The last case also has a low signal-to-noise per pixel and due to having a redshift at 𝑧 > 0.45, it lacks
coverage of H𝛼 and fails to be detected in other broad lines (bottom panel of Figure 17). Visually, one can spot tentative
broad Mg ii and H𝛽 lines and the blue continuum is typical of quasars and Type 1 AGN. Furthermore, a deeper spectrum
to be published with DESI DR1 confirms its QSO spectral type (not shown). Thus, we preserve the QSO spectral type for
these two low-S/N spectra but they are plotted with open red circles in Figures 9, 10 and 18.
 DESI BGS AGN Target Selection 29

Figure 21. Rest-frame DESI spectra of two red QSOs. The unsmoothed spectrum is in gray with a smoothed spectrum overplotted in black.
The error spectrum is shown in yellow. Both spectra present emission lines such as Mg ii, H𝛽, and [O iii] as well as stellar absorption features
along a reddened continuum. Color images come from LS DR9 (middle panels) and HSC DR2 (right-hand panels). The 1.5 arcsec DESI fiber
aperture is marked with a circle on the LS DR9 images.

A.3. Red QSOs

The majority of normal quasars are characterized by blue continuum and colors, which is attributed to emission from the
accretion disk. We indeed found a typical blue quasar spectrum when averaging the spectra in Section 5.3. However, dust
attenuation can impact these colors and lead to red quasars (e.g., Richards et al. 2003; Glikman et al. 2004) or even extremely
red quasars (e.g., Ross et al. 2015; Hamann et al. 2017). These may in turn correspond to special phases of galaxy evolution
such as the blow out phase following a gas-rich major merger (Sanders et al. 1988), according to which dusty gas-rich material is
entrained thereby obscuring the central AGN with a foreground dust screen relative to the line of sight.
We show two example spectra of red quasars in Figure 21. The top panel shows a spectrum which was classified by Redrock as
a galaxy but had a Type 2 label from visual inspection. However, the Mg ii line is clearly a broad line so we relabel it as a QSO
despite the red continuum. The bottom panel shows TARGETID=39633145284267618, which was also classified by Redrock as
a galaxy but had a “Type 1?” flag from visual inspection. A closer look reveal the presence of broad but faint lines like Mg ii
and a broad component to H𝛽. The DESI survey now has a better quality spectrum for this object for DR1 (not shown) that the
Redrock pipeline identifies as a QSO, and the presence of these broad lines is confirmed with a high confidence.
There are a few other candidate red QSOs in other figures (e.g., Figures 17, 20). However, we do not attempt to systematically
differentiate between blue and red quasars in this work and we consider them all as QSOs when computing statistics. DESI
includes a secondary target program dedicated to observing red quasars (Fawcett et al. 2023), which is finding both red and
extremely red cases. Fawcett et al. (2023) created high quality stacked spectra showing similar colors as the examples reported
here (their figure 12).

A.4. Narrow-line Seyfert 1

Three objects that were labeled as Type 2 AGN from VI were found to have a broad H𝛽 or H𝛼 component by FastSpecFit
(filled green triangles in Figures 9 and 18). Their spectra resemble that of Narrow Line Seyfert 1’s (NLSy1) and are displayed in
Figure 22. These spectra are distinct from both regular broad-line Seyfert 1’s and Seyfert 2’s but we had not defined a separate
NLSy1 category during the visual inspection campaign. Early work on NLSy1’s reported H𝛽 line widths in the range 400-1000 km
s−1 (Osterbrock & Pogge 1985), and the commonly used definition is 400 <FWHM(H𝛽)< 2000 km s−1 (which corresponds to
a Gaussian line width 170 < 𝜎 < 850 km s−1 ) with [O iii]/H𝛽< 3 (Osterbrock & Dahari 1983; Goodrich 1989). More recent
studies have also used Mg ii with the same line width threshold to expand the range of redshift for NLSy1 search (e.g., Rakshit
et al. 2021). The three spectra below are consistent with these definitions, and we note that one of them was also identified as a
QSO by the QN classifier (TARGETID=39633413522588456; middle row of Figure 22).
30 S. Juneau et al.

