Disentangling the Physical Origin of Emission Line Ratio Offsets at High Redshift with Spatially Resolved Spectroscopy

Much of our knowledge of galaxy evolution across cosmic time is based on strong nebular emission lines at rest-frame optical wavelengths. Analysis of nebular line emission reveals properties of star formation and ionized gas including the total star formation rate (SFR), metallicity, gas density, and the ionization parameter. Such studies have proven highly productive, with the advent of sensitive multiplexed spectrographs enabling comprehensive emission line studies at high redshifts (e.g., Steidel et al. 2014; Shapley et al. 2015; Momcheva et al. 2016; Sanders et al. 2016a; Kashino et al. 2019).

A key limitation of evolutionary studies is that some properties are typically not measured directly, in particular the gas metallicity (for which we will adopt the standard convention in terms of the oxygen abundance 12 + logO/H). Instead, nearly all studies at high redshift rely on calibrations of strong emission line ratios to estimate the galaxy metallicity (but see, e.g., Sanders et al. 2016b, 2020a; Patrício et al. 2018; Gburek et al. 2019 for small z > 1 samples that have the direct metallicities based on electron temperature T_e ). When applying these strong-line diagnostics, it is implicitly assumed that the high-redshift samples have similar interstellar medium (ISM) physical conditions as star-forming regions in nearby (z ∼ 0) galaxies, which are used to calibrate these methods.

Large spectroscopic samples of z > 1 galaxies have revealed significant redshift evolution in the excitation sequences of strong emission line ratios. Most notably, it is now well established that there is an offset in the Baldwin—Phillips—Terlevich (BPT) diagram of [O iii]/Hβ versus [N ii]/Hα (Baldwin et al. 1981) between z > 1 samples and the locus of z ∼ 0 star-forming galaxies (e.g., Steidel et al. 2014; Shapley et al. 2015; Kashino et al. 2017). This BPT offset suggests an evolution in the underlying physical properties governing emission line production, such as the shape of the ionizing spectrum, gas density, ionization parameter, and chemical abundance pattern (e.g., Kewley et al. 2013a). One consequence of changing ISM conditions is that the translation between strong-line ratios and metallicity will also evolve, such that applying locally calibrated diagnostics at high redshifts will yield systematically biased metallicity estimates (e.g., Bian et al. 2018; Patrício et al. 2018; Sanders et al. 2021). This systematic error will directly affect inferences on galaxy formation and evolution based on the observed chemical evolution of galaxies (e.g., Lilly et al. 2013; Troncoso et al. 2014; Zahid et al. 2014; Sanders et al. 2021).

Several hypotheses for the observed emission-line ratio offsets at high redshift have been suggested (e.g., Kewley et al. 2013a). These include emission mixing from low-luminosity active galactic nuclei (AGN; e.g., Wright et al. 2010) or shocked gas (e.g., Yuan et al. 2012), contributions from diffuse ionized gas (DIG; e.g., Zhang et al. 2016), an enhanced N/O abundance ratio (e.g., Masters et al. 2014; Shapley et al. 2015; Masters et al. 2016), high ionization parameter (e.g., Kewley et al. 2013b), and hard ionizing spectra driven by super-solar α/Fe abundance patterns (e.g., Steidel et al. 2016; Shapley et al. 2019; Sanders et al. 2020a; Topping et al. 2020a; Sanders et al. 2020b). To distinguish among these possibilities—and thereby recover accurate results with emission line studies—we require sensitive physical measurements beyond those typically available from the strongest nebular lines.

A critical step forward is the measurement of gas–phase metallicity and temperature using the direct T_e method based on auroral emission lines. In our earlier work, Jones et al. (2015) presented T_e-based abundance measurements for a sample of star-forming galaxies at z ∼ 0.8 drawn from the DEEP2 Galaxy Redshift Survey (DEEP2; Davis et al. 2003; Newman et al. 2013). Jones et al. (2015) found that the relations between direct metallicity and rest-optical line ratios of [O ii], [O iii], Hβ, and [Ne iii] for the z ∼ 0.8 sample are remarkably consistent with the sample of z ∼ 0 galaxies from Izotov et al. (2006). While this comparison appeared to suggest no evolution in metallicity calibrations over z = 0 − 0.8, a key caveat is that the Izotov et al. (2006) sample comprises extreme local galaxies with high specific SFRs analogous to those of high-redshift galaxies due to selection effects, as noted by Sanders et al. (2020a, 2021).

Comparison to a wider range of local galaxy properties is important for fully understanding the evolution of metallicity calibrations. Indeed, in comparing more representative samples of local galaxies and H ii regions to T_e-based samples at higher redshifts, Sanders et al. (2020a) found evidence that strong-line metallicity calibrations evolve between z ∼ 0 and z ∼ 1.5–3.5. By modeling the spectra of their z ∼ 2 T_e sample, these authors found that the evolution of strong-line calibrations is due to a harder ionizing spectrum driven by α-enhancement of young, massive stars, in agreement with strong-line and rest-UV studies providing similar explanations for the z ∼ 2 BPT diagram offset (Steidel et al. 2016; Strom et al. 2018; Shapley et al. 2019; Sanders et al. 2020b; Topping et al. 2020a; Sanders et al. 2020b; Runco et al. 2021). While these studies based on integrated galaxy spectra have yielded great progress in understanding the drivers of galaxy line ratio evolution, there remain potential contributors that cannot be easily identified in integrated measurements including DIG emission, weak AGN, and shocks.

