A Comprehensive Study of Hα Emitters at z ∼ 0.62 in the DAWN Survey: The Need for Deep and Wide Regions

In order to better constrain the bright end of the Hα LF, medium-deep images in eight pointings flanking the deep COSMOS region were obtained as part of the NOAO survey program 2013B-0236 (PI: Finkelstein; Stevens et al. 2020). We present an overview of these eight fields as well as the deep field in Table 1 and Figure 1. The dithering strategy and readout patterns followed were similar to that of C18. Full details regarding the DAWN survey will be presented in an upcoming paper (J. E. Rhoads et al. 2020, in preparation).

Zoom In Zoom Out Reset image size

Table 1.  Observation Summary of DAWN-COSMOS Fields

Pointing	R.A.	Decl.	Int. Time	FWHM	Deptha
	(J2000)	(J2000)	(hr)	(arcsec)	(5σ, AB)
Deep	10:00:30	+02:14:45	81	1.4	23.6
P1	10:02:13	+01:49:27	3	1.4	22.1
P2	09:58:45	+02:14:37	1	1.5	20.1
P3	09:58:46	+02:41:17	1.5	1.4	20.5
P4	10:00:30	+02:41:15	1.67	1.2	20.3
P5	10:02:14	+02:41:02	2	1.3	21.0
P6	10:02:10	+02:15:06	2.5	1.4	21.6
P7	09:58:46	+01:49:47	2	1.3	21.9
P8	10:00:30	+01:50:25	2	1.2	22.0

Note.

^aDepth measurements were based on $2^{\prime\prime}$ apertures.

Download table as:  ASCIITypeset image

The data reduction was performed using the NEWFIRM pipeline (Swaters et al. 2009), which produced images that were calibrated, sky-subtracted, re-projected, and resampled along with their corresponding bad-pixel masks. The seeing FWHM for each of the pointings varied due to changing weather conditions across different observing nights. The final stacked images had slightly different total integration times across different flanking regions (see Table 1).

We used publicly available Y- and J-band images from the UltraVISTA survey DR3 (McCracken et al. 2012) because they are substantially deeper than the NEWFIRM broadband images that were obtained along with our NB data. These images are one of the deepest near-infrared observations of the COSMOS region covering a total area of ∼1.5 deg², reaching 5σ ($2^{\prime\prime}$ aperture, AB) depths of ∼25 mag in Y and ∼24 mag in the JHKs bands. Since VIRCAM/UltraVISTA images have a higher spatial resolution of 0.15 arcsec pixel⁻¹ compared to NEWFIRM, the broadband images were downgraded to a pixel resolution of 0.4 arcsec pixel⁻¹, using SWARP (Bertin et al. 2002) software, to match the NEWFIRM observations. In each case, the resulting image was inspected for any evidence of astrometric mismatch by overlaying center coordinates of known bright point sources from the Two Micron All Sky Survey catalog (Skrutskie et al. 2006) and blinking between images to check for misalignment. The astrometric alignment between images matched well within a single pixel offset (≲01).

2.3.1. Photometric Calibration

2.3.2. Source Detection and Multi-wavelength Photometry

The sample of candidate line emitters includes various kinds of line emitters such as Hα, Hβ/[O iii]_{λλ4959,5007}, and [O ii]_λ3727. The nature of each source, in terms of their line emission, can be determined using several methods. A robust confirmation would be a match with available spectroscopic-redshift catalogs. However, because of a lack of large number of spectroscopic confirmations, a match with photometric-redshift catalogs would be the next best means to categorize these emission-line sources. NB filters are designed such that they are expected to detect line emitters exquisitely, which have strong and narrow emission lines, potentially with little to no continuum detected in the NB. For sources with faint continuum, it is possible that photometric redshifts might be unreliable or even nonexistent. Therefore, for such sources, a color–color calibration based on spectroscopically confirmed sources (and their broadband photometry) can be used for the classification.

