The PHANGS-HST-Hα Survey: Warm Ionized Gas Physics at High Angular Resolution in Nearby Galaxies with the Hubble Space Telescope

Imaging observations for the PHANGS-HST-Hα Treasury program (Cycle 30, PID 17126; PI: R. Chandar) were conducted starting in 2022 December and allocated 19 orbits. We note that NGC 2835-N was observed just prior to publication, so it will be included in a future image release, and NGC 4535 was ultimately observed as part of a related HST program (PID 17457; PI: Belfiore). Imaging is conducted with the WFC3 camera in either the F658N or F657N filter (more below). PHANGS-Hα footprint coverage is designed to maximize overlap with the five broadband filters observed as part of the PHANGS-HST Treasury Survey (Cycle 26, PID 15654; PI: J. Lee). We restrict the allowed orientation to match that of the Cycle 26 PHANGS-HST broadband observations for cases where such restrictions do not make the visit too challenging to schedule. There are either one or two pointings per galaxy.

There are two WFC3/UVIS narrowband filters that cover the Hα line at the distances of our sample galaxies: F658N and F657N. The F658N filter is preferred because it is significantly narrower, with a width of 27.5 Å or ∼1300 km s⁻¹ (and a pivot wavelength of 6585.6 Å), compared with the F657N filter, which has a width of 121 Å or ∼5500 km s⁻¹ (and a pivot wavelength of 6566.6 Å). Note that even the narrow F658N filter includes the [N ii] λ6583 emission line. We show the transmission curves for the two filters in Figure 3.

Zoom In Zoom Out Reset image size

Both filters have a peak throughput of 25%. We use the narrow F658N filter to observe galaxies with systemic velocities less than ≈1500 km s⁻¹ and the F657N filter to observe those with higher velocities (note that the F657N filter was used to image NGC 1433 even though it has a velocity ≲700 km s⁻¹).

With the WFC3/UVIS camera, we observe a single target in each visit and maximize the integration time in each exposure. A post flash of 17 or 18 e⁻ is applied for each exposure to increase the background counts, which is recommended to mitigate issues due to charge transfer efficiency (CTE) losses. The dither sequence is optimized to cover the WFC3 chip gap. We obtain two exposures at each of two pointings using the “WFC3-UVIS-GAP-LINE” pattern, which is designed to cover the gap between the two WFC3 CCDs. The exposure times for each galaxy are listed in Column (4) of Table 2. We show footprints of the new PHANGS-HST-Hα coverage (red) and that of VLT/MUSE IFU (white) in the Appendix.

Standard procedures are used to reduce and drizzle the F658N and F657N images. The exposures are processed through the Pyraf/STSDAS CALWFC3 or CALACS software, which performs initial data quality flagging, bias subtraction, gain correction, bias stripe removal, correction for CTE losses, dark current subtraction, flat-fielding, and photometric calibration and outputs “FLC” FITS files for each exposure. These files are then aligned and registered with north up and east to the left, sky subtracted, and drizzled (using the DRIZZLEPAC software package) to create a single F658N or F657N image matched to the pixel grid of the PHANGS-HST V-band images (J. C. Lee et al. 2022), with a pixel scale of 004. The PHANGS-HST V-band images used astrometry calibrated using Gaia DR2 sources (Gaia Collaboration et al. 2018; L. Lindegren et al. 2018). The final FITS files are in units of e⁻ s⁻¹, and important use cases are discussed in Section 5.

We used archival narrowband images for NGC 628-C, NGC 1300, NGC 1433, and NGC 1672. These were taken by the following programs: GO-10402 (PI: Chandar) for NGC 628-C, GO-10342 (PI: Noll) for NGC 1300, GO-13773 (PI: Chandar) for NGC 1433, and GO-10354 (PI: Jenkins) for NGC 1672. The specific filter and exposure times are compiled in Table 2 for these data as well.

