Bo\"otes-HiZELS: an optical to near-infrared survey of emission-line galaxies at $\bf z=0.4-4.7$

TL;DR

This survey of approximately 1000 emission-line galaxies across redshifts 0.4 to 4.7 uses multiple narrow-band filters to analyze galaxy evolution, emission-line properties, and the impact of AGN activity.

Contribution

First comprehensive optical to near-infrared emission-line galaxy survey at high redshift with detailed catalogs, redshift confirmation, and analysis of luminosity evolution and line ratios.

Findings

Confirmed strong luminosity evolution of star-forming galaxies from z~0.4 to 5.

Measured increase in [OII]/Hα ratio with redshift, suggesting decreasing dust attenuation.

Identified deviation from Schechter function at high luminosities due to AGN contribution.

Abstract

We present a sample of $\sim 1000$ emission line galaxies at $z=0.4-4.7$ from the $\sim0.7$deg$^2$ High-$z$ Emission Line Survey (HiZELS) in the Bo\"otes field identified with a suite of six narrow-band filters at $\approx 0.4-2.1$ $\mu$m. These galaxies have been selected on their Ly$\alpha$ (73), {\sc [Oii]} (285), H$\beta$/{\sc [Oiii]} (387) or H$\alpha$ (362) emission-line, and have been classified with optical to near-infrared colours. A subsample of 98 sources have reliable redshifts from multiple narrow-band (e.g. [O{\sc ii}]-H$\alpha$) detections and/or spectroscopy. In this survey paper, we present the observations, selection and catalogs of emitters. We measure number densities of Ly$\alpha$, [O{\sc ii}], H$\beta$/{\sc [Oiii]} and H$\alpha$ and confirm strong luminosity evolution in star-forming galaxies from $z\sim0.4$ to $\sim 5$, in agreement with previous results. To demonstrate the usefulness of dual-line emitters, we use the sample of dual [O{\sc ii}]-H$\alpha$ emitters to measure the observed [O{\sc ii}]/H$\alpha$ ratio at $z=1.47$. The observed [O{\sc ii}]/H$\alpha$ ratio increases significantly from 0.40$\pm0.01$ at $z=0.1$ to 0.52$\pm0.05$ at $z=1.47$, which we attribute to either decreasing dust attenuation with redshift, or due to a bias in the (typically) fiber-measurements in the local Universe which only measure the central kpc regions. At the bright end, we find that both the H$\alpha$ and Ly$\alpha$ number densities at $z\approx2.2$ deviate significantly from a Schechter form, following a power-law. We show that this is driven entirely by an increasing X-ray/AGN fraction with line-luminosity, which reaches $\approx 100$ \% at line-luminosities $L\gtrsim3\times10^{44}$ erg s$^{-1}$.

Tables

Table 1. Table 1: Description of the available (archival and new) multi-wavelength data in Boötes-HiZELS, with narrow-band filters highlighted in bold. The abbreviations for the archival surveys are: LBFS – LBT Boötes Field Survey (Bian et al., 2013 ) ; NDWFS – NOAO Deep Wide Field Survey (Jannuzi & Dey, 1999 ) ; IBIS – Infrared Bootes Imaging Survey (Gonzalez et al., 2010 ) . λ c subscript 𝜆 𝑐 \lambda_{c} is the central wavelength of the filter and Δ ​ λ Δ 𝜆 \Delta\lambda is the width between the full width half maxima of the filter transmission. The full width half maximum (FWHM) of the point spread function has been measured as described in § 2.2 . We list the total exposure time per pixel and its variance. For NB921 and z 𝑧 z , 60 % of the coverage has the highest exposure time listed. Depths are measured by measuring the sky value in 100,000 3 ′′ apertures on blank regions in the images as described in § 3.2.2 . The coverage is after masking each individual filter for uncovered regions or regions with insufficient depth.

Filter	Telescope	Survey	${\lambda }_c$	$\Delta \lambda$	FWHM	Exposure time	Dates	Depth	Coverage
			[nm]	[nm]	[′′]	[ks]		[3$\sigma$, AB]	[deg2]
$U$	LBT	LBFS	359	54	1.2			25.3	0.78
NB392	INT	This survey	392	5.2	1.8	12.4$\pm 2.0$	2013 Jun 6-10; 2014 Feb 27, Mar 1-8,27	24.3	0.54
$B_w$	Mayall	NDWFS	464	110	1.4			25.4	0.78
stV	INT	This survey	410	16	1.9	3.8$\pm 0.8$	2013 Jun 6-10; 2014 Mar 2, 5-7; 2016 Jun 11-13	24.1	0.63
$g$	INT	This survey	485	129	1.6	6.0$\pm 0.0$	2016 Jun 6-8	24.9	0.78
NB501	INT	This survey	501	10	1.6	8.4$\pm 0.1$	2015 Apr 11, 12, 16, 17; 2016 Jun 5, 6, 7, 10, 12	24.7	0.74
$R$	Mayall	NDWFS	602	160	1.1			25.0	0.78
$I$	Mayall	NDWFS	754	170	1.1			24.4	0.78
$z$	Subaru	This survey	878	113	0.8	1.0-2.1	2014 May 28, 29	24.3	0.76
NB921	Subaru	This survey	919	13	0.8	2.16-2.52	2014 May 28, 29	24.0	0.46
$Y$	LBT	LBFS	984	42	0.8			23.1	0.78
$J$	NEWFIRM	IBIS	1300	190	1.0			22.9	0.78
$H$	UKIRT	This survey	1600	200	0.8	1.2$\pm 0.1$	2010 April 2-8	22.6	0.78
NBH	UKIRT	This survey	1620	21	0.8	15.4$\pm 1.4$	2010 April 2-6	22.1	0.74
$K$	NEWFIRM	IBIS	2260	280	1.2			22.1	0.78
NBK	UKIRT	This survey	2120	21	1.2	20.2$\pm 0.0$	2010 April 8, 9, 14, July 18-23; 2011 Feb 16, 27, 28, Mar 9, 12, 16, 20, 22-26	22.3	0.73

Table 2. Table 2: Spectroscopically and dual-NB confirmed emission-lines observed in narrow-band filters. We note that spectroscopic redshifts are highly biased towards AGN, and that the spectroscopic redshift distribution does not resemble the real redshift distribution, particularly for fainter line-emitters. Dual-NB redshifts are only available at z = 1.47 𝑧 1.47 z=1.47 and z = 2.23 𝑧 2.23 z=2.23 , see Fig. 1 .

Filter	Emission-line	Redshift	# $z_{\mathrm{spec}}$	# $z_{\mathrm{dualNB}}$
NB392	Mgiiλ2798	0.39–0.41	8	-
	Ciii]λ1909	1.04–1.07	1	-
	Civλ1549	1.51–1.55	2	-
	Ly${\alpha }_{\lambda 1216}$	2.20–2.24	2	5
stV	Mgiiλ2798	0.42–0.50	1	-
	Ciii]λ1909	1.10–1.20	3	-
	Heiiλ1640	1.44–1.56	1	-
	Civλ1549	1.59–1.71	4	-
	Ly${\alpha }_{\lambda 1216}$	2.30–2.45	6	-
NB501	[Oii]λ3727	0.32–0.36	1	-
	Mgiiλ2798	0.76–0.81	1	-
	Civλ1549	2.19–2.27	1	3
	Ly${\alpha }_{\lambda 1216}$	3.06–3.17	5	-
NB921	H${\alpha }_{\lambda 6563}$	0.39–0.41	5	-
	[Oiii]λλ4959,5007	0.82–0.87	1	-
	[Oii]λ3727	1.44–1.48	0	20
	Mgiiλ2798	2.25–2.31	1	3
NBH	H${\alpha }_{\lambda 6563}$	1.44–1.48	6	21
	[Oiii]λλ4959,5007	2.19–2.29	3	16
NBK	H${\alpha }_{\lambda 6563}$	2.21–2.25	8	20

Table 3. Table 3: Colour selection-criteria used to classify line-emitters. These criteria are based on the expected positions of the Lyman- and Balmer-breaks, see for example the B ​ z ​ K 𝐵 𝑧 𝐾 BzK criterion from Daddi et al. ( 2004 ) . These criteria are then fine-tuned using spectroscopic and dual-NB redshifts. In comparison to traditional criteria that use the B 𝐵 B band, we adjust the B w subscript 𝐵 𝑤 B_{w} magnitude for the contribution from flux in the U 𝑈 U band. For samples at z > 2.4 𝑧 2.4 z>2.4 , we also include a colour-criterion that removes very red objects for which the Lyman-break criterion has selected a strong Balmer break; see also Xue et al. ( 2017 ) . The colour-colour selections are illustrated in the figures in Appendix B .

Filter	Emission-line	Colour-criterion
NB392	Ly$\alpha$ $z=2.2$
stV	Ly$\alpha$ $z=2.4$
NB501	Ly$\alpha$ $z=3.1$
NB921	H$\alpha$ $z=0.40$
	[Oiii]/H$\beta$ $z=0.80$
	[Oii] $z=1.47$
NBH	$z>1$	or $z-K>2$
	H$\alpha$ $z=1.47$	$z>1$ &
	[Oiii]/H$\beta$ $z=2.23$	$z>1$ & $J-K>2.1\left(I-J\right)-1$
	[Oii] $z=3.3$	Not H$\alpha$ or [Oiii]/H$\beta$ in NBH &
NBK	$z>1$	or $z-K>2$
	H$\alpha$ $z=2.23$	$z>1$ &
	[Oiii]/H$\beta$ $z=3.2$	$z>1$ &
	[Oii] $z=4.7$	Not H$\alpha$ or [Oiii]/H$\beta$ in NBK & $g-I>1.5$

Table 4. Table 4: Line-identifications of the total ∼ 2000 similar-to absent 2000 \sim 2000 emitters (as described in § 4 ) in the Boötes-HiZELS narrow-band filters.

Filter	Sub-sample	# of sources
NB392	$\Sigma >3$, EW$>30$ Å	57
	Ly$\alpha$ at $z=2.23$	25
stV	$\Sigma >3$, EW$>130$ Å	39
	Ly$\alpha$ at $z=2.4$	16
NB501	$\Sigma >3$, EW$>50$ Å	65
	Ly$\alpha$ at $z=3.1$	32
NB921	$\Sigma >3$, EW$>30$ Å	1161
	H$\alpha$ at $z=0.40$	198
	[Oiii]/H$\beta$ at $z=0.8$	304
	[Oii] at $z=1.47$	277
NBH	$\Sigma >3$, EW$>85$ Å	301
	H$\alpha$ at $z=1.47$	87
	[Oiii]/H$\beta$ at $z=2.23$	72
	[Oii] at $z=3.3$	6
NBK	$\Sigma >3$, EW$>80$ Å	255
	H$\alpha$ at $z=2.23$	77
	[Oiii]/H$\beta$ at $z=3.2$	11
	[Oii] at $z=4.7$	2

Table 5. Table 5: Survey volumes and flux completenesses for the various line-emitters in this survey.

Filter	Emission-line	Volume	50 % completeness
		[105 Mpc3]	[erg s-1 cm-2]
NB392	Ly$\alpha$ $z=2.2$	2.8	$1.3\times {10}^{-16}$
stV	Ly$\alpha$ $z=2.4$	9.5	$4.7\times {10}^{-16}$
NB501	Ly$\alpha$ $z=3.1$	7.2	$1.1\times {10}^{-16}$
NB921	H$\alpha$ $z=0.4$	0.2	$1.0\times {10}^{-16}$
	H$\beta$/[Oiii] $z=0.8$	1.2	$1.0\times {10}^{-16}$
	[Oii] $z=1.47$	1.7	$1.0\times {10}^{-16}$
NBH	H$\alpha$ $z=1.47$	2.5	$1.3\times {10}^{-16}$
	H$\beta$/[Oiii] $z=2.2$	5.2	$1.3\times {10}^{-16}$
NBK	H$\alpha$ $z=2.2$	2.7	$0.5\times {10}^{-16}$

Table 6. Table 6: List of sources that are observed as dual-emitters (line-emitters in at least two narrow-bands). We present the coordinates, spectroscopic and dual-NB redshifts, I 𝐼 I band magnitudes and the list of emission-lines that are detected. X-Ray and LOFAR detected sources are marked. We note that because the coverage in NB392, stV and NB921 is not homogeneous, the number of Ly α 𝛼 \alpha detections at z = 2.23 𝑧 2.23 z=2.23 and [O ii ] at z = 1.47 𝑧 1.47 z=1.47 is likely underestimated. An electronic version of this table is available online.

