Galaxy sample for DES Y1 BAO measurements 1

Dark Energy Survey Year 1 Results: Galaxy Sample for BAO

Measurement
M. Crocce1? , A. J. Ross2 , I. Sevilla-Noarbe3 , E. Gaztanaga1 , J. Elvin-Poole4 , S. Avila5,6 , A. Alarcon1 , K.
C. Chan1,48 , N. Banik7 , J. Carretero8 , E. Sanchez3 , W. G. Hartley9,10 , C. Sánchez8 , T. Giannantonio11,12,13 ,
R. Rosenfeld14,15 , A. I. Salvador3 , M. Garcia-Fernandez3 , J. Garcı́a-Bellido7 , T. M. C. Abbott16 ,
F. B. Abdalla17,10 , S. Allam7 , J. Annis7 , K. Bechtol18 , A. Benoit-Lévy19,10,20 , G. M. Bernstein21 ,
R. A. Bernstein22 , E. Bertin20,19 , D. Brooks10 , E. Buckley-Geer7 , A. Carnero Rosell23,15 , M. Car-
rasco Kind24,25 , F. J. Castander1 , R. Cawthon26 , C. E. Cunha27 , C. B. D’Andrea22 , L. N. da Costa23,15 ,
C. Davis27 , J. De Vicente3 , S. Desai28 , H. T. Diehl7 , P. Doel10 , A. Drlica-Wagner7 , T. F. Eifler29,30 ,
arXiv:1712.06211v2 [astro-ph.CO] 14 Dec 2018

P. Fosalba1 , J. Frieman7,26 , J. Garcı́a-Bellido6 , D. W. Gerdes31,32 , D. Gruen27,33 , R. A. Gruendl25,24 ,

J. Gschwend23,15 , G. Gutierrez7 , D. Hollowood34 , K. Honscheid2,35 , B. Jain22 , D. J. James36 ,
E. Krause30,29 , K. Kuehn37 , S. Kuhlmann38 , N. Kuropatkin7 , O. Lahav10 , M. Lima39,15 , M. A. G. Maia15,23 ,
J. L. Marshall40 , P. Martini2,41 , F. Menanteau24,25 , C. J. Miller31,32 , R. Miquel42,8 , R. C. Nichol5 ,
W. J. Percival5 , A. A. Plazas29 , M. Sako22 , V. Scarpine7 , R. Schindler33 , D. Scolnic26 , E. Sheldon43 ,
M. Smith44 , R. C. Smith16 , M. Soares-Santos7,45 , F. Sobreira15,46 , E. Suchyta47 , M. E. C. Swanson25 ,
G. Tarle32 , D. Thomas5 , D. L. Tucker7 , V. Vikram38 , A. R. Walker16 , B. Yanny7 , Y. Zhang7
(The Dark Energy Survey Collaboration)
1
Institute of Space Sciences (ICE, CSIC) & Institut d’Estudis Espacials de Catalunya (IEEC), Campus UAB, Carrer de Can Magrans, s/n, 08193 Barcelona, Spain
2
Center for Cosmology and Astro-Particle Physics, The Ohio State University, Columbus, OH 43210, USA
3
Centro de Investigaciones Energéticas, Medioambientales y Tecnológicas (CIEMAT), Madrid, Spain
4
Jodrell Bank Center for Astrophysics, School of Physics and Astronomy, University of Manchester, Oxford Road, Manchester, M13 9PL, UK
5
Institute of Cosmology & Gravitation, University of Portsmouth, Portsmouth, PO1 3FX, UK
6
Instituto de Fisica Teorica UAM/CSIC, Universidad Autonoma de Madrid, 28049 Madrid, Spain
7
Fermi National Accelerator Laboratory, P. O. Box 500, Batavia, IL 60510, USA
8
Institut de Fı́sica d’Altes Energies (IFAE), The Barcelona Institute of Science and Technology, Campus UAB, 08193 Bellaterra (Barcelona) Spain
9
Department of Physics, ETH Zurich, Wolfgang-Pauli-Strasse 16, CH-8093 Zurich, Switzerland
10
Department of Physics & Astronomy, University College London, Gower Street, London, WC1E 6BT, UK
11
Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK
12
Kavli Institute for Cosmology, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK
13
Universitäts-Sternwarte, Fakultät für Physik, Ludwig-Maximilians Universität München, Scheinerstr. 1, 81679 München, Germany
14
ICTP South American Institute for Fundamental Research & Instituto de Fı́sica Teórica, Universidade Estadual Paulista, São Paulo, Brazil
15
Laboratório Interinstitucional de e-Astronomia - LIneA, Rua Gal. José Cristino 77, Rio de Janeiro, RJ - 20921-400, Brazil
16
Cerro Tololo Inter-American Observatory, National Optical Astronomy Observatory, Casilla 603, La Serena, Chile
17
Department of Physics and Electronics, Rhodes University, PO Box 94, Grahamstown, 6140, South Africa
18
LSST, 933 North Cherry Avenue, Tucson, AZ 85721, USA
19
CNRS, UMR 7095, Institut d’Astrophysique de Paris, F-75014, Paris, France
20
Sorbonne Universités, UPMC Univ Paris 06, UMR 7095, Institut d’Astrophysique de Paris, F-75014, Paris, France
21
Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA
22
Observatories of the Carnegie Institution of Washington, 813 Santa Barbara St., Pasadena, CA 91101, USA
23
Observatório Nacional, Rua Gal. José Cristino 77, Rio de Janeiro, RJ - 20921-400, Brazil
24
Department of Astronomy, University of Illinois at Urbana-Champaign, 1002 W. Green Street, Urbana, IL 61801, USA
25
National Center for Supercomputing Applications, 1205 West Clark St., Urbana, IL 61801, USA
26
Kavli Institute for Cosmological Physics, University of Chicago, Chicago, IL 60637, USA
27
Kavli Institute for Particle Astrophysics & Cosmology, P. O. Box 2450, Stanford University, Stanford, CA 94305, USA
28
Department of Physics, IIT Hyderabad, Kandi, Telangana 502285, India
29
Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Dr., Pasadena, CA 91109, USA
30
Department of Astronomy/Steward Observatory, 933 North Cherry Avenue, Tucson, AZ 85721-0065, USA
31
Department of Astronomy, University of Michigan, Ann Arbor, MI 48109, USA
32
Department of Physics, University of Michigan, Ann Arbor, MI 48109, USA
33
SLAC National Accelerator Laboratory, Menlo Park, CA 94025, USA
34
Santa Cruz Institute for Particle Physics, Santa Cruz, CA 95064, USA
35
Department of Physics, The Ohio State University, Columbus, OH 43210, USA
36
Event Horizon Telescope, Harvard-Smithsonian Center for Astrophysics, MS-42, 60 Garden Street, Cambridge, MA 02138, UK
37
Australian Astronomical Observatory, North Ryde, NSW 2113, Australia
38
Argonne National Laboratory, 9700 South Cass Avenue, Lemont, IL 60439, USA
39
Departamento de Fı́sica Matemática, Instituto de Fı́sica, Universidade de São Paulo, CP 66318, São Paulo, SP, 05314-970, Brazil
40
George P. and Cynthia Woods Mitchell Institute for Fundamental Physics and Astronomy, and Department of Physics and Astronomy, Texas A&M University, College Station, TX 77843, USA
41
Department of Astronomy, The Ohio State University, Columbus, OH 43210, USA
42
Institució Catalana de Recerca i Estudis Avançats, E-08010 Barcelona, Spain
43
Brookhaven National Laboratory, Bldg 510, Upton, NY 11973, USA
44
School of Physics and Astronomy, University of Southampton, Southampton, SO17 1BJ, UK
45
Department of Physics, Brandeis University, Waltham, MA 02453, USA
46
Instituto de Fı́sica Gleb Wataghin, Universidade Estadual de Campinas, 13083-859, Campinas, SP, Brazil
47
Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, TN 37831
48
School of Physics and Astronomy, Sun Yat-Sen University, Guangzhou 510275, China

MNRAS 000, 1–16 (2017)

Prepared for submission to MNRAS

MNRAS 000, 1–16 (2017) Preprint 17 December 2018 Compiled using MNRAS LATEX style file v3.0

ABSTRACT
We define and characterise a sample of 1.3 million galaxies extracted from the first year of
Dark Energy Survey data, optimised to measure Baryon Acoustic Oscillations in the presence
of significant redshift uncertainties. The sample is dominated by luminous red galaxies located
at redshifts z & 0.6. We define the exact selection using color and magnitude cuts that balance
the need of high number densities and small photometric redshift uncertainties, using the
corresponding forecasted BAO distance error as a figure-of-merit in the process. The typical
photo-z uncertainty varies from 2.3% to 3.6% (in units of 1+z) from z = 0.6 to 1, with number
densities from 200 to 130 galaxies per deg2 in tomographic bins of width ∆z = 0.1. Next we
summarise the validation of the photometric redshift estimation. We characterise and mitigate
observational systematics including stellar contamination, and show that the clustering on
large scales is robust in front of those contaminants. We show that the clustering signal in the
auto-correlations and cross-correlations is generally consistent with theoretical models, which
serves as an additional test of the redshift distributions.
Key words: cosmology: observations - (cosmology:) large-scale structure of Universe