Figure 22. Rest-frame DESI spectra of three candidate NLSy1 galaxies. They were visually identified as possibly being Type 2 AGN but the
line width fitting found a line to the NLSy1 range (170 < 𝜎 < 850 kms). The unsmoothed spectrum is in gray with a smoothed spectrum
overplotted in black. The error spectrum is shown in yellow. All spectra present clear emission lines as well as stellar continuum, and noticeably
the H𝛼 line at 6563 Å is partly blended with the [N ii]𝜆6584 doublet. The bottom two spectra cover Mg ii (2800 Å), which also appear consistent
with a NLSy1 classification. Color images come from LS DR9 (middle panels) and either HSC DR2 or LS DR10 (right-hand panels). The
1.5 arcsec DESI fiber aperture is marked with a circle on the LS DR9 images.

In detail, recent work found that careful spectral fitting might require a third component resulting in a narrow, intermediate
and broad component (Mullaney & Ward 2008). While there are only a few examples shown here from the BGS-AGN survey
validation sample, the full DESI survey is likely well suited to study this subpopulation. NLSy1’s may be at an interesting
evolutionary phase such as young AGN that have recently switched on and are accreting with high Eddington rates (see, e.g.,
review by Komossa 2008).

A.5. Seyfert 2
We use the label Sy2 to refer to Type 2 AGN and QSO candidates and meaning that only narrow lines are visible in their
spectra. The VI campaign originally assigned a Type 2 label to 12 spectra. One of them was relabeled as galaxy while in the
reverse direction one galaxy was relabeled as Type 2 due, in both cases, to their [O iii]/H𝛽 ratio and line width as measured by
FastSpecFit. In addition, three were found to have a narrow broad line consistent with a NLSy1 category introduced above
(Section A.4, bringing the number down to nine. Lastly, one of those was relabeled as a red QSO upon further examination
showing a broad but faint line together with a reddened continuum (top row of Figure 21). So this results in a final number of
eight Sy2.
First, we show 5/8 that were classified as QSO from the QN or MgII classifiers in Figure 23. Those five spectra all have emission
lines with either a red or flat continuum. Some of them display an [Ne iv]𝜆3425 emission line, which is a clear evidence of AGN
due to its high ionization potential of 97 eV. However, we cannot visually rule out whether some of these spectra are host-reddened
Type 1 AGN (i.e., red QSOs) rather than regular Type 2 AGN attributed to small-scale obscuration of the BLR according to the
AGN unification model (e.g., Antonucci 1993; Urry & Padovani 1995, but also see Ramos Almeida et al. (2011)). We count
them as narrow line AGN in this work, but we remind that the modified QSO classification pipeline based on the QN and MgII
 DESI BGS AGN Target Selection 31

Figure 23. Rest-frame DESI spectra visually classified as Seyfert 2 but which were classified as QSO by either the QN or MgII classifier. The
unsmoothed spectrum is in gray with a smoothed spectrum overplotted in black. The error spectrum is shown in yellow. Color images come
from LS DR9 (middle panels) and either HSC DR2 if available or otherwise LS DR10 (right-hand panels). The 1.5 arcsec DESI fiber aperture
is marked with a circle on the LS DR9 images.

classifiers assigns them a QSO spectral type. If we switched those five as well as the three NLSy1, our QSO success rate would
reach (495/519) = 95.4%.
Last, we show the remaining 3/8 that were not classified as QSO by the afterburners in Figure 24. While they all very strong
narrow emission lines, none of them have a [Ne iv]𝜆3425 detection. They could concurrently host a starburst and/or have
low-metallicity gas similarly to Green Pea galaxies (Cardamone et al. 2009), some of which have been found to host an AGN
(Harish et al. 2023).
32 S. Juneau et al.