For spectroscopic matches, we use a master catalog of spectroscopic redshifts compiled from various past surveys covering the COSMOS region: zCOSMOS (Lilly et al. 2009), 10K-Deep Imaging Multi-Object Spectrograph (Hasinger et al. 2018), 3D-Hubble Space Telescope (Brammer et al. 2012; Momcheva et al. 2016), Very Large Telescope /FORS2 observations (Comparat et al. 2015), Complete Calibration of the Color-Redshift Relation (Masters et al. 2017), Fiber-Multi Object Spectrograph–COSMOS (Silverman et al. 2015), Galaxy Environment Evolution Collaboration 2 (Balogh et al. 2014), COSMOS-[O ii] (Kaasinen et al. 2017), Large Early Galaxy Astrophysics Census (Straatman et al. 2018), MOSFIRE Deep Evolution Field (Kriek et al. 2015), PRIsm MUlti-object Survey (Coil et al. 2011; Cool et al. 2013), MMT/Hectospec observations (Prescott et al. 2006), and Magellan/IMACS observations (Trump et al. 2009). All sources with redshifts in the range 0.6 ≤ z_spec ≤ 0.65 were selected as Hα emitters irrespective of their quality flag, since NB-excess-selected sources are a reaffirmation of the measured spectroscopic redshifts.

Using multiwavelength observations from UV to near-IR, the COSMOS2015 (Laigle et al. 2016) catalog contains one of the largest compilations of photometric redshifts in the ∼2 deg² COSMOS field. Given the uncertainties associated with photometric redshifts, all sources with redshifts in the range 0.57 ≤ z_phot ≤ 0.67 were selected as Hα emitters. Figure 3 shows the spectroscopic and photometric-redshift distribution for all NB-excess selected sources. In both the distributions, there are well-defined peaks at z ∼ 0.62, 1.13, and 1.86, corresponding to the line emitters Hα, Hβ/[O iii], and [O ii] detected by our NB1066 filter, respectively. Since the COSMOS2015 catalog is based on a stacked zYJHKs image, the catalog is highly complete relative to our Hα sample, given that our NB1066 filter overlaps with the Y band. Out of the 1163 candidate line emitters, ∼98% of the sample contained redshift estimates, either photometric or spectroscopic, and in some cases, both.

Zoom In Zoom Out Reset image size

Apart from the redshift-based characterization, we also used broadband photometry available from Laigle et al. (2016) to categorize sources based on color–color selection criteria. For our sample, we used the following color–color criteria to select Hα emitters (Figure 4):

$$\begin{eqnarray}&&\begin{array}{l}(r-{i}^{+})\lt 0.75\ \mathrm{and}\ (r-{i}^{+})\lt (B-V)-0.45\\ \mathrm{and}\ (r-{i}^{+})\gt -1.1(B-V)+0.95,\end{array}\end{eqnarray} \tag{ 3 }$$

$$\begin{eqnarray}&&\begin{array}{l}(V-{i}^{+})\gt 0.5\ \mathrm{and}\ (r-{z}^{++})\lt 1.5\\ \mathrm{and}\ (r-{z}^{++})\lt 1.2(V-{i}^{+})-0.15.\end{array}\end{eqnarray} \tag{ 4 }$$

Zoom In Zoom Out Reset image size

The BVri⁺ criteria separates low-redshift sources (z < 0.5) from all other higher-redshift sources in the sample. After excluding these low-redshift sources, the ${{Vi}}^{+}{{rz}}^{++}$ criteria is used to separate z ∼ 0.62 sources, which are mostly Hα, from other high-redshift sources (mostly, [O iii] and [O ii]) in the sample. A drawback of this method is that some interlopers might get wrongly selected as Hα emitters and some genuine Hα sources might lie outside our color–color selection region. However, we can measure the fraction of contaminating as well as missed sources using spectroscopically confirmed sources, and the total contamination fraction remains relatively low (<10%). Finally, 241 sources were selected as Hα emitters, with 111 sources selected based on spectroscopic redshifts, 110 sources selected using photometric redshifts, and 20 unique sources selected using the broadband color–color criteria.