ID	R.A.	Dec.	$z_{\mathrm{dual}-\mathrm{NB}}$	$z_{\mathrm{spec}}$	$I$	Note
B-HiZELS_1	14:33:19.29	+33:34:31.53	2.23	2.23	18.9	Ly$\alpha$+[Oiii] + H$\alpha$.
B-HiZELS_2	14:32:58.85	+33:25:49.33	2.23		19.6	H$\alpha$ + Ly$\alpha$ + Nv, X-Ray detected.
B-HiZELS_3	14:31:41.51	+33:49:11.27	2.23	2.26	20.4	[Oiii] + H$\alpha$ + Civ +Ly$\alpha$, X-Ray detected.
B-HiZELS_4	14:29:40.05	+33:33:32.13	2.23	2.27	20.6	[Oiii] + H$\alpha$ + Mgii, X-Ray detected.
B-HiZELS_5	14:32:01.52	+33:16:59.86	2.23		20.9	[Oiii] + H$\alpha$
B-HiZELS_6	14:30:38.29	+33:20:17.85	2.23		21.3	[Oiii]+H$\alpha$, X-Ray detected.
B-HiZELS_7	14:29:30.87	+34:05:44.85	2.23		21.7	[Oiii]+H$\alpha$
B-HiZELS_8	14:32:13.88	+33:25:57.48	2.23		21.9	[Oiii] + H$\alpha$ + Ly$\alpha$, LOFAR detected.
B-HiZELS_9	14:31:06.64	+33:46:19.18	2.23		22.1	[Oiii] + H$\alpha$
B-HiZELS_10	14:33:07.48	+33:52:42.48	1.47		22.1	[Oii] + H$\alpha$
B-HiZELS_11	14:32:37.05	+33:33:56.18	1.47		22.1	[Oii] + H$\alpha$
B-HiZELS_12	14:30:51.00	+33:43:54.56	2.23		22.3	[Oiii]+H$\alpha$
B-HiZELS_13	14:33:14.40	+33:46:53.17	1.47		22.4	[Oii] + H$\alpha$
B-HiZELS_14	14:32:12.54	+33:22:26.15	1.47		22.4	[Oii] + H$\alpha$
B-HiZELS_15	14:32:32.59	+33:59:03.32	2.23	2.24	22.5	[Oiii] + H$\alpha$ + Civ, X-Ray & LOFAR detected.
B-HiZELS_16	14:30:40.31	+34:03:20.64	2.23		22.8	Ly$\alpha$, Civ and H$\alpha$.
B-HiZELS_17	14:31:33.54	+34:02:48.52	2.23		22.8	[Oiii] + H$\alpha$
B-HiZELS_18	14:32:52.90	+33:39:43.29	1.47		22.8	[Oii] + H$\alpha$
B-HiZELS_19	14:30:16.12	+33:17:09.56	1.47		23.0	[Oii] + H$\alpha$
B-HiZELS_20	14:30:28.28	+33:38:16.75	1.47		23.0	[Oii] + H$\alpha$
B-HiZELS_21	14:30:19.38	+33:37:10.18	1.47		23.1	[Oii] + H$\alpha$
B-HiZELS_22	14:33:15.16	+33:50:09.62	1.47		23.1	[Oii] + H$\alpha$
B-HiZELS_23	14:32:37.05	+33:33:06.64	1.47		23.1	[Oii] + H$\alpha$, LOFAR detected.
B-HiZELS_24	14:30:12.20	+33:51:57.33	2.23		23.3	[Oiii] + H$\alpha$
B-HiZELS_25	14:30:26.29	+33:28:51.17	2.23		23.3	[Oiii] + H$\alpha$
B-HiZELS_26	14:30:39.53	+33:57:09.60	1.47		23.3	[Oii] + H$\alpha$
B-HiZELS_27	14:33:19.61	+33:34:36.66	2.23		23.4	[Oiii] + H$\alpha$
B-HiZELS_28	14:31:50.68	+33:18:44.15	1.47		23.4	[Oii] + H$\alpha$
B-HiZELS_29	14:30:28.56	+33:33:29.07	2.23		23.5	Ly$\alpha$ and H$\alpha$
B-HiZELS_30	14:30:54.03	+33:33:01.26	1.47		23.5	[Oii] + H$\alpha$
B-HiZELS_31	14:31:19.33	+33:26:14.90	1.47		23.5	[Oii] + H$\alpha$
B-HiZELS_32	14:32:48.04	+33:57:18.93	2.23		23.8	[Oiii] + H$\alpha$
B-HiZELS_33	14:30:45.55	+33:23:50.12	1.47		23.8	[Oii] + H$\alpha$
B-HiZELS_34	14:30:29.53	+33:20:49.91	1.47		23.8	[Oii] + H$\alpha$
B-HiZELS_35	14:30:18.58	+34:03:24.88	2.23		23.9	[Oiii] + Ly$\alpha$
B-HiZELS_36	14:31:11.70	+33:41:28.69	1.47		24.0	[Oii] + H$\alpha$
B-HiZELS_37	14:33:21.85	+33:54:50.65	1.47		24.2	[Oii] + H$\alpha$
B-HiZELS_38	14:33:22.55	+33:48:04.57	1.47		24.3	[Oii] + H$\alpha$
B-HiZELS_39	14:29:44.27	+33:50:43.84	1.47		24.3	[Oii] + H$\alpha$
B-HiZELS_40	14:31:57.69	+33:16:37.87	2.23		24.9	[Oii]+H$\alpha$
B-HiZELS_41	14:32:41.50	+33:26:18.45	2.23		25.5	[Oiii] + H$\alpha$
B-HiZELS_42	14:30:39.05	+33:51:51.16	2.23		25.8	[Oiii]+H$\alpha$

Table 7. Table 7: First five entries in the catalog of NB392 line-emitters. Coordinates are in J2000. Line-flux is in 10 − 16 superscript 10 16 10^{-16} erg s -1 cm -2 . EW obs is in Å, where -99 entries mark sources without secure continuum measurement, and have EW > obs 550 {}_{\rm obs}>550 Å. Entries with 99 are undetected in I 𝐼 I band. Flag_Class: 0: Unclassed, 1: LAE at z = 2.2 𝑧 2.2 z=2.2 . A full version of this table is available online.

ID	R.A.	Dec.	Line-flux	EWobs	$I$	Flag_Class
B-HiZELS_NB392_1	217.945	33.310	1.3	136	22.2	0
B-HiZELS_NB392_2	218.017	33.316	1.9	119	20.1	0
B-HiZELS_NB392_3	217.779	33.340	1.2	158	23.6	1
B-HiZELS_NB392_4	218.056	33.366	2.1	-99	99	0
B-HiZELS_NB392_5	217.675	33.386	4.2	56	21.9	0

Table 8. Table 8: First five entries in the catalog of stV line-emitters. Coordinates are in J2000. Line-flux is in 10 − 16 superscript 10 16 10^{-16} erg s -1 cm -2 . EW obs is in Å, where -99 entries mark sources without secure continuum measurement, and have EW > obs 2500 {}_{\rm obs}>2500 Å. Entries with 99 are undetected in I 𝐼 I band. Flag_Class: 0: Unclassed, 1: LAE at z = 2.4 𝑧 2.4 z=2.4 . A full version of this table is available online.

ID	R.A.	Dec.	Line-flux	EWobs	$I$	Flag_Class
B-HiZELS_stV_1	218.239	33.305	64.3	97	17.2	0
B-HiZELS_stV_2	218.021	33.314	13.9	1131	22.6	0
B-HiZELS_stV_3	217.909	33.351	13.5	590	22.5	0
B-HiZELS_stV_4	218.053	33.354	6.4	207	22.3	0
B-HiZELS_stV_5	218.086	33.353	25.7	1079	21.7	1

Table 9. Table 9: First five entries in the catalog of NB501 line-emitters. Coordinates are in J2000. Line-flux is in 10 − 16 superscript 10 16 10^{-16} erg s -1 cm -2 . EW obs is in Å, where -99 entries mark sources without secure continuum measurement, and have EW > obs 550 {}_{\rm obs}>550 Å. Entries with 99 are undetected in I 𝐼 I band. Flag_Class: 0: Unclassed, 1: LAE at z = 3.1 𝑧 3.1 z=3.1 . A full version of this table is available online.

ID	R.A.	Dec.	Line-flux	EWobs	$I$	Flag_Class
B-HiZELS_NB501_1	217.635	33.366	14.9	198	19.9	0
B-HiZELS_NB501_2	217.923	33.820	13.7	64	20.4	0
B-HiZELS_NB501_3	217.619	33.558	7.8	391	23.3	0
B-HiZELS_NB501_4	218.127	33.666	7.2	-99	22.9	1
B-HiZELS_NB501_5	217.668	34.056	5.4	138	22.6	0

Table 10. Table 10: First five entries in the catalog of NB921 line-emitters. Coordinates are in J2000. Line-flux is in 10 − 16 superscript 10 16 10^{-16} erg s -1 cm -2 . EW obs is in Å, where -99 entries mark sources without secure continuum measurement, and have EW > obs 1100 {}_{\rm obs}>1100 Å. Entries with 99 are undetected in I 𝐼 I band. Flag_Class: 0: Unclassed, 1: H α 𝛼 \alpha at z = 0.4 𝑧 0.4 z=0.4 , 2: H β 𝛽 \beta /[O iii ] at z = 0.8 𝑧 0.8 z=0.8 , 3: [O ii ] at z = 1.47 𝑧 1.47 z=1.47 . A full version of this table is available online.

ID	R.A.	Dec.	Line-flux	EWobs	$I$	Flag_Class
B-HiZELS_NB921_1	217.417	33.559	3.9	40	20.5	0
B-HiZELS_NB921_2	217.990	33.277	0.3	142	24.6	0
B-HiZELS_NB921_3	217.831	33.437	0.3	59	23.7	3
B-HiZELS_NB921_4	218.220	33.662	1.3	101	22.8	3
B-HiZELS_NB921_5	217.799	33.691	0.3	70	23.9	3

Table 11. Table 11: First five entries in the catalog of NB H line-emitters. Coordinates are in J2000. Line-flux is in 10 − 16 superscript 10 16 10^{-16} erg s -1 cm -2 . EW obs is in Å, where -99 entries mark sources without secure continuum measurement, and have EW > obs 1200 {}_{\rm obs}>1200 Å. Entries with 99 are undetected in I 𝐼 I band. Flag_Class: 0: Unclassed, 1: H α 𝛼 \alpha at z = 1.47 𝑧 1.47 z=1.47 , 2: H β 𝛽 \beta /[O iii ] at z = 2.2 𝑧 2.2 z=2.2 , 3: [O ii ] at z = 3.3 𝑧 3.3 z=3.3 . A full version of this table is available online.