1 INTRODUCTION that its precision in the measurement of redshifts is limited, pre-

venting the measurement of the Hubble parameter evolution. How-
The use of the imprint of Baryon Acoustic Oscillations (BAO) in
ever, the evolution of the angular distance with redshift is possible,
the spatial distribution of galaxies as a standard ruler has become
through the measurement of angular correlation functions (Seo &
one of the common methods in current observational cosmology
Eisenstein 2003; Padmanabhan et al. 2005; Blake & Bridle 2005;
to understand the Universe. The physics that causes BAO is well
Padmanabhan et al. 2007; Crocce et al. 2011; Sánchez et al. 2011;
understood. Primordial perturbations generated acoustic waves in
Carnero et al. 2012; Seo et al. 2012; de Simoni et al. 2013).
the photon-baryon fluid until decoupling (z ∼ 1100). These sound
Although DES will only measure BAO in the angular distri-
waves lead to the large oscillations observed in the power spectrum
bution of galaxies, a determination of the photometric redshift as
of the CMB anisotropies, but they are also visible in the clustering
precise as possible brings several benefits. It allows a finer tomog-
of matter, and therefore galaxies, as a high density region around
raphy in the mapping of the BAO evolution with the redshift and
the original source of the perturbation, at a distance given by the
makes the analysis cleaner, reducing the correlations between red-
sound horizon length at recombination. This high density region
shift bins. A sample of Luminous Red Galaxies (LRGs) would fit
shows as a small excess in the number of pairs of galaxies sep-
these requirements (Padmanabhan et al. 2005, 2007). LRGs are lu-
arated by ∼ 150 Mpc. Since the sound horizon is very precisely
minous and massive galaxies with a nearly uniform Spectral En-
measured in the cosmic microwave background (Planck Collabora-
ergy Distribution (SED), but with a strong break at 4000 Å in the
tion et al. 2016), the BAO measurements can be used as a standard
rest frame. These features allow a clean selection and an accu-
ruler. This is, therefore, a geometrical probe of the expansion rate of
rate determination of the redshift for this type of galaxies, even in
the Universe, that maps the angular diameter distance and the Hub-
photometric surveys. This selection has been done previously for
ble parameter as functions of the redshift. There have now been
imaging data at z . 0.6 (Padmanabhan et al. 2005). But the BAO
multiple detections of the BAO in redshift surveys (Eisenstein et al.
scale has already been measured with high precision in this redshift
2005; Percival et al. 2010; Ross et al. 2015; Alam et al. 2017; Ata
range (e.g. Alam et al. (2017) and references therein). In order to
et al. 2018; Delubac et al. 2015; Bautista et al. 2017; Percival et al.
go to higher redshifts, the selection criteria need to be redefined.
2001; Cole et al. 2005; Blake et al. 2011; Beutler et al. 2011) and
The 4000 Å feature enters the i band at z = 0.75, and the methods
it is considered as one of the main cosmological probes for current
used in previous selections are not valid anymore.
and planned cosmological projects.
In this paper we describe the selection of a sample of red
A key feature of the BAO method is the fact that the sound
galaxies to measure BAO in DES, that includes, but is not limited
horizon length is large, and, therefore, very deep and wide galaxy
to, LRGs. The selection is defined by two conditions. On the one
surveys are needed in order to reach precise measurements of the
hand, keep the determination of the photometric redshift as precise
BAO scale. But, at the same time, this large scale protects the BAO
as possible. On the other hand, keep the galaxy density high enough
feature from large corrections due to astrophysical and non-linear
to have a BAO measurement that is not limited by shot noise.
effects of structure formation and therefore from systematic errors,
In order to guide our efforts to select an optimized sample for
making BAO a solid probe of the expansion rate of the Universe.
measuring BAO distance scales, we rely on Fisher matrix forecasts.
The Dark Energy Survey (DES) is one of the most important
Seo & Eisenstein (2007) provide a framework and simple formulae
of the currently ongoing large galaxy surveys and, as its name sug-
to predict the precision that one can achieve with a given set of
gests, it is specially designed to attack the problem of the physical
galaxy data. Thus, we will test how Fisher matrix forecasts vary
nature of the dark energy. It will do it using several independent and
given the variations obtained for the number density and estimated
complementary methods at the same time. One of them is the pre-
redshift uncertainty given a set of color-magnitude cuts.
cise study of the spatial distribution of galaxies, and in particular,
This paper, detailing the BAO sample selection, is one of a
the BAO standard ruler. DES is a photometric survey, which means
series describing the supporting work leading to the BAO mea-
surement using DES Y1 data presented in The Dark Energy Sur-
vey Collaboration et al. (2017) (hereafter DES-BAO-MAIN). As
? e-mail: [email protected] part of such series, one paper presents the mock galaxy catalogues,

c 2017 The Authors

 Galaxy sample for DES Y1 BAO measurements 3
Avila et al. (2018) (hereafter DES-BAO-MOCKS). Gaztañaga et al. column in the catalog. Finally, individual objects which have been
(2018) discusses in detail the photo-z validation, and we denote it identified as being problematic by the DESDM processing or by
DES-BAO-PHOTOZ. Chan et al. (2018), from now on DES-BAO- the vetting process carried out by the scientists in the collaboration
θ-METHOD, introduces the BAO extraction pipeline using a to- are flagged when configuring the catalog (this is done through the
mographic analysis of angular correlation functions, while Cama- flags gold column). All data we describe in this and in sub-
cho et al. (2018) presents the study of the angular power spectrum sequent sections are drawn from quantities and maps released as
(hereafter DES-BAO-`-METHOD). Lastly, Ross et al. (2017a), in part of the DES Y1 Gold catalog and are fully described in Drlica-
what follows referred to as DES-BAO-s⊥ -METHOD, introduced a Wagner et al. (2017).
novel technique to infer BAO distances using the three-dimensional The photometry used in this work comes mainly from two dif-
correlation function binned in projected separations. ferent sources:
This paper is organized as follow: in section 2, a description
• the SExtractor (Bertin & Arnouts (1996)) AUTO magni-
of the main features of the DES-Y1 catalogue is given: in section 3,
tudes, which are derived from the best matched elliptical aperture
we give a detailed description of the selection cuts that define the
according to the coadd object elongation and angle in the sky, mea-
data sample that has been used to measure the BAO scale in DES;
sured using the coadded object flux;
section 4 contains a description of the procedure that has been de-
• Multi-Object Fitting (MOF) pipeline, which performs a multi-
veloped and applied in DES in order to ensure the quality of the
epoch and multi-band fit of the shape and per-band fluxes directly
photometric redshift determination, and to determine its relation
on the single epoch exposures for each of the coadd objects, with
with the true redshift; section 5 describes the masking scheme and
additional neighboring light subtraction. This is described in more
the treatment of the variable depth in the survey; section 6 is a de-
detail in Drlica-Wagner et al. (2017).
scription of the analysis and mitigation of observational system-
atic errors on the clustering measurement; and finally, section 7 Using these photometric measurements, we will consider
describes the measured two-point correlation and cross-correlation three different photometric redshift catalogues. Two of them
functions and their evolution with redshift for the selected sample. are built using BPZ (Benı́tez 2000), a Bayesian template-fitting
We finish with our conclusions in section 8. method, and another using a machine learning approach: the Di-
rectional Neighborhood Fitting (DNF) algorithm as described in
De Vicente, Sánchez & Sevilla-Noarbe (2016). They are combined
with the photometric quantities described above and used as fol-
2 DES Y1 DATA
lows:
The BAO galaxy sample we will define in this work makes use of
• BPZ run with AUTO magnitudes (hereafter zBPZ−AUTO ) used
the first year of data (Y1) from the Dark Energy Survey. This pho-
for making the selection of the overall sample.
tometric dataset has been produced using the Dark Energy Camera
• BPZ run with MOF magnitudes (hereafter zBPZ−MOF ) used
(DECam, Flaugher et al. (2015)) observations, processed and cal-
for redshift binning and transverse distance calculation, finally used
ibrated by the DES Data Management system (DESDM) (Sevilla
as secondary catalogue to show the robustness of the analysis.
et al. 2011; Mohr et al. 2012; Morganson et al. 2018) and finally
• DNF run with MOF magnitudes (hereafter zDNF−MOF ) used
curated, optimized and complemented into the Gold catalog (here-
for redshift binning and transverse distance calculation, finally used
after denoted ‘Y1GOLD’), as described in Drlica-Wagner et al.
as our fiducial catalogue.
(2017). For each band, single exposures are combined in coadds
to achieve a higher depth. We keep track of the complex geome- We should note that BPZ with AUTO magnitudes is part of
try that the combinations of these dithered exposures will create at the DESDM data reduction pipeline and is available early on in the
each point in the sky in terms of observing conditions and survey catalogue making. This explains why we used that particular com-
properties. Objects are detected in chi-squared combinations of the bination for sample selection. We did not find, and do not expect,
r, i and z coadds to create the final coadd catalog (Szalay, Connolly the relative optimization of the sample selection and cuts to depend
& Szokoly 1999). much on the particular photo-z catalogue (but the final absolute er-
Y1GOLD covers a total footprint of more than 1800 deg2 ; ror on BAO distance measurement does).
this footprint is defined by a H EALPIX (Górski et al. 2005) map at In Section 4, we summarize the validation performed to select
resolution Nside = 4096 and includes only area with a minimum and characterise the true redshift distributions of the binned sam-
total exposure time of at least 90 seconds in each of the griz bands, ples, which is described in detail in DES-BAO-PHOTOZ.
and a valid calibration solution (see Drlica-Wagner et al. (2017) for Throughout our analysis we assume the redshift estimate of
details). This footprint is divided into several disjoint sub-regions each galaxy to be the mean redshift of the redshift posterior for
which encompass the supernova survey areas, a region overlapping BPZ, or the predicted value for the object in the fitted hyper-plane
stripe 82 from the SDSS footprint (S82; Annis et al. (2014)) and from the DNF code (see De Vicente, Sánchez & Sevilla-Noarbe
a larger area overlapping with the South Pole Telescope coverage (2016). Any potential biases from these estimates are calibrated as
(SPT; Carlstrom et al. (2011)). Figure 1 shows the angular distri- described in Section 4.
bution of galaxies, selected as described in Section 3, that includes
these two areas. A series of veto masks, including masks for bright
stars and the Large Magellanic Cloud among others, reduce the
3 SAMPLE SELECTION
area by ∼ 500 deg2 , leaving 1336 deg2 suitable for LSS study.
Other areas that are severely affected by imaging artifacts or other- In this section, we describe the steps towards the construction of a
wise have a high density of image artifacts are masked out as well. red galaxy dominated sample, optimized for BAO measurements,
Section 5 provides a full account of the final mask used in com- starting from the dataset described in Section 2. The selection is
bination with the final BAO sample. “Bad” regions information is performed over the largest continuous regions of the survey at
propagated to the ‘object’ level by using the flags badregion this point, namely SPT and S82. Objects are selected so that we

MNRAS 000, 1–16 (2017)

4 Crocce et al.

0.50
0
0.45
10
0.40
20 0.35

ng [arcmin 2]
0.30
DEC

30
0.25
40
0.20
50 0.15
0.10
120 90 60 30 0 330 300 270
RA
Figure 1. Angular distribution and projected density of the DES-Y1 red galaxy sample described in this paper, and subsequently used for BAO measurements.
The unmasked footprint comprises the two largest compact regions of the dataset: one in the southern hemisphere of 1203 deg2 , overlapping South Pole
Telescope observations (SPT; Carlstrom et al. 2011), and 115 deg2 near the celestial equator, overlapping with Stripe 82 (S82, Annis et al. 2014). The sample
consists of about 1.3 million galaxies with photometric redshifts in the range [0.6 − 1.0] and constitutes the baseline for our DES Y1 BAO analysis.

Table 1. Complete description of the selection performed to obtain a sample dominated by red galaxies with a good compromise of photo-z accuracy and
number density, optimal for the BAO measurement presented in DES-BAO-MAIN. The redshifts of the resulting catalogue are then computed using different
codes (BPZ and DNF) as described in Sec 2. Therefore, any subsequent photo-z selection can be done either with zphoto from BPZ or DNF.