Using NB1066 and Y-band flux densities, the emission-line fluxes (F_L) and observed equivalent width (EW_obs) for our Hα sample were calculated as follows:

$$\begin{eqnarray}&&{F}_{{\rm{L}}}={\rm{\Delta }}\mathrm{NB}\left(\displaystyle \frac{{f}_{\mathrm{NB}}-{f}_{{\text{}}Y}}{1-({\rm{\Delta }}\mathrm{NB}/{\rm{\Delta }}{\text{}}Y)}\right),\end{eqnarray} \tag{ 5 }$$

$$\begin{eqnarray}&&{\mathrm{EW}}_{\mathrm{obs}}={\rm{\Delta }}\mathrm{NB}\left(\displaystyle \frac{{f}_{\mathrm{NB}}-{f}_{{\text{}}Y}}{{f}_{{\text{}}Y}-{f}_{\mathrm{NB}}({\rm{\Delta }}\mathrm{NB}/{\rm{\Delta }}{\text{}}Y)}\right),\end{eqnarray} \tag{ 6 }$$

where ΔNB and ΔY are the filter widths (FWHM in Å), and f_NB and f_Y are the flux densities (erg s⁻¹ cm⁻² Å⁻¹) for NB1066 and the Y band, respectively. The corresponding line luminosities are derived assuming z = 0.62, which is the median redshift of our Hα sample.

The intrinsic Hα luminosity can be derived using the observed line luminosity after correcting for contamination due to adjacent [N ii]_{λλ6548,6584} lines, as well as attenuation due to dust.

For typical L_* galaxies in the nearby universe, past surveys have adopted corrections based on the typical Hα/[N ii] flux ratio of 2.3 (Kennicutt 1992; Gallego et al. 1997). However, recent NB surveys (Villar et al. 2008; Ly et al. 2011; Sobral et al. 2013) have adopted EW-dependent corrections based on the mean relationship between the rest-frame EW of Hα+[N ii]λ6583 and the Hα/[N ii] ratio.

Unlike previous surveys, the NB filter, NB1066, in DAWN is relatively narrow such that, for any Hα source at z ∼ 0.62, only one of the [N ii] lines is expected to be contaminating the Hα line flux. For example, when Hα is detected on the bluer side of the filter, the redder [N ii]_λ6584 line is expected at the wings of the filter (<50% transmission). However, if Hα is detected on the redder part, the bluer [N ii]_λ6548 line should fall in a reasonably transmissive portion of the filter. Owing to a lack of spectroscopic redshift for each source in our Hα sample, deriving individual correction is fairly difficult given that it is highly impractical to determine which one of the [N ii] lines is responsible for contamination in each source. Therefore, the necessary corrections in this case are intrinsically different compared to other surveys.

For these aforementioned reasons, we derive a luminosity-dependent [N ii] correction as follows: we generate a mock galaxy sample with luminosities and redshifts based on the observed luminosity distribution and a uniform redshift distribution comprising redshifts probed by our NB1066 filter. Assuming a fixed [N ii]/Hα value of 0.43 (Kennicutt 1992; Gallego et al. 1997), we derive Hα luminosities for all sources in the mock sample after convolving them through our NB1066 filter curve (case "A"). In a similar way, we also derive Hα luminosities assuming [N ii]/Hα ∼0 for the mock sample (case "B"). Using the Hα luminosity distribution from case "A" and "B," we calculate their ratio as a function of luminosity, which provides an estimate of the factor by which objects have been overcounted per luminosity bin. The resulting correction factor is applied to the LFs in Section 5.1 which corrects for the presence of [N ii] within our Hα sample.