ID	R.A.	Dec.	Line-flux	EWobs	$I$	Flag_Class
B-HiZELS_NBH_1	218.345	33.265	1.7	188	99	0
B-HiZELS_NBH_2	218.329	33.299	2.1	212	99	0
B-HiZELS_NBH_3	217.990	33.312	1.8	474	25.5	0
B-HiZELS_NBH_4	218.165	33.317	1.8	139	22.1	0
B-HiZELS_NBH_5	217.670	33.344	1.6	180	22.8	0

Table 12. Table 12: First five entries in the catalog of NB K line-emitters. Coordinates are in J2000. Line-flux is in 10 − 16 superscript 10 16 10^{-16} erg s -1 cm -2 . EW obs is in Å, where -99 entries mark sources without secure continuum measurement, and have EW > obs 1250 {}_{\rm obs}>1250 Å. Entries with 99 are undetected in I 𝐼 I band. Flag_Class: 0: Unclassed, 1: H α 𝛼 \alpha at z = 2.23 𝑧 2.23 z=2.23 , 2: H β 𝛽 \beta /[O iii ] at z = 3.2 𝑧 3.2 z=3.2 , 3: [O ii ] at z = 4.7 𝑧 4.7 z=4.7 , 4: z < 1 𝑧 1 z<1 . A full version of this table is available online.

ID	R.A.	Dec.	Line-flux	EWobs	$I$	Flag_Class
B-HiZELS_NBK_1	218.080	33.261	0.8	121	22.3	4
B-HiZELS_NBK_2	218.345	33.265	1.1	-99	99	0
B-HiZELS_NBK_3	217.878	33.276	0.7	410	25.4	0
B-HiZELS_NBK_4	217.991	33.277	1.0	494	25.1	1
B-HiZELS_NBK_5	217.996	33.280	1.0	-99	24.8	1

Equations

$$f_i=\frac{c}{{\lambda }_{i,\mathrm{center}}^2}10^{-0.4(m_i+48.6)},$$

$$E{W}_{\mathrm{obs}}=\Delta {\lambda }_{\mathrm{NB}}\frac{f_{\mathrm{NB}}-f_{\mathrm{BB}}}{f_{\mathrm{BB}}-f_{\mathrm{NB}}\frac{\Delta {\lambda }_{\mathrm{NB}}}{\Delta {\lambda }_{\mathrm{BB}}}}.$$

$$\Sigma =\frac{1-10^{-0.4(BB-NB)}}{10^{-0.4(ZP-NB)}\sqrt{({\sigma }_{\mathrm{box},\mathrm{BB}}^2+{\sigma }_{\mathrm{box},\mathrm{NB}}^2)}},$$

$$f_{\mathrm{line}}=\Delta {\lambda }_{\mathrm{NB}}\frac{f_{\mathrm{NB}}-f_{\mathrm{BB}}}{1-\frac{\Delta {\lambda }_{\mathrm{NB}}}{\Delta {\lambda }_{\mathrm{BB}}}}.$$

$$B{B}_{max}=NB-2.5{\log }_{10}(\frac{{\lambda }_{c,BB}^2\Delta {\lambda }_{NB}}{{\lambda }_{c,NB}^2\Delta {\lambda }_{BB}})+0.5$$

Full text

††pubyear: 2017††pagerange: Boötes-HiZELS: An optical to near-infrared survey of emission-line galaxies at $\bf z=0.4-4.7$–D

1 Introduction

Understanding how and when galaxies grow their stellar mass and in some cases eventually stop forming stars are key goals of galaxy formation theory. However, since it is only possible to observe an individual galaxy at a single epoch, to assess their evolution it is crucial to homogeneously select equivalent samples of galaxies over a wide redshift range. Currently, different epochs in cosmic time are probed by different selections of galaxies. Moreover, the galaxy properties (such as star formation rates and estimates of dust attenuation) are measured with different tracers (e.g. Speagle et al., 2014). Therefore, it is important to understand whether local calibrations can be extrapolated to high redshift. This requires large samples of galaxies with a well understood selection function and a large dynamic range in galaxy properties.

Homogeneously selected samples of star-forming galaxies can be obtained with narrow-band (NB) surveys, that are very efficient in selecting emission-line galaxies. Using different NBs, galaxies can be selected on a particular emission-line across a range of redshifts with well defined line-luminosity and equivalent width limits. For example, in specific windows from the optical to the near-infrared, ground-based NB surveys can select H$\alpha_{\lambda 6563}$ emission-line galaxies up to $z\sim 2.6$ (e.g. Bunker et al., 1995; Malkan et al., 1996; van der Werf et al., 2000; Ly et al., 2007; Geach et al., 2008; Tadaki et al., 2011; Lee et al., 2012; Drake et al., 2013; Sobral et al., 2013; Sobral et al., 2015; Stroe & Sobral, 2015). The H$\alpha$ recombination-line is a reliable tracer of star-formation rate on $>10$ Myr time-scales (Kennicutt, 1998), and is less sensitive to attenuation due to dust than other shorter wavelength tracers (e.g. Garn et al., 2010; Ibar et al., 2013; Stott et al., 2013). At redshifts $z>2.5$, the most commonly used rest-optical emission-lines are challenging to observe (but see e.g. Khostovan et al. 2015), while the rest-frame UV Lyman-$\alpha_{\lambda 1216}$ (Ly$\alpha$) line, intrinsically the strongest emission-line emitted in Hii regions, is efficiently observed up to $z\sim 7$ (e.g. Rhoads et al., 2000; Dawson et al., 2007; Ouchi et al., 2008; Lee et al., 2014; Matthee et al., 2015; Santos et al., 2016), but is extremely sensitive to resonant scattering and dust attenuation (e.g. Hayes, 2015).

Our High-$z$ Emission Line Survey (HiZELS, Geach et al. 2008; Best et al. 2013; Sobral et al. 2013) has been designed to observe multiple emission-lines in different NBs simultaneously. Hence, Sobral et al. (2012) used observations with NB921 at $\approx 920$ nm and NBH at $\approx 1620$ nm to jointly observe [Oii]λλ3726,3729 and H$\alpha$ at $z=1.47$. At $z=2.2$, matched NB surveys have observed (combinations of) Ly$\alpha$+[Oii]+[Oiii]λλ4959,5007+H$\alpha$ at $z=2.2$ (Lee et al., 2012; Nakajima et al., 2012; Oteo et al., 2015; Matthee et al., 2016; Sobral et al., 2017). One of the advantages of this dual (or multiple) NB technique is that identification of the specific emission-line is secure, such that dual-emitters may be used to fine-tune colour selection criteria (in particular in fields where only limited multi-wavelength data are available). Another advantage is that various SFR estimators, for example [Oii] or (to lesser extent) Ly$\alpha$, can be calibrated with joint H$\alpha$ observations.

Here we present the first results from Boötes-HiZELS, which is a survey of a central 0.7 deg2 region in the Boötes field with a suite of six narrow-band filters, split into two sets: three red filters at $\approx$921, 1620 and 2120 nm from HiZELS that select rest-optical lines such as H$\alpha$111We note that narrow-band H$\alpha$ measurements measure the line-flux and EW of the combined H$\alpha$ and [Nii] doublet depending on the precise redshift. Therefore, a correction needs to be applied to measured H$\alpha$ EWs and line-fluxes. For simplicity, we refer to H$\alpha$+[Nii] emitters as H$\alpha$ emitters from now on., H$\beta$/[Oiii]222Typical photometric redshifts are not accurate enough to distinguish between H$\beta$ line-emitters and a line-emitter with one of the [Oiii] lines. Moreover, depending on the specific redshift, we either detect H$\beta$, or one or two of the [Oiii] lines in the narrow-band filter. The majority of H$\beta$/[Oiii] emitters are [Oiii]λ5007 emitters, as this line is typically the stronger line; see Sobral et al. (2015) and Khostovan et al. (2015) for details. Yet, to remind the reader of these caveats, we call these emitters H$\beta$/[Oiii] emitters throughout the paper. and [Oii], complemented by three blue filters at $\approx$ 392, 411 and 501 nm that select Ly$\alpha$ emitters. Using these narrow-band filters, we select samples of emission-line galaxies using their H$\alpha$ line at $z=0.4-2.2$, and from $z=0.8-4.7$ with H$\beta$/[Oiii], [Oii] and Ly$\alpha$. These samples are used as targets for ongoing detailed spectroscopic follow-up studies.

This paper presents the selection and classification of line-emitters, their global properties such as number densities, the number of dual-NB emitters and X-ray detections. We compare our number densities to published luminosity functions for samples in the range $z\approx 0.4-4.7$ and we study [Oii]-H$\alpha$ emitters at $z=1.47$. These emitters can be used to measure whether the observed [Oii]/H$\alpha$ ratio changes with redshift (e.g. Hayashi et al., 2013), which is essential for studies employing [Oii] as a SFR indicator at $z>1$ (e.g. Ly et al., 2012). We also use the available deep X-ray coverage to study the X-ray fractions of line-emitters.

We present the observations and archival data used in this survey in §2. The data reduction, characteristics and catalog production and selection of emitters are presented in §3. §4 presents our procedure for classifying emission-line galaxies. We present the number densities of classed line-emitters and compare these to published luminosity functions in §5. In §6 we investigate the properties of dual-NB line-emitters, the observed [Oii]/H$\alpha$ ratio at $z=1.47$ and the X-ray fractions of HAEs and LAEs. Finally, §7 presents our conclusions.

We adopt a $\Lambda$CDM cosmology with $H_{0}$ = 70 km s*-1Mpc-1*, $\Omega_{\rm M}=0.3$ and $\Omega_{\Lambda}=0.7$. Magnitudes are in the AB system measured in 3*′′* apertures, unless noted otherwise.

2 Observations & data

We observed a 0.7 deg2 region in the Boötes field with six narrow-band filters (NB392, stV, NB501, NB921, NBH and NBK). The Boötes field was chosen for the availability of deep multi-wavelength data (see e.g. Lee et al. 2011; Bian et al. 2012, 2013; Beare et al. 2015) over a relatively large area, avoiding the galactic plane and its observability from La Palma and Hawaii. In addition to publicly available broad-band (BB) imaging in the $U$, $B$, $R$, $I$, $J$ and $K$ bands (described in §2.1), we also obtain the missing wavelength coverage with three broad-band filters $g$, $z$ and $H$ (see Table 1 for an overview).

2.1 Public/archival multi-wavelength data

The Boötes field has been imaged in the optical by the NOAO Deep Wide Field survey333http://noao.edu/noao/noaodeep/ in $Bw$, $R$ and $I$ (Jannuzi & Dey, 1999) and by the LBT Bootes Field Survey in the $U$ and $Y$ bands (Bian et al., 2013). Near-infrared data in the $J$, $H$ and $Ks$ band are available from the Infrared Bootes Imaging Survey (Gonzalez et al., 2010), although we do not use the $H$ band data as our data are deeper. The general characteristics of the archival data used in this paper are listed in Table 1. In addition, the field has been imaged in the X-ray by Chandra (Murray et al., 2005), in the UV by the GALEX Deep Imaging Survey (Martin et al., 2005), in the mid-infrared by Spitzer/IRAC (Ashby et al., 2009) and in the far-infrared by Herschel as part of the Herschel Multi-tiered Extragalactic Survey (Oliver et al., 2012). However, the X-Ray, UV and mid- and far-infrared data are not explicitly used in the selection of line-emitters. In addition, Williams et al. (2016) presented deep low-frequency (150 MHz) radio observations in this field. Finally, spectroscopic follow-up of mostly X-ray selected sources (and hence AGN) has been performed by Kochanek et al. (2012).

2.2 Optical observations

Optical observations in two narrow-band filters (NB392, NB501), a medium band filter (stV) and the $g$ band were performed with the Wide Field Camera (WFC) on the 2.5m Isaac Newton Telescope, part of the Roque de los Muchachos Observatory on the island of La Palma, Spain. WFC has a mosaic of 4 CCDs with a combined field of view of 0.3 deg2 and a 0.33*′′* pixel scale, see Table 1. The survey was designed with four pointings, each with a C-NE-NW-SE-SW dither pattern (with 30*′′* offsets). Individual exposure times for narrow and medium-bands were either 0.2 or 1.0ks, depending on whether the telescope could successfully guide on a star (since the auto-guider CCD is behind the filter, this is challenging for narrow-band filters in extra-galactic fields) and the stability of the weather. The individual exposure times for the $g$ band were 0.6ks.