Keyword Cut Description

Gold observations present in the Gold catalog Drlica-Wagner et al. (2017)

Quality flags badregion < 4; flags gold = 0 Sec.5; Sec.2
Footprint 1336 deg2 (1221 deg2 in SPT and 115 deg2 in S82) Fig. 1 Sec.5
Color Outliers −1 < gauto − rauto < 3 Sec. 3.1
−1 < rauto − iauto < 2.5 Sec. 3.1
−1 < iauto − zauto < 2 Sec. 3.1
[Optimized] Color Selection (iauto − zauto ) + 2.0(rauto − iauto ) > 1.7 Sec. 3.4.1
[Optimized] Completeness Cut iauto < 22 Sec. 3.1
[Optimized] Flux Selection 17.5 < iauto < 19.0 + 3.0zBPZ−AUTO Sec. 3.4.2
Star-galaxy separation spread model i + (5/3) spreaderr model i > 0.007 Sec. 3.2
Photo-z range [0.6 − 1.0] Sec. 4

avoid imaging artifacts and pernicious regions with foreground ob- Additionally, we remove the most luminous objects by making the
jects using the cuts on flags badregion and flags gold de- cut iauto > 17.5 . The cut of Eq. (1) is chosen as a compromise be-
scribed therein. In the rest of this section we go into finer details on tweensurvey area, given that we need to achieve an homogeneous
the flux, color and star-galaxy separation selection. depth, and the number of galaxies in that area. For a given overall
In Table 1, we summarise this sample selection, including ref- flux limit of the galaxy sample (e.g. all galaxies with i 6 22) we
erences to the sections where these cuts are explained. select the regions of the survey that are deeper than that limit (e.g.
i-band 10σ limit depth > 22) and mask everything brighter. In this
way that sample selection should be complete over such footprint.
Clearly, for fainter selections more objects are incorporated into
3.1 Completeness and color outliers cuts
the sample but the area of the survey reaching that depth homoge-
The overall flux-limit of the sample is set as neously is also smaller. Hence there is a compromise between area
and number of objects. In Fig. 2 we show the normalized counts as
iauto < 22. (1)

MNRAS 000, 1–16 (2017)

 Galaxy sample for DES Y1 BAO measurements 5

25
MODEST_CLASS (default classifier in Y1GOLD)
BAO classifier > 0.005 (Extra 3% galaxies vs MODEST)
20 BAO classifier > 0.007 (loss of 3% galaxies vs MODEST)

Star contamination level (%)

BAO classifier > 0.009 (loss of 9% galaxies vs MODEST)
15

10

0
0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00
zphoto
Figure 3. Contamination of galaxy sample from stars as a function of red-
Figure 2. Measurement of the trade off between area and number of objects shift and star-galaxy separation threshold, as measured using galaxy density
as a function of magnitude limit and sample flux limit in Y1GOLD and SV. vs stellar density plots (from a pure stellar sample). The MODEST classifier
For a given iauto -band “threshold” value we select all regions which have is defined in Drlica-Wagner et al. (2017) as the default star galaxy classi-
a deeper limiting magnitude that this value (10σ depth limit > “threshold”) fier (based on spread model and wavg spread model). ‘BAO classifier’
and count the galaxies brighter than the “threshold” value over those re- stands for a cut in spread model i + (5.0/3.0)spreaderr model i. A
gions. These should be complete samples at each threshold value. Number threshold of 0.007 provides an important decrease of contamination with a
counts are shown normalized to their maximum in the figure. minor adjustment in the number of galaxies, which becomes significantly
more severe at higher thresholds for a very similar purity. The redshift bin-
ning here uses zBPZ−AUTO .
a function of the magnitude limit cut. For comparison we include
the same quantity in Science Verification Data, which is deeper than
Y1 but has much smaller area, see Crocce et al. (2016). We would
like to select a sample and footprint that are at once homogeneous
20.0
and with the highest possible number of galaxies. The curve shows Stars selected morphologically
a plateau in the range 22 . iauto . 22.3 where the number counts 17.5
is maximized, with variations of about 5%. But the figure does 15.0
not account for photo-z performance, which degrades rapidly for 12.5
fainter objects (particularly at high redshift) and is of key relevance
10.0
for BAO measurements, as shown below in Sec. 3.4. Therefore we
decided to stay at the bright end of this range (iauto = 22) as an 7.5
overall flux limit of the sample. 5.0
Color outliers which are either unphysical or from special 2.5
samples (Solar System objects, high redshift quasars) are removed 0.0
as well, to avoid extraneous photo-z populations in the sample (see 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 1.00
Table 1).
zphoto
Figure 4. Photometric redshift distribution of stars selected morphologi-
cally and passing the same cuts described in Table 1.The redshift value
3.2 Star-Galaxy Separation zphot is the mean from the pdf of zBPZ−AUTO , which was used for the
overall sample selection in Section 3.
Removing stars from the galaxy sample is an essential step to avoid
the dampening of the BAO signal-to-noise (Carnero et al. 2012)
or the introduction of spurious power on large scales (Ross et al.
2011a). Stellar contamination affects the broad shape of the mea- A detailed follow up analysis of star-galaxy separation is given in
surement and so we want to minimise it to be able to fit the BAO Sevilla-Noarbe et al. (2018). Here instead we decided to modify
template properly. However, it does not appreciably affect the lo- slightly this proposed cut in order to increase the purity of the sam-
cation of the BAO feature, so we do not need to push for 100% ple (from 95% to 97 − 98%), at the cost of losing approximately
purity. Any residual contamination is then taken care of by using 3% of the objects, by making the following selection:
the weighting scheme detailed in Section 6.
spread model i + (5.0/3.0)spreaderr model i > 0.007.
In this work we have used the default star-galaxy clas-
sification scheme described in detail in Sevilla-Noarbe et al. In Fig. 3 we show the estimated star sample contamination
(2018), see also Drlica-Wagner et al. (2017), which is based on for different thresholds of this cut, using the relation between
the i-band coadd magnitude spread model i and its associated galaxy density and a map of stellar density built from Y1GOLD
error spreaderr model i, from SExtractor. This classifier (a methodology that is described in detail in section 6). The error
was developed using as truth tables data from COSMOS (Leau- bars displayed are the fitting errors obtained for the intercept when
thaud et al. 2007), GOOD-S (Giavalisco et al. 2004) and VVDS parametrizing the contamination level using a linear relationship
(Le Fèvre et al. 2005) overlapping Y1GOLD, and subsequently between the galaxy density as a function of stellar density. Note
tested against CFHTLenS (Erben et al. 2013). The combination that a threshold of 0.007 reduces the contamination level to less
spread model i + (5.0/3.0)spreaderr model i > 0.005 is than 5% across the redshift range of interest. In Table 3 we re-
suggested for high-confidence galaxies as a baseline for Y1GOLD. port a consistent or smaller level of stellar contamination, using a

MNRAS 000, 1–16 (2017)

6 Crocce et al.

El B2004a Scd B2004a SB3 B2004a ssp 25Myr z008

Sbc B2004a Im B2004a SB2 B2004a ssp 5Myr z008

2.0 2.0 2.0

BAO cut: (i z) + 2.0(r i) > 1.7 z = 0.1
1.5 1.5 Stellar locus 1.5
z = 1.0
1.0 1.0 1.0
i

z
r

i
0.5 0.5 0.5
z = 0.6
0.0 0.0 0.0
color error color error color error
0.5 0.5 0.5
0.5 0.0 0.5 1.0 1.5 2.0 2.5 0.5 0.0 0.5 1.0 1.5 2.0 1.0 0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
g r r i r z
Figure 5. Evolution of BPZ templates in color-color space. Each dot corresponds to a different redshift in steps of 0.1, ranging from z = 0.0 to z = 2.0. The
shadowed region in the central panel is excluded from the sample. The black dots indicate the position of z = 0.6 (triangles), and z = 1.0 (squares) for the
two reddest templates. Also shown, for reference, is the stellar locus as a purple dashed line. The inset crosses indicate an estimate of the error in the colors,
arising from photometric errors, from a sub-sample of DES Y1 galaxies selected in the range 21 < iauto < 22 (see text for more details).

similar estimation, in the catalogues with MOF photometry, both use them to define cuts in color-color space intended to isolate the
for BPZ and DNF (see Sec. 6). In Fig. 5 we also include in the red templates.
middle figure the track from the stellar locus, which showcases the In real data galaxy colors have an uncertainty due to photo-
reason why the first two redshift bins are more affected by stellar metric errors, which effectively thicken those tracks. In order to
contamination, as it crosses the elliptical templates at these red- provide an estimate for this we computed the errors in the colors
shifts. To further illustrate this, in Fig. 4 we show the distribution for a sub-sample of Y1GOLD galaxies with 21 < iauto < 22 (the
of the mean photometric redshifts for stars (selected using the cri- typical range of magnitudes that we explore below to define the
terion |wavg spread model i| < 0.002, a more accurate variant BAO sample). For each galaxy we estimate the color error adding
of spread model i using single-epoch, suitable for moderate to in quadrature the corresponding magnitude errors1 . The average er-
bright magnitude ranges) showcasing how they will contaminate ror in each corresponding color is shown with a cross at the bottom
preferentially the second redshift bin, following the same trend as right inset label of the three panels of Fig. 5. Their values are 0.128,
shown in Table 3. 0.073, 0.067, 0.076 for (g-r, r-i, i-z, r-z) respectively.
In addition, a model of a red elliptical galaxy spectrum is
shown in Figure 6, redshifted to z = 0.4, 0.8, 1.15, where the
3.3 Selecting Red Luminous Galaxies notable 4000 Å break crosses from g → r, r → i and i → z. This
Next we want to select from Y1GOLD a sample dominated by lu- suggests that for z > 0.6 the strongest evolution in color will be for
minous red galaxies, because their typical photo-z estimates are i−z and r −i, and hence we will focus in these color combinations
more accurate than for the average galaxy population, thanks to the in what follows (that moreover have the smallest error).
4000 Å Balmer break in their spectra. This feature makes redshift Note how the transition of the 4000 Å break from one band to
determination easier even with broad-band photometry (Padman- another abruptly bends the color-color tracks in Figure 5. However,
abhan et al. 2005). In addition we want our BAO sample to cover this applies mainly to elliptical templates, and recent star formation
redshifts larger than 0.6 as there are already very precise BAO mea- will dampen this effect.
surements for z < 0.6, see e.g. Cuesta et al. (2016); Ross et al.
(2017b); Beutler et al. (2017).
We have tested that, while a very stringent selection can be 3.4 Optimization of the color and magnitude cuts for BAO
done to yield minimal photo-z errors, e.g. with the redMaGiC
algorithm (Rozo et al. 2016), it does not lead to optimal BAO Optimizing the actual sample selection for the measurement of
constraints because the sample ends up being very sparse, with BAO in imaging data is considerably different that doing so for
∼ 200, 000 galaxies in Y1GOLD at z > 0.6 (Elvin-Poole et al. spectrospopic data. In the later case one basically needs to maxi-
2017). Instead we will follow an alternative path and apply a stan- mize the area (or volume) provided that n̄P > 1 (where n̄ is the
dard selection in color-color space to isolate red galaxies at high galaxy density and P the power spectrum). For imaging data the
redshift, balancing photo-z accuracy and number density with a photometric redshift accuracy plays a vital role. Worse photo-z er-
BAO figure-of-merit in mind. ror degrades the signal as the galaxy radial separations are smeared
In Figure 5 we show the evolution in redshift of the eight out (this also complicates the definition of survey volume). In turn,
spectral templates used in BPZ, which includes one typical red el- the best photo-z’s are typically obtained for very bright, and low
liptical galaxy, two spirals and five blue irregulars/starbursts (color density, samples. Therefore there is a non-trivial. interplay to max-
coded) based on Coleman, Wu & Weedman (1980) and Kinney imise BAO signal to noise.
et al. (1996). We compute the expected observed DES broad-band In DES-BAO-s⊥ -METHOD we discussed in detail how to
magnitudes for these templates as a function of redshift and show
them in different color-color combinations.The tracks are evolved
from z = 0 to z = 2.0 in steps of 0.1 (marked with dots). We will 1 In turn computed as merr = −2.5(F luxerr /F lux)/ log(10)