Dust obscuration is a significant source of uncertainty in UV and optical measurements of galaxy properties including SFR. Although Hα emission is less affected by dust compared to the UV continuum, correcting Hα luminosities for dust is necessary to accurately measure SFR. Ideally, dust corrections applied to galaxies should be measured individually, for example, based on Balmer decrements (Hα/Hβ), but that requires rest-frame optical/near-IR spectra for each galaxy (e.g., Reddy et al. 2015), which is practically infeasible for large samples.

In the past, some studies (e.g., Sobral et al. 2013) have adopted a simple dust correction of A_Hα = 1 assuming that dust affects all sources in the sample equally, whereas some others (e.g., Ly et al. 2011) have assumed that dust extinction in galaxies depends on their SFR/luminosity (Hopkins et al. 2001) or stellar-mass (Garn & Best 2010) and hence apply corrections accordingly. Sobral et al. (2013) believe that the typical extinction in a galaxy need not necessarily depend on its SFR/luminosity in an absolute manner, but rather depends on the nature of the source (meaning the extent to which it is star-forming or luminous) relative to the normal star-forming galaxy at a particular epoch.

In order to explore the effects of different dust-extinction correction on LF and SFRD, we correct our Hα luminosities following all aforementioned prescriptions and analyze them separately hereafter. For luminosity-dependent extinction correction, we adopted the following relation given by Ly et al. (2012):

$$\begin{eqnarray}&&\mathrm{log}({L}_{\mathrm{obs}})=\mathrm{log}({L}_{\mathrm{int}})-2.36\ \mathrm{log}\left[\displaystyle \frac{0.797\ \mathrm{log}({L}_{\mathrm{int}})-29.1}{2.86}\right],\end{eqnarray} \tag{ 7 }$$

where L_obs and L_int are the observed and intrinsic Hα luminosities (erg s⁻¹), respectively. As mentioned earlier, this extinction correction is based on the SFR-dependent formalism derived by Hopkins et al. (2001), which demonstrates that Hα luminosity directly correlates with SFR, meaning dust reddening will be higher for sources with higher Hα luminosity. In case of stellar-mass-dependent correction, the dust extinction is computed according to the following relation given by Garn & Best (2010):

$$\begin{eqnarray}&&{{\rm{A}}}_{{\rm{H}}\alpha }=0.91X+0.77{X}^{2}+0.11{X}^{3}-0.09{X}^{4},\end{eqnarray} \tag{ 8 }$$

where X = log₁₀(M_*/10¹⁰ M_⊙). For our Hα sample, the stellar-mass estimates available from COSMOS2015 (Laigle et al. 2016) are used to derive the extinction corrections. Since several sources in our sample have stellar masses log(M_*/M_⊙) < 8.5, where the parameterization does not account for such low stellar-mass sources, we assume a fixed dust correction, A_Hα = 0.3 mag, corresponding to the extinction correction derived for a source with log(M_*/M_⊙) ∼ 8.5, for all such sources.

Given our methods of detection and selection of Hα emitters, we have to estimate the incompleteness arising out of this process and apply an appropriate correction for each source in our sample. Based on the procedure suggested by C18, we estimate the completeness fraction of our sample. Briefly put, artificial sources are randomly superimposed on the science image. The standard detection and selection methods are followed to determine the number of sources recovered. A comparison between the number of emission-line detected sources (N_detected) and the number of artificial (N_artificial) plus real (N_real) sources present in the image provides us with a recovery fraction for the sample:

$$\begin{eqnarray}&&\kappa =\displaystyle \frac{{N}_{\mathrm{detected}}-{N}_{\mathrm{real}}}{{N}_{\mathrm{artificial}}}\end{eqnarray} \tag{ 9 }$$

This procedure is repeated once for each bin across a range of luminosities and EW_obs.

Owing to different image depths across the deep and flanking regions, the completeness simulation was performed individually for each region. Simulations were performed for each of the 600 bins across a luminosity range 10^39.7–10^42.7 L_⊙ (Δ ∼ 0.1 dex), and an EW_obs range 0–200 Å (Δ ∼ 10 Å). Since the total number of sources detected in the flanking regions is less than half the number detected in the deep region, the completeness simulation for flanking regions included 5000 artificial sources, whereas the simulation for the deep region included 10,000 artificial sources. The completeness correction thereby computed was applied to each source, depending on its luminosity and EW_obs, within each region.