Figure 24. Rest-frame DESI spectra of three candidate Seyfert 2, which were classified as a galaxy spectral type by Redrock and the QN and
MgII classifiers. The unsmoothed spectrum is in gray with a smoothed spectrum overplotted in black. The error spectrum is shown in yellow.
The expected locations of a subset of emission lines are marked with vertical dotted lines. All spectra present strong emission lines but no [Ne
V] emission and the Mg II doublet is detected once in emission (middle row) and once in absorption (bottom row). Color images come from
LS DR9 (middle panels) and either HSC DR2 if available or otherwise LS DR10 (right-hand panels). The 1.5 arcsec DESI fiber aperture is
marked with a circle on the LS DR9 images.

A.6. Blazars
Blazars are a class including BL Lac objects (with a nearly featureless spectra) and compact radio sources with a flat spectrum
thought to arise when the observer is looking toward the radio jet (Angel & Stockman 1980; Antonucci & Ulvestad 1985). This
chance alignment makes this a rare class of objects, and is usually associated with high variability. Plotkin et al. (2008) classify
optical properties based on SDSS spectra and found a range of spectral shapes and features. The cases shown in Figure 25
resemble the BL Lac in being nearly featureless except that host galaxy stellar absorption lines are noticeable (vertical dashed
red lines). Those three spectra were assigned a galaxy spectral type by Redrock and by the visual inspection campaign. A closer
look showed an unexpected blue rise of the continuum given the otherwise near absence of emission lines so we searched for
the presence of radio emission from the VLASS2.1 which covers 3 GHz and all three have obvious and fairly compact emission,
consistent with a blazar classification. None of the remaining nine galaxy spectral types show such radio emission. We thus
classify those three as the only three blazar candidates but did not systematically search for radio emission in the QSO spectra
even though previous work has reported some blazars with broad optical lines (e.g., Marcha et al. 1996; Shaw et al. 2012).
 DESI BGS AGN Target Selection 33

Figure 25. Rest-frame DESI spectra of three candidate Blazars, which were identified by Redrock and visually as having a GALAXY spectral
type. The unsmoothed spectrum is in gray with a smoothed spectrum overplotted in black. The error spectrum is shown in yellow. All spectra
present weak to no emission lines and have noticeable absorption features as labeled by the vertical dashed lines. Color images come from LS
DR9 (middle panels) and radio maps from VLASS 1.2 (right-hand panels). The 1.5 arcsec DESI fiber aperture is marked with a circle on the
LS DR9 images.
34 S. Juneau et al.

B. STAR AND GALAXY CONTAMINATION

As mentioned in Section 5, there were small percentages of stellar and galaxy contaminants in the BGS AGN sample. In this
Appendix, we show example spectra and color images for those objects.

B.1. Stars
We split the nine stars based on their photometric morphology from the Tractor fitting and according to the visual inspection
confidence. Figure 26 shows the four stars with a point source (PSF) morphological type. They are all bright by definition (PSF
was only allowed for 𝑟 < 17.5) and were all assigned a high confidence.
Figure 27 shows the two stellar contaminants with a non-PSF morphological type and a high redshift confidence rating. In both
cases, there are possibly two (or more) sources in the optical images with different colors. The top row shows a blue spectrum and
target and while there is possibly a second object that contributes to the spectrum, no secondary redshift identification was made.
Zooming in the spectrum interactively suggests subtle absorption lines at the expected location of the Balmer lines but there are no
obvious strong features seen in the noisy spectrum. In contrast, the second row shows a red stellar spectrum with clear absorption
lines and bands. The image shows a central orange object (star) but the nature of the redder extended feature is unknown. Potential
contributions from more than one target to the resulting spectra have not been ruled out but no secondary redshifts were visually
flagged and a more detailed analysis is beyond the scope of this work. The bottom line is that we suspect that these stellar
contaminants were included in the BGS-AGN sample based on the unusual combination of colors and morphology affected by
blending with overlapping neighbors in angular distance, which may be projection effects and not necessarily associated pairs or
multiples.
Lastly, we show the three cases with non-PSF morphological types and with a low confidence redshifts, which also tend to
correspond to a low-confidence classification due to poorer spectral quality (Figure 28). They tend to be faint and the images
show either a possible extended component (in 2/3 cases) or a close projected neighbor with a different color (in 1/3 case). In the
two cases with possible extended components, the spectra are noisy and highly uncertain and the objects appear unlikely to be
stars due to their morphologies. Therefore, no confident redshift or classification could be obtained for them. The third case with
a close projected neighbor may show H𝛼 absorption line consistent with a near zero redshift but remains overall a low confidence
case. Luckily, these uncertain cases represent only 0.6% (3/519) of the sample.