MNRAS 000, 1–16 (2017)

 Galaxy sample for DES Y1 BAO measurements 7
Elliptical galaxy at redshift z = 1.15
Table 2. Sensitivity of the forecasted BAO distance error to variations in
density, photometric redshift errors and survey area. Note that these vari-
ations are considered individually, neglecting their correlations. Baseline
values are those corresponding to the optimal cuts discussed in Sec. 3.4.
Fλ

g r i z
property variation forecasted BAO distance error

Elliptical galaxy at redshift z = 0.80 10% worse photo-z 8% worse

20% worse photo-z 16% worse
10% lower density 3% worse
20% lower density 6% worse
Fλ

10% smaller area 2.8% worse

g r i z

Elliptical galaxy at redshift z = 0.40 form,

(iauto − zauto ) + a1 (rauto − iauto ) > a2 . (2)
The cut was chosen in this form following the discussion in
Fλ

Sec. 3.3 (see Fig. 5), as it allows us to select more likely the red-
g r i z dest galaxies which are the ones with lower uncertainties in their
photometric redshift determination and still present a high enough
4000 5000 6000 7000 8000 9000 10000
Wavelength [Å]
number density.
Samples were produced across a grid of a1 and a2 values,
Figure 6. Elliptical model spectrum used in template-based fitting code calculating the number of galaxies Ngal and a mean width of the
BPZ. Overplotted are the DES response filters g,r,i,z. The template has been photo-z distribution σz /(1 + z) for each sample, after splitting the
redshifted to z = 0.4, 0.8, 1.15, where the notable 4000 Å break crosses galaxy in tomographic bins. For BPZ we estimated σz averaging in
from g → r, r → i and i → z. each tomographic bin the width of the individual redshifts posterior
distributions (PDFs) provided per galaxy.
The BAO forecast using the algorithm of Seo & Eisenstein
fold in the photo-z accuracy into an effective n̄eff 2 . However com- (2007) is then run for the Ngal and σz /(1 + z) of each sample and
puting n̄eff is cumbersome and as complicated as doing an actual final values of a1 and a2 are selected to minimise the forecasted
BAO forecasting. Therefore we decided to follow this later path and BAO uncertainty, finding a balance between galaxy number density
rely on the Fisher matrix forecast formalism described in Seo & and redshift uncertainty. In order to give a sense for the sensitivity
Eisenstein (2007). Provided with a concrete set of color-magnitude of such process, we note there is a slight degeneracy when increas-
cuts we measure in the data the number density and redshift uncer- ing a1 and a2 simultaneously, resulting in similar forecasted BAO
tainty in several tomographic bins within 0.6 6 photo-z 6 1.0, and uncertainties. However deviations from this degeneracy direction
assume a clustering amplitude. We then use the formulae from Seo lead to significant degradation in the forecasted error. For example,
& Eisenstein (2007) to predict the precision that one can achieve doubling a1 leads to a degradation of the forecasted error by ap-
with that set of galaxy data properties. We repeat this process for a proximately 0.01 (from 5% to 6% roughly). The values used in this
different set of cuts until an optimal BAO distance error is achieved. analysis are a1 = 2.0, a2 = 1.7. Figure 5 shows the color cut in
Through this process we fix the clustering amplitude, assum- the central panel, where the shadowed region is excluded from the
ing a galaxy bias of b = 1.6 for all calculations. This is the bias sample.
found in Crocce et al. (2016) for a flux limited sample (i < 22.5)
at redshifts z ∼ 0.9, selected from DES Science Verification (SV)
data. Since that redshift and magnitude are compatible with what 3.4.2 Optimization of the magnitude cut
we expect in this paper, we consider b = 1.6 a representative value.
More precise measurements are expected for more biased samples, To further minimize the forecasted BAO uncertainty, an additional,
but the galaxy bias for any given sample is not known a priori and redshift dependent magnitude cut is applied to the sample as a sec-
the redshift uncertainty and number density are the more dominant ond step. This applies a cut to iauto at low redshift which is stricter
factors. than the global iauto < 22 cut (at lower redshift the sample is suf-
For illustrative purposes we show in Table 2 the variation in ficiently abundant that one can still select brighter galaxies, with
BAO distance error achieved by changing the number density and better photo-z, and still be sample variance dominated). The cut is
photo-z accuracy away from those at the optimal cuts described in the form,
below. We also include the variation with survey area. As pointed
iauto < a3 + a4 z. (3)
before, BAO distance errors are very sensitive to photo-z accuracy.
As with the color cut in Eq. 2, this is designed to find a sample
that balances redshift uncertainty with number density, to minimise
3.4.1 Optimization of the color cut the forecasted BAO error. The BAO forecast error was minimised
Thus, in order to maximize the signal-to-noise of the BAO fore- at the values a3 = 19 and a4 = 3 and this cut was applied to the
casted measurement, a color cut is applied to the sample in the sample. We find that the forecasted error improves by ∼ 15% when
introducing the redshift dependent flux limit as opposed to a global
iauto < 22 cut.
2 Photometric redshift errors leads to n̄eff P < 1 in all cases explored. The final forecasted uncertainty on angular diameter distance

MNRAS 000, 1–16 (2017)

8 Crocce et al.

Table 3. Characteristics of the DES Y1 BAO sample, as a function of red- estimates, or the stacking from the nearest neighbour redshifts from
shift. Results are shown for a selection of the sample in bins according to the training sample, in the case of DNF (henceforth we’ll call these
DNF photo-z (zphot ) estimate in top of the table and BPZ in the bottom, stack N (z)). Figure 7 shows the stack N (z) (yellow histograms)
both with MOF photometry. Here z̄ =< ztrue > is the mean true redshift, in all 4 redshift bins for our fiducial DNF photo-z analysis.
σ68 and W68 are the 68% confidence widths of (zphot − ztrue )/(1 +
ztrue ) and ztrue respectively, all estimated from COSMOS-DES valida-
tion with SVC correction, as detailed in Sec. 4 and Fig. 7. fstar is the esti- 4.1 COSMOS Validation
mated stellar contamination fraction, see Sec. 6
As detailed in DES-BAO-PHOTOZ, we check the performance of
DNF Ngal bias z̄ σ68 W68 fstar each code by using redshifts in the COSMOS field (which are not
part of the training set in the case of DNF), following the procedure
0.6 − 0.7 386057 1.81 ± 0.05 0.652 0.023 0.047 0.004 outlined in Hoyle et al. (2017). These redshifts are either spectro-
0.7 − 0.8 353789 1.77 ± 0.05 0.739 0.028 0.068 0.037
scopic or accurate (σ68 < 0.01) 30-band photo-z estimates from
0.8 − 0.9 330959 1.78 ± 0.05 0.844 0.029 0.060 0.012
0.9 − 1.0 229395 2.05 ± 0.06 0.936 0.036 0.067 0.015
Laigle et al. (2016). Both validation samples give consistent results
in our case because the samples under study are relatively bright.
BPZ Ngal bias z̄ σ68 W68 fstar The COSMOS field is not part of the DES survey. However a
few select exposures were done by DECam which were processed
0.6 − 0.7 332242 1.90 ± 0.05 0.656 0.027 0.049 0.018
by DESDM using the main survey pipeline. We call this sample
0.7 − 0.8 429366 1.79 ± 0.05 0.746 0.031 0.076 0.042
0.8 − 0.9 380059 1.81 ± 0.06 0.866 0.034 0.060 0.015
DES-COSMOS. Because the COSMOS area is small (2 square de-
0.9 − 1.0 180560 2.05 ± 0.07 0.948 0.039 0.068 0.006 grees) and DECam COSMOS images were deeper and not taken
as part of the main DES-Y1 Survey, we need to first resample the
DES-COSMOS photometry to make it representative of the full
DES Y1 samples that we select in our BAO analysis. Hence we
combining all the tomographic bins is ∼ 4.7%. Note that the dis-
add noise to the fluxes in the DES-COSMOS catalog to match the
cussion in this section only has as a goal the definition of the sam-
noise properties of the fluxes in the DES-Y1 BAO sample, this is
ple. The real data analysis with the sample defined here, and the
what we refer to as resampled photometry. Then for each galaxy in
final BAO error achieved, will of course depend in many other vari-
the DES-Y1 BAO sample, we select the galaxy in DES-COSMOS
ables that were not considered up to this point. Such as the quality
whose resampled flux returns a minimum χ2 when compared to the
of photometric redshift errors, analysis and mitigation of systemat-
DES-Y1 BAO flux (the χ2 combines all bands, g, r, i and z). This
ics, use of the full covariance and optimized BAO extraction meth-
is done for every galaxy in the DES-Y1 BAO sample to make up
ods.
the ‘COSMOS-Validation’ catalog, which by construction has col-
Nonetheless we stress that the forecasted error obtained in this
ors matching those in the DES-Y1 BAO sample. The “true” redshift
section matches the one from the analysis of mock simulations, see
is retrieved from the spectroscopic/30-band photo-z of this match.
e.g. DES-BAO-θ-METHOD, and is in fact quite close to the final
We then run the DNF photo-z code over the COSMOS-
BAO error obtained in DES-BAO-MAIN. In the following sections
Validation catalog to select 4 redshift bin samples in the same way
we discuss the various components that will enter the real data anal-
as we did for the full DES-Y1 BAO sample. We use the “true” red-
ysis, starting with the validation of photometric redsfhit errors and
shifts from the COSMOS-Validation catalogs to estimate the N (z)
the estimate of redshift distributions.
in each redshift bin by normalising the histogram of these true red-
shifts.
Results are shown as histograms in Figure 7, which are com-
4 PHOTOMETRIC REDSHIFTS pared to the stack N (z) from the photo-z code, for reference. The
black histograms show large fluctuations which are caused by real
The photometric redshifts used for redshift binning and transverse
individual large scale structures in the COSMOS field. This can
distance computations in our fiducial analyses are derived using the
be seen by visual inspection of the maps. This sampling variance
Directional Neighborhood Fitting (DNF) algorithm (De Vicente,
comes from the relatively small size of the COSMOS validation re-
Sánchez & Sevilla-Noarbe 2016), which is trained with public
gion. There is also a shot-noise component, indicated by the error
spectroscopic samples as detailed in Hoyle et al. (2017). For com-
bars over the black dots, but it is smaller. In the next section, we
parison we also discuss below the Bayesian Photometric Redshift
briefly describe the methodology to correct for this to be able to
(BPZ) (Benı́tez 2000) which we find slightly less performant in
make use of this validation sample effectively.
terms of the error with respect to “true” redshift values (see below).
In both cases we use MOF photometry which provides ∼ 10−20%
more accurate photo-z estimates with respect to the equivalent esti-
4.2 Sample variance correction
mates using SExtractor MAG AUTO quantities from coadd photom-
etry. In this section we summarise the steps taken to arrive at these As detailed in DES-BAO-PHOTOZ we apply a sampling variance
choices, based on a validation against data over the COSMOS field. correction (SVC) to the data and test this method with the Halo-
We recall that throughout this work we use the individual ob- gen mocks described in DES-BAO-MOCKS. In what follows we
ject’s mean photo-z from BPZ (not to be confused with the mean provide a summary of such process and its main results.
value z̄ =< z > of the sample) and the predicted value in the We use the VIPERS catalog (Scodeggio et al. 2016), which
fitted hyper-plane from the DNF code, as our point estimate for spans 24 square degrees to i < 22.5, to estimate the sampling
galaxy redshifts. As for the estimates of the N (z) from the photo-z variance effects in the above COSMOS validation. After correct-
codes, for comparison with our fiducial choice based on the COS- ing VIPERS for target, color and spectroscopic incompleteness we
MOS narrow band p(z), we will use the stacking of Monte Carlo select galaxies in a similar way as done in section 3. We then use the
realisations of the posterior redshift distributions p(z) for the BPZ VIPERS redshifts to estimate the true N (z) distribution of the par-