As noted in the previous section, two other HST programs (besides PID 17126) continue to obtain broadband and Hα imaging of PHANGS galaxies that were not part of the PHANGS-JWST Cycle 1 Treasury sample. Specifically, PID 17457 (PI: F. Belfiore) is completing Hα imaging for the rest of the original 38 PHANGS-HST galaxies. Additionally, PID 17502 (PI: D. Thilker), a Cycle 31 Large Treasury, is obtaining both NUV-U-B-V-I and Hα for PHANGS galaxies observed by JWST in Cycle 2. When completed, these programs will expand science analysis such as described herein to 74 galaxies, each with HST+JWST+ALMA data sets. The PHANGS collaboration is working to obtain MUSE IFU spectroscopy for as many of these as possible.

The PHANGS-MUSE survey has carried out observations with the MUSE integral field spectrograph at the ESO VLT to map the 19 nearby spiral galaxies that form our sample (E. Emsellem et al. 2022). The footprints of the PHANGS-MUSE survey were designed to maximize overlap with the PHANGS-ALMA CO (2–1) maps and are shown in the Appendix (Figure 15). The spatial resolutions of the MUSE observations for each galaxy are compiled in Table 2. PHANGS-MUSE has produced ≈15 × 10⁶ spectra that have a wavelength coverage of 4800–9300 Å with a spectral resolution going from ∼80 km s⁻¹ at the blue end to ∼35 km s⁻¹ at the red end and targeting key optical lines, including Hβ, [O iii] λ5006, [O i] λ6300 [S ii] λλ6716, 6731, Hα, and [N ii] λ6583.

The data reduction process is described in detail in E. Emsellem et al. (2022). The data cubes are publicly available from the ESO archive.³² In this work, we use the PHANGS-MUSE spectra to help anchor the flux calibration of our HST-Hα images and to correct for [N ii] contamination, as described in Section 4.

We generate synthetic images from the MUSE data cubes integrating over the HST F657N, F658N, F555W, and F814W filters associated with the appropriate camera+ filter (WFC3/UVIS or Advanced Camera for Surveys (ACS)/WFC) for each galaxy (see Appendix A.3 for comparison of alternative filter selection).

The wavelength coverage of the MUSE spectra does not fully cover the range of the F555W and F814W transmission curves (see Figure 5). Consequently, a small fraction of the emission remains unaccounted for in the synthetic images. However, the continuum subtraction is performed in units of flux density (rather than flux; see below), so the small missing bandwidth does not significantly affect the final HST-Hα maps.

Zoom In Zoom Out Reset image size

The HST and synthetic MUSE images are all converted to the same units of flux density (erg s⁻¹ cm⁻² Å⁻¹ pixel⁻¹). To convert the HST images from units of electrons s⁻¹ to flux density, we employ the photflam parameter. The photflam value is for a source with a constant flux per unit wavelength in units of erg s⁻¹ cm⁻² Å⁻¹ that produces a count rate of 1 count s⁻¹ and is taken from the header of the HST images. Additionally, we use the SYNPHOT package³⁴ to calculate photplam, or the pivot wavelength (“pivot” parameter), and photbw, or the width of the filter (“rectwidth” parameter). These parameters for all filters are summarized in Table 3.

Next, we smooth and regrid the HST images to match the resolution and pixel grid of the synthetic MUSE images (see Figure 6).

Zoom In Zoom Out Reset image size

We first smooth all HST images using a Gaussian function with a size of

where the resolutions of the MUSE observations (θ_MUSE) for each galaxy are compiled in Table 1 (see E. Emsellem et al. 2022) and the HST resolution (θ_HST) is assumed to be 01. In practice, the point-spread function (PSF) of the HST observations is not necessarily a Gaussian (J. E. Krist et al. 2011) and will fluctuate by ≈002 around this value depending on the filter and pixel coordinates. However, since the HST images have a spatial resolution approximately an order of magnitude higher than the MUSE observations and are, hence, smoothed significantly during this step of the procedure, small variations in the PSF/shape of the beam of the HST images have essentially no impact on the resulting MUSE-resolution-matched HST-Hα maps.