B.2. Galaxies
In this section we display spectra and images of the nine BGS-AGN targets that were visually classified as galaxies and without
any AGN subcategory such as Type 2 or blazar candidates. There remains a lot of variety among both the spectra and the color
images, ranging from star-forming galaxies with blue images and strong narrow emission lines to redder passive galaxies with
spectra dominated by stellar continuum and stellar absorption lines. First, we create a montage for the three bluest galaxies
characterized by strong emission lines (Figure 29). In addition to the obvious emission lines, all three spectra exhibit stellar
absorption features which are especially noticeable in the 3700-4000 Å region (H&K lines, Balmer absorption, etc.). These
features suggest that the host galaxies are fairly massive and may include a range of stellar population ages in addition to ongoing
star formation or starbursting activity.
Next, we display five galaxies with weak or no detectable emission lines in order of increasing redshift (Figure 30). The color
images of the first case indicates a clear overlap with a background galaxy, and it is not clear whether the low redshift of 𝑧 = 0.087
applies to the intended target. The remaining four galaxies in Figure 30 tend to show variations in their color images suggesting
possible multiple components (associated or only in projection). The spectra still tend to show a blue continuum but the weakness
of the emission lines are generally consistent with either no or very weak AGN. Additionally, we verified that there is no VLASS
detection as this could indicate radio AGN like the Blazar candidates (Figure 25).
Lastly, we show the rest-frame spectrum and color images of the single VI-identified galaxy with a redshift 𝑧 > 1 in Figure 31.
The spectrum is characterized by a faint blue continuum with a clear Mg ii doublet absorption feature (∼ 2800 Å) and an obvious
[O ii]𝜆𝜆3726,3729 emission feature though the doublet components are not well resolved. The galaxy seems marginally resolved
in the images with the LS DR9 image showing bright and blue central emission. Given the lack of AGN signatures in the spectrum,
we assume that this emission is due to star forming or starbursting regions in the host galaxy.
 DESI BGS AGN Target Selection 35

Figure 26. Observed frame DESI spectra of the four stellar contaminants with a PSF morphology (left-hand panels). The observed spectrum is
in gray with a smoothed spectrum overplotted in black. The error spectrum is shown in yellow, and is comparatively negligible in the case of
the four bright targets shown here. Color images come from LS DR9 (middle panels) and either HSC DR2 if available or otherwise LS DR10
(right-hand panels). The 1.5 arcsec DESI fiber aperture is marked with a circle on the LS DR9 images.
36 S. Juneau et al.

Figure 27. Similar to Figure 26 but for the observed frame DESI spectra of the two stellar contaminants with a non-PSF morphology and with
a high confidence redshift quality (left-hand panels). Color images are shown from LS DR9 (middle panels) and either HSC DR2 if available
or otherwise LS DR10 (right-hand panels). The images suggest possible two or more blended objects with distinct colors.

Figure 28. Observed frame DESI spectra of the three candidate stellar contaminants with a non-PSF morphology and a low confidence in the
redshift and/or classification (left-hand panels). Color images are shown from LS DR9 (middle panels) and either HSC DR2 if available or
otherwise LS DR10 (right-hand panels). Some images suggest two or more blended objects while the top and third row appear to have a possible
diffuse component but with faint spectra leading to a low confidence visual classification.
 DESI BGS AGN Target Selection 37

Figure 29. Rest-frame DESI spectra visually classified as galaxies. The three cases shown here share common characteristics such as strong
emission lines and overall blue continuum and colors. The observed spectrum is in gray, a Gaussian-smoothed spectrum in black and the error
spectrum is in yellow. Color images come from LS DR9 (middle panels) and HSC DR2 (right-hand panels). The 1.5 arcsec DESI fiber aperture
is marked with a circle on the LS DR9 images.
38 S. Juneau et al.

