A&A 653, A107 (2021)
https://doi.org/10.1051/0004-6361/202140854
Astronomy
&
Astrophysics
c ESO 2021
Preparing for LSST data
Estimating the physical properties of z < 2.5 main-sequence galaxies
G. Riccio1 , K. Małek1,2 , A. Nanni1,2 , M. Boquien4 , V. Buat2 , D. Burgarella2 , D. Donevski3 , M. Hamed1 , P. Hurley6 ,
R. Shirley6,7 , and A. Pollo1,5
1
2
3
4
5
6
7
National Centre for Nuclear Research, ul. Pasteura 7, 02-093 Warszawa, Poland
e-mail: gabriele.riccio@ncbj.gov.pl
Aix Marseille Univ. CNRS, CNES, LAM, Marseille, France
SISSA, Via Bonomea 265, Trieste, Italy
Centro de Astronomía (CITEVA), Universidad de Antofagasta, Avenida Angamos 601, Antofagasta, Chile
Astronomical Observatory of the Jagiellonian University, ul. Orla 171, 30-244 Cracow, Poland
Astronomy Centre, Department of Physics and Astronomy, University of Sussex, Falmer, Brighton BN1 9QH, UK
Instituto de Astrofísica de Canarias, 38205 La Laguna, Tenerife, Spain
Received 22 March 2021 / Accepted 18 June 2021
ABSTRACT
Aims. We study how the upcoming Legacy Survey of Space and Time (LSST) data from the Vera C. Rubin Observatory can be
employed to constrain the physical properties of normal star-forming galaxies (main-sequence galaxies). Because the majority of the
observed LSST objects will have no auxiliary data, we use simulated LSST data and existing real observations to test the reliability
of estimates of the physical properties of galaxies, such as their star formation rate (SFR), stellar mass (Mstar ), and dust luminosity
(Ldust ). We focus on normal star-forming galaxies because they form the majority of the galaxy population in the universe and are
therefore more likely to be observed with the LSST.
Methods. We performed a simulation of LSST observations and uncertainties of 50 385 real galaxies within the redshift range 0 < z <
2.5. In order to achieve this goal, we used the unique multi-wavelength data from the Herschel Extragalactic Legacy Project (HELP)
survey. Our analysis focused on two fields, ELAIS N1 and COSMOS. To obtain the physical properties of the galaxies, we fit their
spectral energy distributions (SEDs) using the Code Investigating GALaxy Emission. We simulated the LSST data by convolving the
SEDs fitted by employing the multi-wavelength observations. We compared the main galaxy physical properties, such as SFR, Mstar ,
and Ldust obtained from the fit of the observed multi-wavelength photometry of galaxies (from the UV to the far-IR) to those obtained
from the simulated LSST optical measurements alone.
Results. We present the catalogue of simulated LSST observations for 23 291 main-sequence galaxies in the ELAIS N1 field and for
9093 galaxies in the COSMOS field. It is available in the HELP virtual observatory. The stellar masses estimated based on the LSST
measurements agree with the full UV to far-IR SED estimates because they mainly depend on the UV and optical emission, which is
well covered by LSST in the considered redshift range. Instead, we obtain a clear overestimate of the dust-related properties (SFR,
Ldust , Mstar ) estimated with the LSST alone. They are highly correlated with redshift. We investigate the cause of this overestimate and
conclude that it is related to an overestimate of the dust attenuation in both UV and near-IR. We find that it is necessary to employ
auxiliary rest-frame mid-IR observations, simulated UV observations, or the far-UV attenuation (AFUV)-Mstar relation to correct for
the overestimate. We also deliver the correction formula log10 (SFRLSST /SFRreal ) = 0.26 · z2 − 0.94 · z + 0.87. It is based on the 32 384
MS galaxies detected with Herschel.
Key words. galaxies: fundamental parameters – galaxies: photometry – infrared: galaxies – galaxies: star formation – surveys
1. Introduction
In the past 20 years, the study of the multi-wavelength emission of galaxies from X-rays to radio was found to be necessary
to properly analyse the physical properties of galaxies. Because
the spectral energy distribution (SED) is the result of a complex interplay of several components, such as old and young
stars, stellar remnants, the interstellar medium, dust, and supermassive black holes (Walcher et al. 2011; Conroy 2013), only
the panchromatic view of galaxies can give the full information about their physical properties. For example, the emission
from the hot interstellar medium, active galactic nuclei (AGN),
or stellar remnants can be observed in the X-ray band (Fabbiano
2006), while the emission of the dust heated by interstellar radiation can be observed in the mid- and far-IR band (Silva et al.
1998; Noll et al. 2009; da Cunha et al. 2010; Hao et al. 2011;
Calzetti et al. 2012; Schreiber et al. 2018; Leja et al. 2018). To
fully comprehend the interactions between these parts, the simultaneous use of different spectral ranges is needed. As broad-band
photometry is much less expensive than spectroscopy in terms of
observation time, modelling the broad-band SED of galaxies has
become one of the most commonly employed methods to evaluate and constrain the physical properties. In this way, properties
such as the star formation rate (SFR) and stellar mass (Mstar ),
which are essential for a complete understanding of galaxy formation and evolution, can be evaluated.
However, modelling the SED can be an intricate problem
because galaxies with very different properties can look similar over some wavelength range: that is, a young dusty galaxy
can appear to be an old dust-free galaxy because they both look
Article published by EDP Sciences
A107, page 1 of 18
A&A 653, A107 (2021)
red in the optical. This is particularly the case when restricted
wavelength ranges, instead of the full SED, are considered. The
full SED is rarely available. This makes estimating the physical
properties with only a limited wavelength range a great challenge
for SED modelling.
In the literature (i.e. Kennicutt 1998; Le Floc’h et al. 2005;
Schreiber et al. 2015; Whitaker et al. 2017) it has been shown
that the ultraviolet (UV) to infrared (IR) SED contains important information about the star formation activity of galaxies. For
example, some knowledge about newborn stars can be directly
inferred from the UV band, making it a very efficient tracer
of the SFR. The region in which these stars are created, however, is highly obscured by dust, which makes them very difficult to observe. Dust is composed of carbonaceous and silicate
grains and absorbs part of the UV emission and re-emits it in the
IR band. For example, Fig. 10 of Buat et al. (2019) shows that
the total SFR of a galaxy is the sum of the SFR obtained from
UV/optical measurements and the SFR estimated from IR data.
Because the role played by dust is so important, it is fundamental for the SED fitting process to introduce attenuation laws that
describe how dust obscures the light emitted by the stars.
The attenuation law developed by Calzetti et al. (1994) for
nearby UV-bright starburst galaxies is by far the most commonly law used in literature. However, other laws such as those
proposed by Charlot & Fall (2000) and Lo Faro et al. (2017)
are widely employed in the SED fitting codes. Małek et al.
(2018) used a combination of UV and IR observations to
determine the best approach to fit SEDs of millions of galaxies from the Herschel Extragalactic Legacy Project (HELP)
across a wide redshift range (0 < z < 6) to obtain homogeneous estimates of the main physical properties. They found
that using three different attenuation laws, the estimate of stellar masses can change by a factor of 2 on average. Similar
results were found in a sample of Ultra-Luminous IR Galaxies
at z ∼ 2 for instance by Lo Faro et al. (2017), for galaxies obtained from the semi-analytic galaxy formation model
GALFORM by Mitchell et al. (2013), and by Burgarella et al.
(2013), who combined UV to IR measurements up to z = 3.6
to calculate the redshift evolution of the total SFR and dust
attenuation. They found that the attenuation increases up to
z = 1.2 and then decreases at higher redshift. The ratio of UV
and far-IR (FIR) emission also serves as an indicator of the
dust attenuation in galaxies (Buat et al. 2005; Takeuchi et al.
2005). All these factors require the combined usage of UV
and IR observations to provide a better understanding of the
star formation history (SFH), SFR, and dust attenuation properties of the galaxies. In order to perform the SED fitting of
galaxies, different methods and codes were developed, such as
STARLIGHT (Cid Fernandes et al. 2005), VESPA (Tojeiro et al.
2007), Hyperz (Bolzonella et al. 2000), Le Phare (Arnouts et al.
1999; Ilbert et al. 2006), PÉGASE.3 (Fioc & Rocca-Volmerange
2019), and COSMOS2020 (Weaver et al. 2021) together with
Bayesian SED fitting codes such as GOSSIP (Franzetti et al.
2008), PROSPECTOR (Leja et al. 2017), Code Investigating
GALaxy Emission (CIGALE; Noll et al. 2009; Boquien et al.
2019), and BayeSED (Han & Han 2014).