MNRAS 000, 1–16 (2017)

 Galaxy sample for DES Y1 BAO measurements 9

MOF-DNF 0.6<Z<0.7 < zstack > =0.657 MOF-DNF 0.7<Z<0.8 < zstack > =0.758
17.5 COSMOS SVC W68=0.046 z=0.005 68=0.023 14 COSMOS SVC W68=0.057 z=0.019 68=0.028
DES-Y1 stack W68=0.048 68=0.022 DES-Y1 stack W68=0.061 68=0.028
COSMOS raw W68=0.047 z=0.0 68=0.022 COSMOS raw W68=0.068 z=0.019 68=0.029
15.0 12
12.5
N(z) (Normalized dN/dz)

N(z) (Normalized dN/dz)

10
10.0 8

7.5 6

5.0 4

2.5 2

0.0 0
0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.6 0.7 0.8 0.9 1.0
True redshift z True redshift z

MOF-DNF 0.8<Z<0.9 < zstack > =0.852 MOF-DNF 0.9<Z<1.0 < zstack > =0.934
COSMOS SVC W68=0.057 z=0.008 68=0.029 COSMOS SVC W68=0.072 z=0.008 68=0.036
14 DES-Y1 stack W68=0.064 68=0.03 12 DES-Y1 stack W68=0.078 68=0.037
COSMOS raw W68=0.06 z=0.002 68=0.027 COSMOS raw W68=0.067 z=0.009 68=0.033
12 10
N(z) (Normalized dN/dz)

N(z) (Normalized dN/dz)

10
8
8
6
6
4
4

2 2

0 0
0.6 0.7 0.8 0.9 1.0 1.1 0.7 0.8 0.9 1.0 1.1 1.2
True redshift z True redshift z

Figure 7. Normalised redshift distributions for our different tomographic bins of DNF-MOF photo-z. Stack N (z) are shown for the full DES-Y1 BAO sample
(yellow histograms). The black histogram (with Poisson error bars) shows the raw 30-band photo-z from the COSMOS-DES validation sample. Magenta lines
shows the same sample corrected by sample variance cancellation (SVC, see text), which is our fiducial estimate. The labels show the values of W68 , σ68 and
∆z =< zstack > − < z > and in each case, see also Table 3.

ent DES-COSMOS sample (before we select in photometric red- 4.3 Photo-z validation results
shifts). The ratio of the N (z) in the DES-COSMOS sample to the
one in VIPERS gives a sample variance correction that needs to be In Table 3 we show the values of σ68 , which corresponds to the
applied to the N (z) in each of the tomographic bins. 68% interval of values in the distribution of (zphoto − ztrue )/(1 +
ztrue ) around its median value, where zphoto is the photo-z from
Figure 7 shows the SVC-corrected version of the raw COS- DNF (zmean above) and ztrue is the redshift from the COSMOS
MOS catalog in magenta. As shown in this figure the resulting validation sample corrected by SVC. We also show W68 and z̄
distribution is much smoother than the original raw measurements which are the 68% interval and mean redshift in the ztrue distribu-
(black histograms). This by itself indicates that SVC is working tion for each redshift bin. The corresponding values for the stack
well. Tests in simulations show that this SVC method is unbiased N (z) and raw N (z) are also shown in the labels of Figure 7. ∆z
and reduces the errors in the mean and variance of the N (z) distri- in the label inset shows the difference ∆z =< zstack > − < z >,
bution by up to a factor of two. Similar results are found for differ- where < zstack > is the mean stack redshifts for DES-Y1, shown
ent binnings in redshift. in the top label.
We have performed an extensive a comparison of the quan-
Notably, the distributions obtained from the stacked N(z) and tities shown in Table 3 computed with different validations sets:
the ones from COSMOS SVC match well overall, although some DES-COSMOS with and without SVC, using N (z) from DNF
discrepancies can be seen, e.g. for the second and fourth bin. stacks, using the COSMOS subsample with spectroscopic redshifts
More quantitative statements are provided below, but in DES-BAO- (as opposed to that with 30-band photo-z). We have also compared
MAIN (Table 5, entry denoted “w(θ) z uncal”) we show these have these N (z) to the one predicted by subset galaxies that have spec-
no impact in our cosmological results. The difference in angular di- tra within the BAO sample over full DES-Y1 footprint. Further-
ameter distance measurements when using either of these two sets more we have performed a validation using a larger spectroscopic
of redshift distributions is less than ∼ 0.25σ. sample in the VIPERS/W4 field (∼ 4 square degrees) which was

MNRAS 000, 1–16 (2017)

10 Crocce et al.
observed in DESY1 and is completely independent from the COS- calibration and star-galaxy separation tests (COSMOS and VVDS-
MOS validation 3 . The results from these different validation sets 14h), which do not contribute to our clustering signal at BAO
is that the means of the redshift distributions hzi (w.r.t to the mean scales (they total 30 deg2 ).
using the stack N (z)) are always within 0.01 except for the sec-
ond tomographic bin where differences are < 0.02 (see also la- • Pixelized maps of the survey coverage fraction were created
bels of Figure 7). The values of W68 are always within 0.01 as at a H EALPIX resolution of Nside = 4096 (area = 0.73 arcmin2 )
well, for all bins. This means that the differences in W68 are within by calculating the fraction of high resolution subpixels (Nside =
15% − 20% (depending on redshift) and hzi is within 1% (2% 32768, area = 0.01 arcmin2 ) that were contained within the original
for the bin [0.7 − 0.8]). In Sec. 4.3 of DES-BAO-θ-METHOD we mangle mask (see Drlica-Wagner et al. (2017) for a description
investigate the impact in derived BAO angular diameter distances of the later). Since our color selection requires observations in all
from systematic errors in the mean and variance of the underly- four griz bands we use the coverage maps to enforce that all pixels
ing redshift distributions. The most important quantity is the mean considered, at resolution 4096, show at least 80% coverage in
of dn/dz. The level of shifts discussed above would induce about each band (this removes 70.7 deg2 with respect to the case where
0.8% systematic error in θBAO , while 20% in the variance would no miminum coverage is required). Furthermore we then use the
have no impact. These are small compared to the statistical errors, minimum coverage across all four bands to down-weight the given
see DES-BAO-MAIN. The validation errors and biases in hzi, σ68 pixel when generating random distributions, see Sec. 7.
and W68 were also studied and we anticipate that they are subdom-
inant for the BAO analysis, which instead is dominated by the lim- • In order to match the global magnitude cut of the sample and
ited size of the DES Y1 footprint. These results will be presented ensure it is complete across our analysis footprint, we select re-
more extensively in DES-BAO-PHOTOZ. gions with 10σ limiting depth of iauto > 22, where the depths are
We also include in that work a comparison with BPZ photo- calculated according to the procedure presented in Drlica-Wagner
z (see also Table 3) and results for different photo-z with coadd et al. (2017).
photometry. The values of W68 and σ68 are always smaller (by 10-
20%) for DNF with MOF photometry, which is therefore used as • Since we want to reliably impose the color cut defined in
our fiducial photo-z sample. Eq. (2) and Table 3, we consider only areas with limiting depth
We finish the section by stressing that the fiducial N (z) used in the corresponding bands large enough to measure it. Given
in the main BAO analysis are the ones from DES-COSMOS with that we are already imposing iauto depth greater than 22, the
SVC (magenta lines in Figure 7). new condition implies keeping only the regions with 10σ limit-
ing magnitudes (2 rauto − zauto ) < 23.7, or equivalently those
with zauto > 2 rauto − 23.7. This removes an additional 53.8 deg2 .
5 ANGULAR MASK
• As a result of our analysis of observational systematics in
We build our mask as a combination of thresholds/constraints on Sec. 6, we identify that galaxy number density in regions of high
basic survey observation properties, conditions due to our particu- z-band seeing shows an anomalous behaviour. To isolate this out
lar sample selection, and restrictions to avoid potential clustering we remove areas with z-band seeing greater than 1 arc-second
systematics. In summary, (this amounts to 71 deg2 , or 5% of the footprint).
• We start by combining the Y1GOLD Footprint and Bad
regions mask, both of which are described in Drlica-Wagner • Lastly we also remove a patch of 18 deg2 over which the
et al. (2017). The Footprint mask imposes minimum total airmass computation was corrupted.
exposure times, valid stellar locus regression4 (SLR) calibration
solutions and basic coverage fractions. The Bad Regions The resulting footprint occupies 1336 deg2 and is shown in Fig. 1.
mask removes at different levels various catalog artifacts, regions
around bright stars and large foreground objects. In particular,
for the later we remove everything with flag bit > 2 in Table 5
of Drlica-Wagner et al. (2017), corresponding to regions around 6 MITIGATION OF OBSERVATIONAL SYSTEMATIC
bright starts in the 2MASS catalogue (Skrutskie et al. 2006). EFFECTS
We have tested for observational systematics in a manner similar to
• We introduce coordinate cuts to select only the wide area Elvin-Poole et al. (2017), which builds upon work in DES Science
parts of the surveys, namely those overlapping SPT (roughly with Verification Data (Crocce et al. 2016) and other surveys (e.g. Ross
300 < RA(deg) < 99.6 and −40 < DEC(deg) < −60) and et al. (2011a); Ho et al. (2012)).
S82 (with 317.5 < RA(deg) < 360 and −1.76 < DEC(deg) < Generically, we test the dependence of the galaxy density
1.79). This removes small and disjoint regions which are part of against survey properties (SPs). We expect there to be no depen-
the Supernova survey and two auxiliary fields used for photo-z dence if SPs do not introduce density fluctuations in our sample
beyond those already accounted for by the masking process. We
3 The completeness of the VIPERS sample depends on galaxy type and has have used the same set of SP maps as in Elvin-Poole et al. (2017),
a color preseleccion to exclude galaxies at z < 0.5. We have included all namely :
the suggested incompleteness factors (Scodeggio et al. 2016), but nonethe-
less have decided to use COSMOS-SVC as our fiducial validation set to
• 10σ limiting depth in band
avoid potential residuals. • full width half maximum of point sources (“seeing”)
4 This is a complementary calibration technique used for the construction • total exposure time
of Y1GOLD making use of the distinct color locus occupied by stars to • total sky brightness,
perform relative additional calibration between bands. • atmospheric airmass,