Figure 5 shows completeness fractions for the deep and flanking regions. We adopt a 20% completeness limit for each luminosity–EW_obs bin while applying corrections for sources in a particular region.

Zoom In Zoom Out Reset image size

Hα LFs for this survey are derived using the V/V_max method (Schmidt 1968). In this work, since the sample of Hα emitters are selected from regions of varying imaging depths, we construct and analyze LFs separately for the deep region, the shallow flanking regions, and the full DAWN–COSMOS region (∼1.5 deg²). Following Section 4.3, the LFs presented hereafter are derived based on both prescriptions of dust correction, for comparison purposes. Unless otherwise specified, the errors for each LF bin are Poissonian, with an additional error of 20% added in quadrature to account for the uncertainty in completeness corrections.

Each LF presented in this work can be modeled based on the typical Schechter profile (Schechter 1976) defined as follows:

$$\begin{eqnarray}&&\phi (L){dL}={\phi }^{* }{\left(\displaystyle \frac{L}{{L}^{* }}\right)}^{\alpha }{\rm{\exp }}\left(-\displaystyle \frac{L}{{L}^{* }}\right)\left(\displaystyle \frac{{dL}}{{L}^{* }}\right).\end{eqnarray} \tag{ 10 }$$

In the log form, this function can be defined as

$$\begin{eqnarray}&&\phi (L){dL}={\rm{ln}}(10)\ {\phi }^{* }{\left(\displaystyle \frac{L}{{L}^{* }}\right)}^{\alpha +1}{\rm{\exp }}\left(-\displaystyle \frac{L}{{L}^{* }}\right)d({\rm{log}}\ L).\end{eqnarray} \tag{ 11 }$$

Many Hα surveys in the past have been able to successfully model their LF using the Schechter function; in further sections, we show that this holds good for our Hα LF as well. We adopt a 20% completeness limit in terms of the Hα LF, for all calculations hereafter.

The best-fit Schechter parameters and their associated 1σ uncertainties are determined using MCMC simulations based on the Metropolis–Hastings algorithm. The simulation involves the following steps: (1) an initial guess for the Schechter parameters is validated against a uniform prior ($-4\lt \mathrm{log}\ {\phi }^{* }\ {\mathrm{Mpc}}^{-3}\lt -1$, $40\lt \mathrm{log}\ {L}^{* }\ \mathrm{erg}\ {{\rm{s}}}^{-1}\lt 44$ and −2 < α < 0). (2) Each iteration determines the goodness of the Schechter fit to the given Hα LF based on the χ² statistic. (3) The entire parameter space for all Schechter parameters is explored over 500,000 iterations and their probability distributions are derived. The median and 1σ estimates from these distributions correspond to the best-fit Schechter parameters and their errors for our Hα LF, respectively.

The LFs derived using the Hα subsample from the deep region alone are presented in Figure 6 (left), which are completeness- and dust-corrected. We derive three different LFs (and their respective Schechter fits) based on the three prescriptions of dust-extinction correction used in this work. In either of the LFs, it is seen that the faint-end slopes are steep and consistent with the canonical value of α = −1.6, which is observed among most Hα LFs from recent NB surveys at z ∼ 0–2 (e.g., Ly et al. 2011; Sobral et al. 2013; Gómez-Guijarro et al. 2016). However, since the volume probed by this region is relatively small (∼0.25 deg²), the brighter Hα population (L${}_{{\rm{H}}\alpha }\gt {10}^{42}$ erg s⁻¹) is sparsely sampled, therefore the bright end of the LF is weakly constrained.

