Aperture and Resolution Effects on Ultraviolet Star-forming Properties: Insights from Local Galaxies and Implications for High-redshift Observations

We present an analysis of the effects of spectral resolution and aperture scale on derived galaxy properties using far-ultraviolet (FUV) spectra of local star-forming galaxies from the International Ultraviolet Explorer (R ∼ 250, field of view (FOV) ∼ 10″ × 20″) and Cosmic Origins Spectrograph on the Hubble Space Telescope (R ∼ 15,000, FOV ∼ 2.″5). Using these spectra, we measured FUV luminosities, spectral slopes, dust attenuation, and equivalent widths. We find that galaxies with one dominant stellar cluster have FUV properties that are independent of aperture size, while galaxies with multiple bright clusters are sensitive to the total light fraction captured by the aperture. Additionally, we find significant correlations between the strength of stellar and interstellar absorption lines and metallicity, indicating metallicity-dependent line-driven stellar winds and interstellar macroscopic gas flows shape stellar and interstellar spectral lines, respectively. The observed line strength versus metallicity relation of stellar-wind lines agrees with the prediction of population synthesis models for young starbursts. In particular, measurements of the strong stellar C iv λλ1548, 1550 line provide an opportunity to determine stellar abundances as a complement to gas-phase abundances. We provide a relation between the equivalent width of the C iv line and the oxygen abundance of the galaxy. We discuss this relation in terms of the stellar-wind properties of massive stars. As the driving lines in stellar winds are mostly ionized iron species, the C iv line may eventually offer a method to probe α-element-to-iron ratios in star-forming galaxies once consistent models with nonsolar abundance ratios are available. These results have important implications for the galaxy-scale, low-resolution observations of high-redshift galaxies from JWST (R ∼ 100–3500).


INTRODUCTION
The advent of space-based observations permitted the study of star-forming galaxies in the ultraviolet (UV) where newly formed stars reach the peak of their spectral energy distributions (SED).Pioneering studies of nearby starforming galaxies were done with the International Ultravio-Since the end of the IUE mission, the Hubble Space Telescope (HST) has been the dominant observatory for UV observations.HST's current spectrographs, the Space Telescope Imaging Spectrograph (STIS) and Cosmic Origins Spectrograph (COS), have greatly expanded and improved upon the legacy of the IUE in the 1150 -3200 Å wavelength range.The vast array of available entrance apertures, gratings, and improved sensitivity of these instruments have enabled transformative science not previously possible with IUE.Nevertheless, there is still scientific value in the IUE data set that has not yet been unlocked, even if the IUE's data quality cannot compete with that of HST in most aspects.Therefore, the goal of the present study is to compare IUE and HST spectroscopic data of nearby star-forming galaxies to gain insight into how the instrumental properties affect the interpretation of the data, as well as understanding the physical processes operating in these galaxies.This is of particular interest because of the limited lifetime of the HST.Exploring the capabilities of the IUE data will help identify the best use of future HST UV science in its limited lifetime, and inform interpretations of low-resolution rest-frame UV spectra of high-redshift galaxies taken with the James Webb Space Telescope (JWST).Kinney et al. (1993) published a comprehensive atlas of spectra of star-forming and active galaxies obtained with IUE while IUE was still accumulating new data and before the final data products were released in 1998.Heckman et al. (1998) analyzed the data set of Kinney et al. to establish trends of various measurements with galaxy parameters.The present work builds on and extends these past works.In particular, our study makes several improvements: (i) New data are included that were not available to Heckman et al. (ii) The final data release permits a homogeneous analysis of the data.(iii) Available COS high-resolution spectra allow us to resolve and remove Galactic foreground absorption, which was not feasible for Heckman et al. (iv) Synthetic galaxy spectra for quantitative analysis of the data have only become available after the work of Heckman et al., allowing a comparison with synthetic spectra in the present work.
Almost all galaxies originally targeted by IUE have also been observed with HST's spectrographs in much greater detail and with superior quality.In this study, we will focus on data collected with COS whose light-collecting power is particularly well-suited for z 0 extragalactic spectroscopy.The present work provides three major scientific advancements in the UV spectral analysis of nearby galaxies: (i) We constructed an atlas of nearby star-forming galaxies with both IUE and HST/COS spectra.Leitherer et al. (2011) previously published an atlas presenting spectra obtained with the HST first-generation UV spectrographs, the Faint Object Spectrograph (FOS), and Goddard High Resolution Spectrograph (GHRS), but no such atlas exists for the latest generation of spectrographs.(ii) We perform a comparative analysis between the IUE and COS spectra that allows us to differentiate galaxy properties on pc and kpc scales.IUE's large 10 ′′ × 20 ′′ entrance aperture provides the large scale (kpc-scale), integrated galaxy properties, while HST/COS' 2. ′′ 5 aperture probes smaller, pc-sized scales.(iii) The data quality of the IUE spectra in terms of signal-to-noise (S/N) and spectral resolution is inferior to that of HST spectra of local galaxies (e.g., Berg et al. 2022) but is comparable to that of spectra in the high-redshift universe obtained with JWST.Recent JWST spectra of z > 7 galaxies have already revealed the potential for deep galaxy evolution studies, but their interpretation requires comparison to detailed analysis that is only possible in local galaxies (e.g., Arellano-Córdova et al. 2022;Trump et al. 2023).The results of this paper, therefore, provide guidance for the planning and interpretation of observations of galaxies close to the era of reionization.
The remainder of the paper is organized as follows: We describe the archival UV and optical spectral observations used in this work in Section 2. The sample selection and its properties are in Section 2.1.The processing of the IUE and HST/COS data is described in Sections 2.2 and 2.3, respectively.A general comparison of the two data sets is performed in Section 2.4.The ground-based optical spectra and the determination of metallicities are discussed in Section 2.5.Section 3 covers our measurements.The determination of the β−slopes is in Section 3.1, followed by the reddening determinations in Section 3.2.Equivalent widths are derived in Section 3.3, with a comparison between IUE and COS in Section 3.4.In Section 4 we study the relation of the derived properties with oxygen abundances.This is done for both the COS and the IUE samples in Section 4.1 and 4.2, respectively.In Section 4.3 we interpret our results.Finally, our conclusions are presented in Section 5.

Sample definition and basic properties
The Mikulski Archive for Space Telescopes (MAST) hosts data from numerous space missions focusing on the UV, optical, and near infrared, including the products of the IUE and HST.We queried MAST for existing UV spectroscopic data from both the IUE and HST missions.As IUE is the much more restricted data set in terms of number and quality of spectra, we initiated our query with the IUE set, guided by the earlier works of Kinney et al. (1993) and Heckman et al. (1998).Next, we queried MAST for existing co-spatial HST/COS and HST/STIS spectra of all galaxies with IUE spectra in order to create a sample having both IUE and HST spectra.The science driver for this requirement is the goal to investigate line profiles in the HST data and compare spectral   2020), (18) Shirazi & Brinchmann (2012).* Metallicity derived using strong-line methods; see Section 2.5 for more details.
resolution effects between the IUE and HST data.The resulting 29 galaxies are hereafter referred to as the "IUE Sample".Of these 29 galaxies, four galaxies have COS M mode spectra, eight have STIS L mode spectra, and 17 have both COS and STIS spectra.COS M mode and STIS L mode have very different resolving power of R ∼ 15, 000 and R ∼ 1, 000, respectively.As our goal is to study the influence of spectral resolution and since R ∼ 1, 000 is insufficient for separating stellar and interstellar lines, we define a sample of galaxies having COS M mode spectra.As a result, we identify 21 galaxies with both IUE and co-spatial COS M mode spectra.
We refer to this sample as the "IUE-COS Sample".We keep the larger IUE Sample in order to maximize the statistics for all IUE-related analysis, whereas all comparisons between IUE and COS are based on the IUE-COS sample.
A comparison of the aperture field of views for IUE and COS observations is shown for the full sample in Figure 1.The IUE apertures have been drawn to indicate the field size; their orientations are arbitrary1 .In most cases, the data used in our study are the superposition of multiple spectra obtained at different orientations.Therefore the actual area covered can be thought of as a rotating IUE aperture.In almost all cases the COS pointings are near the center of the IUE aperture, and the orientation of the IUE aperture is unimportant.NGC 3690 is an exception: it is unclear whether the COS and IUE spectra are co-spatial.We decided to keep this galaxy in our sample to maximize statistics.We searched for archival images emphasizing UV wavelengths whenever available.The majority of the images in this figure were obtained with HST's Advanced Camera for Surveys (ACS), STIS, Wide-Field and Planetary Camera 2 (WFPC2), and Wide-Field Camera 3 (WFC3).Two galaxies (NGC 3991 and NGC 5996) have spherically aberrated HST Faint Object Camera (FOC) images.For UGCA 219 we used a Sloan Digital Sky Survey (SDSS) g-band image.The image of NGC 5457 was obtained with the Jacobus Kapteyn Telescope (JKT) in the U-band.
We compile the relevant basic properties for our galaxy sample from the literature and present them in Table 1.This table gives important properties, including the galaxy numbering we use throughout this paper, official galaxy names, the galaxy central coordinates, morphological types, and Galactic foreground extinctions.The Galactic foreground extinctions were obtained using the recalibrations of Schlafly & Finkbeiner (2011) and range between 0.01 and 0.11.The heliocentric radial velocities (v rad ), which fall between 200 and 5700 km s −1 , are relevant for removing and/or deblending Galactic foreground absorption lines from intrinsic spectral lines.We also performed a literature search in the NASA Extragalactic Database (NED) to obtain individual distance determinations for our galaxy sample.We prioritized highquality distances, D, based on the tip of the red giant branch (TRGB) for all galaxies with D < 13 Mpc.Galaxies at larger distances were assumed to be in the Hubble Flow and, thus, their distances were derived from v rad and the local velocity field model of Mould et al. (2000) using the terms for the influence of the Virgo Cluster, the Great Attractor, and the Shapley Super Cluster.