The main problem of the multi-wavelength fitting technique
is the lack of high-quality IR observations, which is due to instrumental sensitivity and the lower resolution of wide IR cameras or
extremely expensive sub-millimeter observations such as those
with ALMA. In contrast, a very large and high-quality coverage
of the optical part of the spectrum is usually available for widefield surveys and narrow and deep-field imaging with several
ground and space telescopes. With the upcoming Legacy SurA107, page 2 of 18
vey of Space and Time (LSST, Ivezić et al. 2019) from the Vera
C. Rubin Observatory, we will obtain even higher-quality optical
images in the ugrizy bands. The LSST survey will observe about
20 billion galaxies during ten years of observations. Most of
these galaxies will not have any counterpart in the available IR
catalogues. Moreover, IR astronomy often has blending issues,
which makes a precise match of optical and IR sources even
more difficult (Hurley et al. 2017; Pearson et al. 2018).
The LSST will be the largest (8.4 m of the primary mirror)
wide-field ground telescope designed to obtain repeated images
covering the sky that is visible from Cerro Pachón in Chile. The
survey will observe about 30 000 deg2 of the southern sky, covering the wavelength range 320–1050 nm. At the end of the ten-year
survey, it will reach a magnitude depth ∼27.5 in r band and similar
in the other bands. Considering the depth of forthcoming observations, it is expected that the LSST will unveil a significant number
of faint galaxies that have remained undetected in current widearea surveys. These potentially large datasets will raise multifold
questions, such as how we can use only LSST optical observations to obtain estimates of the main physical properties of galaxies, and how realistic and reliable they would be. We investigate
these topics by performing a simulation of LSST observations of
main-sequence (hereafter MS) galaxies that form a nearly linear
relation (in log-log space) between their stellar mass and SFR
(Noeske et al. 2007; Elbaz et al. 2010; Rodighiero et al. 2011;
Speagle et al. 2014; Schreiber et al. 2015; Whitaker et al. 2015;
Pearson et al. 2018). Main-sequence galaxies constitute the dominant population in deep fields such as COSMOS and ELAIS N1 ,
which can reach very faint optical magnitudes (∼29–30 mag).
The LSST is expected to expand the observed MS population of
galaxies to other fields that are not currently covered by deep-field
surveys. For this reason, we decided to focus on the MS galaxies.
The paper is organised as follows. In Sect. 2 we describe
the data and the HELP project. In Sect. 3 we present the sample
selection, outliers, and starburst, and the method we used for this
work. In Sect. 4 we discuss the simulated LSST magnitude and
errors. The same section, together with Sects. 5 and 6, presents
the results. Our conclusions are presented in Sect. 7. Throughout this paper we use the WMAP7 cosmology (Komatsu et al.
2011): Ωm = 0.272, ΩΛ = 0.728, and H0 = 70.4 km s−1 Mpc−1 .
2. Data
The HELP collaboration provides extremely valuable multiwavelength data over the HerMES (Oliver et al. 2012) and the
H-ATLAS survey fields (Eales et al. 2010) and other relevant
Herschel fields. The total area of HELP is 1269.1 deg2 (Oliver
et al., in prep, Shirley et al. 2019). Herschel was equipped with
two imaging instruments, the Photodetector Array Camera and
Spectrometer (PACS; Poglitsch et al. 2010), which observed the
FIR at 100 and 160 µm, and the Spectral and Photometric Imaging Receiver (SPIRE; Griffin et al. 2010), which covered the
250, 350, and 500 µm wavelength ranges.
Surveys that combine a wide range of wavelengths have particular identification issues because the spatial resolution of the
sources is different in different bands. To correct for this issue,
HELP builds a master list catalogue of objects as complete as
possible for each field and uses the near-IR (NIR) sources of this
catalogue as prior information to deblend the Herschel maps.
A detailed description can be found in Shirley et al. (2019). The
tool developed to obtain the photometry of Herschel sources,
XID+ (Hurley et al. 2017), is a probabilistic deblending algorithm that extracts source flux densities from photometry maps
that show source confusion. It uses Bayesian inference to explore
G. Riccio et al.: Preparing for LSST data. Estimating the physical properties of z < 2.5 main-sequence galaxies
Table 1. Filter for the ELAIS N1 and COSMOS fields.
Telescope
CFHT
Subaru
Isaac Newton
PanSTARRS1
UKIRT
VST
Blanco
VISTA
Spitzer
Herschel
Instrument
Elais-N1 filters
COSMOS filters
MegaCam
WIRcam
HSC
Wide Field Cam.
Gigapixel Cam.1
WFCam
OmegaCAM
DEcam
Vircam
IRAC
MIPS
PACS
SPIRE
u, g, r, y, z
u, g, r, i, y, z
H, J, K s
g, r, i, z, N921, y
g, r, i, z, N921, y
u, g, r, i, z
g, r, i, z, y
J, K
3.6, 4.5, 5.8, 8.0 (µm)
24 (µm)
100, 160 (µm)
250, 350, 500 (µm)
the posterior probability distribution and provide probability
density functions (PDFs) for all prior sources, and thus flux
and uncertainties can be estimates. A detailed description can be
found in Hurley et al. (2017). The whole procedure is described
in Notebook and stored in a GitHub repository1 .
We use two HELP fields because many multi-wavelength
data are available within their wide-field coverage: the European Large Area ISO Survey North 1, hereafter ELAIS N1
(Oliver et al. 2000), and the COSMOS field (Laigle et al. 2016).
In addition to data from two PACS and three SPIRE maps,
we used available sets of photometric data at shorter wavelengths for both fields (listed in Table 1 and described below
in Sects. 2.1 and 2.2). Based on true galaxy observations from
these fields, we evaluate the LSST-like observations we used for
further analysis.
2.1. ELAIS N1 field
According to the HELP strategy, all sources detected in any
of the Spitzer IRAC bands were used as a prior for XID+ to
obtain FIR fluxes. XID+ was run on the Spitzer MIPS 24 µm
and Herschel PACS and SPIRE maps. The flux level at which
the average posterior probability distribution of the source flux
becomes Gaussian is 20 mJy for MIPS, 12.5 and 17.5 mJy for
100 µm and 160 µm PACS bands, respectively, and 4 mJy for all
three (250, 350, and 500 µm) SPIRE bands (for more details,
see Hurley et al. 2017; indicating that information from data
dominates those of the prior). In the deblending procedure, the
priors used to compute the fluxes satisfied two criteria: they
must have an IRAC 1 band detection, and they must have been
detected in either the optical or NIR wavelengths to eliminate
artefacts. More information about the catalogue can be found
in Małek et al. (2018), Shirley et al. (in prep.), and on the main
webpage of the HELP project2 .
In addition to the FIR bands, the catalogue is built on a
position cross-match of all the public survey data available in
the optical and mid-IR (MIR) range. This comprises observation from the Isaac Newton Telescope/Wide Field Camera
(INT/WFC) survey (González-Solares et al. 2011), the Subaru
Telescope/Hyper Suprime-Cam Strategic Program Catalogues
(HSC-SSP; Aihara et al. 2018), the Panoramic Survey Telescope
1
2
https://github.com/H-E-L-P/dmu_products
http://hedam.lam.fr/HELP/
g, r, i
J, H, K
u, g, r, i
g, r, z
J, H, K s, y
3.6, 4.5,5.8, 8.0 (µm)
24 (µm)
100, 160 (µm)
250, 350, 500 (µm)
and Rapid Response System (Pan-STARRS; Chambers et al.
2016), the UK Infrared Telescope Deep Sky Survey –
Deep Extragalactic Survey (UKIDSS-DXS) (Swinbank 2013;
Lawrence et al. 2007), the Spitzer Extragalactic Representative
Volume Survey (SERVS; Mauduit et al. 2012), and the Spitzer
Wide InfraRed Extragalatic survey (SWIRE; Lonsdale et al.
2003; Stauffer et al. 2005). We show the list of filters for
ELAIS N1 in Table 1. The whole matching procedure is
described in Shirley et al. (2019).
2.2. COSMOS field
For the COSMOS field, the XID+ analysis was perfomed on
Spitzer and Herschel maps for all the sources with fluxes greater
than 1 µJy in any of the IRAC bands from the COSMOS2015
catalogue (Laigle et al. 2016). The fluxes obtained follow the criterion of goodness defined in XID+ and correspond to a Gaussian posterior distribution of the estimated flux.
Starting with this multi-wavelength catalogue, COSMOS2015 (Laigle et al. 2016), ancillary photometry was added
with a position cross-match with other public survey, again containing optical and MIR observations. Other than the one already
mentioned for ELAIS N1 , this comprises the WIRCam Deep
Survey (WIRDS, WIRcam bands J, H, K s), the VLT Survey
Telescope (VST; Arnaboldi et al. 1998), the Victor Blanco 4 m
Telescope, the Visible and Infrared Survey Telescope for Astronomy (VISTA; Emerson et al. 2006; Dalton et al. 2006), and the
UKIDSS-LAS (WFCAM bands J, H, K) catalogues. (The merging strategy is the same as for ELAIS N1 and is described in
detail in Shirley et al. 2019). The list of filters used for COSMOS survey is shown in Table 1.
Detailed description of both fields (the area, mean depths in
different filters, all raw files and ancillary data, and many others)
can be found on the3 webpage.