We convolve the smoothed HST images to the same pixel grid as the synthetic MUSE images (pixel size of 02) using a bilinear function (making use of the astropy.convolution.convolve_fft function; Astropy Collaboration et al. 2018).

We compare the flux density values in each pixel of the matched-resolution HST and synthetic MUSE images. The best linear fit between these will be used to anchor fluxes in the HST images to the more accurate MUSE values. To determine this relation, we first mask pixels that (1) are contaminated by bright foreground stars (identified from MUSE; B. Groves et al. 2023) and (2) limit the data to a range around zero. The latter was done since in this step we are primarily interested in the calibration of any zero-point offset between the two images and therefore want to limit the dependence of our fitting on the bright emission regions. We limit the data between the 0.01th and 99th percentiles in the HST and synthetic MUSE images. In addition, we bin the flux data into 25 bins, which again limits the impact of outlier pixels on the fit. A similar method was recently employed for the calibration of the PHANGS-JWST images (A. K. Leroy et al. 2023; T. G. Williams et al. 2024). Lastly, a linear least-squares fitting (y = ax + b, where y represents the HST image data points and x represents the MUSE image data points) was performed.

The results of this analysis can be seen in Figure 7, which shows the matched-resolution HST (y-axis) and synthetic MUSE (x-axis) flux densities for the F555W, F658N, and F814W filters toward NGC 628-C, as well as the binning and fitting routines. We apply the background correction from these solutions to the full-resolution HST images as y_bgsub = y − b (i.e., we do not correct for the slope of the fit, but only the background offset).³⁵ The background correction offsets for each galaxy and each filter are given in Table 4.

Zoom In Zoom Out Reset image size

This procedure yields HST-Hα maps with a robust background level anchored to match the MUSE measurements.

The purpose of this step is to subtract the stellar continuum present in the narrowband HST filter images (see Figure 5). To achieve this, we utilize the HST F555W and F814W images, which are dominated by the stellar continuum and have been flux anchored to the synthetic MUSE images, as described in the previous section. We note that our procedure would also work with other filters. For example, the F550M or F547M filter could replace the F555W filter, as demonstrated in the Appendix.

We begin by correcting the observations for foreground extinction due to the Milky Way. This is achieved using the extinction law of J. A. Cardelli et al. (1989) and E(B − V) values from E. F. Schlafly & D. P. Finkbeiner (2011). Although this adjustment results in a relatively minor flux correction (on the order of 10%), it is important for accurate continuum subtraction. Without it, systematic over- or undersubtraction could occur across filters owing to the wavelength dependence of extinction. Additionally, the same foreground extinction correction has been applied to the MUSE DAP Hα maps, which are used to account for the contribution of the [N ii] line, as detailed in E. Emsellem et al. (2022).

To estimate the continuum flux in the narrowband F658N or F657N filter, we calculate weighting factors, W, based on the relative differences in the central wavelengths of the broad- and narrowband filters:

where λ_F555W, λ_F814W, and λ_F65XN are the central “pivot” wavelengths of the F555W, F814W, and either F657N or F658N filters, respectively, determined from their respective transmission filter response curves (see Section 4.2). The continuum image is produced in logarithmic space as

where F_F555W and F_F814W are the pixel fluxes in the F555W and F814W images (in units of erg cm⁻² s⁻¹ Å⁻¹). The continuum-subtracted Hα image is then produced in linear space as

where F_F65XN represents the pixel fluxes in either the F657N or F658N image. As a final step, we convert the continuum-subtracted images from units of flux density to units of flux using the filter bandwidth (from the respective transmission filter response curve; see Section 4.2).

The result of this continuum subtraction routine can be seen in Figure 8. Here the bright nebulae are cleanly separated from the background stellar continuum across the whole disk of the galaxy (in this case NGC 628). There are, however, low-level artifacts that are difficult to remove, e.g., due to diffuse stellar emission toward the center of the galaxy or compact stellar clusters that may have continuum spectra that are not well recovered by only using the (log) interpolation of the two broadband observations. It is also worth noting that artifacts in either of the broadband or narrowband images become more prominent in the continuum-subtracted image, due to the relative intensity of the nebula emission to stellar emission.