IUE Spectra
Here we present the IUE spectra for the galaxies listed in Table 1.We retrieved the merged, extracted MXLO spectral files for the IUE Sample; the properties of the observations are listed in Table 2.The MXLO files are the one-dimensional spectral tables extracted from the twodimensional SILO image files processed with the NEWSIPS pipeline (Garhart et al. 1997).The MXLO tables contain the absolutely calibrated fluxes, wavelengths, data quality flags, and Poisson error spectra.Spectra processed with NEWSIPS show an increase in S/N of 10% -50% in comparison with the previous IUESIPS processing (Nichols & Linsky 1996).The retrieved spectra were resampled to equal wavelength steps of 1.2 Å, i.e., five pixels within the nominal resolution of 6 Å.When multiple co-spatial exposures exist, the individual spectra were coadded using a weighting factor scaled by the square root of the exposure time, i.e., by the Poisson noise.We combined the SWP and LWP/R (when available) exposures for the full far-and near-UV wavelength range.The short-wavelength end of the LWR spectra between 1950 and 2200 Å have lower S/N due to the SiO 2 annihilation coating of the LWR optics (Figure 13 of Bohlin et al. 1980).This spectral region has noticeably lower S/N in the combined spectra.The spectra were then truncated at starting and ending wavelengths of 1150 Å and 3300 Å, respectively.In cases when only SWP data were available, the long-wavelength truncation was set to 1980 Å.

HST/COS Spectra
HST/COS spectra were retrieved from MAST for the 21 galaxies in the IUE-COS Sample, as indicated by checkmarks in Table 1.In particular, we retrieved the highest resolution data available, which comes from the mediumresolution gratings of G130M, G160M, and G185M, and considered all COS apertures that are co-spatial with the IUE aperture.The HST/COS datasets used in this work are  ) and consists of the integrated light within a large aperture (10 ′′ ×20 ′′ ).In contrast, the COS spectrum has significantly higher spectral resolution (λ/∆λ= 15, 000) and higher S/N, but only within the much smaller 2. ′′ 5 COS aperture.Both spectra are normalized at 1450 Å. summarized in Table 3.Note that NGC 5236, NGC 5253, and TOl 1924-416 have COS spectra from multiple pointings, not all of which are located within the IUE aperture.
The pointings outside the IUE aperture were used for studying spectral variations across the surface of the galaxies, but not for comparison with the IUE data.G130M spectra exist for all galaxies in the IUE-COS Sample, while 10 of these galaxies also have G160M spectra, and three galaxies have G160M+G185M data.All raw data were reduced with the CALCOS pipeline (v3.3.10) using the standard twozone extraction technique.However, each HST/COS grating has a different spectral resolution and different observations are taken at different position angles and lifetime positions, all of which must be accounted for when combining COS data.We address these issues by following the coaddition method laid out in Berg et al. (2022).In short, this method involves several steps of (1) joining the segments/stripes of individual grating datasets, (2) coadding any multiples of individual grating datasets, including all cenwave configurations, (3) coadding datasets across gratings, (4) binning the spectra by the nominal 6 pixels, and (5) correcting for Galactic contamination.While the initial flux calibration was performed for each dataset during the initial reduction by CALCOS, relative fluxing was also performed during the coadding process when more than one grating existed.To do so, the G160M spectrum was treated as the flux anchor of each spectrum and the continuum of all other datasets were fit and scaled to G160M at the intercept of their wavelength coverage (or average of the grating separation) when neighboring gratings overlaped (were disjoint) prior to coadding.Coadding steps used a combined normalized data quality weight (using the DQ_WGT array; to filter out or de-weight photons correlated with anomalies/bad data) and exposure-time-weighted calibration curve (to preserve the Poisson count statistics).This weighting method was used for all instances where coadding was performed.For further details of the coadding method, see Berg et al. (2022) and James et al. (2021).
Both the IUE and HST/COS spectra were corrected for Galactic foreground extinction using the values in Table 1 and the extinction law of Cardelli et al. (1989).Given the small values of E(B − V ) MW , the corrections are rather minor.Finally, the spectra were transformed into the rest-frame using the heliocentric radial velocities listed in Table 1.

Comparison of IUE and COS Spectra
The IUE-COS Sample contains galaxies with both IUE and COS observations, allowing visual comparison of the datasets.We find the COS apertures focus on the starforming knots in each galaxy, while the IUE probes the wider galactic environment.In more than half of our sample the IUE aperture fully covers the entire galaxy, including diffuse gas.
As an example, the HST/COS and IUE spectra for one galaxy in our sample, NGC 7714, are plotted in Figure 2, showing substantial, but expected differences.The complete figure set showing the spectra of all 21 galaxies with IUE and HST/COS spectra is available in the online journal.The smaller aperture of COS relative to the IUE 10 ′′ ×20 ′′ aperture records lower UV continuum fluxes.To account for this, we have normalized both spectra at 1450 Å.Now in direct comparison, the higher resolution of the HST/COS spectrum (∼ 0.1 Å resolution compared to the ∼ 6 Å of the IUE spectrum) reveals a significant number of features that are not present in the IUE spectrum, and also more complex line profiles.Specifically, the N V λλ1238,1242, C IV λλ1548,1550, and Si IV λλ1393,1402 lines are shown in the inset windows of Figure 2 to have combination profiles of stellar P-Cygni absorption+emission and extended ISM absorption.These high-resolution profiles allow the stellar continuum to be fit and removed so that uncontaminated ISM absorption can be studied.
On the other hand, the high-resolution data permit studies of the stellar-wind profiles after the removal of the interstellar contribution.The comparison in Figure 2 demonstrates the need for sufficiently high spectral resolution when interpreting stellar and interstellar features.The stellar P Cygni wind lines have a more or less significant interstellar contribution that must be accounted for when modeling the stellar population.Both Milky Way foreground and intrinsic lines may be important, depending on the ion.Lower ionization levels, such as Si IV have a stronger, or even dominant interstellar contribution, whereas higher levels, such as N V or C IV are mostly shaped by stellar winds.