2.3. Total sample
As part of the HELP database, both field catalogues include
photometric redshifts generated using a template fitting method
that is based on the Bayesian combination approach described
3
http://hedam.lam.fr/HELP/dataproducts/dmu31/dmu31_
Field_overviews/
A107, page 3 of 18
A&A 653, A107 (2021)
Table 2. Input parameters for the code CIGALE.
Parameters
Values
Star formation history:
Delayed star formation history + additional burst
e-folding time of the main stellar population model (Myr)
1000, 2000, 3000, 5000, 7000
e-folding time of the late starburst population model (Myr)
5000
Mass fraction of the late burst population
0.001, 0.01, 0.03, 0.1, 0.3
Age (Myr)
1000, 2000, 3000, 4000, 5000, 6500, 10000
Age of the late burst (Myr)
10, 40, 70
Delayed star formation history
e-folding time of the main stellar population model (Myr)
1000, 2000, 3000, 4000, 5000, 6500, 8000
Age (Myr)
500, 1000, 3000, 4000, 5000, 6000, 7000, 8000, 9000,
10000, 12000
Mass fraction of the late burst population
0.0
Single stellar population Bruzual & Charlot (2003)
Initial mass function
Chabrier (2003)
Metallicities (solar metallicity)
0.02
Age of the separation between the young and the old star population (Myr)
10
Dust attenuation law Charlot & Fall (2000)
AV in the Birth Clouds
0, 0.05, 0.1, 0.3, 0.8, 1.2, 1.7, 2.3, 2.8, 3.3, 3.8, 4.0, 4.2
Power law slopes of the attenuation in the birth clouds
−0.7
BC to ISM factor (Av ISM/Av BC)
0.5, 0.8
slope ISM
−0.7
Dust emission:
Draine et al. (2014)
Mass fraction of PAH
1.12, 2.5, 3.19
Minimum radiation field (Umin )
5.0, 10.0, 25.0
Power law slope dU/dM (U α )
2.0, 2.8
Fraction illuminated from Umin to Umax (γ)
0.02
Dale et al. (2014)
AGN fraction
0
Power law slope dU/dM (U α )
2.0
in Duncan et al. (2018). The authors investigated the performance of three photometric redshift template sets as a function
of redshift, radio luminosity, and infrared/X-ray properties over
the NOAO Deep Wide Field Survey Bootes and COSMOS
fields. The three template sets are (1) the default EZY reduced
galaxy set (Brammer et al. 2008), (2) XMM COSMOS templates (Salvato et al. 2009), and (3) the atlas of Galaxy SEDs
(Brown et al. 2014).
The total sample includes 39 329 objects for the ELAIS N1
survey and 14 864 for COSMOS, with FIR detections in at least
two photometric bands with a signal-to-noise ratio (S /N)>3.
This cut is performed to remove objects with unreliable photometry and thus improves the quality of the SED fitting process. We
kept in mind that by employing the selection described above, we
restricted our analysis to only a subsample of galaxies that LSST
will observe, which are objects that are bright in the FIR. Then,
as we show in the next section, we selected only the so-called MS
galaxies observed in the spectral range from UV to FIR because
these are the most commonly observed types of galaxy. The considered bands are u, g, r, i, z, N921, y, J, H, K, Spitzer IRAC 3.6,
4.5, 5.8, and 8.0 µm, Spitzer MIPS 24 µm, and five passbands
from Herschel, two from PACS (100 and 160 µm), and three
from SPIRE (250, 350 and 500 µm) across the ELAIS N1 and
COSMOS fields.
A107, page 4 of 18
3. Method: SED fitting, starbursts, and outlier
detection
3.1. SED fitting with CIGALE
The SED fitting was performed with the Code Investigating
GALaxy Emission4 (CIGALE). For a detailed description of the
code, we refer to Boquien et al. (2019). We provide a brief summary here. CIGALE is a Bayesian SED fitting code that combines
modelled stellar spectra with dust attenuation and emission.
CIGALE preserves the energy balance considering the energy
emitted by massive stars, which is partially absorbed by dust
grains and then re-emitted in the MIR and FIR. The quality of
the fit is expressed by the best χ2 (and a reduced best χ2 defined
as χ2r = χ2 /(N − 1), with N the number of data points). The minimum value of χ2r corresponds to the best model selected from the
grid of all possible computed models from the input parameters.
The physical properties and their uncertainties are estimated as
the likelihood-weighted means and standard deviations.
To obtain the starting MS sample of galaxies with the correct
physical properties to compare them with those obtained with
LSST alone, we ran CIGALE on the ELAIS N1 and COSMOS
samples with the physical modules and parameters reported in
Table 2. We did not use the AGN module (see Appendix B and
4
https://cigale.lam.fr
G. Riccio et al.: Preparing for LSST data. Estimating the physical properties of z < 2.5 main-sequence galaxies
Sect. 3.5). As shown in Małek et al. (2018), this set of parameters corresponds to the set that best fits a large sample of IRdetected galaxies in the 23 HELP fields for the redshift range
0 < z < 6. We used an SFH modelled as a delayed exponential function with an additional exponential burst to select and
remove starburst galaxies from our sample in order to retain MS
galaxies alone. We performed the SED fitting by employing the
modified version of the Charlot & Fall (2000) attenuation law,
which was employed in Małek et al. (2018) for a large sample of
multi-wavelength HELP data. We used the Draine et al. (2014)
dust emission module. A detailed description of each module can
be found in Boquien et al. (2019) and Małek et al. (2018).
To improve the quality of our selection, we selected only
objects with redshift lower than 2.5 from the full sample. Our cut
in redshift was not related with the LSST redshift range because
the photometric redshifts for the LSST will be applied and calibrated over the range 0 < z < 4 for galaxies to r ∼ 27.5
(LSST Science Collaboration 2009). The z < 2.5 is related to
the redshift distribution in the ELAIS N1 and COSMOS fields
that were used in this analysis, and it also restricts us to highquality data and so maximises the accuracy of the estimation of
physical properties. Moreover, we removed all the objects recognized as possible stars by Gaia (flag Gaia > 0 in the database). In
this way, we removed 2 921 objects (7.5% of the sample) from
ELAIS N1 and 887 (6% of the sample) from the COSMOS catalogue. From now on, we refer to the remaining 36 408 and 13 977
galaxies from the ELAIS N1 and COSMOS fields as the real
sample.
3.2. Selection of starburst galaxies
Galaxies can be classified according to their different properties:
morphology, colour, environment, mass, etc. A property that is
often used in the literature is the rate at which stars are forming
out of gas, the SFR. This leads to a definition of three different types of galaxies: passive, normal/MS, and starburst (SB).
The boundaries dividing these classifications are not precisely
defined because different authors use different methods to distinguish starbursts from MS galaxies (i.e. Rodighiero et al. 2011;
Speagle et al. 2014; Elbaz et al. 2018; Donevski et al. 2020). No
universally accepted method exists. Nevertheless, there is agreement that the three groups differ in regard to their evolution and
physical properties, such as SFH, dust and gas content, and others (Silverman et al. 2018; Elbaz et al. 2018).
Most galaxies observed with the LSST will be composed of
active IR galaxies, but the majority of them are likely to be normal, MS–like, or passive galaxies. For example, the current estimate for the contribution of SB galaxies to the full star formation
population is about 5% (Schreiber et al. 2016; Béthermin et al.
2017). However, this contribution increases when we isolate
brighter IR galaxies (e.g., Miettinen et al. 2017).
To interpret possible bias for the physical parameter estimation, we have to ensure that the selection effects do not
produce artificial trends in the analysis. To quantify the accuracy of the physical property estimates of LSST galaxies, we
decided to focus on MS objects alone because we can select a
large number of these galaxies from the HELP data to obtain
a statistically important sample of real and simulated galaxies.
The method we used to separate MS galaxies is described in
detail in Rodighiero et al. (2011). We divided our sample into
four redshift bins (Table 3) because the definition of starbursts
changes with redshift. It was shown by Schreiber et al. (2015)
that the average SFR of star-forming galaxies in the same mass
ranges increases with redshift. In this work starbursts are defined
Table 3. Total number of galaxies and SB percentage for ELAIS and
COSMOS fields in each redshift range.
Redshift range
0–0.5
0.5–1
1–1.5
1.5–2.5
ELAIS N1
SB%
COSMOS
SB%
7718
11 353
8120
9214
0.64%
1.54%
2.96%
3.19%
2629
5081
2526
3733
1.35%
1.65%
2.09%
0.83%
according to their specific SFR distribution (SFR/Mstar , hereafter sSFR). Figure 1 shows that the sSFR follows a Gaussian
distribution. We followed the same definition for starbursts as
Rodighiero et al. (2011), that is, objects with sSFR that lie above
sSFR+3σ, where sSFR is the Gaussian mean of the sSFR distribution. The right panel in Fig. 1 shows the selected starbursts
and the MS galaxies.