Zoom In Zoom Out Reset image size

The final global correction applied to the HST-Hα images accounts for the contribution from the [N ii] λ6548 and [N ii] λ6583 lines, which can fall within the narrowband F658N and F657N filters depending on the HST filter and the velocity of the galaxies (see Figure 5). For this correction, we compare the HST-filtered MUSE images³⁶ with the DAP Hα image to determine the correction factor. This factor is subsequently applied to the full-resolution continuum-subtracted HST observations. Using the MUSE data alone to derive this correction factor, rather than employing the smoothed HST observations, was motivated by the goal of conducting a consistent test using data from the same set of observations.

To do so, we conduct the continuum subtraction analysis (including the MW extinction correction) outlined in Section 4.6 using the broadband and narrowband synthetic MUSE images for each galaxy. This produces MUSE continuum-subtracted emission maps, with the Hα and [N ii] lines convolved with the filter transmission curves for each galaxy (see maps in bottom panels of Figure 8).³⁷

Next, we compare the MUSE continuum-subtracted emission maps and the MUSE DAP Hα emission-line maps (see Figure 9). We mask the outlying pixels with percentiles in both the MUSE continuum-subtracted and DAP Hα maps (1%−99% in this case), bin the remaining points into 20 equally spaced bins, and perform a least-squares fit to the binned data points assuming a linear polynomial (y = ax + b, where y represents the MUSE continuum-subtracted emission data points and x represents the MUSE DAP Hα emission-line data points; as shown in Figure 9). We then divide the full-resolution HST fluxes by this factor (a), to obtain the HST-Hα-only emission-line maps.³⁸

Zoom In Zoom Out Reset image size

It is worth noting that the relative contribution of these lines is somewhat challenging to constrain on a pixel-by-pixel basis across all galaxies. The exact contribution depends on the velocity, how the lines fall within the HST filter transmission curves, and the intrinsic nature and morphology of the specific source. For instance, assuming a global correction factor could introduce systematic errors in the flux with galactocentric radius, particularly in galaxies with strong metallicity gradients. In such cases, this approach might result in subtracting too little flux near the center and too much flux near the edges. Indeed, we find that NGC 4303 has a correction factor slightly below 1, which in principal should not be possible. This may be attributed to this issue, although this galaxy does not exhibit a particularly strong metallicity gradient (see B. Groves et al. 2023). This discrepancy could also arise from imperfect continuum subtraction and/or absorption of the Hα (which is corrected for in the MUSE DAP fits; see E. Emsellem et al. 2022).

Figure 9 shows an example of this procedure for one galaxy, NGC 628, where we find a [N ii] correction factor of 1.35. The final correction factors determined individually for each galaxy are compiled in Table 4. Again, we note that fixing the intercept of the fits to zero makes only a minor change to the correction factors (see Table 6 in the Appendix).

We conduct a number of consistency checks of the resulting images. To do so, we sum the fluxes in the final HST-Hα and MUSE Hα maps within the masks from the MUSE nebula catalogs. In Figure 10, we plot the histogram distributions of this flux ratio for all sources within the catalog. We see that the ratios scatter around 1, which highlights that the correction of the [N ii] contribution in the HST data is reasonable. We note that we do not use the HST data to estimate this correction factor, but rather the synthetic MUSE observations, and despite this, the correlations are close to unity. We fit a Gaussian to the distribution for all galaxies and see that they have a standard deviation of around 30%. Limiting the fit to the brightest regions above 10⁻¹⁵ erg s⁻¹ cm⁻², we see that this deviation decreases to around 15%. We use these as the uncertainty confidence limits on our flux measurements across the HST-Hα images.

Zoom In Zoom Out Reset image size

In Figure 11, we plot the ratio of these fluxes as a function of the MUSE Hα, MUSE Hα velocity, galactocentric radius, and MUSE Hα equivalent width. Across the sample, we see evidence for mild systematic trends with galactocentric radius or velocity in a few cases, likely due to the emission shifting out of the filter (see Figure 16).