This work is concerned with the spatially resolved analysis of a subset of galaxies studied by Jones et al. (2015) to further understand the ionizing sources that power nebular line emission in high-z galaxies. Here we use Hubble Space Telescope (HST) WFC3-IR grism spectroscopy covering the Hα+[N ii], ⁸ [S ii], and [S iii] emission lines for 15 galaxies at ≃ 1 kpc spatial resolution. Spatial mapping of these emission lines enables several key tests of the ionization mechanisms. In particular we seek to distinguish emission powered by star formation (i.e., H ii regions), AGN, shocked gas, and DIG. If AGN (particularly low-ionization nuclear emission-line regions, LINERs) are present and responsible for BPT offsets, we expect to observe an elevated [S ii]/(Hα+[N ii]) ratio in the central ∼1 kpc resolution element. Conversely if BPT offsets are due to shocked gas or DIG, we would expect elevated [S ii]/(Hα+[N ii]) in low surface brightness regions. Contributions from density-bounded H ii regions with escaping ionizing radiation are expected to result in elevated [S iii]/[S ii] ratios in the corresponding spatial regions. The diffraction-limited HST spectroscopic data presented herein therefore allow us to differentiate between these possibilities and discern the origin of redshift evolution in the BPT diagram. Understanding the cause of these line ratio offsets will allow us to better calibrate metallicity measurements between the low- and high-redshift universe.

This paper is structured as follows. In Section 2, we use spatially resolved spectroscopy of example galaxies at z ∼ 0 to verify the methodology that we will apply to HST/WFC3-IR grism spectra. We describe our HST sample and data reduction methods in Section 3. In Section 4 we describe our analysis techniques and discuss one sample object in depth. In Sections 5 and 6 we analyze the various processes that can potentially power the observed emission lines. Finally, in Section 7, we summarize our results and discuss the implications of this work for future research. Where necessary we assume a flat ΛCDM cosmology with H₀ = 69.6 km s⁻¹ Mpc⁻¹, Ω_m = 0.286, and Ω_Λ = 0.714. All magnitudes are on the AB system.

To establish our methodology, we first consider examples of different sources of emission line excitation observed in low-redshift galaxies from the SDSS-IV Mapping Nearby Galaxies at APO (MaNGA) survey (Bundy et al. 2015), which obtained optical integral field spectroscopy (IFS) at ∼1 kpc resolution for ∼10,000 low-redshift galaxies. In this section we use MaNGA IFS data to demonstrate how various ionization scenarios can be clearly distinguished using the diagnostics available with HST grism data. Our primary diagnostic is the line ratio [S ii]/(Hα+[N ii]), combined with the Hα+[N ii] surface brightness and spatial location within a galaxy.

We utilized the MaNGA catalogs and emission-line flux maps produced using the PIPE3D pipeline ⁹ (Sánchez et al. 2016a, 2016b). For each galaxy, we selected all spaxels with S/N ≥ 3 in Hα, Hβ, [O iii]λ5007, [N ii]λ6584, [S ii]λ6716, and [S ii]λ6731. We then calculated the line ratios [O iii]/Hβ, [N ii]/Hα, [S ii]/Hα, and [S ii]/(Hα+[N ii]), noting that a reddening correction is not required due to the close proximity of the wavelengths of the lines in each ratio. Following Belfiore et al. (2016), we classified spaxels as primarily ionized by star formation (SF), Seyfert-like AGN, or as low-ionization emission regions (LIERs), based on cuts in the [O iii]/Hβ versus [S ii]/Hα diagram using the demarcations of Kewley et al. (2001, 2006). LIERs that are nuclear (e.g., LINERs) can be attributed to low-luminosity AGN, while LIERs that are extended throughout the disk are thought to originate from shocked gas and/or DIG emission (Belfiore et al. 2016). We furthermore computed the observed Hα+[N ii] luminosity surface brightness (Σ_Hα ) using the Hα+[N ii] fluxes, spaxel sizes, and measured redshifts.

We produced maps of [S ii]/(Hα+[N ii]) and SF versus LIER versus AGN classification, as well as plots of [S ii]/(Hα+[N ii]) versus Σ_Hα (where Σ_Hα represents the Hα+[N ii] surface brightness) for many MaNGA targets in order to evaluate how well these observables can distinguish between the various ionization mechanisms that we will use for our subsequent HST analysis. Illustrative examples of four MaNGA targets are shown in Figure 1. In each row, the left panel displays a map of the [S ii]/(Hα+[N ii]) ratio, the middle panel shows the ionization mechanism classification for each spaxel, and the right panel presents [S ii]/(Hα+[N ii]) as a function of Σ_Hα , with the properties of the central 1 kpc indicated by a star. For purposes of Figure 1, we adopt a single classification for shocks, DIG, LIER, and LINER (i.e., DIG/LI(N)ER) as these origins are not clearly distinguished by the emission line ratios alone. In our analysis herein we adopt the following conventions: concentrated nuclear LIER emission is considered a signature of LINER AGN (e.g., Figure 1(b)), while spatially extended low surface brightness LIER emission is considered a signature of shocks and DIG (e.g., Figure 1(c)). In our analysis, we distinguish between LINER and DIG/LIER regions by their spatial distribution and Hα+[N ii] surface brightnesses: LINER emission is nuclear in origin and thus will be concentrated in the central regions of the galaxies, typically with higher Hα+[N ii] surface brightness whereas DIG/LIER regions are located throughout the disk and in galaxy outskirts, in regions with predominantly lower Hα+[N ii] surface brightnesses.

Zoom In Zoom Out Reset image size

2.1. Weak AGN

2.2. Shocks and Diffuse Ionized Gas

2.3. Escaping Ionizing Radiation

2.4. Star Formation

Having established the utility of spatially resolved line ratio mapping with z ∼ 0 galaxies, we will now apply the approach from Section 2 to HST grism spectroscopy at higher redshift (z ∼ 0.8). Despite the larger cosmological distance, HST delivers ∼1.0 kpc spatial resolution, which is comparable to the low-redshift examples used to verify our methodology.