MNRAS 000, 1–16 (2017)

 Galaxy sample for DES Y1 BAO measurements 11
all of them in each of the four bands griz, in addition to Galac-
tic extinction and stellar contamination (refer to Elvin-Poole et al.
(2017) for a detailed explanation on how the stellar density map is
1.15 0.6 < z < 0.7
constructed from Y1GOLD data). We find that the relevant system- 0.7 < z < 0.8
atics are stellar density, PSF FWHM, and the image depth. We out- 1.10 0.8 < z < 0.9
0.9 < z < 1.0
line the tests that reveal this and how we apply weights to counter
1.05
their effect in what follows.

ngal /hngal i
We found the most important systematic effect, in terms of 1.00
its impact on the measured clustering, to be the stellar density. In 0.95
the top panel of Fig. 8 we find positive trends when comparing the
number density of our ‘galaxy’ sample as a function of the stellar 0.90
number density (nstar ). Our interpretation is that there are stars in 0.85
our sample. Assuming these contaminating stars follow the same
spatial distribution as the stars we use to create our stellar density 0.80
map, this stellar contamination will produce a linear relationship 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
between the density of our galaxy sample and the stellar density. In stellar density (arcmin−2 )
this scenario, the value of the best-fit trend where the number den-
sity of stars, nstar , is 0 is then the purity of the sample. We find the
results are indeed consistent with a linear relationship, as illustrated
in the top panel of Fig. 8. The stellar contamination, fstar , that can 1.10
0.6 < z < 1.0
be determined from these plots is listed in Table 3. The stellar con-
tamination varies significantly with redshift, as expected given the 1.05
proximity of the stellar locus to the red sequence as a function of ngal /hngal i
redshift. Thus, we measure the stellar contamination in ∆z = 0.05 1.00
bin widths and use a cubic spline interpolation in order to obtain the
stellar contamination at any given redshift. This allows us to assign
0.95
a weight to each galaxy given by,
w(fstar (z)) = ((1 − fstar (z)) + nstar fstar (z)/hnstar i)−1 , (4) 0.90

where nstar is the stellar density that depends on angular location

and hnstar i is the mean stellar density over the DES-Y1 footprint. 0.8 0.9 1.0 1.1 1.2
Note that we repeat the fitting procedure for each photo-z cata- mean i-band seeing (arcseconds)
logue, hence redshift here means either zDNF−MOF or zBPZ−MOF .
From Fig. 8 it seems that the measurements are a bit noisy. However
this procedure helps us resolve the peak in the stellar contamination 1.15
of five per cent at z ∼ 0.78. The uncertainty on each fit is ∼ 0.01,
which is consistent with the scatter we find in the values of fstar 1.10
per bin. The spline simply interpolates between the best-fit values.
1.05
We also add weights based on fits against relationships with
ngal /hngal i

the mean i-band PSF FWHM (seeing, which we denote as si ) and 1.00
the g-band depth (dg ). For the seeing, we do not find a strong de- 0.95
pendence on redshift and thus use the full sample to define the see-
ing dependent weight 0.90
−1
w(si ) = (As + Bs si ) , (5) 0.85

where As and Bs are simply the intercept and slope of the best- 0.80
fit linear relationship, shown in the middle panel of Fig. 8. The 22.00 22.25 22.50 22.75 23.00 23.25 23.50 23.75 24.00
coefficients we use are Ai = 0.782 and Bi = 0.0625. For the 10σ g-band depth limit (magnitudes)
g-band depth, we fit linear relationships in redshift bins ∆z = 0.1
and again use a cubic spline interpolation in order to obtain a weight Figure 8. The galaxy density vs. potential systematic relationship used to
at any redshift define weights that we apply to clustering measurements. Top panel: The
galaxy density versus stellar density in four photometric redshift bins. The
w(dg , z) = (C(z) + dg (1 − C(z))/hdg i)−1 , (6)
linear fits are used to determine the stellar contamination. The χ2 values for
where C(z) is the interpolated result for the value of the linear- the fits are 9.7, 10.0, 3.5, and 14.3 (8 degrees of freedom). Middle panel:
fit where dg = 0. The relationships as a function of redshift and The galaxy density versus the mean i-band seeing for our full sample. The
the linear best-fit models are shown in the bottom panel of Fig. 8. inverse linear fit is used to define weights applied to clustering measure-
ments. The χ2 is 7.7 (8 degrees of freedom) and the coefficients are 0.788
The total systematic weight, wsys , is thus multiplication of the three
and 0.0618. Bottom panel: The galaxy density versus g-band depth in four
weights
photometric redshift bins. The coefficients are interpolated as a function of
wsys = w(fstar (z))w(si )w(dg , z). (7) redshift and used to define weights to be used in the clustering measure-
ments. The χ2 values for the fits, given 8 degrees of freedom, are 7.7, 8.9,
The dependencies we find are purely empirical as we lack any 12.7, and 6.1. The slopes are (-0.0256, 0.0320, 0.103, 0.0609).

MNRAS 000, 1–16 (2017)

12 Crocce et al.
more fundamental understanding for how these correlations de-
velop. They must result from the complicated intersection of our
color/magnitude selection and the photometric redshift algorithm,
that are not perfectly captured by our mask. Besides the relations
with different observing properties (airmass, seeing, dust, exposure
time) are also very correlated what makes physical interpretation
very complicated.
In the following section, we test the impact of these weights on
the measured clustering, and determine their total potential impact.
In DES-BAO-MAIN , we show that the weights have minimal im-
pact on the BAO scale measurements and that our treatment is thus
sufficient for such measurements. Our treatment is not as compre-
hensive as Elvin-Poole et al. (2017), and thus further study might
be required when using the sample defined here for non-BAO ap-
plications.

7 TWO-POINT CLUSTERING
In this section we describe the basic two-point clustering properties
of the samples previously defined. We concentrate on large-scales
where the BAO signal resides, and the sample using zDNF−MOF
photometric redshifts which is the default one used in DES-BAO-
MAIN.
We compute the angular correlation function w(θ) of the sam-
ple, split into four redshift bins, using the standard Landy-Szalay
estimator (Landy & Szalay 1993),
DD(θ) − 2DR(θ) + RR(θ) Figure 9. Top panel shows the impact of the systematic weights on each
w(θ) = (8) redshift bin, shown by the differential angular correlations, with and with-
RR(θ)
out weights applied, relative to the uncertainty. One can see that the weights
as implemented in the CUTE software5 (Alonso 2012), where make the biggest difference for the 0.7 < z < 0.8 bin, which is the redshift
DD(θ), DR(θ) and RR(θ) refer to normalized pair-counts of range with the greatest stellar contamination. The thick solid line displays
Data (D) and Random (R) points, separated by an angular aper- the BAO feature in similar units, (wBAO − wno BAO )/σw , for the second
tomographic bin as an example (different bins show similar BAO strength
ture θ. Random points are uniformly distributed across the foot-
but displaced slightly in the angular coordinate). The systematic weights
print defined by our mask (albeit downsampled following the frac-
only modify the underlying smooth shape, and do not have a sharp feature
tional coverage of each pixel, described in Sec. 5), with an abun- at BAO scales. Bottom panel shows the ratio of correlations for each bin,
dance twenty times larger than that of the data in each given bin. which provides additional information on the absolute size of the correc-
For the fits and χ2 values quoted in this section we always con- tions (in this case we only plot up to scale with no zero crossings of w).
sider 16 angular-bins linearly spaced between θ = 0.45 deg and
θ = 4.95 deg, matching the scale cuts in the BAO analysis us-
ing w(θ) of DES-BAO-MAIN. We compute pair-counts in angular
aperture bins of width 0.3 deg in order to reduce the covariance BAO feature at this scales we also display in thick solid black line
between the measurements. The covariance matrix is derived from the theoretical angular correlation function with and without BAO,
1800 Halogen mocks, described in detail in DES-BAO-MOCKS. for the second tomographic bin for concreteness, relative to the sta-
The expected noise in the inverse covariance from the finite tistical errors. The corrections are all at the same level (or smaller)
number of realisations (Hartlap, Simon & Schneider 2007) and the than the expected BAO signal.
translation of that into the variance of derived parameters (Dodel- The weights have the largest impact in terms of clustering am-
son & Schneider 2013) is negligible given the size of our data vec- plitude for the redshift bin 0.7 < z < 0.8, which is the redshift
tor (16 angular measurements per tomographic redshift bin) and range with the largest stellar contamination (∼ 4%, see Table 3), al-
the number of model parameters (one bias per bin). For instance though never exceeding one σw . For the remaining bins the change
the increased error in derived best-fit biases
p in any given bin would in the correlation functions are within 1/4 of σw . We can assess
be sub-percent. The change in the full χ2 is ∼ 3.7% (16x4 data- quantitatively the total potential impact of the weights by calculat-
points, see the discussion below). We therefore neglect these cor- ing χ2sys = ∆w(θ)t C −1 ∆w(θ); the square-root of this number is
rections in this section. an upper bound in the impact, in terms of number of σ’s, that the
Figure 9 shows the impact of the systematic weights on the weights could have on the determination of any model parameter.
measured angular clustering in terms of the difference ∆w between In the range 0.45 deg < θ < 4.95 deg, with 16 data-points,
the pre-weighted correlation function w and the post-weighted one we find χ2sys = 0.1, 1.35, 0.2 and 0.5 respectively for each tomo-
wweighted , relative to the statistical error σw (i.e. neglecting all co- graphic bin separately (showing that for example best-fit bias de-
variance). To compare this against the expected amplitude of the rived solely from the 2nd tomographic bin can be shifted by more
than one sigma if weights are uncorrected for). More interestingly,
for the four bins combined and including the full covariance matrix,
5 https://github.com/damonge/CUTE we find χ2sys = 1.35. This implies a maximum impact of 1.16σ in

MNRAS 000, 1–16 (2017)

 Galaxy sample for DES Y1 BAO measurements 13

Figure 10. Angular correlation function in four redshift bins, for galaxies selected with zDNF−MOF . Symbols with error bars show the clustering of galaxy
sample corrected for the most relevant systematics. Dashed line displays a model using linear theory with an extra damping of the BAO feature due to
nonlinearities, and a linear bias fitted to the data (whose best fit value is reported in the inset labels). We consider 16 data-points and one fitting parameter in
each case (dof=15). Note that the points are very covariant, which might explain the visual mismatch in the first tomographic bin that nonetheless retains a
good χ2 /dof.