Observations in the NB921 narrow-band filter and the $z$ filter were performed with Suprime-Cam (S-cam) on the 8.0m Subaru telescope of the National Astronomical Observatory of Japan. S-cam consists of a mosaic of 10 CCDs with a combined field of view of 0.255 deg2 with a 0.2*′′* pixel scale. We imaged the field with the $z$ (NB921) filter with five (three) pointings. For NB921, we used individual 360s exposures dithered either 7 (2 pointings) or 6 (1 pointing) times. For $z$, we used individual 150s exposures of the same pointings as NB921 dithered 14 times, and 3 times 100s in the other two pointings. Observations were done sequentially to avoid contamination of the emission-line sample by variable sources and/or supernova (i.e. Matthee et al., 2014).

2.3 Near-infrared observations

Near-infrared observations in $H$, NBH and NBK were performed with WFCAM on the UK Infrared Telescope (UKIRT) on Mauna Kea as part of the High-$z$ Emission Line Survey (HiZELS, e.g. Sobral et al. 2013). WFCAM has a “paw-print” configuration of four CCDs, with a total field of view of 0.21 deg2 and a 0.4*′′* pixel scale. The field was imaged with 4 pointings in a dither sequence of 14 exposures with small offsets. Due to the high sky background in the near-infrared, the individual exposure times were 10s, 100s, and 60s for $H$, NBH and NBK, respectively, to avoid saturation. In order to obtain the final depth, this dither sequence was typically repeated 9, 11 and 24 times for the respective filters.

3 Data reduction & Catalogue production

3.1 Data reduction

3.1.1 Optical

We reduce data from the INT/WFC with a custom-made pipeline based on python described in detail in Stroe et al. (2014) and Sobral et al. (2017) and we reduce Subaru/S-Cam data similarly with SDFred2 (Ouchi et al., 2004). In summary, we first bias subtract individual frames using a master bias from the median stack of bias frames for the corresponding night. We then create a master flat by median combining twilight flats and use it to flat-field the individual frames. Subsequently, we measure the FWHM of the PSF using unsaturated stars in individually reduced frames with SExtractor (Bertin & Arnouts, 1996), and reject frames with PSF FWHM above the chosen target FWHM listed in Table 1. This is particularly important for some exposures with the INT/WFC that have been observed in poor conditions. We then match the PSF of remaining frames before combining frames to the common mosaic by smoothing the images with a gaussian kernel.

3.1.2 Near-infrared

Near-infrared data from UKIRT/WFCAM have been reduced using PfHiZELS; see Sobral et al. (2009) and Sobral et al. (2013) for full details. The steps are similar to the steps in the optical data reduction, except for dark subtraction instead of bias subtraction and the master flat that is based on an iterative self-flat method using the science frames themselves, instead of relying on twilight flats; see Sobral et al. (2013).

3.1.3 Astrometric alignment

The reduced frames are astrometrically registered to the 2MASS point source catalog (Skrutskie et al., 2006) with Scamp (Bertin, 2006). Frames are then co-added, resampled to a pixel-scale of 0.33*′′* and mapped to the mosaic pointing with Swarp (Bertin, 2010). We apply the same method for public data described in §2.1.

While extracting initial catalogues, we encountered significant astrometric distortions of up to 1.2*′′* in the edges of the cameras of the public $B_{w}$, $R$ and $I$ data. These distortions significantly affect dual-mode photometry (described below in §3.2.1). In order to obtain a more accurate astrometric solution for these data, we used the Scamp software to remap the images to the SDSS DR7 astrometry (Abazajian et al., 2009). The astrometric differences between 2MASS and SDSS are minimal and no significant distortions affecting our photometry have been noticed after this correction.

3.1.4 Photometric calibration

We set the photometric zero-point ($ZP$) of the images to an arbitrary common $ZP=30$ by matching the Mag-auto photometry in the combined images to the following available data: $g$ and $z$ are calibrated to SDSS and $H$ and $K$ to 2MASS. We then use the broad-bands available to calibrate the narrow-bands in the two following steps: we first calibrated NB921 to $z$, NBH and NBK are calibrated to $H$ and $K$, NB501 to $g$ from SDSS, and NB392 and stV to $Bw$. After this first step, we have to correct for the fact that most narrow-band central wavelengths are not in the center wavelength of the broad-band filters, which leads to a bias in line-flux measurements due to gradients in the continuum. This can be resolved by using colour information in adjacent broad-bands. For NB921, NBH and NBK we follow the corrections described in Sobral et al. (2013) and for NB392 we use the corrections from Matthee et al. (2016). We derive the following correction for stV: stV${}_{\rm cor}=$ stV$-0.23(U-B_{w})+0.24$. For sources undetected in $U$ and $B_{w}$ we apply the median correction of $+0.04$. We do not apply a correction for NB501 as it is close to the center of the $g$ band.

3.2 Catalogue production

3.2.1 Photometry

Photometry of the optical-NIR filters listed in Table 1 is performed with SExtractor in dual-image mode. We create six catalogues, each with one of the six narrow-bands as detection image. Photometry is measured within circular apertures with a diameter of 3*′′. For each narrow-band we measure the narrow-band and the corresponding broadband magnitudes from images with their PSF matched to the narrow-band imaging. We also measure the magnitudes in all broad-bands with their PSF matched to the $g$ band PSF (1.6′′* FWHM). The measurements with the PSF from the NB are used to select line-emitters and compute emission-line properties such as line-flux and equivalent width. The other measurements are used for colour-colour selections.

We have produced a mask for each narrow-band individually, where we mask regions around bright, saturated stars, CCD bleeding, cross-talk in near-infrared detectors and regions with low S/N or incomplete coverage (e.g. Sobral et al., 2009; Santos et al., 2016).

3.2.2 Depths

We estimate the depth of images by measuring the standard deviation of the total counts in 100,000 apertures with a diameter of 3*′′* placed at random (but avoiding sources) locations in our images. 3$\sigma$ depths range from $\sim 25$ AB magnitude in blue broad-band filters to $\sim 22$ AB magnitude in the near-infrared filters, see Table 1.

3.3 Selecting Line-emitters

Line-emitters are selected based on two criteria: the narrow-band excess (the equivalent width, EW) must be high enough and the excess must be significant. For the narrow-band filters NB392, stV, NB501, NB921, NBH and NBK we use the corresponding broad-band filters $U$, $B_{w}$, $g$, $z$, $H$ and $K$, see Fig. 2. In order to convert the photometric narrow-band excess to observed EW, we convert magnitudes ($m_{i}$) to flux densities in each filter ($f_{i}$) with the standard AB magnitude convention:

[TABLE]

where $c$ is the speed of light and $\lambda_{i,\rm center}$ is the central wavelength in each filter. Next, we use the following equations to convert the narrow-band and their corresponding broad-bands to EW:

[TABLE]

Here, $f_{\rm NB}$ and $f_{\rm BB}$ are the flux-densities in the narrow-band and broad-band and $\Delta\lambda_{\rm NB}$ and $\Delta\lambda_{\rm BB}$ the filter-widths. In Eq. 2, the numerator is the difference in narrow-band and broad-band flux and the denominator is the continuum level, corrected for the contribution from the flux in the narrow-band. For sources without broad-band detection we set the EW to a lower limit. This lower limit ranges from 550 Å for NB392 and NB501 to 2500 Å for stV depending on the width of the NB filter and the depth of the BB data used to measure the continuum. The lower limit is around $\approx 1200$ Å for near-infrared NBs.

The excess significance ($\Sigma$) quantifies whether a certain narrow-band excess is due to errors in the narrow-band and broad-band photometry or not. Hence, we follow the methodology presented in Bunker et al. (1995) and the equation from Sobral et al. (2013) to compute $\Sigma$:

[TABLE]

where $BB$ is the broadband magnitude used for the continuum estimate, $NB$ is the narrow-band magnitude and $ZP$ is the zero-point of the images. $\sigma_{\rm box}$ is the root mean squared (rms) of background aperture values in the data of the respective filters (see §3.2.2).

The line-flux is computed using:

[TABLE]

We select line-emitters among narrow-band selected sources in all six narrow-band filters with the criterion that $\Sigma>3$. However, because each narrow-band has different filter characteristics, we do not apply a homogeneous excess (EW) selection threshold. For each filter, we apply the criterion that the observed EW is three times the standard scatter in observed EWs for sources detected at $>15\sigma$. This means that we apply EW$>30,130,50,30,85,80$ Å for NB392, stV, NB501, NB921, NBH and NBK respectively.

Before obtaining our final list of line-emitters, in each filter, we visually inspect all the sources in the narrow-band images for remaining spurious sources such as artefacts from bright stars, cosmic rays or mis-identifications by SExtractor. This can happen when the noise properties vary strongly locally, which is the case in small regions of the coverage by the NB392, stV and NB921 filters.

4 Classifying line-emitters

Fig. 1 shows the redshift ranges where our narrow-band filters sample the brightest emission-lines seen in normal star-forming galaxies and AGN. By a combination of design and coincidence, the HiZELS narrow-band filters coincide with several different emission-lines at specific redshifts. At $z=1.47$, the NB921/NBH combination is sensitive to the [Oii] and H$\alpha$ lines (Sobral et al., 2012). At $z=2.23,$ the Ly$\alpha$, [Oiii] and H$\alpha$ lines fall in the NB392/NBH/NBK combination444In this case, it is certain that the emission-line in NBH is [Oiii] and not H$\beta$.. At $z=2.23$, the NB501 filter is also sensitive to Civ emission. The redshifts of line-emitters detected in several narrow-bands (dual-emitters) can be estimated accurately and we refer to them as $z_{\rm dual-NB}$ in the remainder of this paper. As can be seen in Table 6, dual-emitters are found as faint as $I\approx 25$, three magnitudes fainter than typical available spectroscopic redshifts, which were mostly for X-ray selected sources.

Line-emitters that have no existing spectrosopic redshift or are not detected as dual-emitters are classified using colour-colour selections tuned to identify Lyman- and Balmer-breaks at various redshift intervals. For the blue narrow-bands, we use colour selections to identify Ly$\alpha$-emitters. For the red filters we use colour selections to identify H$\alpha$ emitters, H$\beta$/[Oiii] emitters and [Oii] emitters (see also similar selections in Sobral et al. 2013; Khostovan et al. 2015).

The main strategy to devise colour-criteria has been as follows: after removing stars (due to atmospheric features in the blue or the near-infrared, stars may be picked up with a narrow-band excess) using their $uJK$ colours (e.g. Muzzin et al., 2013), we start with colour selections from the literature, which we slightly modify using the spectroscopically confirmed line-emitters and the dual-emitters. Colour-criteria are stated explicitly below and listed in Table 3. We summarise the number of emitters and classified line-emitters in Table 4.

4.1 Line-emitters in NB392

The narrow-band NB392 has specifically been designed to conduct a Ly$\alpha$ survey with a matched volume coverage to H$\alpha$ emitters identified with the HiZELS NBK filter (H2S1) at $z=2.23$ in order to study the Ly$\alpha$ escape fraction and its dependencies on galaxy properties, as described in detail in Matthee et al. (2016) and Sobral et al. (2017).