Ancillary Optical Spectra
Optical spectra of star-forming galaxies can be used to derive the physical conditions of the nebular gas, the total chemical abundances, and current conditions such as star formation rate (SFR).In this work, we are particularly concerned with how UV spectral properties change as a function of gas-phase metallicity, where the metallicity (or oxygen abundance) is used to trace the metal enrichment of the ionized gas as a proxy for galaxy evolution in star-forming regions.Accurate oxygen abundances are derived via the socalled the direct method or T e -method, which requires the detection of an inherently-faint auroral line (e.g., [O III] λ4363, [N II] λ5755, or [S III] λ6312).We searched the literature for oxygen abundances derived via the direct-method (or T esensitive method) for each galaxy.The compilation of the metallicities for the IUE Sample is included in Table 1 and will be used for our interpretation of the observed UV spectra.In summary, 23 of the 29 galaxies in the IUE Sample have published direct-metallicity determinations.For two galaxies, NGC 3390 and NGC 3991, we used archival optical spectra from the SDSS to derive the metallicity.This provides the opportunity to improve the measurements of the metallicity of these galaxies previously derived from strongline methods (Heckman et al. 1998).The SDSS spectra of NGC 3690 and NGC 3991 show the presence of [N II] λ5755 and [S III] λ6312, respectively, allowing the computation of the electron temperature.Therefore, we have determined the physical conditions and metallicity following the procedure described in Arellano-Córdova et al. (2022), which includes the correction of Galactic extinction, the subtraction of the underlying population, the emission line fitting, and the reddening correction using the Cardelli et al. (1989) reddening law.For a summary of this procedure, we have fitted the emission lines with Gaussian profiles using the Python package LMFIT2 .We constrained the offset from line centers and the line width.To calculate the flux error, we used the expression reported in Berg et al. (2013) and Rogers et al. (2021).Finally, to derive the electron density and temperature, and metallicity, we use the PyNeb package (version 1.1.14)(Luridiana et al. 2015) with the atomic data also reported in Arellano-Córdova et al. (2022).For the six remaining galaxies NGC 3049, NGC 3256, NGC 3310, NGC 3351, NGC 5996 and NGC 7552 we need to rely on alternative methods to derive the metallicity using the emission spectra reported in the literature.For three of those galaxies the metallicity was determined using the calibration of Dopita et al. (2016).Three galaxies have COS spectra available (NGC 3049,NGC 3256,and NGC 7552).This calibration is based on photoionization models and uses a linear fit between [N II] λ6584 and [S II] λλ6717, 6731 to estimate the metallicity.
For NGC 3310, NGC 3351, and NGC 5996, we report metallicities derived using the C-method of Pilyugin et al. (2012) Determining a characteristic metallicity of massive galaxies is challenging.However, Moustakas & Kennicutt (2006) showed that the metallicities inferred from integrated spectra of disk galaxies correlates well with the characteristic gasphase abundance, as determined by the H II region abundance measured at 0.4R 253 .Moreover, 86% of our sample with COS and IUE spectra are dwarf galaxies with a relatively homogeneous spatial distribution of metals within 1 kpc scales (Annibali & Tosi 2022).Thus, we assume the integrated abundances adopted here are representative of our galaxy sample and allow for a safe comparison with other galaxies properties.

UV SPECTRAL MEASUREMENTS
Now that we have established the different physical scales and spectra resolution probed by the IUE and HST/COS spectra, we can begin to investigate their effects on properties measured from the UV spectra.Below we describe our uniform measurements of the UV β−slope and absorption feature equivalent widths (EWs).

Beta-Slope Measurements
An important property in characterizing UV spectra is the slope of the FUV stellar continuum, in a given wavelength interval, otherwise referred to as the β−slope.The β−slope is only weakly sensitive to the stellar properties of a young population, whose spectral energy distribution is in the Raleigh-Jeans regime at these wavelengths.In contrast, dust attenuation strongly affects the UV continuum.Therefore, the β−slope is commonly used to derive the dust reddening.Ad-ditionally, the β−slope has also been theoretically predicted to correlate with the escape of ionizing continuum photons (Zackrisson et al. 2013), which was recently observationally confirmed by Chisholm et al. (2022).
Typically, the β−slope is measured over a sufficiently wide UV wavelength range, centered at a wavelength around 1500 Å.Assuming a power-law model fit to the continuum such that F λ ∝ λ β , we measure the β−slope using a least-squared first-degree polynomial fit to log-wave versus log-flux data.Specifically, we use the featureless continuum-windows recommended by Calzetti et al. (1994) of 1268-1274, 1309-1316, 1342-1371, 1407-1420, 1563-1583, 1677-1740, 1760-1833, 1866-1890, 1930-1950, and 2400-2580 Å to mask out undesirable portions of the spectra.Given that most of our COS spectra only have G130M coverage, most of the COS spectra are fit with the first four windows only.Note that it is important to exclude continuum blueward of 1250 Å in the β−slope fit because this wavelength regime can probe the peak flux (and thus turnover) of massive stars in the FUV.On the other hand, wavelengths longward of ∼ 1800 Å should also be excluded for metal-rich galaxies where there can be significant contributions from the broad 2200 Å dust feature.Longward of 2500 Å additional emission from older, less massive stars may contribute, and the β−slope may become sensitive to the details of the starformation history.Therefore the wavelength range between 1250 Å and 1850 Å is the "sweet spot" for determining the β−slope.We use a bootstrap Monte Carlo method with 3000 iterations of adding a normal distribution of the error fluctuations to the observed data.The final β−slope fits to the COS and IUE spectra are included in the plots shown in Figure 2 (see online version for the complete figure set).
The derived β-slope values for both the IUE and COS spectra are listed in Table 4.We test the reliability of the IUE measurements by comparing our measurements with results published in the literature.Kinney et al. (1993), Calzetti et al. (1994), andHeckman et al. (1998) measured β in samples with significant overlap with our sample.We calculated the mean differences between the β-slope values measured by them and by us for the common galaxies and found ∆β (this work)-(literature) = −0.213± 0.096, 0.108 ± 0.059, and 0.098 ± 0.249 for the differences with Kinney, Calzetti, and Heckman, respectively.These small differences suggest that our measurements are consistent with the previous studies and are, therefore, robust.The resulting β−slopes range from −2.6 to +0.5, which is consistent with our sample containing young stellar populations with dust attenuation.
We compare the IUE and COS β−slopes in the left panel of Figure 3.In general, the two samples are consistent within the errors of the measurements, with a standard deviation of σ = 0.43.We find that β−slopes measured from the bluest windows only are nearly identical across most of the sample.
This suggests that we can measure robust β−slopes, regardless of the aperture size, when considering only the youngest stellar populations (which are best probed by the bluer FUV wavelengths).This is because the COS apertures are centered on the young clusters, and the full light of these clusters are easily contained within the IUE aperture.
There are a few strong outliers from the β IUE vs β COS trend in Figure 3, namely (5) IRAS 08339+6517, (8) NGC 3125, (15) NGC 3690, (18) NGC 4214, and (23) NGC 5236.The COS spectra for both (5) IRAS 08339+651 and (15) NGC 3690 show bluer slopes.Because IRAS 08339+651 is the most distant galaxy in our sample, the IUE aperture captures the extended light profile of the galaxy, while the COS aperture contains the bright center.Thus, the physical regions covered by IUE and COS are among the most divergent.As for NGC 3690, the COS pointing is only in one of the star-forming knots outside of the other bright regions in the center of this galaxy.This offset of the COS aperture towards the edge of NGC 3690 may explain the different β-slope we obtain for this galaxy.The coadded COS spectrum of ( 23) NGC 5236 includes four individual pointings that probe significantly different β-slopes (∆β ≈ 2.1).On the other hand, the COS spectra for (8) NGC 3125 and (18) NGC 4214 seem to have redder slopes than their corresponding IUE spectra.The COS and IUE spectra for NGC 4214 are somewhat different visually but agree within the uncertainties of the two spectra.However, the COS and IUE spectra for NGC 3125 differ by a larger amount.Given that we do not know the true orientation of the IUE aperture, it seems likely that the IUE aperture for NGC 3125 shown in Figure 1 was actually rotated ∼ 45 deg clockwise such that it captured multiple young star-forming clusters and resulting in a bluer integrated continuum slope.
In the right panel of Figure 3 we show a comparison of the average flux at 1500 Å, as determined from our β-slopes using y = β × x + α.The data for both COS and IUE are listed in Table 4.Despite the general agreement in β−slopes, the relative fluxes between the COS and IUE spectra are significantly biased to higher values in the IUE spectra.This trend is expected given that the area encompassed by the IUE aperture is 40× larger than that of COS, and so collects much more light from these extended galaxies.This skew is consistent with our assertion that the IUE spectra are capturing most of the light from the galaxies in our sample.The points that deviate the most from the 1:1 line also show the largest differences between IUE and COS β-slope measurements (∆β).This suggests a trend of increasing ∆β for larger differences of F λ1500 .
In Figure 4 we plot the β−slopes versus F λ1500 values for the galaxies with multiple COS aperture measures (I Zw 18, NGC 5236, NGC 5253, and TOL 1924-416) to investigate how this trend varies between stellar clusters of a given galaxy.We plot the individual measurements as semitransparent points and the measurements from the coadded spectra as solid points.Overall we see that the individual β−slopes show a large range (up to ∆β ≈ 2), and so fall far off the 1:1 relation with the IUE measurements.On the other hand, the measurements from the coadded spectra are in much closer agreement with the IUE measurements for all four galaxies.This aligns with our previous analysis of the β-slopes and highlights that the IUE slopes are different due to the separate stellar clusters going into the integrated light.In the right hand panel of Figure 4 we plot the continuum flux at 1500 Å for the four galaxies with multiple HST/COS apertures and this again skews towards the high IUE values, as expected.