Figure 30. Rest-frame DESI spectra visually classified as galaxies. The three cases shown here share common characteristics such as strong
emission lines and overall blue continuum and colors. The observed spectrum is in gray, a Gaussian-smoothed spectrum in black and the error
spectrum is in yellow. Color images come from LS DR9 (middle panels) and either HSC DR2 if available or otherwise LS DR10 (right-hand
panels). The 1.5 arcsec DESI fiber aperture is marked with a circle on the LS DR9 images. Galaxies are displayed in order of increasing
redshifts from top to bottom.
 DESI BGS AGN Target Selection 39

Figure 31. Rest-frame spectrum of the only 𝑧 > 1 target visually classified as a galaxy (TARGETID=39627927016442968). The spectrum is
shown both unsmoothed (grey) and smoothed with a Gaussian kernel (𝜎 = 3; black line). It is characterized by a faint blue continuum, a MgII
doublet absorption feature (2800 Å), and the [O ii] doublet emission line (3727 Å). Color images are shown for the LS DR9 and LS DR10 as
labeled and measure 10 arcseconds on a side. The DESI 1.5 arcsecond fiber aperture is marked with a circle.

REFERENCES
Ahumada, R., Allende Prieto, C., Almeida, A., et al. 2020, ApJS, DESI Collaboration, Adame, A. G., Aguilar, J., et al. 2023a, arXiv
249, 3, doi: 10.3847/1538-4365/ab929e e-prints, arXiv:2306.06307, doi: 10.48550/arXiv.2306.06307
Alexander, D. M., & Hickox, R. C. 2012, NewAR, 56, 93, —. 2023b, arXiv e-prints, arXiv:2306.06308,
doi: 10.1016/j.newar.2011.11.003 doi: 10.48550/arXiv.2306.06308
Alexander, D. M., Davis, T. M., Chaussidon, E., et al. 2023, AJ, Dey, A., Schlegel, D. J., Lang, D., et al. 2019, AJ, 157, 168,
165, 124, doi: 10.3847/1538-3881/acacfc doi: 10.3847/1538-3881/ab089d
Angel, J. R. P., & Stockman, H. S. 1980, ARA&A, 18, 321, Elvis, M., Wilkes, B. J., McDowell, J. C., et al. 1994, ApJS, 95, 1,
doi: 10.1146/annurev.aa.18.090180.001541 doi: 10.1086/192093
Antonucci, R. 1993, ARA&A, 31, 473, Fabian, A. C. 2012, ARA&A, 50, 455,
doi: 10.1146/annurev.aa.31.090193.002353 doi: 10.1146/annurev-astro-081811-125521
Antonucci, R. R. J., & Ulvestad, J. S. 1985, ApJ, 294, 158, Farr, J., Font-Ribera, A., & Pontzen, A. 2020, JCAP, 2020, 015,
doi: 10.1086/163284 doi: 10.1088/1475-7516/2020/11/015
Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, Fawcett, V. A., Alexander, D. M., Brodzeller, A., et al. 2023,
MNRAS, 525, 5575, doi: 10.1093/mnras/stad2603
A&A, 558, A33, doi: 10.1051/0004-6361/201322068
Filbert, S., Martini, P., Seebaluck, K., et al. 2023, arXiv e-prints,
Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., et al.
arXiv:2309.03434, doi: 10.48550/arXiv.2309.03434
2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f
Fitzpatrick, M. J., Olsen, K., Economou, F., et al. 2014, in Society
Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al.
of Photo-Optical Instrumentation Engineers (SPIE) Conference
2022, ApJ, 935, 167, doi: 10.3847/1538-4357/ac7c74
Series, Vol. 9149, Observatory Operations: Strategies,
Bailey, S. J., et al. 2023, in prep.
Processes, and Systems V, ed. A. B. Peck, C. R. Benn, & R. L.
Baldwin, J. A., Phillips, M. M., & Terlevich, R. 1981, PASP, 93, 5,
Seaman, 91491T, doi: 10.1117/12.2057445
doi: 10.1086/130766
Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2018,
Bassani, L., Dadina, M., Maiolino, R., et al. 1999, ApJS, 121, 473,
A&A, 616, A1, doi: 10.1051/0004-6361/201833051
doi: 10.1086/313202
Glikman, E., Gregg, M. D., Lacy, M., et al. 2004, ApJ, 607, 60,
Bennert, N., Falcke, H., Schulz, H., Wilson, A. S., & Wills, B. J. doi: 10.1086/383305
2002, ApJL, 574, L105, doi: 10.1086/342420 Goodrich, R. W. 1989, ApJ, 342, 224, doi: 10.1086/167586
Busca, N., & Balland, C. 2018, arXiv e-prints, arXiv:1808.09955, Górski, K. M., Hivon, E., Banday, A. J., et al. 2005, ApJ, 622, 759,
doi: 10.48550/arXiv.1808.09955 doi: 10.1086/427976
Cardamone, C., Schawinski, K., Sarzi, M., et al. 2009, MNRAS, Guy, J., Bailey, S., Kremin, A., et al. 2023, AJ, 165, 144,
399, 1191, doi: 10.1111/j.1365-2966.2009.15383.x doi: 10.3847/1538-3881/acb212
Chaussidon, E., Yèche, C., Palanque-Delabrouille, N., et al. 2023, Hahn, C., Wilson, M. J., Ruiz-Macias, O., et al. 2023, AJ, 165, 253,
ApJ, 944, 107, doi: 10.3847/1538-4357/acb3c2 doi: 10.3847/1538-3881/accff8
DESI Collaboration, Aghamousa, A., Aguilar, J., et al. 2016, arXiv Hainline, K. N., Hickox, R. C., Greene, J. E., et al. 2014, ApJ, 787,
e-prints, arXiv:1611.00036. https://arxiv.org/abs/1611.00036 65, doi: 10.1088/0004-637X/787/1/65
DESI Collaboration, Abareshi, B., Aguilar, J., et al. 2022, AJ, 164, Hamann, F., Zakamska, N. L., Ross, N., et al. 2017, MNRAS, 464,
207, doi: 10.3847/1538-3881/ac882b 3431, doi: 10.1093/mnras/stw2387
40 S. Juneau et al.