3.1. Sample

3.2. Keck Near-infrared Spectroscopy

We obtained follow-up near-infrared (near-IR) spectroscopy of 23 galaxies from the T_e sample in order to measure the Hα, [N ii], and [S ii] emission lines, using the Keck/Near Infrared Spectrograph (NIRSpec) and Keck/MOSFIRE spectrographs (T. Jones et al. 2021, in preparation), used to measure the BPT offsets of our sample. While weather conditions prevented us from securing near-IR spectra of all targets, the observed subset is representative of the T_e sample. Relative to the broader z ≃ 0.8 star-forming galaxy population, this sample has large emission-line equivalent width (EW) and specific SFRs, relatively blue continuum colors, and luminosities representative of the overall sample observed by the DEEP2 survey (Jones et al. 2015). In brief, data were reduced using the LONGSLIT_REDUCE pipeline written by George Becker in the case of NIRSPEC, and the MOSFIRE Data Reduction Pipeline in the case of MOSFIRE. Both pipelines perform wavelength calibration, instrument signature removal, sky subtraction, and 1D spectral extraction. MOSFIRE 1D spectra were extracted using the bmep program (Freeman et al. 2019) following the methods used in the MOSFIRE Deep Evolution Field (MOSDEF) survey (Kriek et al. 2015). Standard star observations taken on the same nights were used for telluric correction. Emission line fluxes were measured using single Gaussian fits (as in Jones et al. 2015). Figure 2 shows an example Keck/MOSFIRE spectrum and the best-fit emission line profiles, with good spectral resolution of the [N ii] and [S ii] doublets. We selected the 15 galaxies with the largest [S ii] emission fluxes for the HST grism spectroscopy, which forms the basis of this study. Availability of moderate resolution ground-based spectra is an important aspect, as this provides the [N ii]/Hα ratio for our targets (which is not available from HST grism spectra alone). Although the [S ii] flux selection may bias the HST subsample in terms of emission line ratios, we find that the 15 HST targets are representative of the parent sample in terms of location on the [N ii] BPT diagram. Figure 3 shows both the [O iii]/Hβ versus [N ii]/Hα and [S ii]/Hα BPT diagrams for the HST sample, color coded by their offset from the [N ii] BPT diagram (hereafter referred to as the BPT offset) with negative offsets lying below the typical z ∼ 0 star-forming locus. Each target's BPT offset is calculated as the minimum distance from the z ∼ 0 star-forming locus on the [O iii]/Hβ versus [N ii]/Hα BPT diagram, with a median offset of ∼0.08 dex. In both diagrams, our targets (z ∼ 0.8) show a variety of offsets from where z ∼ 0 H ii regions typically lie. This work explores the cause of the [N ii] BPT offset by spatially mapping emission lines to distinguish between various potential mechanisms that may drive emission at high redshift.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.3. HST Observations and Data Reduction

The targets were observed using HST's Wide Field Camera 3 (WFC3) via program GO-15077. Each target was observed for a single orbit, with four dithered G141 grism exposures plus two direct imaging exposures with the F140W filter, with a median exposure time of ∼2100 s per target. Photometric magnitudes are calculated from direct images and we assume a systematic error floor of 0.02 magnitudes (see Table 1). We use the Grism Redshift and Line Analysis software (Grizli ¹⁰ ; Brammer 2019) to reduce the HST slitless grism spectroscopy data. Grizli offers a complete data reduction pipeline, beginning with preprocessing both the raw direct image and grism exposures, forward modeling their entire field of view, fitting redshifts through a spectral template synthesis, extracting both 1D and 2D spectra (e.g., Figure 4), and outputting spatially resolved emission-line maps and integrated emission-line fluxes (given for our targets in Table 2). Of particular importance for this work is the level of contamination from overlapping spectra and zeroth-order images, which occurs with slitless grism data and can introduce errors in emission line measurements. In all cases the telescope orientation angle was constrained to avoid an overlap of bright objects with our primary target spectra, and we additionally inspect the output spectra to check for contamination residuals. We do not find any cases where contamination significantly affects the primary target spectra that form the basis of this work.

Zoom In Zoom Out Reset image size

Table 2. Emission Line Measurements

Target ID	Integrated Hα+[N ii] a	Integrated [S ii] a	Integrated [S iii] a	Keck [N ii]/Hαb	Keck [O iii]/Hβc
	10−16 erg s−1 cm−2	10−16 erg s−1 cm−2	10−16 erg s−1 cm−2
DEEP2-13016475	7.69 ± 0.16	0.68 ± 0.14	0.81 ± 0.09	0.019 ± 0.002	6.16 ± 0.06
DEEP2-13043682	3.90 ± 0.21	1.00 ± 0.17	0.49 ± 0.17	0.062 ± 0.008	4.64 ± 0.08
DEEP2-13043716	15.24 ± 0.29	1.78 ± 0.21	2.80 ± 0.29	0.147 ± 0.003	2.77 ± 0.05
DEEP2-14018918	4.38 ± 0.20	0.46 ± 0.17	0.53 ± 0.21	0.021 ± 0.006	6.19 ± 0.08
DEEP2-21021292	4.12 ± 0.20	0.71 ± 0.17	0.22 ± 0.18	0.055 ± 0.007	3.80 ± 0.07
DEEP2-21027858	4.96 ± 0.13	0.65 ± 0.11	0.72 ± 0.30	−0.027 ± 0.017 d	4.91 ± 0.11
DEEP2-22020856	3.30 ± 0.17	0.41 ± 0.15	0.27 ± 0.24	0.045 ± 0.006	5.20 ± 0.08
DEEP2-22028402	27.67 ± 0.31	3.68 ± 0.26	4.49 ± 0.35	0.147 ± 0.001	2.49 ± 0.03
DEEP2-22022835	2.57 ± 0.15	0.08 ± 0.13	0.70 ± 0.34	−0.003 ± 0.012 d	5.22 ± 0.12
DEEP2-22032252	2.22 ± 0.16	0.33 ± 0.14	0.18 ± 0.09	0.123 ± 0.023	5.37 ± 0.11
DEEP2-22044304	5.44 ± 0.17	0.85 ± 0.15	0.63 ± 0.25	0.054 ± 0.003	5.51 ± 0.04
DEEP2-41059446	4.74 ± 0.31	1.24 ± 0.28	1.40 ± 0.38	0.031 ± 0.006	4.90 ± 0.13
DEEP2-42045870	3.58 ± 0.25	0.82 ± 0.23	0.06 ± 0.47	0.036 ± 0.005	4.16 ± 0.07
DEEP2-31046514	4.03 ± 0.24	1.01 ± 0.22	0.55 ± 0.33	0.043 ± 0.014	4.85 ± 0.12
DEEP2-31047144	5.29 ± 0.15	0.56 ± 0.13	0.50 ± 0.58	0.046 ± 0.010	6.45 ± 0.13

Notes.