Zoom In Zoom Out Reset image size

To produce HST-Hα error maps, we use the inverse variance (1/V) images for the individual filters (see J. C. Lee et al. 2022). These maps incorporate contributions from the detector and the Poisson noise toward the target, modulated by the number of integrations drizzling into an output pixel and the effective exposure depth after masking, respectively. We convert the inverse variance image into a map of the uncertainties in units of counts s⁻¹ (i.e., ) and then into flux density (erg cm⁻² s⁻¹ Å⁻¹ arcsec⁻²; see Section 4.2). To produce errors for the continuum-subtracted narrowband images, we propagate the individual pixel errors in all the bands as

and then carry this forward to the continuum-subtracted flux density as

This is converted into units of flux (see Section 4.5) and corrected by the [N ii] factor (see Section 4.6). We note that these error maps do not account for uncertainties in the zero-point offset or in the [N ii] correction factor but are scaled accordingly by the correction factor such that they are representative of the errors in the final Hα images. In Table 5 we provide some statistics of the noise across the HST-Hα error maps.

As the final part of our data reduction pipeline, we post-process the images to remove remaining artifacts such as cosmic-ray residuals, strong negative features caused by oversubtraction of the continuum level, and foreground stars. These post-processing steps provide an optimal “artifact free” image set, although some biases may be introduced in the noise.

4.8.1. Cosmic-Ray Correction

4.8.2. Continuum Oversubtraction

4.8.3. Foreground Star Removal

4.8.4. Mosaic Image for NGC 628

HST-Hα emission, seen as the pinkish-red emission in Figure 1, shows the variety of star-forming environments encompassed by the PHANGS-HST-Hα sample. There are several galaxies with compact star-forming rings (including NGC 1512, NGC 1672, and NGC 4321). A number of the spirals in our sample have nuclear star formation, with clumpy Hα associated with nuclear star-forming regions (e.g., NGC 1385, NGC 5068, and NGC 7496). In the barred galaxies (e.g., NGC 1512, NGC 1672, NGC 3627, NGC 7496) it is very obvious that star formation does not occur in the bar, but rather in the center and at the ends of the bar (e.g., A. Fraser-McKelvie et al. 2020). Our sample also includes galaxies with flocculent and grand-design spiral arms. In some galaxies, weak interarm Hα emission is observed, where weak star formation is taking place (e.g., NGC 4254).

The new maps also show a wide range of source morphologies on smaller scales. There are many individual compact and unresolved H ii regions, as well as large complex ones. There are superbubbles (A. T. Barnes et al. 2023; E. J. Watkins et al. 2023) and partial shells with a range of sizes, as well as diffuse Hα emission.

The flux-calibrated emission-line PHANGS-HST-Hα maps will support a number of science projects. The current data release includes the reduced drizzled images and the flux-calibrated, continuum-subtracted Hα maps (more below), which are available via MAST,⁴⁰ as well as at the Canadian Astronomy Data Centre as part of the PHANGS archive.⁴¹

1. 1.  
   We release reduced, drizzled images in units of flux density, which have been aligned and drizzled onto the same pixel grid as the five broadband PHANGS-HST images from J. C. Lee et al. (2022). These images include both line and stellar continuum emission and are ideal for including in spectral energy distribution (SED) fitting to determine the age and reddening of star clusters (see Section 5.3). For these, we will release the DRC FITS images of the individual pointings and mosaicked DRC FITS images for the galaxies that have multiple pointings.
2. 2.  
   We also release the flux-calibrated, continuum-subtracted Hα maps in units of erg s⁻¹ cm⁻¹ arcsec⁻², which were created as described in Sections 4.1−4.7; images that have gone through the additional post-processing steps are available upon request. These images are ideal for identifying individual H ii regions, even very small and faint ones that are not detected in the ground-based MUSE/IFU observations (Section 3.2). They also provide a unique opportunity to measure the radii of H ii regions (Section 5.5), which is quite challenging from the ground and will lead to significantly improved calculations of key stellar feedback processes such as internal pressure from radiation, thermal gas, and winds. These images also reveal diffuse ionizing gas in spiral galaxies, including the warm ionized gas leaking out of H ii regions (Section 5.6).