To further test the reliability of the SB selection, we compared our distribution and the position of starbursts with the
distribution found in Béthermin et al. (2017) (hereafter: B17),
a catalogue of simulated galaxies. B17 was built on IR/submillimeter data, and it is one of only a few models that is
able to simultaneously match the total IR number counts and
the evolution of the sSFR. It simulates a 2 deg2 field including
physical clustering from a dark matter simulation, and is thus
perfectly suited for comparison purposes. Figure 2 shows the
comparison of the SB distribution derived in this paper with the
sample of simulated SBs from B17 (cyan distribution). The simulated SB sample extends to sSFR values that are lower than
the sSFR range obtained from our analysis. The discrepancy
arises because two different selection methods were used in our
work and in B17. On the one hand, B17 randomly drew the SFR
of each source using a continuous log-normal distribution (in
agreement with the observational results, e.g., Rodighiero et al.
2011) and then used the Schreiber et al. (2015) definition of the
MS to select the galaxies, with an additional offset correction
for galaxies at z < 0.5. Specifically, B17 defined MS objects
as those belonging to the distribution centred on 0.87 × SFRMS
with 0.2 dex of width and SBs as those belonging to the distribution centred at 5.3 × SFRMS with 0.3 dex of width. On the other
hand, our selection is based on the statistical analysis presented
in Rodighiero et al. (2011), who used the sSFR distribution over
a broad redshift range. This makes our selection more discrete,
while the selection of B17 is continuous. Figure 2 shows that the
purity of our selection is very high, but at the same time, it can
be incomplete for less active galaxies.
We removed 763 SB galaxies from ELAIS N1 and 228 from
COSMOS. Table 3 shows the fraction of SB galaxies in the
ELAIS N1 and COSMOS field. Our findings for the ELAIS N1
field agree with the literature because it is expected that the fraction of SBs rises from 1% at low redshift to about 3% at higher
redshift, and remains flat thereafter (i.e. Béthermin et al. 2012).
3.3. Selection of passive galaxies
The SFH with two or more stellar populations is suitable to
fit active galaxies with moderate or high SFR. For this reason,
we analysed the remaining objects without SBs employing the
delayed SFH in the SED fitting, which is more suitable for normal MS galaxies (Ciesla et al. 2016). All parameters we used
for the SED fitting are listed in Table 2. We again performed the
SED fitting to obtain real physical properties of the sample of
A107, page 5 of 18
A&A 653, A107 (2021)
Fig. 1. Left panel: example of the sSFR distribution of ELAIS N1 galaxies in the redshift range 1 < z < 1.5 obtained with the delayed SFH plus
an additional burst. The orange line respresents the Gaussian fit, and the dashed black line corresponds to the division of starburst (SB galaxies are
located on the right side of the line) and MS galaxies. The division is located 3σ away from the centre of the Gaussian. Right panel: MS distribution
for the same redshift bin. Magenta circles represent SB galaxies selected from the sSFR distribution shown in the left panel. The dashed black line
represents the Speagle et al. (2014) MS.
Fig. 2. sSFR distribution in four different redshift bins obtained with delayed SFH plus an additional burst. Open histograms are the galaxy
distributions (MS+SB) derived in this work, the dashed black line is the division between SBs (right side of the line) and MS (left side of the line)
in our sample, and the cyan full histograms represent the simulated SB sample from B17. The division is located at about 3σ from the Gaussian
centre. The sample is the same as for Fig. 1, but the binning is different.
real MS galaxies. These values were then used to simulate the
LSST observations (see Sect. 4.1).
To ensure the purity of the MS sample, we additionally
removed possible passive galaxies. As for the starburst evaluation, many different methods were employed in the literature to
select red passive galaxies, that is, an UVJ and NUVrK colour
A107, page 6 of 18
diagram analysis (Williams et al. 2009; Arnouts et al. 2013), a
division based on sSFR (Vulcani et al. 2015; Salim et al. 2016,
2018), or unsupervised machine learning (Siudek et al. 2018).
We decided to follow the method described by Salim et al.
(2018) by removing all objects with log10 (sSFR[yr−1 ]) < −11.
In this way, we removed 340(1%) and 63(0.5%) galaxies from
G. Riccio et al.: Preparing for LSST data. Estimating the physical properties of z < 2.5 main-sequence galaxies
Table 4. Number of AGNs selected based on the MIR features for the ELAIS N1 and COSMOS fields.
Method
Stern et al. (2005)
Donley et al. (2012)
ELAIS N1
COSMOS
Total sample
1269 (3.48%)
497 (1.36%)
1334 (9.50%)
291 (2.08%)
2603 (5.16%)
788 (1.56%)
Notes. The last column shows the total number (and percentage) of AGNs in the full sample.
the ELAIS N1 and COSMOS field, respectively. The almost
negligible number of passive galaxies in the HELP sample is
related to our initial sample selection, which required at least
two Herschel measurements with S /N > 3.
3.4. AGN contribution
Taking advantage of an IRAC detection for all galaxies included
in our analysis, we used MIR detections to determine how
numerous the AGN population in our sample is. We employed
two different selection criteria based on MIR photometry (IRAC
bands) analysis. They are explained in detail in Stern et al.
(2005) and Donley et al. (2012). Figure B.1 shows the IRAC
colour-colour selection using the methods of Donley et al.
(2012) (upper panels) and Stern et al. (2005) (lower panels).
Using both criteria, we find a negligible number of AGN in
comparison to the final sample (1.56% and 5.16% for the criterion of Donley et al. 2012 and Stern et al. 2005, respectively;
see Table 4 for detailed information for both fields). The redshift distribution of selected AGNs is shown in Fig. B.2. For
consistency with the cuts made previously, we removed AGNs
from our sample. We decided to use a conservative approach, and
we removed all 2603 possible AGNs found with the method of
Stern et al. (2005) because this selection includes all the AGNs
that are detected with the approach of Donley et al. (2012).
Fig. 3. Redshift distribution of ELAIS-N1 and COSMOS samples based
on the approach of Duncan et al. (2018).
at a level of 4%, which agrees with what was found in previous
works (Ilbert et al. 2009; Hildebrandt et al. 2010; Duncan et al.
2018). The final redshift distributions of both samples are shown
in Fig. 3.
4. Estimation of the LSST physical properties
3.5. Outlier selection
Because the number of galaxy free parameters is large and
unknown, a simple χ2r selection cannot reliably remove the
majority of the outliers from our sample. Along with a χ2r selection, we used an estimation of the physical properties Ldust and
Mstar (see Appendix A) in order to eliminate possible outliers
and to ensure a high quality of the SED fitting. A similar procedure was used by Małek et al. (2018) for the HELP ELAIS N1
field. Based on these criteria, we removed 2117 galaxies from
ELAIS N1 and 640 from the COSMOS field. This is 5.81% and
4.75%, respectively.
3.6. Final sample
To obtain the final sample of normal star-forming galaxies,
we removed possible starbursts (Sect. 3.2), passive galaxies
(Sect. 3.3), and possible AGNs (Sect. 3.4). We also performed
additional cleaning using outlier selection (Sect. 3.5) to remove
all galaxies with possibly incorrect photometry, or incorrect
matches of UV-optical and FIR measurements. At the end of
the process, 31 936 objects were left for ELAIS N1 (87% of
the total sample) and 11 716 galaxies for COSMOS (84% of the
total number). Furthermore, in order to validate the photometric
redshift estimates used for these objects, we performed a comparison with spectroscopic redshift estimates, which are available for ∼5000 galaxies in the ELAIS N1 and COSMOS fields.
Following the Duncan et al. (2018) definition of critical outliers
|∆z|
( 1+z
> 0.2), we find that the fraction of outliers in our sample is
s
In the following section we discuss the LSST data and uncertainties from simulations and the estimation of the physical properties of the galaxies obtained by performing the SED fitting of
(i) the fiducial input parameters plus LSST data alone and (ii)
the fiducial input parameters plus LSST data coupled with other
observations.
4.1. Estimating LSST-simulated data and uncertainties
Simulating LSST data has been a popular topic in the past years
because of the upcoming start of the survey (Ivezić et al. 2019).
We derive an ‘LSST-like catalogue’ from the best fit of the observational data as described in Sect. 2. In this way, we are able to
quantify the difference between estimating the physical properties based on the LSST measurements alone and the UV-to-FIR
wavelength of the real, observed objects. Considering the depth
reached with ten years of survey data, it is very likely that LSST
will observe objects that are not visible with the current groundbased survey telescopes, and this work will be a starting point
to learn how these objects should be treated with SED fitting
methods.