Stellar Reddening
As discussed above, the slope of the UV continuum is primarily sensitive to dust reddening.Therefore, we use β-slope to derive the dust reddening for both the COS and IUE spectra and evaluate any differences between them.We determine the stellar reddening with two methods: (1) using empirical relationships between β-slopes and E(B − V ) and (2) using stellar population synthesis (SPS) models to fit the continuum.We use two different empirical relationships.First, we use the fit derived by Reddy et al. (2018): and then use the BPASS continuous star formation models with the β-slope values derived in Section 3.1.We also use the relationship derived by Chisholm et al. (2022) using a linear combination of single-burst Starburst99 models Note that both relations were originally derived for galaxies at low and high redshift using data from COS, but it is important to note that in Equation 2 the galaxies used we further away so the apertures contained all the light from the galaxy.Therefore, the assumed attenuation laws used in these works are applicable when both dust absorption and scattering are important.This may not always be the case for very nearby galaxies that fill the 2.5 ′′ COS aperture.However, for the sake of consistency, we opt against switching between different attenuation laws for different galaxies.
For the second method, we split our sample into two groups: metal-poor galaxies with 12+log(O/H) < 8.40 and metal-rich galaxies with 12+log(O/H) ≥ 8.40.We then create a grid of SPS models, adopting a pair of Starburst99 models (Leitherer et al. 2014), one metal-rich (Z = 0.014 or Z ⊙ ) and one metal-poor (Z = 0.002 or 0.14Z ⊙ ), and applying the reddening curve of Calzetti et al. (2000) with a range of E(B − V ) values between 0 and 1.We resample  the Starburst99 models to match the dispersion of the IUE and COS spectra.Once this is done, we perform a χ 2 minimization between the model and observed spectra, using windows that only contain continuum, and adopt the E(B−V ) value that produced the smallest value.
In Table 4 we summarize the results found from the continuum analysis of the IUE and COS spectra.The values in this table are derived by using Eq.(1).A comparison of the resulting E(B − V ) values for the COS spectra from the empirical β−slope method (column 7 of Table 4) and the SPS continuum-fitting method is shown in the top panel of  4. We list only one set of measurements because there is very little difference between the sets of values.The trends in this figure indicate that all three of the E(B − V ) determinations used in this work are consistent.We find a strong 1:1 correlation between the β-slope and model E(B−V ), suggesting the two methods are equivalent and relatively insensitive to the model assumptions of each method.
The bottom panel of Figure 5 shows the comparison of the E(B − V ) values derived from SPS continuum-fitting for the COS and IUE spectra.The overall trend is still in 1:1 agreement, but with a significantly larger scatter of σ = 0.116.This again reflects the effects of the different aperture sizes collecting light from different galaxy areas and, thus, the spectra having different shapes.Figure 5 (bottom) mirrors the trend in Figure 3 (left).If the dispersion is in fact due to physical differences in the spatial distribution of the dust and/or stellar populations observed, then the dispersion informs the potential uncertainty in E(B−V ) values derived with different apertures.

Equivalent Width Measurements
In this section, we test how the wavelength resolution and aperture size differences between the IUE and COS spectra affect our ability to characterize the strengths of ISM and stellar spectral features.To do so, we measure EW values of 11 different spectral lines (when available): S II λ1253, Si II λ1260, O I λ1302, C II λ1335, Si IV λλ1393,1402, C IV λλ1548,1550, Fe II λλ1608,1611, and Al II λ1671.Since the IUE spectra have lower spectral resolution than those of COS, it is often not possible to disentangle absorption lines that are close together, including contamination from Milky Way lines in the lowest-redshift galaxies.We correct the IUE absorption line measurements for Milky Way contamination using a hybrid approach.First, we measure the EW of the Milky Way foreground lines in the COS spectra (where they are sufficiently separated) and then subtract them from the corresponding IUE measurement.This hybrid approach has the distinct advantage of being able to improve our measurements from low-resolution spectra.We correct for this The trend is consistent with a 1:1 relationship but shows some scatter between the two apertures.
by measuring the EW of the Milky Way foreground lines in the COS spectra and then subtracting that value in the corresponding IUE measurement.This is an advantage of the hybrid approach, with high-and low-resolution spectra that we are using in this work.In measuring the EWs, it is first necessary to normalize the continuum.Given the complex nature of these spectra, we choose to do local normalizations around each line of interest rather than attempt a global normalization.Specifically, we carefully select windows of the continuum on both sides of a given absorption line and characterize it with a least-squares linear fit.An example of this fit and the subsequently normalized continuum of the IUE spectrum of NGC 7714 are shown in Figure 6.
In order to perform an appropriate comparison with IUE and take advantage of the superior resolution of COS, we generate two sets of EW measurements: 1. Broad Sample: measurements of both COS and IUE absorption features using broad integration windows appropriate for the low-resolution IUE spectra.2. Narrow Sample: measurements of the COS absorption features only using narrow integration windows customized to the individual lines in each spectrum.
For the Broad Sample, we use broad integration windows designed to capture the full extent of the line wings at IUE resolution.We list the line complexes of interest below, with their line centers and integration window widths: For the Narrow Sample, we defined the limits of integration as the point where the absorption line returned back to the normalized continuum (F λ = 1).
The Broad Sample EWs in the COS and IUE spectra are measured using a straight integration technique in the Interactive Data Language (IDL) software.Broad Sample errors ∆EW are estimated using the method from Stetson & Pancino (2008), which is based on Cayrel (1988): where δλ is the spectral resolution and S/N is the average S/N over the whole spectrum, obtained from performing the β-slope fit in Section 3.1.
As an initial test, we compared the IUE to the COS data smoothed to the same lower spectral resolution of IUE in order to determine the effects of aperture size for those 21 galaxies with both IUE and COS spectra.Then we compared the original high-resolution COS spectra to the smoothed COS spectra of these galaxies in order to determine the effects of spectral resolution.While we find significant scatter between the measurements for individual galaxies, there is no systematic trend.As an example, we give the measurements for the detected lines in the spectra of NGC 7714.The quoted values are EWs in Å for IUE, COS original, and COS smoothed to the resolution of IUE, respectively: S II λ1253+Si II λ1260: (2.23 ± 0.40, 2.53 ± 0.01, 2.32 ± 0.01), O I λ1302+Si II λ1304: (2.70 ± 0.44, 3.75 ± 0.01, 3.18 ± 0.01), C II λ1335: (2.83 ± 0.54, 3.18 ± 0.01, 3.22 ± 0.01), Si IV λλ1393,1402: (6.96 ± 0.70, 6.00 ± 0.01, 5.89 ± 0.01), and C IV λλ1548,1550: (6.91 ± 0.70, 11.3 ± 0.10, 9.45 ± 0.01).As we find no benefit in utilizing the smoothed COS spectra, we proceed with the analysis of the original highresolution COS spectra for comparison with the IUE EWs.The Broad Sample EWs for both IUE and COS thus obtained are reported in Table 5.
EWs are measured for the Narrow Sample using a Bootstrap Monte Carlo simulation with 3000 iterations.For each iteration, a new spectrum is generated, drawn from the normal distribution of values with a center and width corresponding to the flux and 1-σ uncertainty, respectively, at each wavelength.The EW of each iteration is measured using the Numpy.trapzfunction in PYTHON, which integrates along the wavelength axis using the composite trapezoidal rule.The final EW value and uncertainty is taken as the av-erage and standard deviation of the resulting distribution calculated, respectively.The Narrow Sample EWs for COS are reported in Table 6.