^a Measured from HST WFC3-IR grism spectra. ^b [N ii]λ6584/Hα, measured from MOSFIRE or NIRSPEC spectra, as described in Section 3.2. ^c Measured from DEEP2 spectra as described in Jones et al. (2015). Here [O iii] refers to the λ5007 line but is calculated as 2.98 × the [O iii]λ4959 flux, as [O iii]λ5007 is redshifted beyond the spectral coverage in several targets. ^d These objects have negative [N ii] fluxes in their fits. Figures using these data show 2σ upper limits.

Download table as:  ASCIITypeset image

In this paper we are mainly concerned with spatially resolved nebular emission-line fluxes. Our analysis uses the Hα+[N ii] and [S ii] emission line maps directly from Grizli, where both the Hα+[N ii] lines and the [S ii]λ λ 6717,6731 doublet are blended due to the grism's low spectral resolution. When both lines of the [S iii]λ λ 9068,9531 doublet are available, we use a weighted average of the doublet for our [S iii] measurement, using the fixed ratio of 2.44:1 between the stronger (λ9531) and weaker (λ9068) lines. When [S iii]λ9531 is redshifted out of the G141 grism wavelength coverage and thus cannot be detected, we use only the [S iii]λ9068 line multiplied by a factor of 2.44 (i.e., expressed as the expected intensity of the λ9531 line). Figures 5 and 6 show examples of HST data products, showing direct images of the entire sample and spatially resolved emission-line maps for one target (DEEP2-13043716), respectively. The pixel scale of the emission line maps is 01 and the angular resolution is 013 (FWHM). Emission line maps for the remainder of the sample can be found in the Appendix.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

This work utilizes the Hα+[N ii], [S ii], and [S iii] emission line maps output from Grizli to study spatially resolved emission-line ratios. We are specifically interested in regions with elevated [S ii]/(Hα+[N ii]) or [S iii]/[S ii], which are indicative of potential effects of weak AGN, shocks, and DIG, or escaping ionizing radiation as demonstrated in Section 2. These effects can plausibly cause the observed deviation of high-redshift galaxies from the z ∼ 0 star-forming locus in BPT diagrams.

We define the spatial extent of each target from its F140W direct image by applying a surface brightness cutoff for each system of ∼25.5 magnitudes/arcsec². This cutoff probes into the outer regions of our targets, including essentially all detected line emission, while mitigating the amount of noise included from larger radii. This area corresponds to ∼ 20 − 140 independent spatial resolution elements per target (median of 40), indicating that our galaxies are resolved and that we have the appropriate resolution to distinguish the spatial structure and trends with surface brightness. The targets are however so sufficiently compact that there is no significant blending between the Hα+[N ii] and [S ii] lines (see Figures 4 and 5), such that we obtain clean maps of both over the same spatial regions.

To increase the signal-to-noise ratio (S/N) for spatially resolved analysis, we utilize two methods of binning our data before calculating the line ratios of interest, namely [S ii]/(Hα+[N ii]) and [S iii]/[S ii]. First we binned the data by Hα+[N ii] surface brightness in order to reach the lowest surface brightness regions that are not necessarily spatially contiguous. This is motivated by our objective to search for possible DIG or LIER emission (e.g., Figure 1(c) shows prominent DIG at low surface brightness regions scattered throughout the galaxy). Second, to examine spatially correlated regions, we use the Voronoi binning method of Cappellari & Copin (2003). Voronoi binning assigns spatially adjacent pixels to bins with a predetermined S/N requirement, and allows us to distinguish between the central and outer regions of galaxies more clearly. We required a median S/N threshold of 10 for Hα+[N ii] in each Voronoi bin.

Here we briefly discuss results for the galaxy DEEP2-13043716 demonstrated in Figures 7 and 8, as an example of what can be learned from the grism spectroscopy. This galaxy is offset by ∼0.1 dex from the z ∼ 0 star-forming locus in the BPT diagram, slightly above the median offset in our sample (Table 1). The ratio [S ii]/(Hα+[N ii]) ≃ 0.12 is remarkably constant across an order of magnitude in surface brightness, thus ruling out LINER AGN, shocks, and DIG as the dominant forms of emission on kiloparsec scales. The [S iii]/[S ii] ratio for this object is broadly constant with values ∼1–2, although the error bars are large for regions of lower surface brightness. However in our regions of interest (i.e., those with the highest surface brightness), the [S iii]/[S ii] ratio remains flat and we can rule out a large contribution of density-bounded H ii regions to the observed emission. The values are comparable to local high-excitation H ii regions and z ∼ 1.5 main-sequence galaxies (e.g., Sanders et al. 2020b). In sum, this target lacks the clear distinguishing features in the line ratio maps which we would expect for LINER AGN, shocks, DIG, or escaping ionizing radiation, thus ruling out strong contributions from these sources. In contrast, this target exhibits relatively uniform [S ii]/(Hα+[N ii]) and [S iii]/[S ii] ratios across all surface brightnesses which are consistent with that expected from excitation by newly formed stars, suggesting that star formation may be the primary driver of its emission lines in all well-detected spatial regions. We perform a quantitative analysis for the complete sample in Section 5.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

In this section we explore the possible sources of energy giving rise to the observed emission lines in our sample. Our approach here is to consider evidence for each emission mechanism based primarily on aggregate characteristics of the sample as a whole, as well as relevant individual objects (Appendix). We note that the general results of this section also hold if we consider only the subset of galaxies with largest offsets in the BPT diagram (e.g., > 0.1 dex offset from Table 1).