We simulated the observed fluxes in the six LSST bands
(ugrizy; the filter response curve is provided by the LSST developers team, Ivezić et al. 2019). To obtain LSST fluxes, we ran
CIGALE by fitting the photometric measurements and providing
the LSST filter response curves for the code. We used a CIGALE
module (called fluxes) that is specifically designed to estimate
the fluxes in the defined filters. We computed LSST fluxes from
A107, page 7 of 18
A&A 653, A107 (2021)
Fig. 4. Magnitude errors vs observed apparent magnitudes for simulated
LSST observations from the ELAIS N1 sample. Simulations are cut to
the maximum LSST limit. The choice to vary simulation parameters
allows us to obtain different errors for similar magnitude values, as we
would expect in the case of real observations. The brightest end we
reached is 13 mag, but for clarity of the plot, we cut it at 21 mag.
the best-fit model of each object. We included in our sample all
the galaxies that will be detected in all bands at the depth of the
ten-year survey: u < 26.1, g < 27.4, r < 27.5, i < 26.8, z < 26.1,
y < 24.9. In this way we discard 8645 objects (23% of the total
sample) from ELAIS N1 and 2623 (19% of the total sample)
from COSMOS.
To incorporate an LSST-like observational uncertainty in our
catalogue, we must take random phenomena into account that
might occur during a real observation, such as a change in the
sky seeing or the number of visits. The predicted magnitude
errors that we converted into flux errors following the conversion provided in the LSST manual (Ivezić et al. 2019) depend
on the galaxy magnitude, the sky seeing, and the total survey
exposure time in a given filter. We used the LSST simulation
software package CatSim5 to calculate magnitude errors. The
error evaluation was based on Eq. (5) of Ivezić et al. (2019) and
took variations in the photometry due to hardware and observational components (e.g., detector, darksky, and atmosphere) into
account. The evaluated random error was then divided by the
square root of the number of visits during the survey. The LSST
manual provides mean values for all these components.
To mimic the real conditions, we added a value that was
randomly chosen from a Gaussian distribution centred on the
provided mean value to the average value of each component
provided in the LSST manual, which had a standard deviation
the 10% of the mean. In this process, we varied the number of
visits and sky seeing. The assumed standard conditions for these
components are 0.8 arcsec for the seeing and a uniform progression that assumes a total of 56, 80, 184, 184, 160, and 160 visits
in filters ugrizy, respectively, in ten years, where each visit is 30
seconds of integration time.
At the end, to further mimic a possible divergence of the
‘real’ observed flux from the simulated value evaluated from
the best-fit SED, we again added a value randomly chosen
from a Gaussian distribution centred on 0 to the best-fit SED
with the standard deviation of the flux error calculated before.
Figure 4 shows the magnitude errors as a function of the simulated observed magnitude for our sample of galaxies. We only
selected objects that would be observed in all six bands accord-
To estimate the main physical properties of the LSST sample, we
ran CIGALE on simulated LSST observations and uncertainties
employing the same modules and parameters as were used for
the HELP MS sample (Table 2, with delayed SFH). Figure 5
shows two example SEDs of the same galaxy at redshift 0.92,
obtained with the UV-FIR and LSST photometric-only data set,
respectively. For this specific case, we found an agreement of
estimated stellar masses (Mstar real = 6.05 × 1010 ± 5.62 × 109 M ,
Mstar LSST = 5.59×1010 ±1.80×1010 M ). Instead, the SFR calculated for the LSST-like photometric data alone is highly overestimated (by a factor of six) with respect to the real value obtained
by employing the UV-FIR data set (SFRreal = 11.9±2.16 M yr−1
and SFRLSST = 67.4 ± 45.9 M yr−1 ). Moreover, the residuals
for the LSST are very small (but never null), as we can easly
find a model that almost perfectly fits just six observations in
the optical part. However, the IR part of the SED, and so the
5
6
https://www.lsst.org/scientists/simulations/catsim
A107, page 8 of 18
Fig. 5. Comparison of SEDs evaluated for the same object at z ∼ 1
using the full UV-to-IR observations (upper panel) and the LSST optical
bands alone (bottom panel). The blue square represents the observed
fluxes, and red dots represent the fluxes predicted by the model. The
magenta lines delimit the division between optical and IR bands, which
is taken into account by CIGALE. The IR emission is not constrained
for the LSST estimate.
ing to our simulation. As a result, we reached the LSST magnitude limit only for the u band. The final catalogue contains
simulated LSST fluxes and uncertainties for 23 291 galaxies in
the ELAIS N1 field and 9093 in the COSMOS field. The catalogues, together with the photometric redshifts and HELP IDs,
are available at the HELP virtual observatory6 .
4.2. Fiducial parameters and LSST data alone
https://herschel-vos.phys.sussex.ac.uk/
G. Riccio et al.: Preparing for LSST data. Estimating the physical properties of z < 2.5 main-sequence galaxies
Fig. 6. Main-sequence (SFR vs. Mstar ) relation for the ELAIS N1 and COSMOS fields in four redshift bins. In blue we show the LSST-like sample,
and in black the real sample. The solid black line represents the MS by Speagle et al. (2014), and the dashed lines mark the loci two times above
and below the MS. This plot shows a clear SFR overestimate obtained using LSST bands alone, which tends to disappear at higher redshift ranges.
dust emission module, is completely unconstrained (we used the
same dust emission module from Draine et al. 2014 as was used
for the original HELP data with the same grid of parameters).
The relation of SFR and Mstar in the four redshift bins is
shown in Fig. 6. In this figure, we compare the MS relation
obtained for the LSST-like sample with the relation obtained
from the full UV-FIR SED fitting. We show the MS from
Speagle et al. (2014) as a reference. At low redshift, the LSST
estimation fails to probe low-SFR objects, and this leads to a
clear division between the respective MS relations, which overlap at higher redshifts, however. In Appendix D we discuss the
scatter between our sample and the MS law. Figure 7 shows this
overestimation as a function of redshift separately for ELAIS N1
and COSMOS (left upper panel). We also plot in the same figure
the ratio of the LSST-derived stellar mass, Ldust , and Mstar and
those from the full UV-IR SED fitting. We obtain an overestimation of the dust related properties (SFR, Ldust , and Mstar )
but the values of Mstar are comparable. The overestimation of
the SFR is strongly dependent on the redshift. The ratio of
the stellar masses is evenly distributed around zero, leading to
comparable results between the two runs because stellar masses
mostly rely on optical data. This result holds for the ELAIS N1
and COSMOS fields and shows that there is no dependence on
the field.
These results can be explained when we consider how the
physical properties are evaluated by the Bayesian method. This
is basically done through a likelihood estimation. Each model
in the grid of models built from the starting input parameters will have an associated likelihood taken as exp(−χ2 /2)
that is used as weight to estimate the physical parameters (the
likelihood-weighted mean of the physical parameters attributed
to each model) and the related uncertainty (see Sect. 4.3 of
Boquien et al. 2019). Fitting just LSST optical observations
results in high likelihood values even for templates that do
not reflect the real physical properties of the modelled galaxy.
Because the SFR is partly estimated from the UV emission of
the massive young stars in star-forming regions and because this
emission is attenuated by the dust and is re-emitted in the IR
band, we find that the lack of information about the UV and MIR
rest-frame wavelengths for the LSST sample causes CIGALE
to overestimate the attenuation. Overestimated attenuation also
results in an overestimation of the SFR.
A107, page 9 of 18
A&A 653, A107 (2021)
SFR
Dash line: fitted equation ELAIS-N1
Points:Median in z bins COSMOS
1.2
1.0
1.0
log10(Mstarratio)
log10(SFRratio)
0.8
0.6
0.4
0.2
0.0
0.2
0.4
0.0
0.5
1.0
1.5
redshift
2.0
0.6
0.4
0.2
0.0
0.4
2.5
Ldust
0.0
0.5
1.0
1.0
0.8
0.8
0.6
0.4
0.2
0.0
0.2
1.0
1.5
redshift
2.0
2.5
2.0
2.5
Mdust
1.2
log10(Mdustratio)
log10(Ldustratio)
0.8
0.2
1.2
0.4
Mstar
1.2
0.6
0.4
0.2
0.0
0.2
0.0
0.5
1.0
1.5
redshift
2.0
2.5
0.4
0.0
0.5
1.0
1.5
redshift
Fig. 7. Ratios of different physical properties obtained from fitting the simulated LSST data alone and from the UV-FIR SED (e.g., SFRratio =
SFRLSST /SFRUV−FIR ) as a function of redshift for the ELAIS N1 and COSMOS fields. The properties obtained from the UV-FIR SED fitting are
considered as the true values. From the upper left panel we show clockwise the SFR, Mstar , Ldust , and Mstar comparisons. The dashed lines represent
polynomial fits performed on the samples. The points are the median values in each redshift bin, with the median absolute deviation as errors. A
ratio equal to zero corresponds to a perfect agreement between the estimates obtained based on the LSST-like sample and the real values. The
distributions calculated for ELAIS N1 and COSMOS are comparable within the errors.
is better probed because the LSST bands cover a larger portion
of the UV rest-frame spectra, where dust attenuates more effectively, and a better constraint of the SFR is provided. As shown
in Fig. 7, the differences in the estimated SFR become negligible
for z & 1.3.