We select 57 line-emitters with an excess criterion of EW${}_{\rm obs}>30$ Å ($U$-NB392 $>0.45$). For LAEs at $z=2.23$ this corresponds to EW${}_{0}>9$ Å (it is possible to go to such low EWs because the width of NB392 is very narrow). Although Ly$\alpha$ surveys at $z\approx 2-3$ typically invoke a higher EW criterion of $\sim 25-30$ Å (e.g. Ouchi et al., 2008; Nakajima et al., 2012), we found that such a selection results in missing the most luminous LAEs at $z=2.2$ in the COSMOS and UDS fields (Sobral et al., 2017). This is because these sources are typically AGN, which have bright Ly$\alpha$ emission on top of a bright UV continuum.

Using the spectroscopy available from AGES (Kochanek et al., 2012), we find 8 Mgii emitters at $z=0.4$ and 5 line-emitters at $z>1$ (including two LAEs at $z=2.2$), see Table 2. By matching the sample of line-emitters with the samples of line-emitters in NBH and NBK (see below), we add four other robust LAEs at $z=2.2$. Other line-emitters are classed using the criteria described in Table 3. We spectroscopically identify four interlopers ($\approx 15\pm 7$ % contamination, similar to the $10\pm 4$ % from Sobral et al. 2017). These comprise two Civ emitters at $z=1.53$ and two AGN for which we measure Lyman-Werner and Lyman-Continuum radiation in the NB392 filter at $z=3.16$ and $z=3.57$. We also identify two dual-emitters that are missed by the colour-colour selection (see Fig. 9). This results in a final sample of 25 LAEs at $z=2.2$.

4.2 Line-emitters in stV

The stV medium-band filter is used to identify LAEs at $z\approx 2.4$. Because the width of the filter is relatively broad, it is sensitive to line-emitters over a larger redshift space (and thus covers a larger volume), at the cost of being only sensitive to lines with high EW. We apply a selection criterion of EW${}_{\rm obs}>130$ Å (which corresponds to $B_{w}-$stV$>0.54$). We note that due to the width of the filter, it is possible that multiple lines contribute to the observed EW and line-flux, such as the combination of Ly$\alpha$+Nv.555For example, for Type I AGN, Nv/Ly$\alpha$ is typically $\approx 3$ % (e.g. Vanden Berk et al., 2001), while for type II AGN (such as narrow-line Seyferts) Nv/Ly$\alpha$ can be as high as 50 % (typically $\approx 20$ %, e.g. Alexandroff et al. 2013). It is therefore not straightforward to interpret measured EWs and line-fluxes and caution must be taken.

We find a total of 39 line-emitters, of which 15 have spectroscopic redshifts, see e.g. Table 2. As expected, these are dominated by LAEs at $z=2.3-2.45$, but also contains high-ionization lines as Ciii] and Civ at $z\approx 1.1-1.7$. After removal of one spectroscopic contaminant (a Civ emitter at $z=1.613$) selected with the colour-criteria described in Table 3 and illustrated in Fig. 9), we obtain a sample of 16 LAEs.

4.3 Line-emitters in NB501

The NB501 filter is used to select Ly$\alpha$ emitters at $z=3.1$. We apply EW${}_{\rm obs}>50$ Å ($g-$NB501$>0.45$), corresponding to a Ly$\alpha$ rest-frame EW of $>12$ Å.

We find a total of 65 line-emitters, of which only four have an archival spectroscopic redshift. This is because the majority of these line-emitters are faint with line-fluxes below $2\times 10^{-16}$ erg s*-1* cm*-2*. One spectroscopic confirmed line-emitter is a LAE, two are Civ emitters at $z=2.24$ and $z=2.26$ (these are the dual-emitters B-HiZELS_3 and B-HiZELS_15, Table 6, also detected as line-emitters in several other bands) and one is possibly [Nev] at $z=0.426$.

LAEs are selected as relatively blue $U$ drop-outs (e.g. Hildebrandt et al., 2009) as described in Table 3 and illustrated in Fig. 10). One Civ emitter at $z=2.24$ is mis-classed as a LAE and is removed from the sample. This leads to a sample of 31 LAEs. We note that our EW criterion of $EW_{0}>12$ Å is somewhat lower than the typical criterion used for selections of LAEs (EW${}_{0}>25$Å, e.g. Ouchi et al. 2008; Yamada et al. 2012). However, more than 90 % of the identified LAEs have EW${}_{0}>25$Å. Contrary to the properties of LAEs at $z\sim 2$, the additional LAEs with low EW are all faint in their UV continuum. This indicates an evolution in the properties of luminous LAEs from $z=2-3$, with an increasing Ly$\alpha$ EW0 with redshift at fixed Ly$\alpha$ luminosity. Very recently, four additional LAEs from this sample have been confirmed at $z=3.1$ from our spectroscopic follow-up campaign (to be presented in Sobral et al. in prep), including the brightest LAE in our sample with a Ly$\alpha$ luminosity of $\approx 10^{43.8}$ erg s*-1* ($\sim 10\times L^{\star}$ at $z=3.1$, Ouchi et al. 2008) in a 3*′′* aperture and an EW0 of $\sim 150$ Å. To the current surface brightness limit, it is extended over $\sim 5^{\prime\prime}$ (40 kpc), and it may thus be classed a Ly$\alpha$ blob (e.g. Matsuda et al., 2004; Prescott et al., 2008; Dey et al., 2016). This follow-up spectroscopy also identifies two interlopers: a red [Oii] emitter at $z=0.35$ and a Mgii emitter at $z=0.81$, that we have removed from the sample.

4.4 Line-emitters in NB921

While the NB921 filter has been used to select LAEs at $z=6.6$ (e.g. Matthee et al., 2015), it is also used to select H$\alpha$, H$\beta$/[Oiii] and [Oii] emitters at lower redshift (e.g. Ly et al., 2007; Drake et al., 2013; Sobral et al., 2013; Khostovan et al., 2015). We select 1161 line-emitters with the excess criterion of EW$>30$ Å (corresponding to $z-$NB921$>0.3$).

Since our sample of line-emitters is selected from relatively deep narrow-band imaging (compared to the other narrow-bands in this survey), it is dominated by sources with fluxes fainter than $2\times 10^{-16}$ erg s*-1* cm*-2* ($>94$ % of line-emitters), down to fluxes of $2\times 10^{-17}$ erg s*-1* cm*-2*. Because of this, the number of spectroscopic redshifts is limited to only seven, of which five are H$\alpha$ emitters at $z=0.4$. However, the number of galaxies with a robust redshift due to emission-lines in multiple narrow-bands is significantly higher (23), see Table 2.

Among our sample of line-emitters, we use colour-criteria (Fig. 11; based on Sobral et al. 2013) to identify H$\alpha$ emitters at $z=0.40$, H$\beta$/[Oiii] emitters at $z=0.83$ and [Oii] emitters at $z=1.47$, see Table 3. We select 198 H$\alpha$ emitters, 304 H$\beta$/[Oiii] emitters and 277 [Oii] emitters (see Table 4). Two Mgii emitters at $z=2.26$ are mis-identified as [Oii] emitter, while one dual-emitter at $z=1.47$ is mis-identified as H$\beta$/[Oiii] emitters.

Due to their faintness, 359 out of the 1161 line-emitters are not detected in a sufficient number of broadbands required for classification and can thus not be classed. We expect that most of these sources are faint H$\alpha$, H$\beta$/[Oiii] or [Oii] emitters. Based on the fraction of emitters in different classifications as a function of line-flux, we expect an increasing fraction of [Oii] emitters at low line-fluxes (see also Sobral et al. (2012)). As illustrated in the left panel of Fig. 3, the majority of sources indeed has $B_{w}-R$ colours similar to [Oii] emitters (but could not be classed due to their faintness in other broadband filters). We discuss this ‘identification-incompleteness’ further in §5.1.3.

4.5 Line-emitters in NBH

We select line-emitters detected in the NBH filter with a narrow-band excess of EW$>85$ Å (corresponding to $H-$NB${}_{H}>0.3$). Since the near-infrared detectors of UKIRT/WFCAM contain significant amounts of crosstalk, we perform careful visual inspections of our sample of line-emitters, resulting in a sample of 301 line-emitters.

While the sample of line-emitters includes nine sources with spectroscopic redshifts (six H$\alpha$ at $z=1.4$ and three H$\beta$/[Oiii] at $z=2.2$), the NBH line-emitters are particularly suitable for identifying line-emitters with the dual-NB technique. These robustly identified line-emitters are used to adapt the colour selection criteria from Sobral et al. (2013) for the data available in this field. Using the criteria listed in Table 3, we select 87 H$\alpha$ emitters at $z=1.47$, 72 H$\beta$/[Oiii] emitters at $z=2.2$ and 6 [Oii] emitters at $z=3.3$, see Table 4. 99 sources are classed as low-redshift interlopers and 39 sources are too faint to be detected in the required broad-bands. Prior to the final identification, four spectroscopically confirmed H$\beta$/[Oiii] emitters were classed as H$\alpha$ emitters, while five H$\alpha$ emitters were classed as H$\beta$/[Oiii] emitter. These mis-identified emitters are typically very luminous and likely AGN, such that their colours are anomalous (see Fig. 12). There are no such identified contaminants among the fainter dual-emitters.

4.6 Line-emitters in NBK

We select line-emitters detected in the NBK filter with EW$>80$ Å (corresponding to $K-$NB${}_{K}>0.23$). In total, after visual inspections, we find 255 line-emitters, of which 18 have a spectroscopic redshift (including eight H$\alpha$ at $z=2.23$) and 20 are dual-emitters (all H$\alpha$ at $z=2.23$). Based on these robust redshifts and the colour selection criteria listed in Table 3, we select 77 H$\alpha$ emitters at $z=2.2$, 11 H$\beta$/[Oiii] emitters at $z=3.2$ and two [Oii] emitters at $z=4.7$ (see Table 4). 110 line-emitters are at $z<1.5$ and 55 line-emitters are too faint to be classified. We have not identified any spectroscopically confirmed contaminants before the final classification. However, the colour-colour criteria missed two X-ray detected dual-emitters at $z=2.23$ as illustrated in Fig. 13.

5 Number densities

5.1 Method

We measure the number densities of LAEs at $z=2.2,2.4$ and $z=3.1$, H$\alpha$ emitters at $z=0.4,1.47$ and $z=2.23$, H$\beta$/[Oiii] emitters at $z=0.8,2.2,3.2$ and [Oii] emitters at $z=1.47,3.3,4.7$ as a function of their line-luminosity, in narrow luminosity bins (0.2-0.3 dex in this analysis). The luminosity is calculated using the line-flux (§3.3) and assuming the luminosity distance corresponding to the redshift of peak filter transmission for the relevant emission-line. We calculate the comoving volume for each line/filter combination using the redshifts of half peak transmission, see Table 5. For H$\beta$/[Oiii], we compute the volume following Khostovan et al. (2015), who uses only the volume probed by the [Oiii]λ5007 line. We refer to this work and Sobral et al. (2015) for a detailed discussion on the contribution of H$\beta$ and [Oiii]λ4959. Luminosity-binned number densities were calculated by dividing the number of sources in each bin by the comoving volume, and then correcting these (as described in the following subsections) for the effects of the filter profile, flux incompleteness and identification incompleteness. Uncertainties on these number densities were estimated using Poissonian errors. To be conservative, we add in quadrature 20 % of the flux-completeness correction and 20 % of the identification-incompleteness correction (in the case of the red narrow-bands) to the error of each bin. We only show bins with $>40$ % flux-completeness.

5.1.1 Filter profile correction

As described in Khostovan et al. (2015) and Sobral et al. (2013); Sobral et al. (2015), observed number densities have to be corrected for the fact that the filter transmission curves are not a perfect top-hat. Because of this, luminous sources may be observed as faint sources if they lie at a redshift corresponding to the wings of the filter. Furthermore, at fixed flux-limit, fainter sources can only be observed over a smaller volume than more luminous sources. Following the method described in these papers, we compute the number density corrections using a simulation. This simulation assumes that sources are distributed randomly in redshift space and computes their observed luminosities based on the relevant filter transmission. We then obtain a volume correction for each luminosity bin. These corrections typically increase the number densities of the most luminous bins by at most 0.3 dex, while the number densities of fainter bins stay constant, or are decreased by at most 0.05 dex.