5.1. Weak AGN

5.1.1. Seyfert AGN

As noted in Section 2, we cannot distinguish Seyfert AGN by looking at [S ii]/(Hα+[N ii]) levels because Seyfert AGN-dominated nuclei exhibit the same range of [S ii]/(Hα+[N ii]) values as the spaxels dominated purely by star formation (see Figure 1(a)). However, we can still explore whether these offsets are driven by significant Seyfert AGN activity by estimating what percent of the flux must be attributed to AGN in order to cause the offsets observed in our HST sample.

Figure 9 shows how points along the Kewley et al. (2013a) z ∼ 0 star-forming locus would be offset when increasing fractions of the flux are from AGN, assuming 0% Seyfert AGN contribution along the star-forming locus and 100% Seyfert AGN contribution at the tip of the AGN branch in the [N ii] BPT diagram, assuming fiducial Seyfert line ratios of log([N ii]/Hα) = 0 and log([O iii]/Hβ) = 1 (e.g., Kewley et al. 2006). Using this metric, BPT offsets in our sample can be explained by a relatively small contribution of ≲10% of their Hα flux coming from AGN. The percentages of Hα flux in the central resolution elements of our sample range from 10%–40%. Therefore it is plausible that Seyfert AGN could be responsible for the observed offsets, if such emission represents a large fraction of flux from the central 1 kpc². If such AGN emission is common, we would expect to see signatures in IFS data that resolve the [N ii]/Hα lines, or in X-ray or radio observations. We note that, given the luminosities of our sample, we would need relatively deep X-ray or radio data to determine if there is significant AGN emission. However, sensitive IFS surveys at similar redshift indicate low fractions of AGN activity at stellar masses of ≲ 10¹⁰ M_☉ and integrated [N ii]/Hα < 0.2 ratios of our sample (with AGN signatures in <10% of galaxies; e.g., Förster Schreiber et al. 2019). Data with adaptive optics in particular can clearly distinguish elevated nuclear [N ii] emission from AGN at the level required to explain offsets in our sample, but is rarely seen at these masses (e.g., Jones et al. 2013; Leethochawalit et al. 2016; Hirtenstein et al. 2019). Thus we view Seyfert AGN activity as an unlikely explanation for the BPT offsets.

Zoom In Zoom Out Reset image size

5.1.2. LINER AGN

We can distinguish contributions from LINER-type emission by looking for elevated [S ii]/(Hα+[N ii]) in the nuclear regions (as shown in Figure 1(b)). Figure 10 shows all data for the aggregate sample, where we can see that the nuclear and highest surface brightness regions are relatively flat with values of [S ii]/(Hα+[N ii]) ≲ 0.2, well below that expected for LINER emission. In order to explain the BPT offsets in our sample, LINER emission would have to account for ∼25% of the Hα flux in the central resolution element as found for Seyfert AGN. For a fiducial mix of 75% star formation (with [S ii]/(Hα+[N ii]) = 0.15) and 25% AGN (with [S ii]/(Hα+[N ii]) = 0.6), we would expect a total [S ii]/(Hα+[N ii]) ∼0.26 in the nucleus. This is clearly inconsistent with the majority of our sample, thus ruling out LINER emission as a dominant source even within the central resolution elements of our targets.

Zoom In Zoom Out Reset image size

Within Figure 10 there are four individual targets which show elevated [S ii]/(Hα+[N ii]) ≳ 0.3 in their nuclear regions: DEEP2 IDs 21021292, 22022835, 41059446, and 42045870. Separate data for these galaxies is presented in the Appendix along with the rest of the sample. In three cases the elevated line ratios are not statistically significant ( ≲ 1σ) suggesting that the ratios are likely due to noise fluctuations rather than AGN. While these nuclear line ratios are indicative of LINER AGN activity, all four galaxies in question have negligible offsets from the z ≃ 0 BPT star-forming locus: −0.001, 0.001, 0.016, and −0.03 dex, respectively (all consistent with zero offset).

We conclude that there is little evidence for LINER emission in the grism spectra, with only one object plausibly showing the expected LINER signature at > 1σ significance. In particular there is no sign of LINER activity among the galaxies exhibiting substantial offsets in the BPT diagram, despite sufficient angular resolution and sensitivity to detect the expected signal. Therefore, we can confidently rule out LINER-like line emission as the cause of BPT offsets in our sample.

5.2. Shocks and Diffuse Ionized Gas

To determine if shocks or DIG may contribute significantly to the emission properties of our sample, we next look for evidence of elevated [S ii]/(Hα+[N ii]) outside of the most luminous star-forming regions in galaxy outskirts and in regions of low Hα+[N ii] surface brightness. Although shocked gas and DIG emission show similar signatures in our diagnostic, they have different effects on different emission lines and therefore on BPT diagram offsets. Shocks can cause an elevated [N ii]/Hα ratio, for example as observed in a lensed z ≃ 1 arc by Yuan et al. (2012). DIG-dominated regions are not likely to cause offsets on the [N ii] BPT diagram, but have a significant effect on the [S ii] flux and corresponding BPT diagram (e.g., Sanders et al. 2017). We note that while DIG may not be responsible for causing [N ii] BPT offsets, we still seek to quantify the role of DIG emission at moderate redshifts in our spatially resolved sample.

Looking at the entire sample in the Appendix, the [S ii]/(Hα+[N ii]) ratio appears flat in the lowest Hα+[N ii] surface brightness regions, to within the statistical uncertainties. While there are a few stray bins with higher [S ii]/(Hα+[N ii]) (∼one bin per galaxy), the uncertainties with these data points are much higher and thus do not suggest significant contributions from shocks or DIG. The integrated data agree with the spatially resolved bins, showing offsets along the z ∼ 0 SF/AGN dividing line, with no clear signs of offset toward typical shock/DIG emission regions (see Figure 3). While there is one galaxy more substantially offset toward the DIG/LIER region in [S ii]/Hα, its [N ii]/Hα ratio does not show a significant BPT offset.