Harish, S., Malhotra, S., Rhoads, J. E., et al. 2023, ApJ, 945, 157, Meisner, A. M., Lang, D. A., & Schlegel, D. J. 2018, Research
doi: 10.3847/1538-4357/acb99c Notes of the American Astronomical Society, 2, 202,
Harrison, C. M., Costa, T., Tadhunter, C. N., et al. 2018, Nature doi: 10.3847/2515-5172/aaecd5
Astronomy, 2, 198, doi: 10.1038/s41550-018-0403-6 Moustakas, J. 2023, FastSpecFit: Fast spectral synthesis and
Heckman, T. M., & Best, P. N. 2014, ARA&A, 52, 589, emission-line fitting of DESI spectra, Astrophysics Source Code
doi: 10.1146/annurev-astro-081913-035722 Library, record ascl:2308.005. http://ascl.net/2308.005
Hviding, R. E., Hainline, K. N., Rieke, M., et al. 2022, AJ, 163, Mullaney, J. R., Alexander, D. M., Fine, S., et al. 2013, MNRAS,
224, doi: 10.3847/1538-3881/ac5e33 433, 622, doi: 10.1093/mnras/stt751
Jarrett, T. H., Cohen, M., Masci, F., et al. 2011, ApJ, 735, 112, Mullaney, J. R., & Ward, M. J. 2008, MNRAS, 385, 53,
doi: 10.1088/0004-637X/735/2/112 doi: 10.1111/j.1365-2966.2007.12777.x
Juneau, S., Dickinson, M., Alexander, D. M., & Salim, S. 2011, Myers, A. D., Moustakas, J., Bailey, S., et al. 2023, AJ, 165, 50,
ApJ, 736, 104, doi: 10.1088/0004-637X/736/2/104 doi: 10.3847/1538-3881/aca5f9
Juneau, S., Bournaud, F., Charlot, S., et al. 2014, ApJ, 788, 88, Napolitano, L., Pandey, A., Myers, A. D., et al. 2023, AJ, 166, 99,
doi: 10.1088/0004-637X/788/1/88 doi: 10.3847/1538-3881/ace62c
Juneau, S., Jacques, A., Pothier, S., et al. 2024, arXiv e-prints, Nikutta, R., Fitzpatrick, M., Scott, A., & Weaver, B. A. 2020,
arXiv:2401.05576, doi: 10.48550/arXiv.2401.05576 Astronomy and Computing, 33, 100411,
Kauffmann, G., Heckman, T. M., Tremonti, C., et al. 2003, doi: 10.1016/j.ascom.2020.100411
MNRAS, 346, 1055, doi: 10.1111/j.1365-2966.2003.07154.x Osterbrock, D. E., & Dahari, O. 1983, ApJ, 273, 478,
Kewley, L. J., Nicholls, D. C., & Sutherland, R. S. 2019, ARA&A, doi: 10.1086/161385
57, 511, doi: 10.1146/annurev-astro-081817-051832 Osterbrock, D. E., & Ferland, G. J. 2006, Astrophysics of gaseous
Khachikian, E. Y., & Weedman, D. W. 1974, ApJ, 192, 581, nebulae and active galactic nuclei
doi: 10.1086/153093 Osterbrock, D. E., & Pogge, R. W. 1985, ApJ, 297, 166,
Komossa, S. 2008, in Revista Mexicana de Astronomia y doi: 10.1086/163513
Astrofisica Conference Series, Vol. 32, Revista Mexicana de Plotkin, R. M., Anderson, S. F., Hall, P. B., et al. 2008, AJ, 135,
Astronomia y Astrofisica Conference Series, 86–92, 2453, doi: 10.1088/0004-6256/135/6/2453