In order to obtain a useful function with which to correct
for the overestimation of the SFR, we performed a polynomial
fit on the SFR ratio distribution of ELAIS N1 and COSMOS
combined,
log10 (SFRratio ) = 0.26 · z2 − 0.94 · z + 0.87,
Fig. 8. LSST coverage of an example SED at different redshifts indicated in the panel.
The fluxes observed by the LSST in the optical at high redshift are in the UV rest-frame. Because the UV wavelengths trace
young stellar populations, the estimates of the SFR from the SED
fitting with CIGALE, significantly improves as a consequence.
Figure 8 shows an example of an SED superimposed with LSST
coverage at different redshifts. Even at z = 1, the LSST bands
are almost entirely shifted to the rest-frame UV bands, which
range between 0.01 and 0.38 µm. As a result, the dust attenuation
A107, page 10 of 18
(1)
where SFRratio stands for SFRLSST /SFRreal . However, this formula is highly dependent on the input parameters.
To determine whether the complex set of parameters used
for the Draine et al. (2014) model is responsible for the overestimation of the SFR, we ran the whole analysis and adopted
the Dale et al. (2014) model. Dust emission in this model is
dU
parametrised by a single parameter α defined as dM
= U α , where
M is the dust mass heated by a radiation field at intensity U. We
used the parameter α = 2 to better describe the stellar emission
from MS galaxies. The module also allows adding an optional
AGN component, which we set to 0 for this test. Comparing the
LSST physical properties with the real one using the Dale et al.
(2014) module, we obtain the same overestimation of the SFR
and Ldust as was obtained with the Draine et al. (2014) dust emission. We were not able to compare dust mass because it is not an
output parameter of the Dale et al. (2014) module. As we found
G. Riccio et al.: Preparing for LSST data. Estimating the physical properties of z < 2.5 main-sequence galaxies
SFR
1.2
1.0
log10(SFRratio)
log10(SFRratio)
0.8
0.6
0.4
0.2
0.0
0.2
0.6
0.4
0.2
0.0
0.2
0.0
0.5
1.0
1.5
redshift
2.0
0.4
2.5
SFR
1.25
0.0
log10(Mdustratio)
0.50
0.25
0.00
0.25
1.0
1.5
redshift
2.0
2.5
2.0
2.5
LSST
Points:Median in z bins LSST+SPIRE
1.0
0.75
0.5
Mdust
1.2
LSST
Points:Median in z bins LSST+GALEX
1.00
log10(SFRratio)
LSST
Points:Median in z bins LSST+SPIRE
1.0
0.8
0.4
SFR
1.2
LSST
Points:Median in z bins LSST+IRAC
0.8
0.6
0.4
0.2
0.0
0.2
0.50
0.0
0.5
1.0
1.5
redshift
2.0
2.5
0.4
0.0
0.5
1.0
1.5
redshift
Fig. 9. Upper left panel: SFR ratio, defined as in Fig. 7, estimated with LSST observations alone (red) and with LSST+IRAC observations (blue).
The two samples disagree at low redshift, but the comparison improves at high redshift, highlighting the redshift dependence of the results. With
the addition of MIR observations, the LSST-only estimates are consistent with the UV-IR estimates. Upper right panel: same comparison, but
adding FIR SPIRE observations. In this case we did not fully remove the overestimation, which highlights that the problem lies in the lack of restframe MIR data. Lower left panel: same as in the other panels, but adding GALEX observations. We obtain a slight decrease in the overestimation.
For this comparison we considered the ∼3000 objects that have a counterpart in the GALEX database. Lower right panel: Mstar ratio evaluated
with LSST observations alone (red) and with LSST+SPIRE observations (blue). The results are consistent with the ‘real’ ratio with a very low
scatter.
no improvement by employing a simpler model, we decided to
keep the results obtained with Draine et al. (2014) for homogeneity with the results published by HELP project (Małek et al.
2018; Shirley et al., in prep.).
4.3. Fiducial parameters, LSST, and ancillary data
A different approach to correct for the overestimation consists
of applying the SED procedure on the LSST data together with
other available observations in different bands (e.g., MIR Spitzer
bands, FIR Herschel Spire bands). Figure 9 shows the results for
SFR ratios obtained by adding IRAC MIR and SPIRE FIR observations. We expect that by adding the rest-frame NIR part of the
SED, the attenuation of the old stellar population will be better
constrained, while the MID and FIR mainly constrain the dust
emission from the star-forming regions and the missing SFR,
which is hidden by dust. The upper left panel of Fig. 9 shows
that when MIR observations alone are added, the overestimation
of the SFR is fully corrected, regardless of the considered redshift range.
The combined use of UV and LSST data might also be
expected to correct for the overestimation of the SFR at low redshift. We tested this hypothesis by adding the UV observations
from GALEX and performing the SED fitting. We performed
a 1.5 arcsec cross-match with the HELP catalogue and identified ∼3000 galaxies with a GALEX counterpart. Figure 9 (bottom left panel) shows the comparison of the SFR estimation for
LSST-like and the UV–FIR data set of these 3000 galaxies. We
conclude from this plot that by adding GALEX observations,
we obtain a slightly lower overestimation in general, which is
still consistent with previous results, however, meaning that the
observed UV fluxes are not enough to completely correct for
the differences. Furthermore, for 0.5 < z < 1.5, we obtain a
slight underestimation of the parameter. Nevertheless, we stress
that this result can be biased by the low number of GALEX
counterparts of the LSST-like catalogue and the low quality of
GALEX observations for higher-redshift sources. To confirm
this statement, we simulated GALEX near-UV (NUV) and farUV (FUV) observations for the whole sample (we refer to it
as GALEXtrue ), again using the CIGALE module fluxes, and
A107, page 11 of 18
A&A 653, A107 (2021)
SFR
Dash line: fitted equation CF
Points:Median in z bins Calzetti
1.2
1.0
log10(SFRratio)
employed it together with the LSST to estimate the physical
properties. We decided to cut objects with z > 1.5 because after
this limit, both NUV and FUV GALEX bands probe emissions
below the Lyman break. Figure C.1 shows the comparison of the
SFR evaluated in this way and the SFR evaluated using LSST
observations alone. We find a clear correction of the overestimation, highlighting the great effect that UV observations have
on the SFR estimation. We confirm that the overestimation is
partially due to the lack of the direct tracer of the young stellar
population. Unfortunately, no new UV missions are planned in
the near future, therefore we do not expect a UV coverage that
large enough to be used in conjunction with the LSST in order
to correctly estimate the physical parameters.
Figure 9 (bottom right panel) also shows the correction of
Mstar estimates that we obtained by adding the SPIRE FIR observations. The good agreement of the LSST plus SPIRE and the
real estimates arises because the FIR emission is a direct probe
of the dust mass.
0.8
0.6
0.4
0.2
0.0
0.2
0.0
0.5
1.0
1.5
redshift
2.0
2.5
Fig. 10. SFR overestimation as a function of redshift for the
Charlot & Fall (2000) (red) and Calzetti et al. (2000) (cyan) attenuation
laws.
5. Testing different input parameters in CIGALE
As discussed in Sect. 3.1, the set of parameters we employed
for the analysis presented so far corresponds to the best set to
fit the large sample of objects in the area of ∼1300 deg2 of the
HELP field. However, we also investigated how much the results
obtained by fitting the LSST simulated data alone are dependent on the CIGALE input parameters. In particular, we tested
possible variations of the derived galaxy physical quantities as
a function of the input radiation field, PAH fraction, and dust
attenuation law. To test how the variations in the dust attenuation laws change our results, we refitted the UV to IR photometry and rederived the LSST simulated data that were later refitted
by adopting a different attenuation law.
5.1. Dust emission and mass
Dust continuum emission is only determined by the energy balance, and therefore it only depends on the amount of absorbed
radiation. As a consequence, the total dust emission is not
affected by the parameter Umin , but is only sensitive to the
total absorbed radiation. In contrast, we find that Mstar is not
constrained by using the LSST data alone, and its estimate is
largely affected by the Umin parameter. When Umin is changed,
the amount of radiation that irradiates the dust is modified, but
the amount of dust emission is unaltered. As a consequence,
higher input values of Umin are translated into lower Mstar and
vice versa. High values of the Umin (Umin = 25) parameter yield
an underestimation of Mstar . As a consequence, because the
shape of dust emission is completely unconstrained using the
LSST coverage alone, the Bayesian method cannot evaluate the
Umin parameter and assigns an average value from all considered
input parameters to all galaxies. This results in a strong dependence of dust mass estimates on the input parameters. When the
dust emission is constrained using SPIRE observations, the Umin
parameter is well evaluated by the Bayesian method, and so is
the dust mass.
The use of different attenuation laws also changes the estimates of dust emission and mass because the radiation absorbed
and the re-emitted by dust grains is modified. By using either
the Calzetti et al. (2000) and Charlot & Fall (2000) attenuation
laws, we obtain an overestimation of the dust luminosity over the
entire redshift range. The trends of the dust luminosity ratios and
the redshift are different, however. When we adopt Calzetti et al.