1607±20 1670±15
NOTE-Comparison of the FUV EWs measured from the small-aperture (2.′′ 5) COS data and the large-aperture (10 ′′ × 20 ′′ ) IUE data.Given the low spectral-resolution of the IUE spectra (∼ 6 Å), we use the broad integration windows listed in the bottom row for both the COS and IUE spectra.The limits set in place in the table are as follows: if the error was less than 0.01 we made the error measurement 0.01, if the flux was greater than 3 σ, we kept the original measurement and flux, and if the flux was greater than only 2σ, we set this as a lower limit.

EWs of High-vs Low-Ionization Lines
Comparing the velocity structure and equivalent widths of low-and high-stellar ionization lines provide important diagnostics of the physical gas conditions, such as the ionization structure and relative gas flows.In this sense, ions with similar ionization potentials are expected to be entrained in the same gas and so their absorption profiles should scale together (see, e.g., Chisholm et al. 2016).Additionally, lowand high-ionization lines have been observed to trace one another kinematically (e.g., Chisholm et al. 2016), but do not necessarily have to.However, the interpretation of ISM and stellar absorption features can be impacted by low spectral resolution that washes out and blends fine details.
We, therefore, now turn our attention to the differences between the EW measurements of ISM absorption features in the IUE and COS spectra.In order to provide a consistent comparison between galaxy-scale and cluster-scale measurements, we use the Broad Sample measurements in this analysis.In these broad integration windows, most of the measured absorption features are blended line complexes.Specifically, we investigate blends of S II λ1253+Si II λ1260, O I λ1302+Si II λ1304, Si IV λλ1393,1402, C IV λλ1548,1550, and Fe II λλ1608,1611.Additionally, the high-ionization Si IV and C IV lines are combinations of ISM and stellar-wind features.
In the top panel of Figure 7 we analyze the galaxy-scale versus cluster-scale EW measurements of low-ionization lines.In general, we find relatively good agreement be- 0.66 ± 0.01 2.25 ± 0.01 4.45 ± 0.02 3.04 ± 0.01 0.55 ± 0.01 1.36 ± 0.01 4.55 ± 0.01 Fe II λ1608.450.30 ± 0.02 NOTE-It is important to note that some lines could not be measured separately, particularly for O I+Si II λλ1302,04.In the case of blended lines both measurements will have around the same value as we are measuring approximately the same EW.The limits set in place in the table are as follows: if the error was less than 0.01 we made the error measurement 0.01, if the flux was greater than 3 σ, we kept the original measurement and flux, and if the flux was greater than only 2 σ, we set this as a lower limit.
tween both the IUE and COS EWs, with an average scatter of σ = 0.23 Å.However, the individual complexes show a range of trends, with Fe II λλ1608,1611 showing the largest dispersion and values generally skewed to larger IUE values.Interestingly, the high-EW end of the C II λ1335 and O I λ1302+Si II λ1304 trends are generally skewed to larger COS EWs relative to the IUE values.On the other hand, both Fe II λλ1608,1611 and O I λ1302+Si II λ1304 EWs are skewed toward higher IUE values at the low EW end of the trend 4 .This could result from poorer detections of faint lowionization lines in IUE spectra, where the lower spectral resolution tends to broaden and wash out weak absorption features.
4 There are only four measurements in our sample for the Fe II λλ1608,1611 lines so this skew towards the IUE at low EWs most likely comes from the lack of COS spectral coverage for some of our galaxies at this wavelength.
In the bottom panel of Figure 7 we plot the high-ionization ions, Si IV λλ1393,1402 and C IV λλ1548,1550, measured from COS and IUE.We find an increased scatter for the highionization trends relative to the low-ionization trends, with an average scatter of σ = 0.38 Å.Similar to the trends observed for some of the low-ionization ions, we find that the Si IV trend deviates from the 1:1 line with a flatter slope.However, while we expect the high-ionization ions should trace the same gas, the C IV profile is markedly different from the Si IV trend.Our C IV EW measurements only sample values greater than ∼ 4Å, so we are not able to access the low-EW trend, but see large scatter about the 1:1 line at the high-EW end.This may be due to the strong P-Cygni stellar-wind features observed in many of the COS spectra.As a result, these complex profiles are smeared out by IUE, reducing the integrated absorption profile, and skewing the trend towards larger COS values.In our sample of galaxies with C IV we see one galaxy that stands out with an uncharacteristic EW Overall, we find that the high-ionization ions, with their more complex line profiles, have greater dispersion between  the IUE and COS measurements than the low-ionization states, which emphasizes the importance of high spectral resolution for robust EW measurements.We also find a generally flatter than 1:1 trend present for both high-and lowionization ions that divides the skew towards higher IUE or COS EW values around 4 Å.However, the statistical significance of this trend is not high enough to permit firm conclusions.

METALLICITY DEPENDENCE
In this section, we investigate the behavior of the measured EWs with metallicity.We study the Narrow Sample and the Broad Sample separately.The Narrow Sample takes advantage of the superior COS resolution and S/N, which allows EW (Å) 12+log(O/H) Figure 8. Equivalent width measurements versus gas-phase metallicity for the Narrow Sample (see Section 3.3 for description).Equivalent widths were determined using integration windows appropriate to each line in a given COS spectrum.The solid blue lines are our best linear fits to the observed distributions.In comparison, our fit for C II in the middle left panel is in good agreement with the trend of Faisst et al. (2016) (solid red line), demonstrating the increasing trend between absorption line EW and gas-phase metallicity.
one to correct for line blending and contamination.However, this sample is unsuitable for a direct comparison with EWs measured in the low-resolution IUE spectra.Therefore, we investigate the metallicity dependence of the EWs measured in the IUE spectra in conjunction with the Broad Sample COS measurements.

The Narrow Sample
In Figure 8 we analyze the trend between the gas-phase metallicity and EW for each ion.The values for 12+log(O/H) are from Table 1 (see Section 2.5).In order to isolate trends for individual ion features, we use the Narrow Sample, which allows us to minimize the contamination by neighboring lines, as well as by Milky Way and geocoronal features.For the case of the O I+Si II λλ1302,04 line complex, the lines are blended and cannot be measured separately.For such blended lines, we use a single integrated EW measurement of the blended profile.For ions with isolated multiplet lines, we plot the line with the strongest oscillator strength (strongest absorber) in Figure 8.Additionally, the Si IV λλ1393,1402 and C IV λλ1548,50 lines show strong stellar P-Cygni profiles in many galaxies and so cannot be deblended.Therefore we opt against considering these two line complexes in the Narrow Sample, and defer a study of their properties to the discussion of the Broad Sample.
For each trend in Figure 8, we fit the individual relations with a first-order polynomial using the NumPy.polyfitfunction in Python.The best fits are shown as solid blue lines and the resulting polynomial coefficient values are given in Table 7.We perform the fits for all spectral lines considered.While we find correlations for some lines (e.g., C II λ1335), none are found for others, such as Fe II λ1608, which we attribute to the small number of data points.In general, we find EW increases with metallicity, as expected from the increase in the respective elemental abundances.

EW (Å)
12+log(O/H) Figure 9. Broad Sample equivalent width measurements of low-ionization species from both the COS and IUE spectra versus gas-phase metallicity.Equivalent widths were obtained using broad integration windows (see Section 4.2).The dashed lines are linear fits to the observed distributions.For comparison, the solid blue lines are the best fits to the corresponding Narrow Sample EWs shown in Figure 8.The increasing trend between EW and metallicity is seen for both the Narrow and Wide samples, but with vertical offsets due to the effects of the assumed integration windows.We note that the C II λλ1335 trend is in good agreement with that from Faisst et al. (2016) (red line).

EW (Å) 12+log(O/H)
Figure 10.Broad Sample equivalent width measurements of high-ionization species from both the COS and IUE spectra versus gas-phase metallicity.Equivalent widths were obtained using broad integration windows (see Section 4.2).The dashed lines are linear fits to the observed distributions.For comparison, the solid blue lines are the best fits to the corresponding Narrow Sample EWs shown in Figure 8.We also compare to the trends from Faisst et al. (2016) (red lines), with generally good agreement.Theoretical expectations from Starburst99 stellar-wind models are shown (dotted lines), revealing a stronger ISM contribution relative to the stellar contribution for Si IV than C IV. Interestingly, some of the small C IV EWs may indicate gas-phase metallicities that are enhanced relative to the stars.
In order to test our EW versus nebular metallicity trends, we compare them to the results from Faisst et al. (2016).Specifically, we prioritize the comparison of C II λ1335 because it is an isolated line and less likely to be saturated.We plot the Faisst et al. (2016) derived relation as a solid red line in the middle right panel of Figure 8.In general, our fit to the COS data is consistent with their trend but extends to higher EWs at high metallicities.This good agreement gives confidence to our other fits measured without literature values to compare to.