While we do not detect strong DIG or shocked gas emission in the individual galaxies, we look to the combined sample for bulk DIG emission across all targets. In Figure 11 we fit our observed aggregate sample [S ii]/(Hα+[N ii]) as a function of surface brightness (binning by Hα+[N ii] surface brightness for an increased S/N at the lowest surface brightnesses) with DIG fractions from Oey et al. (2007). We assume a functional form

$$\begin{eqnarray}&&{f}_{\mathrm{DIG}}=-1.50\times {10}^{-14}\times {{\rm{\Sigma }}}_{{\rm{H}}\alpha }^{1/3}+0.748\end{eqnarray} \tag{ 1 }$$

as found by Sanders et al. (2017), treating the line ratios in SF and DIG regions as free parameters. We expect higher [S ii]/(Hα+[N ii]) at lower surface brightnesses due to the increasing DIG fraction. Although the Sanders et al. (2017) fit parameterizes f_DIG as a function of Hα surface brightness, using the blended Hα+[N ii] in this analysis should not strongly affect our results. We have verified that accounting for the contribution of [N ii] surface brightness, with a plausible range of [N ii]/Hα ratios in DIG-dominated and star-forming regions, yields results consistent within our 1σ uncertainties.

Zoom In Zoom Out Reset image size

Indeed, with this formalism we find evidence for the bulk effect of weak DIG emission at ∼ 2σ significance (Figure 11), with best-fit parameters of [S ii]/(Hα+[N ii]) = 0.29 ± 0.09 for pure DIG emission and [S ii]/(Hα+[N ii]) = 0.11 ± 0.01 for pure star-forming regions. Using these best-fit values with the total [S ii] and Hα+[N ii] fluxes in all bins, we estimate the DIG fraction for our sample as a whole to be ∼22%. We note that the best-fit DIG line ratio of [S ii]/(Hα+[N ii]) for a pure DIG region is somewhat lower than expected based on nearby galaxies, although compatible given the uncertainty (Belfiore et al. 2016; Zhang et al. 2016; Sanders et al. 2017). If the true DIG line ratio is higher than given by this fit, then the DIG fraction of the sample would be lower than implied above (e.g., a typical DIG line ratio of 0.4–0.5 would imply only f_DIG ∼ 10% for the sample). Considering this wide range of plausible DIG line ratios values, our estimated DIG fractions remain consistent with Sanders et al. (2017), who suggest that DIG contributes ∼0%–20% of the total flux at these surface brightnesses. We conclude that while we do marginally detect weak DIG (at 2σ significance) or shocked gas emission in our sample, it is not a dominant source of emission in these targets and is not responsible for their offsets in the BPT diagram.

5.3. Escaping Ionizing Radiation

5.4. Star Formation

Analysis of our sample described in Section 5 strongly suggests that the dominant source of line emission in all cases is from H ii regions powered by star formation. Here we discuss the implied star formation properties of our sample and consider which physical properties of H ii regions might be responsible for offsets in the BPT diagram compared to z ∼ 0 galaxies.

The surface brightness of our sample spans Σ_Hα ∼ 0.5–5 ×10⁴¹ erg s⁻¹ kpc⁻² for well-detected regions, with total luminosities L_Hα ∼ 10⁴² erg s⁻¹ (uncorrected for extinction; two objects have significantly larger luminosities). The emission is extended over several kiloparsecs in all cases with at most ∼40% of the flux being found within a single resolution element. Accounting for ∼1–2 magnitudes of extinction (based on Balmer lines; Jones et al. 2015), this corresponds roughly to SFR densities Σ_SFR ∼ 1 − 10 M_☉yr⁻¹ kpc⁻² and total SFR ∼10–50 M_☉yr⁻¹ (assuming a Chabrier (2003) initial mass function). These values are characteristic of moderately massive (M_* ∼ 10⁹ M_☉) star-forming galaxies at z > 1 (Speagle et al. 2014; Shivaei et al. 2015).

Line ratios and positions in the BPT diagram are known to vary systematically with total emission-line luminosity and surface brightness. At the relatively low luminosities (or equivalently SFRs) typical of spatially resolved z ∼ 0 spiral galaxies, lower Σ_Hα correlates with offsets toward higher [N ii]/Hα, [S ii]/Hα, and [O iii]/Hβ which can be explained by increasing fractions of DIG (e.g., Oey et al. 2007; Masters et al. 2016; Zhang et al. 2016; Sanders et al. 2017; Shapley et al. 2019). In contrast, the global trend is opposite, where higher integrated L_Hα correlates with offsets toward higher [N ii]/Hα and/or [O iii]/Hβ as seen in the BPT diagram for both z ∼ 0 and z > 1 galaxies (e.g., Brinchmann et al. 2008; Cowie et al. 2016; Masters et al. 2016). Our sample is clearly characteristic of the latter case, having luminosities typical of the z > 1 galaxies analyzed by Cowie et al. (2016). While our sample is not large enough to establish an explicit relationship in BPT offset with luminosity, we can nonetheless explore whether the observed BPT offsets of the HST targets follow the same trends as found in larger populations. The luminosity of our sample (typical L_Hβ ∼ 10^41.5 erg s⁻¹) corresponds to offsets in [N ii]/Hα and [O iii]/Hβ of ∼0.1–0.2 dex according to Cowie et al. (2016). This is in good agreement with the median offsets solely in [N ii]/Hα or [O iii]/Hβ (0.19 and 0.7 dex, respectively) in our sample. Thus the BPT offsets in Table 1 are commensurate with trends in global properties seen at both low and high redshifts. As such, the physical mechanisms responsible for the offsets in this sample are likely to be the same as in broader galaxy populations.