(2000), we obtain a constant slight overestimation of the dust
A107, page 12 of 18
mass, centred around 0.3 dex throughout the whole redshift
range, and when we use Charlot & Fall (2000), the dust mass is
overestimated for local galaxies and underestimated for redshift
greater than ∼1.
5.2. Dust attenuation laws and star formation history
We find that the SFR to is not affected by the PAH fraction or the
input radiation field (Umin and α parameters) because the input
quantities only shape dust emission. The SFR is instead affected
by our choice of the attenuation law. When the Calzetti et al.
(2000) modified attenuation law is employed, we obtain the
results shown in Fig. 10. The results using the Calzetti et al.
(2000) attenuation curve agree well with those obtained with the
Charlot & Fall (2000) prescription at low redshift, but change
in shape at higher z. The overestimation with the Calzetti et al.
(2000) curve is more constant throughout the redshift range, but
it decreases faster when the Charlot & Fall (2000) prescription
is used.
Furthermore, we also examined the dependence of SFR estimates on the SFH module. Performing the entire process using
a delayed plus additional burst SFH, we obtain an even higher
overestimation when estimated using LSST observations alone,
which again decreases at higher redshift.
We also inspected changes in the SFR difference when the
best values were used instead of Bayesian values for the LSSTlike sample. We find a consistency between the SFR from the
LSST best fit and the one estimated from the UV-FIR best fit,
finding log10 (SFRLSST,best /SFRUV−FIR,best ) very well distributed
around zero with a scatter of 0.05 dex. This result shows a good
agreement of the best templates of the two runs. It can be understood as follows: The fluxes for the LSST-like run were calculated based on the best template of the UV-FIR run. We confirm
that the differences we found must be sought in the Bayesian
analysis, which tends to overestimate the attenuation when it is
employed on LSST data alone. Unfortunately, being estimated
directly from the SED that best fits our data, the best-fit value
has several drawbacks that make it unsuitable for this type of
analysis. For example, it ignores the degeneracies that can be
encountered because models with equally good fits can have very
different properties. Moreover, the best fit in itself does not provide information about the uncertainties.
G. Riccio et al.: Preparing for LSST data. Estimating the physical properties of z < 2.5 main-sequence galaxies
Fig. 11. AFUV as a function of stellar mass estimated from the UV-FIR
SED fitting (blue density plot) compared with the AFUV estimated by
employing Mstar from the relation of Bogdanoska & Burgarella (2020).
Fig. 12. Difference between AFUV estimated from AFUV –Mstar relations:
Bogdanoska & Burgarella 2020 (green) and Eq. (2) with the coefficients
reported in Table 5 (blue) and AFUV estimated from the UV-FIR SED
fitting as a function of redshift.
6. Application of the A FUV –M star relation to correct
for the SFR overestimation
The IR excess is defined as IRX = log(LIR /LUV ), where LIR stands
for total integrated luminosity in the IR, and LUV is the UV luminosity
derived from flux measured with a filter, such as GALEX or estimated
through an SED process.
7
SFR
1.25
LSST
LSST + AFUV (This work)
LSST + AFUV (Bogdanoska et al. 2020)
1.00
0.75
log10(SFRratio)
Previous results show major miscalculations of the SFR when
LSST data alone are used. Due to the lack of information in
the UV and MIR part of the spectrum, the SED fitting results
in an overestimation of the attenuation, which leads to the general overestimation of the SFR. At the same time, as it is probed
mainly from the optical emission of the galaxy, Mstar seems to
be well estimated using LSST data alone. When we take into
account that Mstar is the result of the previous star formation
activity of the galaxy, which is responsible for producing the
dust, it could be used as a promising tracer of the dust content.
Several works have recently explored possible relations of
Mstar and dust attenuation (Xu et al. 2007; Martin et al. 2007;
Buat et al. 2009; Bogdanoska & Burgarella 2020). Most of them
suggest a possible linear relation between AFUV and log10 (Mstar )
over a wide mass range (9 ≤ log10 (Mstar ) ≤ 12). According to
the literature, this relation is highly dependent on redshift.
Recently, Bogdanoska & Burgarella (2020, hereafter BB20),
modelled a single parameter linear function, assuming a nonzero constant dust attenuation for low-mass galaxies (Eq. (6) in
BB20). They used a sample of galaxies based on the selection
criterion that requires IR excess7 (IRX) to be directly calculated
from the IR-to-UV ratio or by SED fitting. This selection can
introduce a bias in the local Universe and above redshift 2–3
due to the IR detection (at high redshift, only very dusty and
massive galaxies are detected, while in the local universe, the
IR-detected galaxies are rather rare). BB20 found that the AFUV –
Mstar relation cannot be described with a simple linear function,
and they concluded their work with a new relation between AFUV
and Mstar as a function of redshift. In this section we try to use
the AFUV –Mstar relation provided by BB20 to estimate the AFUV
(we refer to it as AFUV BB20 ) of the LSST sample. The procedure
is as follows: (1) from the LSST data we estimated the Mstar ,
(2) using Eqs. (5) and (6) from BB20, we calculated AFUV BB20 ,
and finally, (3) we used AFUV BB20 as a prior of the new LSST
CIGALE run.
Figure 11 shows the AFUV BB20 compared with the AFUV
obtained from the full UV-FIR SED fitting as a function of
Mstar . We find that the estimates AFUV BB20 are substantially lower
0.50
0.25
0.00
0.25
0.50
0.75
1.00
0.0
0.5
1.0
1.5
redshift
2.0
2.5
Fig. 13. SFR ratio, defined as in Fig. 7, as a function of redshift. The
result from the SED fitting employing LSST data alone is shown in red.
Green and blue lines represent results from the SED fitting, adding as
prior the AFUV from the relations of Bogdanoska & Burgarella (2020)
and Eq. (2), respectively.
than those obtained with the SED fitting process. The difference between AFUV BB20 and AFUV UV−FIR is shown in Fig. 12
(green line). The underestimation is accentuated for low-redshift
objects. As expected, employing the AFUV BB20 as prior in the
SED fitting process along with LSST observations results in an
underestimation of the SFR, as shown in Fig. 13, where the
ratios of the true SFR and the one derived by different methods
are shown. We suspect that the reason for the substantial difference between the two estimates of AFUV can be traced back to
the choice of the sample, as the sample used in BB20 is more
general, while in our case, we focused on IR-bright galaxies to
ensure the highest quality of the UV-FIR SED fitting process.
It is clear from these results that we cannot directly empoly the
relation of Bogdanoska & Burgarella (2020) to correct for the
SFR overestimation for our sample.
We decided to incorporate the general idea presented in
Bogdanoska & Burgarella (2020) and to use the AFUV –Mstar
relation in order to correct the SFR for an LSST sample of
data. We built a simplified relation that represented our sample
of IR-bright main-sequence galaxies by following a procedure
A107, page 13 of 18
A&A 653, A107 (2021)
Table 5. Obtained a and b coefficients from fitting Eq. (2) in the four
redshift bins.
Redshift
a
b
0–0.5
0.5–1
1–1.5
1.5–2.5
0.41 ± 0.02
0.44 ± 0.03
0.72 ± 0.03
0.83 ± 0.04
−1.39 ± 0.21
−0.49 ± 0.30
−3.42 ± 0.34
−5.19 ± 0.39
similar to that described by Bogdanoska & Burgarella (2020).
For this purpose, we fit the AFUV estimated from the UV-FIR
SED fitting as a function of the log10 (Mstar ). We used four redshift bins (0–0.5, 0.5–1.0, 1.0–1.5, and 1.5–2.5) to include the
redshift dependence of the AFUV –Mstar relation. Linear, powerlaw, and exponential functions were tested to obtain the best fit,
but we found a negligible difference between them. To be as
consistent as possible with the results obtained in the previous
works, we decided to use the linear function in the form
AFUV−LSST = a · log10 (Mstar−LSST ) + b.
(2)
From the fitting process, we obtained a and b coefficients for
each redshift bin. Table 5 shows all coefficients together with
the uncertainties.
The blue line in Fig. 12 shows the difference between
the AFUV−LSST calculated with our four linear relations and
the one estimated from the UV-FIR SED fitting as a function of redshift. Our relation reproduces the far-UV attenuation
(AFUV) derived from the fitting of the full SED better than
the Bogdanoska & Burgarella (2020) relation. By employing the
AFUV−LSST as a prior in the SED fitting along with LSST observations, we obtain the blue relation shown in Fig. 13. This figure
shows that the SFR overestimation is fully corrected for when
the AFUV−LSST prior is employed. This result also proves that
knowing the AFUV –Mstar relation for a given sample of galaxies, it is possible to estimate SFR without the IR counterpart.
This requires prior knowledge of the sample, which is often not
available. We are aware that the relation constructed in this work
may be applicable only to our sample, or at most to IR-bright
normal star-forming galaxies, but further generalisation of our
results is beyond the scope of this paper. However, considering
the extreme usefulness of this relation for future surveys such
as the LSST, we are planning to extend our analysis to a more
general sample of galaxies in the next work.