Figure 11. Angular cross-correlation functions of the four tomographic bins in 0.6 < zphoto < 1.0, see Fig. 10, for galaxies selected according
p to
zDNF−MOF . The model prediction shown with dashed lines assumes a bias equal to the geometric mean of the auto-correlation fits, i.e. bij = bi bj ,
and is basically proportional to the overlap of redshift distributions, which are shown in the bottom right panel.

MNRAS 000, 1–16 (2017)

14 Crocce et al.
a derived global parameter such as the angular diameter distance
measurement. This maximum threshold is well above the actual
impact of the weights in DA /rs found in DES-BAO-MAIN, which
is 0.125σDA /rs (see Table 5 in that reference). We consider this an
indication that the particular shape of the BAO feature is not easily
reproducible by contaminants, and is therefore largely insensitive to
such corrections, which is consistent with previous analyses (Ross
et al. 2017b).
Figure 10 displays the auto-correlation function (including ob-
servational systematic weights) of 4 tomographic bins of width
∆zphoto = 0.1 between 0.6 6 zphoto 6 1.0. Data at z > 0.8
appear to show significant BAO features. Best fit biases, derived
1σ errors and their corresponding χ2 values are reported as inset
panels and in Table 3. The model displayed assumes linear the-
ory and the MICE cosmology6 (Fosalba et al. 2015; Crocce et al.
2015), with an extra damping of the BAO feature, see DES-BAO-
θ-METHOD for details. The χ2 /dof are all of order ∼ 1 or better,
showing that these are indeed good fits given the covariance of the
data. In Table 3 we also report best fit bias values for a split of the Figure 12. Three-dimensional correlation function binned in projected pair
sample into four tomographic bins using the BPZMOF photo-z, separations. We use projected separations because radial pairs are damped
showing no discrepancies. due to photo-z mixing. The dashed line is the best fit model assuming linear
As a further test of the clustering signal, as well as the tails of bias and a smeared BAO feature, as discussed in detail in DES-BAO-MAIN.
the photo-z distributions, we show in Fig. 11 the cross-correlation
between different bins. The overploted models were derived using
the redshift distributions of the corresponding bins and assume a
bias equal to the geometric mean of the tomographic bins,
Z Z
wij (θ) = b2ij dz dz̃ni (z)nj (z̃)D(z)D(z̃)ξ(rθ ) (9)
2 × 3 and 3 × 4) are driven by the non-diagonal structure of the
where rθ2 = r(z)2 + r(z̃)2 − 2r(z)r(z̃) cos θ and b2ij = bi bj . covariance matrix rather than a mismatch between the best-fit bias
In Eq. (9) we denote ξ the spatial correlation function computed of the cross-correlation bij compared to the geometrical mean of
in linear theory at z = 0. The error bar displayed and the re-
√ for 2 × 3 the best-fit bias
the auto-correlation biases. For example,
ported χ2 values are obtained with a theoretical covariance ma- from w2×3 is only 2% larger than b2 b3 (and the corresponding
trix designed to match the Halogen mocks covariance of the auto- χ2 change sub-percent). On the other hand, the χ2 of the cross-
correlations (i.e. matching the bias and shot noise and area of correlation drops to 0.4 if we only consider a diagonal covariance
the mocks). Detailed formulae and tests of this theory covariance matrix. Similarly χ23×4 drops to 1.28 from 2 using a diagonal co-
are given in a companion paper, DES-BAO-θ-METHOD (see also variance matrix. Overall, we conclude there is a fairly good match
Crocce, Cabré & Gaztañaga (2011); Ross et al. (2011b); Salazar- between the implications of the overlap of redshift distributions and
Albornoz et al. (2014)). However when we test the χ2 values of the cross-correlation clustering signal.
the auto-correlations against the best-fit model7 using this theory In Figure 12 we show ξ(sperp ) which is the three-dimensional
covariance instead of the one derived from the mocks we find con- correlation function binned only in projected physical separations.
siderably larger χ2 values: ri ≡ χ2i,theory−cov /χ2i,mocks−cov = To compute this correlation we converted (photometric) redshift
1.46, 1.37, 1.37, 1.47 for auto-correlations in bin i = 1 to 4, re- and angles to physical distances assuming MICE cosmology. This
spectively. We propagate this uncertainty to the cross-correlations yields a three-dimensional map of the galaxies in comoving coordi-
√
by dividing χ2ij,theory−cov by ri rj . nates. Random points are distributed in this volume with the same
Overall the cross-correlations show a good match to the angular distribution as the angular mask defined in section 5, and
model, which is sensitive to the tails of the redshift distributions and used for w(θ), and drawing redshifts randomly from the galaxies
the geometric mean bias. The χ2 /dof are ∼ 1. The non-adjacent themselves. Pair counts are then computed and binned in projected
bin 1 × 3 (where the expected clustering signal is negligible) shows separations. A full detail of such procedure is given in DES-BAO-
an excess correlation on very large-scales. This most probably in- MAIN as well as in Ross et al. (2017a). The modeling displayed in
dicate a residual systematic and not a problem of the photo-z dis- Fig. 12 projects the real space three-dimensional correlation func-
tributions. tion into photometric space assuming Gaussian photometric red-
The large χ2 values in some of the cross-correlations (bins shift errors per galaxy, provided in Table 3 as σ68 . It also assumes
a linear bias betweeen the galaxies and the matter field.
6
The bias recovered from the three-dimensional projected clus-
We make this choice throughout the DES-Y1 BAO analysis because the tering at a mean redshift of 0.8 is b = 1.83 ± 0.06, consistent with
MICE N-Body simulation was used to calibrate the Halogen mock galaxy
the one from w(θ) tomography. In addition we stress that this clus-
catalogues. MICE cosmology assumes a flat concordance LCDM model
with Ωmatter = 0.25, Ωbaryon = 0.044, ns = 0.95, σ8 = 0.8 and
tering estimate includes all cross-correlations of the data. The fact
h = 0.7. that it is matched by the theory modeling, which in turn includes a
7 The best-fit bias and error from the theory covariance or the mocks one characterisation of the redshift distributions per galaxy, represents
are consistent with each other, however the χ2 values are only so to about also an additional consistency check of reliability of the photomet-
40%. ric redshifts.

MNRAS 000, 1–16 (2017)