Age and mass are two of the most basic properties of a star cluster. Accurate determinations of the ages and masses of clusters are critical to address many fundamental questions related to stellar feedback and star and cluster formation and evolution. However, low-metallicity, dust-free ancient clusters are difficult to distinguish from young, dust-rich clusters or from intermediate-age moderately dusty clusters when only broadband optical observations are used. Breaking this age–reddening–metallicity degeneracy is sufficiently challenging that ≈80% of likely ancient globular clusters were incorrectly dated using five broadband filters to ages ≲100 Myr by the original PHANGS-HST pipeline (e.g., J. A. Turner et al. 2021; B. C. Whitmore et al. 2023; M. Floyd et al. 2024), and a significant fraction of intermediate-age (≈20–500 Myr) clusters were best fit by an extremely young age of 1 Myr and moderate to high reddening (e.g., D. Thilker et al. 2025, in preparation; K. Henny et al. 2025, in preparation).

Knowing whether Hα line emission is present or absent can significantly improve cluster age estimates (e.g., R. Chandar et al. 2016; G. Ashworth et al. 2017; B. C. Whitmore et al. 2023). While including a short-wavelength broadband filter (such as in the NUV and/or U band) is important, the line emission measured in a narrowband filter, like the F658N or F657N filters used in this program, is one of the most direct ways to distinguish very young (≲6 Myr), gas-rich clusters from older, gas-free clusters. Figure 12 shows median SEDs for samples of very young, gas- and dust-rich clusters (shown in blue) and for ancient, dust-free clusters (shown in red) in 10 PHANGS-HST-Hα galaxies (K. Henny et al. 2025, in preparation). While dust-rich young and dust-free old clusters can have red broadband optical colors (decreasing flux density at shorter wavelengths), the young clusters have strong Hα line emission that clearly distinguishes them from the older clusters. Improved age, reddening, and mass estimates from SED fitting that includes Hα photometry from the images presented here will be presented in D. Thilker et al. (2025, in preparation) and K. Henny et al. (2025, in preparation).

Zoom In Zoom Out Reset image size

For some clusters with no Hα emission, it can still be challenging to separate those with ages ≈7–10 Myr and moderate reddening from somewhat older (≈20–50 Myr) clusters with little reddening. In these cases, the presence or absence of near-infrared emission from red supergiants with ages ≈7–10 Myr in NIRCAM photometry further improves age-dating (K. Henny et al. 2025, in preparation).

Warm ionized gas in spiral galaxies originates from H ii regions, supernovae/SNRs, and PNe. Ultraviolet photons can escape from these nebulae and ionize gas far from their birth sites, giving rise to DIG. Identifying nebulae, particularly compact ones, and separating them from DIG in nearby galaxies can be challenging in imaging and IFU/spectroscopic data with resolution on the order of 1″, typical of ground-based observations.

In star-forming galaxies, ionized gas nebulae have a larger range in sizes and more complex morphologies than star clusters. Figures 14(a)–(e) show an example of a large ionized gas region in NGC 1672. Panel (a) highlights two dominant star clusters and the stellar population in a ∼kiloparsec section of NGC 1672 in a composite BVI image from the PHANGS-HST survey (Lee et al. 2022). Panel (b) shows Hα contours in this region from the PHANGS-MUSE map and identifies the level of the DIG. The proximity of the DIG to H ii regions in these images indicates that this gas is comparable to the DIG observed in the Milky Way using radio recombination lines that trace denser, ionized gas (e.g., GDIGS survey; L. D. Anderson et al. 2021). This is distinct from more diffuse gas, commonly referred to as the warm ionized medium that is seen with Hα emission in the Milky Way with the Wisconsin Hα Mapper (L. M. Haffner et al. 2003; D. Krishnarao 2019). Panel (c) shows the HST continuum-subtracted Hα, with the nebulae defined by MUSE indicated by the white contours. Panels (d) and (e) show that with high-resolution HST-Hα images we can dissect complex nebulae and identify the stellar sources (clusters and massive stars, shown as the blue stars) that produce ionizing photons and drive the larger and smaller shells and bubbles (represented by the black curved lines). Panel (f) highlights the range of shell-like structures possible even in more compact nebulae, showing Hα in the left panel of each image pair and BVHα in the right panel.