5.1.2 Detection flux-completeness

The flux-completeness of our selection is measured as a function of line-flux as follows: for the relevant line, we select galaxies that are not selected as a line-emitter, but do fulfil the colour criteria from §4. We then artificially add line-flux (starting from $10^{-18}$ erg s*-1* cm*-2* in steps of 0.05 dex) and re-compute the line excess and excess significance for each step. After each step, we measure the fraction of sources that would be selected as line-emitter with the added line-flux. We tabulate the 50 % completeness in Table 5. Most narrow-band selections are 50 % complete at $\sim 1\times 10^{-16}$ erg s*-1* cm*-2*, with the exception of stV, which is only sensitive to brighter emission-lines.

5.1.3 Identification incompleteness

For the red narrow-bands, we also take into account that the broad-band data is not deep enough for a robust classification of all faintest line-emitters, which we call identification-incompleteness. We estimate corrections for this effect as follows: for each narrow-band filter, we measure the fraction of line-emitters that is classed as either H$\alpha$, H$\beta$/[Oiii] or [Oii] emitter or as lower redshift source, as a function of line-flux and assume that this fraction can be extrapolated to the line-fluxes of the sources that are not classable.

We find that for line-emitters in NB921 the fraction of classed H$\alpha$, H$\beta$/[Oiii] and [Oii] emitters is $25\pm 5$, $35\pm 5$ and $40\pm 4$ % respectively at fluxes $<6.3\times 10^{-17}$ erg s*-1* cm*-2* and $42\pm 10$, $38\pm 10$ and $20\pm 8$ % respectively for fluxes between $6.3\times 10^{-17}$ erg s*-1* cm*-2* and $4\times 10^{-16}$ erg s*-1* cm*-2*, see Fig. 3. This is expected, as sources with fainter fluxes are expected to be at higher redshift. For line-emitters in NBH, the corresponding fractions are $25\pm 5$, $25\pm 7$ and $4\pm 2$ % at fluxes below $4\times 10^{-16}$ erg s*-1* cm*-2*. This means that $\sim 40$ % of the unclassed sources is likely at $z<1$. Above this flux, all sources are classed. These fractions are in agreement with the estimate from Sobral et al. (2012). Finally, for line-emitters in NBK, we estimate that below a flux of $1.5\times 10^{-16}$ erg s*-1* cm*-2* a fraction of $40\pm 8$ % of the line-emitters is H$\alpha$, $7\pm 3$ % is H$\beta$/[Oiii] , while $55\pm 6$ % is at low redshift. All sources with a larger line-flux have been classed. We note that maximally 40 % of the sources in a flux bin are unclassed. This maximum occurs in the faintest bin of the NB921 line-emitters (see Fig. 3). The typical fraction of unclassed sources at the discussed flux levels is 20 %. We use the estimates described above to obtain the identification-incompleteness for each luminosity bin for the relevant emission-line.

5.2 Comparison with previous surveys

5.2.1 Lyman-$\alpha$ emitters at $z=2.2-3.1$

We show the measured luminosity function (LF) of LAEs at $z=2.2-2.4$ and $z=3.1$ in Fig. 4. The depth of our data allows us to constrain the LF to $\approx L^{\star}$. We find good agreement with earlier results from Sobral et al. (2017) and those from Konno et al. (2016) in different survey fields and slightly different colour-criteria (see also An et al. 2016). All surveys indicate that the Ly$\alpha$ LF at $z\approx 2.2$ deviates from a Schechter function at bright luminosities.666 The Schechter (1976) shape of the luminosity function is expressed as $\Phi(L)$ d$L$ = $\Phi^{\star}(\frac{L}{L^{\star}})^{\alpha}e^{-\frac{L}{L^{\star}}}d(\frac{L}{L^{\star}})$, where $\Phi^{\star}$ is the characteristic number density, $L^{\star}$ the characteristic luminosity and $\alpha$ the faint-end slope. We note that all LAEs at $z=2.2$ with a luminosity above $10^{44}$ erg s*-1* are either spectroscopically confirmed or have a dual-NB redshift. As discussed in Sobral et al. (2017), the power-law behaviour of the luminosity function is likely due to the contribution of AGN in addition to the normal Schechter function. Indeed, most ($\sim 80\pm 40$ %) luminous LAEs (L${}_{\rm Ly\alpha}>10^{43}$ erg s*-1*) are AGN (either due to X-Ray detection or Civ detection in NB501). We fully explore this in §6.4.

Due to its larger probed cosmic volume, the stV filter is mostly sensitive to very luminous LAEs. Although all LAEs with a luminosity above $10^{44}$ erg s*-1* at $z=2.4$ are spectroscopically confirmed, we expect that the sample with luminosities $10^{43-44}$ erg s*-1* is contaminated, as the spectroscopic follow-up at these fluxes is not complete. Most importantly, we expect contaminants to be emission lines that are associated with AGN activity such as Civ, Ciii] and Heii at $z=1.15-1.65$ (e.g. Stroe et al., 2017), which are challenging to identify with these colour-colour selections. We estimate the contamination at these flux levels by mimicking the selection of this survey in a similar medium-band in the COSMOS field (IA427, Santos et al. in prep). We select LAEs at $z=2.5$ with the same criteria (including broad-band depths) and estimate the number of interlopers using the most recent photometric redshifts (Laigle et al., 2016) and a compilation of spectroscopic redshifts. We find that at luminosities $\sim 10^{43-44}$ erg s*-1* there is a non-negligible contamination due to Civ, Ciii] and Heii emitters of 20$\pm 10$ %. At higher luminosities the contamination decreases to 4$\pm 4$ %. The plotted number densities are corrected for these contamination rates. We combine the $z=2.2-2.4$ data to fit a power-law function to the number density of LAEs at the bright end (L${}_{\rm Ly\alpha}>10^{43}$ erg s*-1*), which results in: log${}_{10}(\Phi)=27.5^{+7.3}_{-7.4}-0.74^{+0.17}_{-0.17}$log10(LLyα), with a reduced $\chi^{2}$ of 1.1. This fit is slightly shallower than the power-law fitted by Sobral et al. (2017) based on a smaller volume, but consistent within 1$\sigma$.

We estimate the contribution of broad-line Type I AGN to the Ly$\alpha$ LF at $z\approx 2.2$ based on the UV LF of Type I AGN at $2.0<z<2.5$ from Bongiorno et al. (2007) and the typical UV slope and Ly$\alpha$ EW of these AGN. Assuming f${}_{\lambda}\propto\lambda^{-1.5}$ and Ly$\alpha$ EW${}_{0}=$ 80 Å (e.g. Vanden Berk et al., 2001; Hunt et al., 2004), we convert the number densities as a function of $M_{1450}$ to number densities as a function of Ly$\alpha$ luminosity. As shown in Fig. 4, it is clear that the number density of LAEs is higher than the number density of Type I AGN at fixed Ly$\alpha$ luminosity. This indicates that only a fraction of the luminous LAEs are likely Type I AGN. This estimate suggests that at Ly$\alpha$ luminosities $10^{43-44}$ erg s*-1* the fraction of Type I AGN is only $\sim 10$ %, while the fraction is $\sim 20-30$ % at higher Ly$\alpha$ luminosities. Because the AGN LF at the faintest UV magnitudes is relatively flat, these fractions do not depend strongly on the assumed values of the Ly$\alpha$ EW or UV slope. The low Type I AGN fraction indicates that the majority of luminous LAEs are narrow-line Type II AGN (see also §6.4), or star-forming galaxies.

At $z=3.1$ we find that the Ly$\alpha$ LF agrees well with that from Ouchi et al. (2008), who performed a deep Ly$\alpha$ survey over a similar area. Compared to similar NB501 data in the COSMOS field (with deeper ancillary data and relatively more spectroscopic follow-up; Matthee et al. in prep), we also find that the fraction of line-emitters (with similar line-flux and EW distributions) that are classed as LAE is similar: $51\pm 9$ % in Boötes against $46\pm 7$ % in COSMOS. This also confirms evolution in L*⋆* between $z=2.2-3.1$ from L${}_{\rm Ly\alpha}\approx 4\times 10^{42}$ erg s*-1* to L${}_{\rm Ly\alpha}\approx 9\times 10^{42}$ erg s*-1*. The number density of LAEs in the brightest bin (three out of the four sources in this bin are spectroscopically followed-up and confirmed) lies above the Schechter fit from Ouchi et al. (2008) (similar to the actual data-points from that survey), indicating the presence of AGN among these luminous sources, similar to $z\sim 2$. Indeed, we find evidence for AGN activity for most LAEs in the most luminous bin, either due to an X-ray detection or due to the detection of high ionisation emission-lines as [Neiv]λ2424 or broad Siiv and Civ absorption features in the spectrum (Sobral et al. in prep).

5.2.2 H$\alpha$ emitters at $z=0.4-2.2$

We show the number densities of H$\alpha$ emitters as a function of their observed H$\alpha$ luminosities. The H$\alpha$ luminosities are corrected for the contribution due to [Nii] following a method based on observed H$\alpha$+[Nii] EW (Sobral et al., 2012). H$\alpha$ luminosities are not corrected for attenuation due to dust and we compare our results with dust-uncorrected values from the literature. In case dust-uncorrected values are not provided, we convert dust-corrected values back using the prescriptions outlined in the relevant papers.

As illustrated in Fig. 5, we find that the H$\alpha$ LF at $z=0.4,1.47$ and $z=2.23$ are generally in good agreement with the number densities in the UDS+COSMOS parts of the HiZELS survey from Sobral et al. (2013). Within the errors, the LF agrees well with the fitted relations from Drake et al. (2013) and Ly et al. (2007). At $z=2.23$, the number densities complement the number densities at fainter luminosities from Lee et al. (2012). This confirms strong evolution in L${}^{\star}_{\rm H\alpha}$ from $z=0.4-2.23$. We note that at $z=0.4$ there could be some contamination at the faintest luminosities due to identification-incompleteness (see §5.1.3). At $z=1.47$ the luminosities seem to be systematically higher by $\approx 0.1-0.15$ dex, increasing slightly with luminosity. This offset can partly be explained by different apertures used in the photometry. While Sobral et al. (2013) uses 2*′′* apertures for all measurements above $z>0.5$, we use 3*′′* measurements. Redoing the measurements with 2*′′* apertures (smaller symbols in Fig. 5) results in good agreement at our faintest luminosities and reasonable agreement (within the error bars) at higher luminosities. At faint luminosities, the luminosity difference between the two apertures is typically 0.14 dex, while it is typically 0.1 dex at high luminosities. The highest luminosity bins at $z=1.47-2.23$ show number densities diverging from a Schechter function, contributing to the difference as well. The sources in these bins are all spectroscopically confirmed or have dual-NB redshifts and most are X-ray detected, as indicated in the corresponding panels. We discuss this in more detail in§6.4.

Compared to the grism results at $0.3<z<0.9$ from Colbert et al. (2013), the number densities at $z=0.4$ are offset (mostly in terms of luminosity). This can simply be explained by the evolution in the typical H$\alpha$ luminosity between $z=0.4$ and the median redshift of the Colbert et al. (2013) sample of $z\approx 0.6$, as log $L^{\star}_{\rm H\alpha}$ increases with $0.45\times z$ over this redshift range (Sobral et al., 2013). The number densities at $z=1.47$ are in good agreement with the grism results at $0.9<z<1.5$, even though the median grism redshift is $z\approx 1.2$. This indicates that there is little evolution in $L^{\star}_{\rm H\alpha}$ between $z=1.47$ and $z\approx 1.2$.