 Galaxy sample for DES Y1 BAO measurements 15
8 CONCLUSIONS already yield a very interesting counter-part to the high precision
low-z BAO measurements already existing.
This paper describes the selection of a sample of galaxies, opti-
mised for BAO distance measurements, from the first year of DES
data. By construction, this sample is dominated by red and lumi-
nous galaxies with redshifts in the range 0.6 < z < 1.0. We have ACKNOWLEDGMENTS
extended the selection of red galaxies beyond that of previously
MC acknowledges support from the Spanish Ramon y Cajal
published imaging data used for similar goals in SDSS by Padman-
MICINN program. MC and EG have been partially funded by
abhan et al. (2005) to cover the higher redshift and deeper data
AYA2015-71825. AJR is grateful for support from the Ohio State
provided by DES.
University Center for Cosmology and AstroParticle Physics.. KCC
We compute the expected magnitudes of galaxy templates in
acknowledges the support from the Spanish Ministerio de Econo-
the four DES filters and identify the (i − z) and (z − i) color space
mia y Competitividad grant ESP2013-48274-C3-1-P and the Juan
to select red galaxies in the redshift range of interest. The actual
de la Cierva fellowship. This work has made use of CosmoHub, see
selection in color and magnitude is defined using the BAO dis-
Carretero et al. (2017). CosmoHub has been developed by the Port
tance measurement figure-of-merit as a guiding criteria. Remark-
d’Informació Cientı́fica (PIC), maintained through a collaboration
ably, the resulting forecast matches the results obtained in DES-
of the Institut de Fı́sica d’Altes Energies (IFAE) and the Centro
BAO-MAIN with the final analysis. The global flux limit of the
de Investigaciones Energéticas, Medioambientales y Tecnológicas
sample is iauto < 22, although we later introduce a sliding mag-
(CIEMAT), and was partially funded by the “Plan Estatal de In-
nitude cut to limit ourselves to brighter objects towards lower red-
vestigación Cientfica y Técnica y de Innovación” program of the
shifts.
Spanish government.
We consider three different photo-z catalogues, with two dif- We are grateful for the extraordinary contributions of our
ferent photometric determinations. We showed that the typical CTIO colleagues and the DECam Construction, Commissioning
photo-z uncertainty (in units of 1 + z) goes from 2.3% to 3.6% and Science Verification teams in achieving the excellent instru-
from low to high redshift, for DNF redshifts using MOF photom- ment and telescope conditions that have made this work possible.
etry, and slightly worse for BPZ with MOF photometry. Hence The success of this project also relies critically on the expertise and
the former constitutes our primary catalogue in DES-BAO-MAIN, dedication of the DES Data Management group.
while the later is used for consistency. Redshift estimations based Funding for the DES Projects has been provided by the
on COADD photometry turned out to be worse than those derived U.S. Department of Energy, the U.S. National Science Founda-
from MOF photometry by 10%−20%. Our final sample is made of tion, the Ministry of Science and Education of Spain, the Sci-
1.3 million red galaxies across 1336 deg2 of area, largely contained ence and Technology Facilities Council of the United Kingdom, the
in one compact region (SPT). Higher Education Funding Council for England, the National Cen-
We study and mitigate, when needed, observational system- ter for Supercomputing Applications at the University of Illinois at
atics traced by various survey property maps. Of these, the most Urbana-Champaign, the Kavli Institute of Cosmological Physics at
impactful is the stellar contamination, which we find nonetheless the University of Chicago, the Center for Cosmology and Astro-
bound to < 4%. Also i-band mean seeing and g-band depth are Particle Physics at the Ohio State University, the Mitchell Institute
relevant. We define weights to be applied to the galaxies when for Fundamental Physics and Astronomy at Texas A&M Univer-
computing pair counting to remove the relations between galaxy sity, Financiadora de Estudos e Projetos, Fundação Carlos Chagas
number density and large scale fluctuations in those survey prop- Filho de Amparo à Pesquisa do Estado do Rio de Janeiro, Con-
erties. We show that none of these corrections have an impact on selho Nacional de Desenvolvimento Cientı́fico e Tecnológico and
BAO measurements, mainly because they can eventually modify the Ministério da Ciência, Tecnologia e Inovação, the Deutsche
the broad-shape of the correlation functions but do not introduce a Forschungsgemeinschaft and the Collaborating Institutions in the
characteristic localised scale as the BAO. Dark Energy Survey.
Lastly we characterised the two-point clustering of the sam- The Collaborating Institutions are Argonne National Labora-
ple, which is then used in DES-BAO-MAIN to derived distance tory, the University of California at Santa Cruz, the University of
constraints. We find the auto-correlations to be consistent with a Cambridge, Centro de Investigaciones Energéticas, Medioambien-
bias that evolves only slightly with redshift, from 1.8 to 2. The bias tales y Tecnológicas-Madrid, the University of Chicago, Univer-
derived from the tomographic analysis is consistent with the one fit- sity College London, the DES-Brazil Consortium, the University
ted to the whole sample range with the 3D projected distance anal- of Edinburgh, the Eidgenössische Technische Hochschule (ETH)
ysis. Furthermore we investigate the cross-correlation between all Zürich, Fermi National Accelerator Laboratory, the University of
the tomographic bins finding clustering amplitudes matching ex- Illinois at Urbana-Champaign, the Institut de Ciències de l’Espai
pectactions, although with poor χ2 -values in some cases. Overall (IEEC/CSIC), the Institut de Fı́sica d’Altes Energies, Lawrence
this is a further test of the assumed redshift distributions. Berkeley National Laboratory, the Ludwig-Maximilians Univer-
This paper serves the purpose of enabling for the fist time BAO sität München and the associated Excellence Cluster Universe, the
distance measurements using photometric data to redshifts z ∼ 1. University of Michigan, the National Optical Astronomy Observa-
These measurements achieve a precision comparable to those con- tory, the University of Nottingham, The Ohio State University, the
sidered state-of-the-art using photometric redshift to this point (Seo University of Pennsylvania, the University of Portsmouth, SLAC
et al. 2012), as well as those from WiggleZ (Blake et al. 2011), National Accelerator Laboratory, Stanford University, the Univer-
which are both limited to z ∼ 0.65. These BAO results are pre- sity of Sussex, Texas A&M University, and the OzDES Member-
setend in detail in DES-BAO-MAIN. While this paper was com- ship Consortium.
pleted, the third year of DES data was made available to the collab- Based in part on observations at Cerro Tololo Inter-American
oration, totalling 3 to 4 times the area presented here, and similar or Observatory, National Optical Astronomy Observatory, which is
better depth. Hence we look forward to that analysis, which should operated by the Association of Universities for Research in As-

MNRAS 000, 1–16 (2017)

16 Crocce et al.
tronomy (AURA) under a cooperative agreement with the National Drlica-Wagner A. et al., 2017, ArXiv e-prints 1708.01531
Science Foundation. Eisenstein D. J. et al., 2005, ApJ, 633, 560
The DES data management system is supported by the Na- Elvin-Poole J. et al., 2017, ArXiv e-prints 1708.01536
tional Science Foundation under Grant Numbers AST-1138766 Erben T. et al., 2013, MNRAS, 433, 2545
Flaugher B. et al., 2015, AJ, 150, 150
and AST-1536171. The DES participants from Spanish institu-
Fosalba P., Crocce M., Gaztañaga E., Castander F. J., 2015, MNRAS, 448,
tions are partially supported by MINECO under grants AYA2015-
2987
71825, ESP2015-66861, FPA2015-68048, SEV-2016-0588, SEV- Gaztañaga et al., 2018, in prep.: DES BAO PHOTOZ
2016-0597, and MDM-2015-0509, some of which include ERDF Giavalisco M. et al., 2004, ApJL, 600, L93
funds from the European Union. IFAE is partially funded by the Górski K. M., Hivon E., Banday A. J., Wandelt B. D., Hansen F. K., Rei-
CERCA program of the Generalitat de Catalunya. Research leading necke M., Bartelmann M., 2005, ApJ, 622, 759
to these results has received funding from the European Research Hartlap J., Simon P., Schneider P., 2007, A&A, 464, 399
Council under the European Union’s Seventh Framework Pro- Ho S. et al., 2012, ApJ, 761, 14
gram (FP7/2007-2013) including ERC grant agreements 240672, Hoyle B. et al., 2017, ArXiv e-prints 1708.01532
291329, and 306478. We acknowledge support from the Australian Kinney A. L., Calzetti D., Bohlin R. C., McQuade K., Storchi-Bergmann
T., Schmitt H. R., 1996, ApJ, 467, 38
Research Council Centre of Excellence for All-sky Astrophysics
Laigle C. et al., 2016, ApJS, 224, 24
(CAASTRO), through project number CE110001020.
Landy S. D., Szalay A. S., 1993, ApJ, 412, 64
This manuscript has been authored by Fermi Research Al- Le Fèvre O. et al., 2005, A&A, 439, 845
liance, LLC under Contract No. DE-AC02-07CH11359 with the Leauthaud A. et al., 2007, ApJS, 172, 219
U.S. Department of Energy, Office of Science, Office of High En- Mohr J. J. et al., 2012, in Proc. SPIE, Vol. 8451, Software and Cyberinfras-
ergy Physics. The United States Government retains and the pub- tructure for Astronomy II, p. 84510D
lisher, by accepting the article for publication, acknowledges that Morganson E. et al., 2018, PASP, 130, 074501
the United States Government retains a non-exclusive, paid-up, ir- Padmanabhan N. et al., 2005, MNRAS, 359, 237
revocable, world-wide license to publish or reproduce the published Padmanabhan N. et al., 2007, MNRAS, 378, 852
form of this manuscript, or allow others to do so, for United States Percival W. J. et al., 2001, MNRAS, 327, 1297
Percival W. J. et al., 2010, MNRAS, 401, 2148
Government purposes.
Planck Collaboration et al., 2016, A&A, 594, A13
This paper has gone through internal review by the DES col-
Ross A. J. et al., 2017a, MNRAS, 472, 4456: DES BAO s⊥ METHOD
laboration. The DES publication number for this article is DES- Ross A. J. et al., 2017b, MNRAS, 464, 1168
2017-0305. The Fermilab pre-print number is FERMILAB-PUB- Ross A. J. et al., 2011a, MNRAS, 417, 1350
17-585. Ross A. J., Percival W. J., Crocce M., Cabré A., Gaztañaga E., 2011b, MN-
RAS, 415, 2193
Ross A. J., Samushia L., Howlett C., Percival W. J., Burden A., Manera M.,
2015, MNRAS, 449, 835
REFERENCES Rozo E. et al., 2016, MNRAS, 461, 1431
Alam S. et al., 2017, MNRAS, 470, 2617 Salazar-Albornoz S., Sánchez A. G., Padilla N. D., Baugh C. M., 2014,
Alonso D., 2012, ArXiv e-prints 1210.1833 MNRAS, 443, 3612
Annis J. et al., 2014, ApJ, 794, 120 Sánchez E. et al., 2011, MNRAS, 411, 277
Ata M. et al., 2018, MNRAS, 473, 4773 Scodeggio M. et al., 2016, ArXiv e-prints 1611.07048
Avila S. et al., 2018, MNRAS, 479, 94: DES BAO MOCKS Seo H.-J., Eisenstein D. J., 2003, ApJ, 598, 720
Bautista J. E. et al., 2017, A&A, 603, A12 Seo H.-J., Eisenstein D. J., 2007, ApJ, 665, 14
Benı́tez N., 2000, ApJ, 536, 571 Seo H.-J. et al., 2012, ApJ, 761, 13
Bertin E., Arnouts S., 1996, A&AS, 117, 393 Sevilla I. et al., 2011, ArXiv e-prints 1109.6741
Beutler F. et al., 2011, MNRAS, 416, 3017 Sevilla-Noarbe I. et al., 2018, ArXiv e-prints 1805.02427
Beutler F. et al., 2017, MNRAS, 464, 3409 Skrutskie M. F. et al., 2006, AJ, 131, 1163
Blake C., Bridle S., 2005, MNRAS, 363, 1329 Szalay A. S., Connolly A. J., Szokoly G. P., 1999, AJ, 117, 68
Blake C. et al., 2011, MNRAS, 415, 2892 The Dark Energy Survey Collaboration et al., 2017, ArXiv e-prints
Camacho H. et al., 2018, ArXiv e-prints 1807.10163: DES BAO ` 1712.06209: DES-BAO-MAIN
METHOD
Carlstrom J. E. et al., 2011, PASP, 123, 568
Carnero A., Sánchez E., Crocce M., Cabré A., Gaztañaga E., 2012, MN-
RAS, 419, 1689
Carretero J., et al., 2017, PoS, EPS-HEP2017, 488
Chan K. C. et al., 2018, MNRAS, 480, 3031: DES BAO θ METHOD
Cole S. et al., 2005, MNRAS, 362, 505
Coleman G. D., Wu C.-C., Weedman D. W., 1980, ApJS, 43, 393
Crocce M., Cabré A., Gaztañaga E., 2011, MNRAS, 414, 329
Crocce M. et al., 2016, MNRAS, 455, 4301
Crocce M., Castander F. J., Gaztañaga E., Fosalba P., Carretero J., 2015,
MNRAS, 453, 1513
Crocce M., Gaztañaga E., Cabré A., Carnero A., Sánchez E., 2011, MN-
RAS, 417, 2577
Cuesta A. J. et al., 2016, MNRAS, 457, 1770
de Simoni F. et al., 2013, MNRAS, 435, 3017
De Vicente J., Sánchez E., Sevilla-Noarbe I., 2016, MNRAS, 459, 3078
Delubac T. et al., 2015, A&A, 574, A59
Dodelson S., Schneider M. D., 2013, PRD, 88, 063537

MNRAS 000, 1–16 (2017)