Many other works have further explored the main driving mechanisms of the more extreme ISM conditions at z > 0. Such studies have invoked a variety of properties at high redshift such as H ii regions having higher ionization parameters than at z ∼ 0 (e.g., Kewley et al. 2013a, 2013b), an enhanced N/O abundance (e.g., Masters et al. 2014; Shapley et al. 2015), and high α/Fe abundance ratios (e.g., Steidel et al. 2016; Shapley et al. 2019; Sanders et al. 2020a, 2021). A key result of our spatially resolved analysis is that the emission line ratios appear nearly constant across individual galaxies. That is, offsets in the BPT diagram appear to be linked to physical conditions in the H ii regions. Given the limited number of emission lines in HST grism data, we cannot clearly distinguish between various mechanisms proposed in the literature on spatially resolved scales. However, the results of this work support treating the integrated emission-line fluxes of high-z galaxies as originating from H ii regions ionized by massive stars, for purposes of modeling and interpreting their spectra. As an example, the photoionization modeling framework from Sanders et al. (2021)—adopting the T_e-based O/H measurements for our sample (Jones et al. 2015)—indicates that the BPT offsets in our sample are caused by super-solar O/Fe abundances. In our future work, we will apply a rigorous photoionization modeling analysis of this sample using stellar population synthesis models as input (e.g., Binary Population and Spectral Synthesis (BPASS)).

Increased O/Fe abundances are generally expected for galaxies at high redshifts, given the shorter enrichment timescale for α-elements such as O compared to Fe. Recent work has suggested that O/Fe may typically reach ∼ 4 times the solar value at z ∼ 2 (e.g., Steidel et al. 2016; Jones et al. 2018; Sanders et al. 2021, 2020b). Such α enhancement results in harder stellar ionizing spectra at fixed O/H abundance, which drives offsets in the BPT diagram. While we cannot measure O/Fe directly in our sample, we expect it to correlate with young ages and hence high specific star formation rates (sSFR). Figure 12 explores this idea, using Hα EW as a proxy for sSFR (and hence possibly α enhancement). To correct for the dependence of Hα EW along the z ∼ 0 BPT locus, we first fit for a relationship between mean Hα EW of the z ∼ 0 sample and the parameter O3N2 ($=\mathrm{log}$([O iii]/Hb)$-\mathrm{log}$([N ii]/Hα)), using the rest-frame Hα EW. We note that lines of constant O3N2 are roughly orthogonal to the BPT star-forming sequence in the region of interest here, such that normalization at fixed O3N2 is appropriate for comparison with our BPT offset measurements. We find a linear relationship of

$$\begin{eqnarray}&&\mathrm{log}(H\alpha \,\mathrm{EW}[z\sim 0])=0.762+0.779\times {\rm{O}}3{\rm{N}}2\end{eqnarray} \tag{ 2 }$$

(with Hα EW measured in angstrom) to be a good fit for star-forming galaxies in the Sloan Digital Sky Survey (SDSS), spanning the O3N2 > 1.2 values of our sample.

Zoom In Zoom Out Reset image size

Figure 12 plots the normalized Hα EW (using the mean relationship in Equation (2)) at fixed O3N2, revealing a clear correlation between normalized Hα EW and offsets in the BPT diagram for the SDSS sample (as has been noted previously by, e.g., Brinchmann et al. 2008). Moreover, the higher redshift data—shown for both our HST sample at z ∼ 0.8 and MOSDEF galaxies at z ∼ 2.3 (Reddy et al. 2018; Sanders et al. 2018)—show increased BPT offsets at fixed EW. The median displacement above the SDSS best-fit line appears to increase with redshift (0.02 dex for the HST sample, 0.08 dex for the MOSDEF sample), although there is no clear correlation within the MOSDEF sample. This suggests a smooth evolution of increased BPT offset with increasing redshift at fixed Hα EW beyond the relation at z ∼ 0 (consistent with the expectation of increasing α enhancement at high redshifts compared to z ∼ 0).

While our analysis supports α-enhanced abundance patterns, we note that this is indistinguishable from the case where ISM metallicity is higher than that of ionizing stars, with solar abundance patterns. Recent ISM enrichment from massive stars may show significant deviations from the mean galaxy metallicity and may be localized to ≲ 1 kpc spatial scales (Krumholz & Ting 2018). While the stars responsible for ionizing H ii regions are generally thought to have similar composition as the ISM, our data do not provide direct measurements of the stellar abundance. We also note that Sanders et al. (2020b) suggest α enhancement even among z ∼ 2 galaxies which are not offset from the z ∼ 0 BPT locus, such that additional properties beyond abundance patterns must play a significant role.

We note that there are many other possibilities that may drive BPT offsets beyond abundances. In particular, effects of binary stellar evolution (leading to stripped stars or high-mass X-ray binaries), turbulence (e.g., Gray & Scannapieco 2017), and higher ionization parameters (e.g., Jaskot et al. 2019) may also drive high-redshift galaxies off of the z ∼ 0 star-forming locus. Nonetheless, our results are generally consistent with offsets driven by moderate α enhancement, if indeed the Hα EW correlates with young ages and α-enhanced abundance patterns. We conclude that photoionization modeling of H ii regions ionized by massive stars will be a reasonable tool for further interpreting these results since there are no significant signatures from other ionizing sources beyond star formation.

In this paper we present and analyze spatially resolved emission-line maps of 15 galaxies at z ∼ 0.8 in the DEEP2 fields, observed with HST grism spectroscopy. Our targets show moderate offsets from the z ∼ 0 locus of star-forming galaxies on the [O iii]/Hβ versus [N ii]/Hα BPT diagram, comparable to those seen in typical z ∼ 2 galaxies. While line emission in the sample appears to be dominated by star formation, we analyzed Hα+[N ii], [S ii], and [S iii] maps in order to distinguish signatures of AGN, shocks/DIG, or escaping ionizing radiation in our sample to determine whether these mechanisms may be responsible for the observed BPT diagram offsets. Our main results are as follows.