Low-Ionization Species
In Figure 9 we repeat our analysis of the EWs of lowionization species versus gas-phase metallicity for the Broad Sample measurements for both the COS and IUE spectra.
The IUE dataset has more data points for two reasons: (i) There are 29 versus 21 galaxies and (ii) the wavelength coverage is larger.Several galaxies lack COS G160M spectra where Fe II λλ1608,1611 and Al II λ1671 are located; this results in significantly stronger trends for the IUE dataset compared to that of the COS dataset.While the data in Figure 9 include blends of multiple absorption features, the Broad Sample also allows us to examine how larger integration windows and lower spectral resolution affect the measured EWs for the same instrument aperture.As we did for the Narrow Sample in Figure 8, we fit a first-order polynomial to each dataset, and over plot the fits derived for the Narrow Sample.All five spectral features considered show a positive correlation with 12+log(O/H).The fit coefficients are listed in Table 7. Overall, the Narrow Sample and Broad Sample COS measurements display the same trends with 12+log(O/H), but significant discrepancies are seen between the COS and IUE Broad Sample Fits.
The top two left column plots of Figure 9 show our most significant trends: S II λ1253+Si II λ1260 (left top) and C II λ1335 (left middle).The Broad Sample is skewed towards higher EW values in both cases.In the case of S II λ1253+Si II λ1260 this trend can be understood in terms of the integration window, which includes both lines in the Broad Sample measurements but only Si II λ1260 in the Narrow Sample measurements.For the C II λ1335 trend in the middle right panel of Figure 9, the Broad Sample fit (dashed line) is offset to larger EWs than the Narrow Sample fit (solid blue line).This offset is likely due to the inclusion of the C II * fine-structure line in the Broad Sample integration window, but not in the Narrow Sample window.There is also significant Milky Way contamination of C II λ1335 in the Broad Sample EW measurements that contribute to this high offset.We note that Milky Way contamination was particularly difficult to remove from C II λ1335 due to the blended nature of this line in the IUE Spectra.Therefore, we only remove Milky Way contamination for galaxies where we can disentangle the Milky Way component in the low-resolution IUE Spectra.

High-Ionization Species
The high-ionization counterpart to Figure 9 is shown in Figure 10, where strong correlations are seen for Si IV λλ1393,1402 and C IV λλ1548,50.The latter doublet is stellar-wind dominated, whereas the former has contributions from both stellar-wind and interstellar lines.We fit a firstorder polynomial to both the COS and IUE datasets for both ions.Both features were also studied by Faisst et al. (2016), whose 2nd-order polynomial relations (red solid lines) agree rather well with our best fits.
To examine the Si IV and C IV trends further, we investigate the theoretical stellar Si IV λλ1393,1402 and C IV λλ1548,1550 profiles as a function of stellar metallicity using synthetic UV spectra from the Starburst99 code (Leitherer et al. 2014).We adopt the library of theoretical spectra derived from WM-Basic model atmospheres (Leitherer et al. 2010) and use the same integration windows as in the corresponding observed spectra to measure EWs.The Geneva 1992-94 evolutionary tracks with high mass loss cover a metallicity range of 7.6 < 12+log(O/H) < 9.2 (Meynet et al. 1994).This range is consistent with the relevant metallicity range of our sample, where the lowest metallicity galaxies do not have significant absorption features.We assume a standard young population forming constantly over 20 Myr with a standard Kroupa initial mass function (IMF) and power-law exponents of 1.3 and 2.3, producing mass boundaries of 0.1 M ⊙ , 0.5 M ⊙ , and 100 M ⊙ , respectively (Kroupa 2008).In order to investigate the impact of IMF variations, we modify the high-mass exponent to 1.3 and 3.3 for an IMF more or less skewed towards massive stars, respectively.Varying the IMF has little effect on the predicted relation.We also test the influence of using different evolutionary tracks available in Starburst99 and found no significant change.
The predicted stellar model is shown in Figure 10 as a dark dashed line.In general, the theoretical stellar trend underpredicts the observed Si IV features, likely due to the larger relative contribution from ISM absorption.On the other hand, the stellar model is in excellent agreement for large EWs of C IV, but the observed data points seem to fall off at lower metallicities.Below 12+log(O/H) ≈ 8.0, nebular emission contributes to C IV λλ1548,50 (and other lines).This is reflected in the negative EWs in the figure.While a proper comparison with stellar models would require correction for this nebular contribution, this trend may still be diagnostically useful.
The observed trend of C IV EWs versus nebular oxygen abundance in Figure 10 follows the predicted stellar relation remarkably well over the metallicity range for most of the sample (12+log(O/H) > 8).This trend can be understood in terms of the metallicity-dependent stellar wind properties of massive stars, with some deviation, when large contributions of ISM absorption are present.Specifically, while (29) has the largest C IV EW offset above the theoretical trend, it is also offset to larger than average C II EWs for its metallicity, indicating a large ISM absorption component.In contrast, the observed Si IV λλ1393,1402 EWs are systematically higher than the theoretical values.This offset is not surprising, however, as the Si IV stellar features are generally weaker then those of C IV, and so the strong EWs indicate a more significant relative ISM contribution.
The strong empirical correlation between EW and 12+log(O/H) and the agreement with model predictions suggest that the EW of C IV λλ1548,50 can be used to estimate the gas-phase metallicity.The empirical relation can be expressed as 12 + log(O/H) = (0.075 ± 0.008) × EW + (7.956 ± 0.063), (4) where the EW here refers to the C IV EW.For instance, a galaxy with a measurement of EW(C IV) ≈ 5 Å would have an estimated gas-phase oxygen abundance of 12+log(O/H) ≈ 8.3, i.e., similar to that of the Large Magellanic Cloud (LMC).We emphasize that this relation has been derived for a local galaxy sample and, therefore, application to other galaxy samples, e.g., at high-z would require further verification.
Our broad measurement windows and the low spectral resolution of the data do not permit removal of the interstellar components originating within galaxies.As opposed to the case of Si IV, the relative interstellar contribution to the stellar C IV is small but still not negligible, a point raised by Crowther et al. (2006).In principle, this effect is accounted for in our empirical calibration, but the strength of the interstellar components relative to the stellar C IV may be different in a different galaxy sample.This issue could be mitigated by utilizing stellar N V λλ1238,42; owing to its larger ionization potential of 77 eV, the ISM contribution to N V is much smaller.However, N V λλ1238,42 is only present in very young stellar populations (≲ 5 Myr; see, e.g., Chisholm et al. 2019) and, in the present sample, is blended with strong Lyman-α absorption, so is not analyzed here.
Alternatively, deviations from the stellar model in Figure 10 could also indicate non-equal stellar and nebular metallicities or non-solar α/Fe ratios (e.g., Steidel et al. 2016).Since stellar winds are most sensitive to the Fe opacity in their atmospheres, an enhanced α/Fe abundance would divert points below the trend due to seemingly higher nebular oxygen abundances compared to the inferred stellar abundance.On the other hand, deficient α/Fe abundances would drive points above the trend.Despite these complications, the dispersion in the C IV EW trend is still relatively small for metallicities of 12+log(O/H)>8.0, indicating that the C IV EW can be used as a gas-phase metallicity diagnostic.