7. Conclusions
We performed a reliability check of the physical properties estimation of MS galaxies by employing simulated LSST observations. For this purpose, we selected 50 135 and 15 754 objects
from the ELAIS N1 and COSMOS fields, respectively, of the
Herschel Extragalactic Legacy Project (HELP), in order to build
the starting set of data to simulate observed LSST fluxes and to
obtain reliable estimates of the physical properties of galaxies.
An important part of our analysis was the sample selection.
We selected only galaxies from the so-called main sequence
by removing all possible SBs from the sample using the same
method as Rodighiero et al. (2011), and we removed passive
galaxies using the method described in Salim et al. (2018). Furthermore, we also removed galaxies that contain AGN according
to the selection of Stern et al. (2005). We also cleaned the sample
from all non-typical galaxies by implementing additional quality criteria on physical properties following Małek et al. (2018):
A107, page 14 of 18
for our analysis we removed all galaxies for which Ldust and
Mstar estimated from the full SED fitting (from UV to FIR) were
different from those obtained from only the optical or only the
infrared part of the spectrum. At the end of the sample selection,
we selected 43 652 galaxies. This is 86% of the total sample.
We used this sample of MS galaxies as a prior to calculate
the corresponding LSST fluxes in the ugrizy bands. We used the
LSST simulation software package CatSim in order to simulate
the uncertainties on the photometric measurements. We took the
possible effects due to the hardware and observational components (e.g., detector, dark sky, and atmosphere) into account. We
then estimated the main physical properties of galaxies by performing the SED fitting of the simulated LSST data by employing the same sets of modules and parameters as for the originally
used HELP galaxies (Shirley et al. 2019, and in prep.).
We found that Mstar is well estimated by the LSST–like
data set. At the same time, SFR, Ldust are overestimated using
the LSST–like sample alone, while Mstar is completely unconstrained and dependent on the input parameters. The overestimation of the SFR is redshift dependent and clearly decreases
with redshift. It disappears at about redshift ∼1. We found the
relation that can correct the overestimation for the SFR parameter: log10 (SFRratio ) = 0.26 · z2 − 0.94 · z + 0.87.
We determined the photometric data that can be combined
with the LSST data to remove the overestimation. In our analysis we used not simulated but real and sometimes uncompleted
data to fully mimic the auxiliary data for the LSST because we
do not expect to have better UV or FIR data soon. We found
that the most efficient way to correct for the overestimation of
the SFR is adding mid-IR observations (IRAC data), while Mstar
is corrected for by adding the far-IR bands (SPIRE data). The
addition of UV observations from GALEX does not correct the
differences. Our findings suggest that the main problem of the
pure LSST-like sample in the local Universe will be the inability to mimic the real attenuation for the old and young stellar
populations.
By testing the input parameters of CIGALE, we found that
the SFR overestimation is preserved using different attenuation laws that are commonly employed in the literature (e.g.,
Calzetti et al. 2000; Charlot & Fall 2000), but its trend as a function of the redshift changes. The estimate of Mstar is instead
found to be dependent on both the input radiation field (Umin )
and the attenuation law and is unconstrained if LSST data alone
are employed for the SED fitting.
In Sect. 6 we showed that another efficient way to correct for
the SFR is by exploiting a prior knowledge of AFUV , if this is
available. We stress that the further analysis of the AFUV –Mstar
relation can be useful for future surveys and help to properly
estimate main physical parameters of galaxies without IR observations. As future work, we plan to extend this test to different
SED fitting methods and HELP fields to confirm the systematics
of our results.
Acknowledgements. We would like to thank William Pearson for his help
in understanding our object’s behaviour in the SFR-Mstar diagram. G. R.,
K. M., A. N., M. H. and A. P. acknowledge support from the National Science
Centre (UMO-2018/30/E/ST9/00082, UMO-2018/30/M/ST9/00757 and UMO2020/38/E/ST9/00077). Authors are grateful for the support from Polish Ministry of Science and Higher Education through a grant DIR/WK/2018/12. We
also thank the anonymous referee who has helped clarify and improve various
aspects of this article.
References
Aihara, H., Arimoto, N., Armstrong, R., et al. 2018, PASJ, 70, S4
Arnaboldi, M., Capaccioli, M., Mancini, D., et al. 1998, Messenger, 93, 30
G. Riccio et al.: Preparing for LSST data. Estimating the physical properties of z < 2.5 main-sequence galaxies
Arnouts, S., Cristiani, S., Moscardini, L., et al. 1999, MNRAS, 310, 540
Arnouts, S., Le Floc’h, E., Chevallard, J., et al. 2013, A&A, 558, A67
Béthermin, M., Daddi, E., Magdis, G., et al. 2012, ApJ, 757, L23
Béthermin, M., Wu, H.-Y., Lagache, G., et al. 2017, A&A, 607, A89
Bogdanoska, J., & Burgarella, D. 2020, MNRAS, 496, 5341
Bolzonella, M., Miralles, J. M., & Pelló, R. 2000, A&A, 363, 476
Boquien, M., Burgarella, D., Roehlly, Y., et al. 2019, A&A, 622, A103
Brammer, G. B., van Dokkum, P. G., & Coppi, P. 2008, ApJ, 686, 1503
Brown, M. J. I., Moustakas, J., Smith, J. D. T., et al. 2014, ApJS, 212, 18
Bruzual, G., & Charlot, S. 2003, MNRAS, 344, 1000
Buat, V., Iglesias-Páramo, J., Seibert, M., et al. 2005, ApJ, 619, L51
Buat, V., Takeuchi, T. T., Burgarella, D., Giovannoli, E., & Murata, K. L. 2009,
A&A, 507, 693
Buat, V., Ciesla, L., Boquien, M., Małek, K., & Burgarella, D. 2019, A&A, 632,
A79
Burgarella, D., Buat, V., Gruppioni, C., et al. 2013, A&A, 554, A70
Calzetti, D., Kinney, A. L., & Storchi-Bergmann, T. 1994, ApJ, 429, 582
Calzetti, D., Armus, L., Bohlin, R. C., et al. 2000, ApJ, 533, 682
Calzetti, D., Liu, G., & Koda, J. 2012, ApJ, 752, 98
Chabrier, G. 2003, PASP, 115, 763
Chambers, K. C., Magnier, E. A., Metcalfe, N., et al. 2016, ArXiv e-prints
[arXiv:1612.05560]
Charlot, S., & Fall, S. M. 2000, ApJ, 539, 718
Cid Fernandes, R., Mateus, A., Sodré, L., Stasińska, G., & Gomes, J. M. 2005,
MNRAS, 358, 363
Ciesla, L., Boselli, A., Elbaz, D., et al. 2016, A&A, 585, A43
Conroy, C. 2013, ARA&A, 51, 393
da Cunha, E., Charmandaris, V., Díaz-Santos, T., et al. 2010, A&A, 523, A78
Dale, D. A., Helou, G., Magdis, G. E., et al. 2014, ApJ, 784, 83
Dalton, G. B., Caldwell, M., Ward, A. K., et al. 2006, in Society of Photo-Optical
Instrumentation Engineers (SPIE) Conference Series, eds. I. S. McLean, M.
Iye, et al., SPIE Conf. Ser., 6269, 62690X
Donevski, D., Lapi, A., Małek, K., et al. 2020, A&A, 644, A144
Donley, J. L., Koekemoer, A. M., Brusa, M., et al. 2012, ApJ, 748, 142
Draine, B. T., Aniano, G., Krause, O., et al. 2014, ApJ, 780, 172
Duncan, K. J., Brown, M. J. I., Williams, W. L., et al. 2018, MNRAS, 473, 2655
Eales, S., Dunne, L., Clements, D., et al. 2010, PASP, 122, 499
Elbaz, D., Hwang, H. S., Magnelli, B., et al. 2010, A&A, 518, L29
Elbaz, D., Leiton, R., Nagar, N., et al. 2018, A&A, 616, A110
Emerson, J., McPherson, A., & Sutherland, W. 2006, Messenger, 126, 41
Fabbiano, G. 2006, ARA&A, 44, 323
Fioc, M., & Rocca-Volmerange, B. 2019, A&A, 623, A143
Franzetti, P., Scodeggio, M., Garilli, B., Fumana, M., & Paioro, L. 2008, in
Astronomical Data Analysis Software and Systems XVII, eds. R. W. Argyle,
P. S. Bunclark, & J. R. Lewis, ASP Conf. Ser., 394, 642
González-Solares, E. A., Irwin, M., McMahon, R. G., et al. 2011, MNRAS, 416,
927
Griffin, M. J., Abergel, A., Abreu, A., et al. 2010, A&A, 518, L3
Han, Y., & Han, Z. 2014, ApJS, 215, 2
Hao, C.-N., Kennicutt, R. C., Johnson, B