doi: 10.48550/arXiv.0710.3326 Raichoor, A., Moustakas, J., Newman, J. A., et al. 2023, AJ, 165,
Lacy, M., Baum, S. A., Chandler, C. J., et al. 2020, PASP, 132, 126, doi: 10.3847/1538-3881/acb213
035001, doi: 10.1088/1538-3873/ab63eb Rakshit, S., Stalin, C. S., Kotilainen, J., & Shin, J. 2021, ApJS,
Lamastra, A., Bianchi, S., Matt, G., et al. 2009, A&A, 504, 73, 253, 28, doi: 10.3847/1538-4365/abd9bb
doi: 10.1051/0004-6361/200912023 Ramı́rez-Pérez, C., Pérez-Ràfols, I., Font-Ribera, A., et al. 2023,
Lan, T.-W., Tojeiro, R., Armengaud, E., et al. 2023, ApJ, 943, 68, arXiv e-prints, arXiv:2306.06312,
doi: 10.3847/1538-4357/aca5fa doi: 10.48550/arXiv.2306.06312
Lang, D., Hogg, D. W., & Mykytyn, D. 2016, The Tractor: Ramos Almeida, C., Levenson, N. A., Alonso-Herrero, A., et al.
Probabilistic astronomical source detection and measurement. 2011, ApJ, 731, 92, doi: 10.1088/0004-637X/731/2/92
http://ascl.net/1604.008 Richards, G. T., Hall, P. B., Vanden Berk, D. E., et al. 2003, AJ,
Levi, M., Bebek, C., Beers, T., et al. 2013, arXiv e-prints, 126, 1131, doi: 10.1086/377014
arXiv:1308.0847. https://arxiv.org/abs/1308.0847 Ross, N. P., Hamann, F., Zakamska, N. L., et al. 2015, MNRAS,
Madau, P., & Dickinson, M. 2014, ARA&A, 52, 415, 453, 3932, doi: 10.1093/mnras/stv1710
doi: 10.1146/annurev-astro-081811-125615 Ruiz-Macias, O., Zarrouk, P., Cole, S., et al. 2020, Research Notes
Mainzer, A., Bauer, J., Cutri, R. M., et al. 2014, ApJ, 792, 30, of the American Astronomical Society, 4, 187,
doi: 10.1088/0004-637X/792/1/30 doi: 10.3847/2515-5172/abc25a
Marcha, M. J. M., Browne, I. W. A., Impey, C. D., & Smith, P. S. —. 2021, MNRAS, 502, 4328, doi: 10.1093/mnras/stab292
1996, MNRAS, 281, 425, doi: 10.1093/mnras/281.2.425 Sanders, D. B., Soifer, B. T., Elias, J. H., et al. 1988, ApJ, 325, 74,
Marconi, A., Risaliti, G., Gilli, R., et al. 2004, MNRAS, 351, 169, doi: 10.1086/165983
doi: 10.1111/j.1365-2966.2004.07765.x Schlafly, E. F., Meisner, A. M., & Green, G. M. 2019, ApJS, 240,
Mateos, S., Alonso-Herrero, A., Carrera, F. J., et al. 2012, 30, doi: 10.3847/1538-4365/aafbea
MNRAS, 426, 3271, doi: 10.1111/j.1365-2966.2012.21843.x Schlafly, E. F., Kirkby, D., Schlegel, D. J., et al. 2023, arXiv
Meisner, A. M., Lang, D., Schlafly, E. F., & Schlegel, D. J. 2021, e-prints, arXiv:2306.06309, doi: 10.48550/arXiv.2306.06309
Research Notes of the American Astronomical Society, 5, 168, Shaw, M. S., Romani, R. W., Cotter, G., et al. 2012, ApJ, 748, 49,
doi: 10.3847/2515-5172/ac1684 doi: 10.1088/0004-637X/748/1/49
 DESI BGS AGN Target Selection 41