5.2.3 H$\beta$/[Oiii] emitters at $z=0.8-3.2$

We also compare our idenfication of H$\beta$/[Oiii] at $z=0.8,2.2,3.2$ with the analysis from Khostovan et al. (2015) in the COSMOS and UDS fields.

As illustrated in Fig. 5, the number densities of H$\beta$/[Oiii] emitters at $z=0.8$ are in good agreement with those from Khostovan et al. (2015), and higher than those from Ly et al. (2007) and Drake et al. (2013), which could be due to cosmic variance or systematics such as different apertures and estimates of volume and completeness. Using a large 10 deg2 H$\alpha$, H$\beta$/[Oiii] and [Oii] survey, Sobral et al. (2015) estimated empirically that the uncertainty in $L^{\star}$ and $\Phi^{\star}$ due to cosmic variance over the volume probed in these surveys at $z=0.8$ is $\approx 40-50$ %. Such variance could easily explain the observed differences. At $z=2.2$, the number densities are systematically higher in luminosity compared to the literature results. Similarly as in §5.2.2, we find that this is partly due to an aperture effect. As expected, the difference in luminosities measured with different apertures is slightly smaller at $z=2.2$ than at $z=1.47$, with a typical luminosity difference of 0.11 dex. As a consequence, even when matching apertures to Khostovan et al. (2015), we still find an offset at fainter luminosities, which we attribute to cosmic variance. Although the number of H$\beta$/[Oiii] emitters at $z=3.2$ is limited, their number densities agree well with those from Khostovan et al. (2015).

Unlike the H$\alpha$ number densities, the number densities of H$\beta$/[Oiii] emitters are in good agreement with those from Colbert et al. (2013), which could indicate that there is less evolution in the H$\beta$/[Oiii] luminosity function than in the H$\alpha$ luminosity function between $0.7<z<1.5$. However, this could also be due to the contribution of H$\beta$ emitters (see Sobral et al. 2015 and Khostovan et al. 2015 for detailed discussions). Except for luminosities $>10^{43}$ erg s*-1*, the number densities of H$\beta$/[Oiii] emitters at $z=0.8$ are a factor 30 to 100 higher (at fixed [Oiii] luminosity) than the number densities of Type II AGN at $z\sim 0.71$ (Bongiorno et al., 2010).

5.2.4 [Oii] emitters at $z=1.47,3.3,4.7$

Fig. 6 compares the number density of [Oii] emitters at $z=1.47,3.3,4.7$ to the other published results. At $z=1.47$, the number densities agree reasonably well with Drake et al. (2013) and Khostovan et al. (2015), except for the faintest bin, although identification-incompleteness is significant in this bin. At the bright end, the Schechter fit from Ly et al. (2007) indicates a lower number density, potentially due to a lack of bright sources in a small survey volume. We note that the sources in the brightest bin are not spectroscopically confirmed and thus could be interlopers (for this reason, these points have a lighter colour in Fig. 6). Although the number of [Oii] emitters in our samples at $z=3.3$ and $z=4.7$ is limited to a handful, their number densities are slightly higher than Khostovan et al. (2015), potentially indicating either contamination or cosmic variance. A further hint of contamination in this sample is that some have relatively low colour-excess (Fig. 2), which is unexpected for high-redshift sources, although this would also be consistent with a drop in typical EWs for [Oii] emitters (Khostovan et al., 2016).

6 Properties of line-emitters

6.1 Dual-emitters

In total, we detect 42 line-emitters that are line-emitters in multiple narrow-bands. We list the coordinates, redshifts and $I$ band magnitude of these line-emitters in Table 6. The majority (20) of dual-emitters are H$\alpha$+[Oii] emitters at $z=1.47$ detected in NB921 and NBH, followed by 17 H$\alpha$+[Oiii] emitters at $z=2.23$, of which two are also detected in Ly$\alpha$, three in Civ and one in Mgii. Three galaxies are identified as LAE in NB392 at $z\sim$2.2–2.3 and are also detected in either NBH ([Oiii]) or NBK (H$\alpha$), of which one is also detected in NB501 (Civ).

6.2 The most luminous H$\alpha$ emitters at $\bf z=2.23$

B-HiZELS_1 is the most luminous H$\alpha$ emitter known from HiZELS at $z=2.23$ (i.e. Sobral et al., 2016) and also detected as Ly$\alpha$ and [Oiii] emitter. Although its high luminosity (L${}_{\rm H\alpha}=7.9\times 10^{43}$ erg s*-1*) suggests that it is an AGN, it is X-ray undetected (L${}_{X}<3\times 10^{44}$ erg s*-1*, or $<L^{\star}_{X}$, La Franca et al. 2005). There is also a dual-emitter (B-HiZELS_27), detected in [Oiii] and H$\alpha$ at the same redshift only 6*′′* away (a projected distance of $\sim 50$ kpc). This source has an estimated H$\alpha$ EW0 of $\gtrsim 400$ Å and [Oiii] EW${}_{0}\gtrsim 375$ Å. This places it at the very high end of the H$\alpha$ EW distribution at $z=2.2$ (Fumagalli et al., 2012; Sobral et al., 2014), and the galaxy is thus likely a low mass extreme emission line galaxy (e.g. van der Wel et al., 2011). These strong emission-lines could indicate that this galaxy may be undergoing high interaction-induced SFR combined with little extinction due to dust, or is a shocked gas cloud.

The second most luminous H$\alpha$ emitter at $z=2.2$ is B-HiZELS_15 at $z=2.244$ (L${}_{\rm H\alpha}=3.6\times 10^{43}$ erg s*-1*). We also detect [Oiii], H$\alpha$ and Civ emission-lines (there are no Ly$\alpha$ observations at its position). Although B-HiZELS_15 also has a neighbouring galaxy (at $z=2.242$ and a projected separation of $\approx 50$ kpc) and the H$\alpha$ luminosity is only a factor two lower than that of B-HiZELS_1, several other properties are different. B-HiZELS_15 is X-ray detected, has a higher H$\alpha$ EW0 (360 Å versus 120 Å) and is more than three magnitudes fainter in the optical and NIR continuum. This indicates a diversity in the properties of luminous H$\alpha$ emitters, similar to the results from Sobral et al. (2016).

6.3 [OII]-H$\alpha$ view at $\bf z=1.47$

One strength of the Boötes-HiZELS survey is our sample of dual-emitters which can be used to study the relation between different star-formation rate indicators at $z=1.47$ and $z=2.23$, such as [Oii], H$\alpha$ and continuum tracers such as the rest-frame UV, FIR and radio. Compared to H$\alpha$, the [Oii] emission-line and UV continuum are more sensitive to dust attenuation and effects from metallicity and gas density (e.g. Kennicutt, 1998; Jansen et al., 2001; Ly et al., 2012), which may all evolve with redshift. We exploit this sample to derive the observed [Oii]/H$\alpha$ ratio at $z=1.47$ and compare it to a reference sample from SDSS at $z=0.1$, to test claims based on smaller samples by e.g. Hayashi et al. (2013) and Sobral et al. (2012).

We combine the sample of dual-emitters at $z=1.47$ in Boötes with those from HiZELS in the UDS and COSMOS field (see Sobral et al. 2013 for details), and remove any source that is detected in the X-rays. In total, this results in a sample of 340 dual-emitters at $z=1.47$. The majority of these are dominated by faint emitters observed in the deeper imaging in COSMOS. H$\alpha$ luminosities are corrected for the contribution from the adjacent [Nii] doublet using the relation with EW described in Sobral et al. (2012) (see also Sobral et al. 2015 for a spectroscopic validation). As a comparison sample at low redshift, we use a sample of emission-line measurements from a sample of star-forming galaxies at $z\approx 0.1$ drawn from data from SDSS DR7 (Abazajian et al., 2009) as described in Sobral et al. (2012). In short, a sample of 16414 galaxies were selected at $0.07<z<0.1$ with observed H$\alpha$ luminosity $>10^{40.6}$ erg s*-1* and H$\alpha$ EW $>20$ Å. For consistency with our sample at $z=1.47$, we do not remove AGN using the BPT diagnostic (Baldwin et al., 1981). Aperture corrections to emission-line measurements have been done following Garn & Best (2010) based on the ratio between the stellar mass in the fiber and the total stellar mass. We note that these corrections do not change line-ratios.

In Fig. 7, we show the observed [Oii]/H$\alpha$ ratio as a function of observed H$\alpha$ luminosity, normalized by the typical H$\alpha$ luminosity ($L^{\star}_{\rm H\alpha}$) at the specific redshift, both for the sample of dual-emitters and the local comparison sample. After correcting for the evolution in the typical H$\alpha$ luminosity of a factor of $\approx 6$, the distribution of H$\alpha$ luminosities is remarkably similar. By computing the median ratio in 100,000 bootstrap resamples of the data, we measure [Oii]/H$\alpha=0.40\pm 0.01$ with 95 % confidence intervals at $z=0.1$ (slightly lower than the measurement of 0.45 in Kennicutt 1998) and [Oii]/H$\alpha=0.55\pm 0.07$, such that there is a slight increase of the median value with redshift (although the increase is within the observed scatter of $\approx 0.2-0.3$ dex; see also Hayashi et al. 2013). We note that our survey may miss the galaxies with lowest [Oii]//H$\alpha$ ratio, in particular for the faintest H$\alpha$ emitters, which may result in a bias towards finding a higher [Oii]/H$\alpha$ ratio at $z=1.47$. However, if we restrict the analysis to brighter sources ($>0.5\times L^{\star}_{\rm H\alpha}$), we find [Oii]/H$\alpha=0.52\pm 0.05$ at $z=1.47$, while the SDSS results remain unchanged. This indicates that this selection effect is likely not driving the differences. A one dimensional Kolmogorov-Smirnov (KS) test of the observed [Oii]/H$\alpha$ ratios confirms that the distributions are significantly different, with a KS-statistic of 0.20 ($\approx 10^{-10}$ significance) and a P-value of $4\times 10^{-11}$. This indicates that, even though the spread in values is relatively large ($\approx 0.4-0.5$ dex), the median observed [Oii]/H$\alpha$ ratio increases slightly, but statistically significantly, between $z=0.1-1.47$.

We test whether the observed difference can be caused by systematic errors. At $z=1.47$ there is a systematic uncertainty due to the relative filter transmissions at the different wavelengths, that leads to an increase in the scatter and a small bias towards higher [Oii]/H$\alpha$ values. Based on the simulation that is discussed in detail in Sobral et al. (2012), we estimate that this systematic increase is only of the order of $\approx 5$ %, insufficient to explain the offset of the median ratio. If we remove AGN in the SDSS sample using the BPT criterion as defined in Kauffmann et al. (2003), we find [Oii]/H$\alpha=0.42\pm 0.01$. Finally, if we fully mimic the H$\alpha$ measurement (and its correction for the contribution of the [Nii] in the narrow-band), we measure [Oii]/H$\alpha=0.45\pm 0.01$. Thus, none of these effects can explain the observed difference, but they further highlight that the evolution is small, and thus only our large statistical sample can measure evolution.