Interpretation of the Observed Relations
The trends seen in Figures 9 and 10 may seem surprising, as these spectral lines are deeply saturated, at which point they become insensitive to chemical abundance.In the saturation limit, the observed EWs of the interstellar lines lie on the flat part of the curve-of-growth, where b is the Dopper line-broadening parameter and N ion is the column density of the corresponding ion.In this limit, the EW is relatively insensitive to the column density (N ion ) and becomes mainly dependent on velocity via the Doppler parameter (b).Therefore, the metallicity dependence of saturated interstellar lines, such as, e.g., C II λ1335, can be understood in terms of macroscopic turbulence affecting b (e.g., Heckman et al. 1998).The observed trend of ISM line strengths versus oxygen abundance originates from mechanical energy input from powerful stellar-winds and supernovae.As a consequence, more metal-rich galaxies have more luminous starbursts with stronger winds and higher supernova rates, which cause more macroscopic turbulence.
The progression of EW with metallicity in Figures 9 and 10 also reflects an increase in galaxy sizes: metal-rich galaxies tend to be more luminous, more massive, and larger in size than metal-poor dwarfs (i.e., the luminosity-metallicity and mass-metallicity relationships, e.g, Skillman et al. 1989;Tremonti et al. 2004;Berg et al. 2012).This may indicate that the underlying cause of the correlation between oxygen abundance and EW is broadening by increased galactic rotation with galaxy mass.Most of our sample galaxies were also studied by Heckman et al. (1998) who demonstrated that EWs of the interstellar lines also correlate with the rotation velocities derived from the H I 21 cm line widths.However, the correlation is much weaker than that found in Figures 9 and 10, suggesting that galactic rotation is not the prime mechanism responsible for the line broadening.More importantly, the measured EWs would require rotation velocities significantly larger than those obtained from typical H I line widths.We, therefore, conclude that macroscopic turbulence and galactic-scale outflows are primarily responsible for the correlation of EW of the interstellar lines with oxygen abundance.
The trends with abundance for the stellar+ISM lines of Si IV λλ1393,1402 and C IV λλ1548,1550 are even stronger than those of the interstellar lines.Like the interstellar lines, the stellar-wind+ISM lines are deeply saturated.Therefore, the abundance dependence cannot be primarily due to direct changes in the wind column densities.Further insight can be gained by studying wind models for individual stars.
In Figure 11 we plot synthetic spectra from PoWR atmosphere models (Gräfener et al. 2002;Hamann & Gräfener 2003;Sander et al. 2015) for a fiducial O-supergiant (T eff = 40, 000 K, log g = 4.0) assuming the wind mass-loss rates of Vink et al. (2001).The C IV profile is stronger for the solar-abundance MW model (solid red line) than for the 20% solar SMC profile (solid blue line).Going one step further, we might expect the abundance pattern to deviate from the standard solar abundance pattern, where differences in the relative C abundance could affect the observed C IV profile.However, variations of the relative carbon abundance over a range of solar to 0.2 solar leave the C IV λλ1548,1550 line strength almost unchanged, as demonstrated by the Cdeficient MW profile (dashed red line: 0.2 solar C abundance, solar abundances for all other elements).
Alternatively, we can examine the role of the stellar wind in shaping the C IV line profile.The blue dashed line in Figure 11 shows a solar abundance profile but with a weaker SMC-like wind strength that looks similar to the standard SMC profile (solid blue line).Therefore the profile shape is mostly driven by the wind properties, which are largely determined by the Fe opacity for O-supergiants of these metallicities, and not directly by the relative abundances.More specifically, such hot-star winds are driven by radiation pres-sure from numerous strong and weak spectral lines predominantly located in the extreme-UV below the Lyman edge at 912 Å. Winds from stars with Milky Way, the LMC, and SMC abundances are mainly driven by spectral lines from Fe-group elements (Abbott 1982;Kudritzki et al. 1987;Vink et al. 2022), which largely determine the resulting wind properties.Since the wind properties have a stronger effect on the line profiles than the relative abundances, the Si IV and C IV relations seen in Figures 9 and 10 could partly, or even mostly, reflect a relation between stellar Fe and nebular O.
So far, we assume a given mass-loss recipe in our calculations rather than predicting the wind parameters selfconsistently from the abundances.However, even if the absolute scaling of the wind mass-loss rate and terminal velocity should change, the general metallicity scaling has been confirmed in various different wind modeling approaches (e.g., Vink et al. 2022).Our test calculations also show that relative abundances of individual elements have some impact on the derived EWs, so we will need more extensive calculations, including a full population synthesis, to test our interpretation.In particular, we emphasize an important underlying assumption made in the theoretical Starburst99 models in Figure 11.All element ratios in the stellar-wind models are solar for all values of 12 + log (O/H).The theoretical relation is expected to change if the abundance of elements driving stellar winds, i.e.Fe, were modified relative to oxygen.The agreement between the models and the data therefore suggests that our galaxy sample has an O/Fe ratio that is consistent with the solar value.On the other hand, (Steidel et al. 2016) found evidence of strongly enhanced O/Fe ratios in a sample of strongly star-forming galaxies at z ≈ 2.4 and interpret this result as due to oxygen abundance enhancement by core-collapse supernovae.Their sample of galaxies may not follow the predicted relation in Figure 11.Consistent population synthesis models incorporating stellar models with nonsolar abundance ratios may provide an opportunity to study any anomalous O/Fe relation in star-forming galaxies.

CONCLUSIONS
We present an analysis of the effects of spectral-resolution and aperture scales on derived FUV galaxy properties.The rest-frame FUV is fundamental to our understanding of starforming galaxies, as it simultaneously provides a unique window on massive stellar populations, chemical evolution, feedback processes, and reionization.The recent launch of JWST has already revealed how restframe UV spectroscopy traces galaxy evolution into the early universe, but we lack a sufficient understanding of how aperture and resolution affects the interpretation of these galactic properties.We, therefore, constructed an atlas of FUV archival spectra of local star-forming galaxies with multiple aperture sizes and spectral resolutions for comparison.In order to examine observations that mimic the anticipated galaxy-scale, lowresolution observations of high-redshift galaxies from JWST (R ∼ 100 − 3, 500) we used large-aperture (10 ′′ × 20 ′′ ) spectra from IUE (R ∼ 250) and compared to the stellar cluster-scale (2.′′ 5), high-resolution (R ∼ 15, 000) spectra from COS on board the Hubble Space Telescope (HST).
We examined how FUV-derived properties were affected by the galaxy-scale aperture and low-resolution spectra of the IUE versus the stellar cluster-scale aperture and highresolution of the COS spectra.We find that the overall effect of the aperture size difference is non-consequential for galaxies whose light is dominated by a single, bright stellar cluster, while the effect of spectral resolution is strongest when measuring the EWs of interstellar absorption features: • Using featureless regions of the FUV continuum, we measured β-slopes and found that they were generally consistent between the different aperture measurements, except when multiple bright stellar clusters populated the IUE field of view.
• We then measured the reddening due to dust, E(B −V ), of the stellar continua using two methods: (1) converting directly from the β-slope measurements and (2) using a minimization routine to fit starburst99 stellar population synthesis models to the observed spectra.We found that the two methods were consistent within their uncertainties.We also find that the E(B − V ) values agreed within their uncertainties across different aperture sizes and spectral resolutions.Similar to the β-slopes, we find little difference between cluster-scale and galaxy-scale measurements for galaxies dominated by a single ongoing burst.
• Aperture size starts to play a more significant role in the measurement of equivalent widths.We examined trends for the EWs of ions at different ionization states versus metallicity.For both low-and high-ionization states, we measure trends with slopes < 1 such that EW measurements skew to higher IUE values at the low-EW end and towards higher COS values at the high-EW end.While we find a weak high-EW trend around 4 Å, it currently lacks the statistical significance needed to draw a robust conclusion.
• We determined oxygen abundances for our sample and established correlations with UV properties.We found significant correlations between the strength of stellar and interstellar lines and the oxygen abundance despite these lines being heavily saturated.These correlations can be understood in terms of metallicity-dependent line-driven stellar-winds and interstellar macroscopic gas flows shaping the stellar and interstellar spectral lines, respectively.The observed line-strength versus metallicity relation of stellar-wind lines agrees with the prediction of population synthesis models for young starbursts.Measurements of the strong C IV λλ1548,1550 line in particular provides an opportunity to determine stellar abundances as a complement to gas-phase abundances from nebular emission lines.
The application of these results to JWST observations of high-redshift galaxies implies that integrated, galaxy-scale properties can similarly characterize the overall galaxy and the dominating stellar cluster regions well.However, care will need to be taken in interpreting interstellar and stellar absorption features from the low-resolution JWST spectra to avoid confusion from contaminating features, such as determinations of the massive star population masses and outflow properties.This will be crucial for understanding the evolution of galaxies from the early universe extrapolated out to the local Universe.
We would like to express our gratitude to the referee for many thoughtful comments that improved the clarity and impact of this work.The HST and IUE data presented in this paper were obtained from the Mikulski Archive for Space Telescopes (MAST) at the Space Telescope Science Institute.The specific observations analyzed can be accessed via doi:10.17909/r6n6-3m45.Funding for the creation and distribution of the SDSS Archive has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Aeronautics and Space Administration, the National Science Foundation, the U.S. Department of Energy, the Japanese Monbukagakusho, and the Max Planck Society.The SDSS Web site is http://www.sdss.org/.The Participating Institutions are The University of Chicago, Fermilab, the Institute for Advanced Study, the Japan Participation Group, The Johns Hopkins University, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, Princeton University, the United States Naval Observatory, and the University of Washington.Comparison of IUE and coadded COS rest-frame FUV spectra for the galaxies in our sample.The IUE spectra are relatively low resolution (λ/∆λ∼ 300) and consists of the integrated light within a large aperture (10 ′′ ×20 ′′ ).In contrast, the COS spectra have significantly higher spectral resolution (λ/∆λ= 15, 000) and higher S/N, but only within the much smaller 2. ′′ 5 COS aperture.All IUE and coadded COS spectra that extend past 1450 Å are normalized at 1450 Å, while coadded COS spectra with shorter wavelength coverage are normalized at 1350 Å.For galaxies with multiple COS pointing, each spectrum is normalized by the coadded COS spectrum normalization, allowing relative differences in shape and absolute flux to be compared.The β-slope fits to the IUE and coadded COS spectra are overplotted as purple and blue lines respectively.