Somerville, R. S., & Davé, R. 2015, ARA&A, 53, 51, Yan, R., Ho, L. C., Newman, J. A., et al. 2011, ApJ, 728, 38,
doi: 10.1146/annurev-astro-082812-140951 doi: 10.1088/0004-637X/728/1/38
Stern, D., Assef, R. J., Benford, D. J., et al. 2012, ApJ, 753, 30,
Yang, J., Wang, F., Wu, X.-B., et al. 2016, ApJ, 829, 33,
doi: 10.1088/0004-637X/753/1/30
Urry, C. M., & Padovani, P. 1995, PASP, 107, 803, doi: 10.3847/0004-637X/829/1/33
doi: 10.1086/133630 Yang, J., Fan, X., Gupta, A., et al. 2023, arXiv e-prints,
Vanden Berk, D. E., Richards, G. T., Bauer, A., et al. 2001, AJ, arXiv:2302.01777, doi: 10.48550/arXiv.2302.01777
122, 549, doi: 10.1086/321167
Zhang, K., & Hao, L. 2018, ApJ, 856, 171,
Wang, F., Wu, X.-B., Fan, X., et al. 2016, ApJ, 819, 24,
doi: 10.3847/0004-637X/819/1/24 doi: 10.3847/1538-4357/aab207
Weiner, B. J., Papovich, C., Bundy, K., et al. 2007, ApJL, 660, Zhou, R., Dey, B., Newman, J. A., et al. 2023, AJ, 165, 58,
L39, doi: 10.1086/517925
doi: 10.3847/1538-3881/aca5fb
Wright, E. L., Eisenhardt, P. R. M., Mainzer, A. K., et al. 2010, AJ,
140, 1868, doi: 10.1088/0004-6256/140/6/1868 Zou, H., Zhou, X., Fan, X., et al. 2017, PASP, 129, 064101,
Wu, X.-B., Hao, G., Jia, Z., Zhang, Y., & Peng, N. 2012, AJ, 144, doi: 10.1088/1538-3873/aa65ba
49, doi: 10.1088/0004-6256/144/2/49