A higher observed [Oii]/H$\alpha$ ratio is expected when there is less attenuation due to dust, since [Oii] is attenuated more than H$\alpha$ (e.g. Reddy et al., 2015). For example, it could be that galaxies at $z=1.47$ are less dusty. Indeed, if we restrict the sample of local galaxies to those with A${}_{\rm H\alpha}<1.3$ (median A${}_{\rm H\alpha}=0.82$, compared to a median A${}_{\rm H\alpha}=0.91$ for the full sample), we find a similar observed [Oii]/H$\alpha$ ratio of $0.49\pm 0.01$. However, results from Herschel stacking of H$\alpha$ emitters at $z=1.47$ (Thomson et al. 2017; see also Ibar et al. 2013) indicate that their extinction properties are similar to local galaxies, with a similar relation between stellar mass and AHα (Garn & Best, 2010). We note that because the samples are matched in L/L${}^{\star}_{\rm H\alpha}$ there are likely no significant mass differences between the samples (e.g. Sobral et al., 2014). A more detailed analysis of the extinction properties of this sample is beyond the scope of this paper. Finally, another explanation is that the SDSS fiber-measurements are biased towards higher extinction (and thus lower [Oii]/H$\alpha$ ratios), because they measure the line-ratios in the central 3-4 kpc of galaxies, that are observed to be dustier/more evolved (e.g. Sánchez et al., 2014), while the 3*′′* measurements at $z=1.47$ measure flux out to radii of 8-13 kpc (depending on UDS/COSMOS or Boötes). Therefore, the observed offset between $z=0.1$ and $z=1.47$ could also be an observational effect due to dust, age and/or metallicity gradients within galaxy. These observational issues can be overcome with large IFU or matched NB surveys in the local Universe, as for example the J-PAS project (Benitez et al., 2014).

6.4 X-ray fraction & the power-law component of the luminosity function

We investigate the AGN fractions of H$\alpha$ emitters at $z=1.47$ and $z=2.23$ and LAEs at $z=2.4$ by matching our samples to source catalogs from X-ray (Chandra, 0.5–7.0 keV, depth $7.8\times 10^{-15}$ erg s*-1* cm*-2*, 1*′′* resolution and matching radius, Kenter et al. 2005). In addition, we include the H$\alpha$ emitters at $z=1.47$ and $z=2.23$ in the COSMOS field from Sobral et al. (2013) and match these with the Chandra COSMOS point source catalog (0.5–7.0 keV, $5.7\times 10^{-16}$ erg s*-1* cm*-2*, Elvis et al. 2009); see also Calhau et al. (2017) for their detailed properties. These X-ray flux limits correspond to luminosity limits of 0.3–3$\times 10^{44}$ erg s*-1*. If such an X-ray luminosity would have its origin in star formation, it would require SFRs of $>10^{3.5-5}$ M*⊙* yr*-1* (e.g. Lehmer et al., 2016), which is unlikely. This clearly indicates an AGN origin of all X-ray detections discussed in this section.

In total, we detect 21 HAEs in the X-ray, of which 10 are at $z=1.47$ and 11 are at $z=2.23$ and half of the X-ray detected HAEs are in Boötes. We find that the detection rate depends strongly on the H$\alpha$ luminosity, see Fig. 8. This has also been observed using spectroscopic follow-up by Sobral et al. (2016), who found that the majority ($80\pm 30$ %) of luminous HAEs are broad-line AGN. Relatively independent of redshift, roughly half of the most luminous HAEs are X-ray detected. Note that the other half may easily be undetected due to the short duty cycle of X-ray AGN (e.g. Shankar et al., 2009; Fiore et al., 2012), although it could also indicate that roughly half of the luminous HAEs are optically thick to X-rays. By combining the data-points above L*⋆* at $z=1.47-2.23$, we find a best fit $f_{X}=0.57^{+0.15}_{-0.15}$log10(L${}_{\rm H\alpha}/10^{43}$ erg s*-1*) + 0.25${}^{+0.06}_{-0.06}$, that we illustrate in Fig. 8.

We combine the sample of LAEs at $z=2.2-2.4$ identified with the NB392 and the stV filter to investigate the X-ray fraction of LAEs as a function of luminosity. Out of the 41 LAEs, eight are X-ray detected (L${}_{X}\gtrsim 3\times 10^{44}$ erg s*-1*). The X-ray fraction of LAEs increases strongly with line luminosity, from $\approx 0$% at L${}^{\star}_{\rm Ly\alpha}$ to $\approx 100$ % at $\gtrsim 3\times 10^{44}$erg s*-1*. We note that the Ly$\alpha$ luminosities at which the X-ray fraction exceeds 20 % correspond to the luminosities at which the number densities start to deviate from a Schechter function, as observed in Konno et al. (2016), Sobral et al. (2017) and this work (Fig. 4). For both H$\alpha$ and Ly$\alpha$, the X-ray fraction increases above L*⋆. For Ly$\alpha$, we find a best fit relation of: $f_{X}=0.41^{+0.18}_{-0.18}$log10*(L${}_{\rm Ly\alpha}/10^{43}$ erg s*-1*) above L${}_{\rm Ly\alpha}>10^{43}$ erg s*-1*.

7 Conclusions

We presented the first results from the Boötes-HiZELS survey, which uses six narrow-bands to select emission-line galaxies from $z=0.4-4.7$ in a 0.7 deg2 region in the Boötes field. We described the observations, data-reduction, extraction of catalogs and selection of line-emitters, and how multi-wavelength data has been used to classify different populations of line-emitters. The main results are:

We identify 362 candidate H$\alpha$ emitters (HAEs) at $z=0.4,1.47,2.23$, 387 H$\beta$/[Oiii] emitters at $z=0.8,2.23,3.3$, 285 [Oii] emitters at $z=1.47,3.3,4.7$ and 73 Ly$\alpha$ emitters (LAEs) at $z=2.23,2.3,3.1$. 2. 2.

Using a suite of matched narrow-band filters, we identify 42 galaxies with emission-lines in multiple narrow-bands, providing 22/18 new robust redshift identifications of [Oii]/H$\alpha$ and [Oiii]/H$\alpha$ emitters at $z=1.47/2.23$, without pre-selection on AGN activity or $I$ band magnitude, see §6 and Table 6. 56 additional line-emitters have a spectroscopic redshift. 3. 3.

In general, the number densities of line-emitters as a function of luminosity we derive agree remarkably well with luminosity functions observed in other survey fields (§5, Figures 4, 5 and 6), confirming strong evolution in L${}^{\star}_{\rm H\alpha}$ from $z=0.4-2.2$ and evolution in L${}^{\star}_{\rm Ly\alpha}$ from $z=2.2-3.1$. 4. 4.

We confirm the result from Konno et al. (2016) and Sobral et al. (2017) that the luminosity function of LAEs at $z\approx 2.2$ diverges from a Schechter function at the bright end, L${}_{\rm Ly\alpha}\gtrsim 10^{43}$ erg s*-1*. At these luminosities, the luminosity function follows log${}_{10}(\Phi)=27.5-0.74$log10(LLyα). Such a departure from a Schechter function is also clearly observed at the highest H$\alpha$ luminosities (L${}_{\rm H\alpha}\gtrsim 10^{43.5}$ erg s*-1*) at $z=2.2$ (§5.2 and §5.2.2). 5. 5.

Combining our sample of dual-emitters with those from the COSMOS and UDS fields from HiZELS, we compare the observed [Oii]/H$\alpha$ ratio of 340 star-forming galaxies at $z=1.47$ with those from a reference sample in the local Universe (§6.3). We measure a median ratio of [Oii]/H$\alpha=0.40\pm 0.01$ at $z=0.1$ and [Oii]/H$\alpha=0.55\pm 0.07$ ([Oii]/H$\alpha=0.52\pm 0.05$ if we restrict the sample to sources with slightly higher S/N, see Fig. 7). The $\approx 0.1$ dex offset can potentially be attributed to a lower dust attenuation at $z=1.47$, or biases in the fiber-measurements in the local Universe, which measure the ratio at the central 3-4 kpc of galaxies, while the measurements at $z=1.47$ are integrated over $\approx 10$kpc. 6. 6.

By exploiting Chandra X-Ray data, we show that the H$\alpha$ and Ly$\alpha$ luminosities at which the number densities start to diverge from pure Schechter form at similar luminosities to where the X-ray fractions start to increase, from $\sim 20$ % to $\sim 100$ % (Fig. 8). We also show that, under basic assumptions, the majority of luminous LAEs are not broad-line Type I AGN (Fig. 4), and more likely narrow-line Type II AGN.

The sample of identified line-emitters can be used to study various properties of star-forming galaxies. In particular, the relatively large sample of H$\alpha$ emitters at $z=1.5-2.2$ can be used to test various SFR indicators (H$\alpha$, rest-UV, radio, FIR) in future papers.

Acknowledgments

We thank the referee for their constructive comments which helped improve the quality of this work. JM acknowledges the support of a Huygens PhD fellowship from Leiden University and JM and DS acknowledge financial support from a NWO/VENI grant awarded to David Sobral. IRS acknowledges support from STFC (ST/L00075X/1), the ERC Advanced Grant DUSTYGAL (321334) and a Royal Society/Wolfson Merit Award. HR acknowledges support from the ERC Advanced Investigator program NewClusters 321271. PNB is grateful to STFC for support via grant ST/M001229/1. B.D. acknowledges financial support from NASA through the Astrophysics Data Analysis Program (ADAP), grant number NNX12AE20G. We thank Ana Afonso, João Calhau, Leah Morabito, Iván Oteo, Sérgio Santos and Aayush Saxena for their assistance with observations. This work is based on observations obtained using the Wide Field Camera (WFCAM) on the 3.8m United Kingdom Infrared Telescope (UKIRT), as part of the High-redshift(Z) Emission Line Survey (HiZELS; U/CMP/3 and U/10B/07), using Suprime-Cam on the 8.2m Subaru Telescope as part of program S14A-086 and using the WFC on the 2.5m Isaac Newton Telescope, as part of programs 2013AN002, 2013BN008, 2014AC88, 2014AN002, 2014BN006, 2014BC118 and 2016AN001, using ISIS and AF2+WYFFOS on the 4.2m William Herschel Telescope, as part of programs 2016AN004 and 2016BN011 and using DEIMOS on the 10m Keck II Telescope as part of program C267D and on observations made with ESO Telescopes at the La Silla Paranal Observatory under ESO programme IDs 098.A-0819 and 179.A-2005. This work made use of images and/or data products provided by the NOAO Deep Wide-Field Survey (Jannuzi & Dey, 1999) which is supported by the National Optical Astronomy Observatory (NOAO). NOAO is operated by AURA, Inc., under a cooperative agreement with the National Science Foundation. We have benefited greatly from the public available programming language Python, including the numpy, matplotlib, pyfits, scipy and astropy packages (Astropy Collaboration et al., 2013), the astronomical imaging tools SExtractor, Swarp and Scamp and the Topcat analysis program (Taylor, 2013).

Appendix A Photometric consistency check

In our catalogue production steps (§3.2), we only include objects in our catalogues that have a physically plausible NB excess. Because the BB covers the same wavelength as the NB, flux in the NB must also be observed in the BB, otherwise the object is either unreal (such as a cosmic ray, artefacts, etc.) or variable (such as variable stars, supernovae or AGN). It is fairly straightforward to compute the faintest possible BB magnitude given a NB magnitude:

[TABLE]

in this equation, $\lambda_{c,X}$ is the central wavelength of filter $X$, and $\Delta\lambda_{X}$ the width of filter $X$. We conservatively add 0.5 magnitude to take into account uncertainties in the photometry and relative filter transmissions. We remove any source for which the excess is larger than $BB_{max}-NB$. For example, for NB921 and $z$ band, this equation results in $BB_{max}=NB+3.4$. This means that if a source has a NB921 magnitude of 20, it must have a $z$ magnitude of 23.4 or brighter. It is possible that the implied $BB_{max}$ is below the background. In that case, we exclude sources for which the BB is not detected at 2$\sigma$. This consistency check removes most spurious objects such as cosmic rays and detector artefacts.

Appendix B Colour-colour selections

In this section we illustrate the colour-colour selection criteria outlined in Table 3. In all figures (Fig. 9, 10, 11, 12 and 13), line-emitters with secure redshift-identifications from spectroscopy or dual-NB detections are shown with larger symbols. X-ray detected AGN are shown with a star symbol.

Appendix C Catalogue of dual-emitters

Appendix D Example catalogs of line-emitters