• 1.  
  AGN. We achieve ∼1 kpc spatial resolution to distinguish possible contributions from faint AGN. We find that AGN would need to account for a large fraction of emission (≳25% of Hα flux) within the central resolution element in order to explain the observed magnitude of BPT offsets. In the case of LINER AGN, we would expect significantly elevated [S ii]/(Hα+[N ii]) ratios in the nuclear regions, which are not observed. In particular the objects in our sample with the largest BPT offsets are inconsistent with significant emission from LINER AGN. For Seyfert AGN, the [S ii]/(Hα+[N ii]) ratio is not necessarily different from that of H ii regions, and so we cannot clearly distinguish their presence in the sample. However if Seyfert AGN emission is indeed widespread, it would be easily distinguishable in IFS surveys via high [N ii]/Hα ratios in galaxy nuclei, which are not commonly observed in galaxies with mass and metallicity comparable to this sample. We therefore conclude that substantial Seyfert AGN contributions are unlikely, and LINER AGN are clearly ruled out as causes of BPT offsets.
• 2.  
  Shocks and DIG. We examine how [S ii]/(Hα+[N ii]) varies spatially and as a function of surface brightness in order to distinguish contributions from shocks and DIG. There is no clear evidence of shock/DIG emission in individual galaxies, however we find some evidence of bulk DIG emission at 2σ significance across the entire sample. We estimate the DIG fraction in our sample to be ∼22%, in line with the expected fractions based on the resolved surface brightnesses of our sample, and thus conclude that DIG or shocked gas emission are not the primary emission sources in our targets.
• 3.  
  Escaping ionizing radiation. In principle the [S iii]/[S ii] ratio can identify locations of density-bounded H ii regions with escaping ionizing radiation. We demonstrate for one object that high ionizing escape fractions are unlikely, although in general the [S iii] signals are too weak to examine this possibility. Nonetheless future deep grism spectroscopy mapping the [S iii] lines, for example with JWST, holds promise for further study.
• 4.  
  Star formation. All targets in our sample exhibit spatially extended emission with high surface brightness and line ratios indicative of star-forming H ii regions. It is clear that star formation is responsible for the vast majority of emission-line flux in our targets. Having found that shocks, DIG, and LINER emission cannot explain the observed BPT offsets, we conclude that these line ratio offsets are most likely caused by different physical characteristics of H ii regions and/or young stars compared to typical star-forming galaxies at z ∼ 0. We consider trends in BPT offset with global properties of z ∼ 0 star-forming galaxies, and find that the z ∼ 0.8 sample follows similar relations. In particular the BPT offsets are consistent with expectations based on Balmer line luminosities and EWs. The BPT offsets thus appear to be connected to high SFR and sSFR, which are typical of high-z galaxies. Although not conclusive, the trend with sSFR supports recent work suggesting that α-enhanced abundance patterns cause BPT offsets at high redshift.

Overall, the spatially resolved data in our sample appear to be consistent with emission driven by star formation, despite their integrated line ratios being offset from the z ∼ 0 BPT star-forming locus. This highlights an important outstanding question in galaxy evolution studies: how do the physical properties of H ii regions at high redshifts differ from those observed locally? In future work we intend to perform a thorough photoionization modeling analysis of this sample with a full suite of emission lines to further probe this question and discern between the properties that may drive line ratio offsets, such as high ionization parameters, enhanced N/O abundance, or high α/Fe. A key result of this work is that, since emission is powered by star formation, models of photoionization by massive stars should be appropriate for interpreting the spectra of high-z galaxies with similar BPT offsets. Likewise, direct-method abundance measurements (e.g., Jones et al. 2015; Sanders et al. 2020a) and other nebular emission probes should indeed reflect the physical conditions of H ii regions.

Looking forward, we can soon extend this type of analysis to a wider range of redshifts and a larger sample size. JWST will provide increased S/N along with excellent spatial resolution through grisms, Integral Field Unit (IFU), or slit-stepping to distinguish between the properties of nuclear regions, high surface brightness regions, and galactic outskirts. This will allow further exploration of DIG emission and trends with surface brightness, and to use [S iii] as a diagnostic of density-bounded H ii regions with escaping ionizing radiation at high redshifts. Additionally, this analysis is not limited by spectra that cannot resolve Hα and [N ii]; as shown with the MaNGA data in Figure 1, we can clearly distinguish different physical mechanisms using the [S ii]/(Hα+[N ii]) metric without high spectral resolution. Space-based grism spectroscopy is thus a powerful technique that can additionally benefit from multiplexing. While we found no strong evidence of properties beyond star formation in our sample, a larger spatially resolved sample can provide more certainty for high-redshift galaxies as a broader population. However if high-redshift galaxy emission is truly dominated by star formation, integrated emission-line measurements are sufficient to discern between the various proposed properties within H ii regions. Further understanding of the origin of emission line excitation at high redshift will improve our ability to calibrate metallicity measurements between the low- and high-redshift universe, and propel our knowledge of galaxy evolution.

We thank the anonymous referee for a constructive report which substantially improved the content and clarity of this manuscript. This work is based on observations with the NASA/ESA Hubble Space Telescope obtained by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Incorporated, under NASA contract NAS5-26555. Support for Program number GO-15077 was provided through a grant from the STScI under NASA contract NAS5-26555. Support for R.L.S. was provided by NASA through the NASA Hubble Fellowship grant No. HST-HF2-51469 awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Incorporated, under NASA contract NAS5-26555. C.L.M. acknowledges support from the National Science Foundation under grant AST-1817125. M.C.C. acknowledges support from the National Science Foundation under grants AST-1815475 and AST-1518257.

Emission line maps and [S ii]/(Hα+[N ii]) line ratio analyses for the galaxy DEEP2-13043716 are shown in Figure 7, and interpretation of this example object is discussed in Section 4. Equivalent data are shown in Figure 13, and the complete figure set (15 images) is available online.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View figure set (15 panels)

Tarball of figure set images