Zoom In Zoom Out Reset image size

Using the Hα subsample from just the flanking regions, the derived LFs are shown in Figure 6 (right). In contrast to the deep region, the flanking regions contain sample-significant numbers of bright Hα sources (L_Hα ≥ 10⁴² erg s⁻¹). Unlike in the case of deep region, the bright end of the Schechter fit, which essentially depicts the break from the power-law form of the Schechter function, is better constrained for these flanking-region LFs. On the contrary, the faint end is hardly constrained owing to the lack of faint-luminosity sources, which is expected given the shallow exposures. Therefore, the Schechter parameters for these LFs, especially the faint-end slope (α), should not be viewed as significant implications of this work.

For the full Hα sample (deep and flanking regions combined), the resulting LFs and their Schechter fits are shown in Figure 7 in comparison with Hα LFs from previous surveys at various redshifts, z ∼ 0–2 (Ly et al. 2007, 2011; Sobral et al. 2013). The best-fit Schechter parameters are given in Table 2. Considering the empirical relation given for L* and ϕ* as a function of redshift from Sobral et al. (2013), our values are consistent with the expected value from this relation at z ∼ 0.62. We find that the characteristic luminosity, as well as the normalization parameters, are higher compared to those at lower redshifts. However, the faint-end slope is significantly steeper in the case of the LF based on constant dust correction compared to the LF with luminosity-dependent or stellar-mass-dependent dust correction. Using Hopkins et al. (2001) dust correction, previous studies have suggested that the faint-end slope tends to flatten out due to an increase in the number density on the bright end resulting from a higher correction applied for brighter sources (e.g., Villar et al. 2008; An et al. 2014). Our faint-end slope using the same dust correction is consistent with those values. On the other hand, studies employing constant dust correction, A_Hα = 1, observe a steeper faint-end slope, α = −1.6, in their LFs (e.g., Ly et al. 2007; Sobral et al. 2009, 2013, who argue that there is mild dependence of extinction on observed luminosity (Sobral et al. 2012) and a median correction of ∼1 mag holds good on average for large samples. Assuming constant dust correction, the best-fit faint-end slope of our LF, α = −1.62, is in agreement with these studies.

Zoom In Zoom Out Reset image size

Table 2.  Schechter Parameters of the LF and SFR Density Estimate for z ∼ 0.62 Hα Emitters in DAWN–COSMOS

Region	Dust correction	L*	ϕ*	α	log ${ \mathcal L }$	log ρSFR
	(AHα)	(erg s−1)	(Mpc−3)		(erg s−1 Mpc−3)	(M⊙ yr−1 Mpc−3)
COSMOS	Luminosity-dependent	${42.31}_{-0.18}^{+0.27}$	$-{2.80}_{-0.34}^{+0.22}$	$-{1.39}_{-0.14}^{+0.13}$	${39.68}_{-0.06}^{+0.06}$	$-{1.47}_{-0.06}^{+0.06}$
	Stellar-mass-dependent	${42.36}_{-0.13}^{+0.35}$	$-{2.91}_{-0.43}^{+0.28}$	$-{1.48}_{-0.14}^{+0.16}$	${39.69}_{-0.07}^{+0.08}$	$-{1.46}_{-0.07}^{+0.08}$
	Constant	${42.24}_{-0.21}^{+0.39}$	$-{2.85}_{-0.42}^{+0.31}$	$-{1.62}_{-0.16}^{+0.18}$	${39.76}_{-0.09}^{+0.08}$	$-{1.39}_{-0.09}^{+0.08}$

Download table as:  ASCIITypeset image

Using the best-fit Schechter parameters, the total Hα luminosity density can be calculated as follows:

$$\begin{eqnarray}&&{ \mathcal L }={L}^{* }{\phi }^{* }{\rm{\Gamma }}(2+\alpha ).\end{eqnarray} \tag{ 12 }$$

Following this, the SFRD can be calculated using the standard calibration of Kennicutt (1998): ${\rho }_{{\mathsf{SFR}}}=7.9\times {10}^{-42}{ \mathcal L }$, where ${ \mathcal L }$ is calculated by fully integrating down the LFs.