Figure 3 .
Figure 3. Left: Comparison of the IUE and HST/COS β−slope measurements.There is relatively little scatter between the two sample despite their large differences in aperture sizes.Right: The flux at 1500 Å in units of 10 −15 ergs s −1 cm −2 Å −1 , for both the IUE and HST COS spectra.As expected, the integrated UV flux through the IUE apertures is larger than for the physically smaller COS apertures.The color coding indicates the differences between β measured for the IUE and the HST/COS data.Note that all data with ∆β > 0.2 have yellow colors.The labels next to the data points are the galaxy identifiers used in Table1.

Figure 4 .
Figure 4. Comparison of the stellar continuum properties measured from IUE spectra versus HST/COS spectra for galaxies with multiple COS pointings.Left: Comparison of β−slopes.The points labeled by a number only are the coadded spectra of all the COS pointings from an individual galaxy.The points labeled by a number+letter are the individual COS pointings.All points are color-coded by the galaxies as identified in the insert to the right panel.Individual COS pointings show significant scatter in their β−slopes, but the coadded COS spectra all agree more closely with the IUE β−slope values.Right: Comparison of the flux at 1500 Å is in units of 10 −15 erg s −1 cm −2 Å −1 .Similar to Figure 3, the larger aperture of the IUE spectra results in larger continuum fluxes at 1500 Å.
Figure 5.This figure shows the E(B − V ) values obtained with both Eqs.(1) and (2).Only the former values are listed in Table

Figure 5 .
Figure 5. Top: E(B − V ) values derived for the COS spectra from the SPS continuum-fitting method (Model) versus the empirical β−slope method."The β-slope values derived using Reddy et al. (2018) are plotted as light blue circles and the values derived using Chisholm et al. (2022) are plotted as teal triangles.The two methods show a tight agreement, demonstrating that these methods are consistent.Bottom: Comparison of the SPS continuum-fitting derived E(B − V ) values for the COS spectra versus the IUE spectra.The trend is consistent with a 1:1 relationship but shows some scatter between the two apertures.

Figure 6 .
Figure 6.A demonstration of the continuum normalization process for the IUE spectrum of NGC 7714.The top panel shows the initial unnormalized spectrum around S II λ1253 and Si II λ1260.The red portions of the spectra are the windows used to fit a linear continuum around each absorption feature, with the resulting fit plotted in black.The bottom panel shows the subsequent normalized spectrum.

Figure 7 .
Figure 7.Comparison of the broad EW measurements from the IUE and COS spectra.The top panel shows the low-ionization state ions, which include S II+Si II λλ1253.60,O I+Si II λλ1302,04, C II λ1335, Fe II λλ1608,11, and Al II λ1671.The lower panel shows the high-ionization state ions: Si IV λλ1393,1402 and C IV λλ1548,1550.
Al II λ1670 Narrow COS −7.115 +1.0202 Broad COS −6.453 +0.945 IUE −10.273 +1.412 NOTE-Polynomial fits to the trends between EW and metallicity shown in Figures 8, 9 and 10.The fits used are first-order polynomials of the function EW = p 0 + p 1 Z.

Figure 11 .
Figure 11.Theoretical C IV 1548, 1550 profiles for a representative O supergiant with Teff = 40, 000 K and log g = 4.0.The profiles were obtained with PoWR model atmospheres using different abundances and abundance ratios.Solid red: standard solar abundances; solid blue: 0.2 Z⊙, approximating SMC abundances; dashed red:Z⊙ for all elements except for carbon, which is 20% solar; dashed blue: 0.2 Z⊙ for all elements, except carbon and oxygen, which are solar.
made use of the NASA/IPAC Extragalactic Database (NED), which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
work has been provided by NASA through grant No. AR-13878 from the Space Telescope Science Institute, which is operated by AURA, Inc., under NASA contract NAS5-26555.
funding by the Deutsche Forschungsgemeinschaft (DFG -German Research Foundation) in the form of an Emmy Noether Research Group -Project-ID 445674056 (SA4064/1-1,PI Sander).AACS is further supported by funding from the Federal Ministry of Education and Research (BMBF) and the Baden-Württemberg Ministry of Science as part of the Excellence Strategy of the German Federal and State Governments.

Figure
Figure12.Comparison of IUE and coadded COS rest-frame FUV spectra for the galaxies in our sample.The IUE spectra are relatively low resolution (λ/∆λ∼ 300) and consists of the integrated light within a large aperture (10 ′′ ×20 ′′ ).In contrast, the COS spectra have significantly higher spectral resolution (λ/∆λ= 15, 000) and higher S/N, but only within the much smaller 2. ′′ 5 COS aperture.All IUE and coadded COS spectra that extend past 1450 Å are normalized at 1450 Å, while coadded COS spectra with shorter wavelength coverage are normalized at 1350 Å.For galaxies with multiple COS pointing, each spectrum is normalized by the coadded COS spectrum normalization, allowing relative differences in shape and absolute flux to be compared.The β-slope fits to the IUE and coadded COS spectra are overplotted as purple and blue lines respectively.

Table 1 .
UV Galaxy Sample Properties R.A., Decl.Morph.E(B −V ) MW v rad Adopted distances for the sample are listed in Column 6, with corresponding distance determination method and reference listed in Column 7. Note that TRGB is used for the tip of the red giant branch method and HF is used for the Hubble flow.Column 8 in this table gives the resulting linear scales for each galaxy.The spatial scales covered by the COS and IUE entrance apertures, therefore, range from tens of pc to several kpc.Columns 9 and 10 list the luminosity and best measurement of the gas-phase oxygen abundance, with corresponding references.
NOTE-Properties of the present sample.The galaxies in this work are UV bright, nearby galaxies with high quality UV spectral observations from both the IUE and HST and cover a range of properties.Column 1 of this table gives the galaxy name.The R.A. and Decl.coordinates are listed in Column 2 and the morphological type is given in Column 3. The Galactic foreground extinctions in Column 4 are taken from Schlafly & Finkbeiner (2011) who utilized SDSS data to recalibrate the earlier dust-emission based values of Schlegel et al. (1998).Heliocentric radial velocities (v rad ) are listed in Column 5.

Table 4 .
IUE and COS Spectral Properties We compare properties derived from the COS and IUE spectra for each of the 29 galaxies.Columns 2-3 show the COS and IUE continuum flux values measured at 1500 Å.For galaxies with more than one COS aperture, we include the continuum for the resulting combined spectrum as the default comparison to the IUE, as well as the continuum for each individual aperture, which are listed below each galaxy.Columns 4-5 list the measured β-slope values and uncertainties derived from a boot-strap Monte Carlo least squares linear fit to the COS and IUE stellar continua.Stellar continuum windows free of contamination in the range of 1250 Å < λ < 1850 Å were used to avoid the stellar continuum turnover at bluer wavelengths and possible contamination from the broad 2200 Å dust bump, which can affect the continuum as blue as ∼ 1850 Å.

Table 5 .
Broad Sample Equivalent Widths

Table 6 .
Equivalent Widths of the COS Narrow Sample