Although most of the Hα luminosity density can be due to active star formation in galaxies, some contribution is usually attributed to active galactic nucleus (AGN) activity as well. Studies in the past have found AGN contamination to be ∼10%–15% for Hα samples at redshifts z < 2 (e.g., Villar et al. 2008; Ly et al. 2011; Sobral et al. 2013). One way to account for AGN contamination is to look for X-ray-identified sources matching with our Hα sample. The COSMOS2015 catalog contains X-ray sources drawn from XMM-COSMOS (Cappelluti et al. 2007; Hasinger et al. 2007; Brusa et al. 2010) and Chandra-COSMOS (Elvis et al. 2009; Civano et al. 2012, 2016) surveys. The X-ray luminosity limit is ∼10⁴² erg s⁻¹ at z ∼ 0.6 (Marchesi et al. 2016) and assuming a typical X-ray to Hα ratio, log(L_X/${L}_{{\rm{H}}\alpha }$) ∼ 1–2 (Ho et al. 2001; Panessa et al. 2006; Shi et al. 2010), any AGN-powered Hα emitter from either flanking or deep region should be detected in X-rays. We find five X-ray matches for our Hα sample, which suggests that AGN contamination is ∼2% of the total sample and ∼4% of the flanking-region subsample (where it is expected to be complete for X-ray-luminous AGNs). However, since these X-ray surveys are flux-limited, these matches alone are not a representative of the AGN contamination in our sample.

Another method to assess AGN contamination is to use the mid-IR color criterion based on the differing spectral energy distributions (SEDs) of star-forming galaxies and AGNs around the rest-frame 1.6 μm bump. For AGNs, the SED is a rising power law after the bump due to the presence of emission from polycyclic aromatic hydrocarbons and silicate grains. Using Ks-band and IRAC CH1 (3.6 μm) photometry from COSMOS2015, we measure [Ks − 3.6] color for our sample, where redder colors ([Ks − 3.6] > 0.27) represent AGNs and bluer colors are mostly star-forming galaxies. Following this criterion, we find 11% of our sample to be AGN-contaminated.

After correcting for AGN contamination, we estimate the SFRD at z ∼ 0.62 to be log ${\rho }_{{\mathsf{SFR}}}=-1.47$ for luminosity-dependent dust correction, log ${\rho }_{{\mathsf{SFR}}}=-1.46$ for stellar-mass-dependent dust correction, and log ${\rho }_{{\mathsf{SFR}}}=-1.39$ for constant dust correction (Table 2). For DAWN-COSMOS, the cosmic variance uncertainties were estimated to be around 23% based on calculations from Driver & Robotham (2010). Figure 8 shows a comparison of SFRD as a function of redshift for DAWN and other Hα based surveys (e.g., Ly et al. 2007, 2011; Morioka et al. 2008; Shioya et al. 2008; Villar et al. 2008; Sobral et al. 2009, 2013; Westra et al. 2010; Stroe & Sobral 2015; Gómez-Guijarro et al. 2016, Khostovan et al. 2020). We also compare our SFRD estimate to the empirical fits given by Sobral et al. (2013) and Madau & Dickinson (2014). Using Hα samples at z ∼ 0.4, 0.8, 1.47, 2.23 from HiZELS, Sobral et al. (2013) provide an empirical fit for SFRD as a function of redshift, log ${\rho }_{{\mathsf{SFR}}}=-2.1/(z+1)$. In Madau & Dickinson (2014), an empirical fit for SFRD is derived based on measurements from a host of recent UV and IR galaxy surveys. Within 1σ uncertainties, our SFRD estimates are consistent with these fits, as shown in Figure 8. The ${\rho }_{{\mathsf{SFR}}}$ based on luminosity-dependent dust correction is slightly lower, mostly due to the fact that the correction is unequal across the sample given the luminosity-dependent relation and also that some of the faintest Hα emitters in the sample require no correction, according to this dust-correction method.

Zoom In Zoom Out Reset image size