Different methods can be used to identify nebulae. We will present a new catalog of thousands of ionized gas nebulae in an accompanying work (A. Barnes et al. 2025, in preparation) based on the PHANGS-HST-Hα flux-calibrated images described in Section 4. We start from the PHANGS-MUSE nebula catalogs produced by B. Groves et al. (2023) in order to include flux measurements from the PHANGS-MUSE IFU spectra of key emission lines (including Hα, Hβ, [O iii], and [N ii]).

Assuming the expansion of an H ii region from a point source into a homogeneous medium, the pressure P at the inner working surface of the bubble walls scales with radius r as follows: (1) P_gas ∝ r^−3/2 for the thermal pressure exerted by the warm ionized gas that fills the bubble cavity, (2) P_rad ∝ r⁻² for the intense radiation field produced by the massive stars in the cluster, and similarly (3) P_wind ∝ r⁻² for the mechanical pressure from strong stellar winds (see A. G. G. M. Tielens 2010; B. T. Draine 2011; R. S. Klessen & S. C. O. Glover 2016; D. Rahner et al. 2017, for more details).

Previously, A. T. Barnes et al. (2021) constrained these internal pressure terms in over 5000 H ii regions using ground-based MUSE/IFU spectroscopy. However, the best ground-based observations only provide upper limits on the radii of ∼90% of the H ii regions in our sample galaxies, since they are unresolved at the ∼50–100 pc resolution provided by available ground-based observations. With only weak upper limits on their radii, the pressure calculations have large uncertainties, of about two orders of magnitude, as shown by the red and orange bars in Figure 13.

Zoom In Zoom Out Reset image size

PHANGS-HST-Hα observations allow us to directly measure the radii of H ii regions, as well as the mass and age of the ionizing cluster. Direct size measurements decrease the uncertainties in the pressure estimates from a factor of ≈100 to ≈2–3, as shown by the blue data points in Figure 13. The new calculations will allow the dominant source of feedback to be identified in large samples of young clusters as a function of mass and age and across different galactic environments. The method for measuring H ii region sizes will be presented in Barnes et al. (2025, in preparation).

Ionizing photons leak out from H ii regions and produce diffuse ionized emission (DIG) far from their birth sites. The PHANGS-HST-Hα observations are ideal for directly measuring the amount of ionizing radiation leaking out of H ii regions, f_esc, right at the source. In Figures 14(e) and (f), we show examples of diffuse Hα emission escaping from H ii regions with a variety of structures, some with open shells and some without. Measurements of Hα flux outside of H ii regions (e.g., along the outer curved shell wall as shown in panel (e)) can be compared with that measured inside H ii regions to quantify f_esc starting at very early ages.

Zoom In Zoom Out Reset image size

These local measurements of f_esc and H ii region morphology can be used to test predictions of how f_esc depends on the molecular gas density of the local environment, which will be measured from the ALMA CO (2−1) maps. For example, simulations (Kim et al. 2019) indicate that H ii regions that form in clouds with high surface density spend a longer time (relative to the freefall time) in the embedded phase, during which the escape fraction of ionizing radiation is predicted to be small. Simulations also predict a strong (anti)correlation of the instantaneous f_esc with extinction (which we will estimate from the MUSE-based Balmer decrement) and show that f_esc sensitively probes gas/dust substructure driven by turbulence that creates channels for photons to emerge. The high-resolution, near-infrared images from JWST for this same sample of galaxies will also shed light on the presence of holes versus dense gas.

In Sections 4 and 4.4, we determined accurate background levels for the HST-Hα images and derived correction factors for the [N ii] emission-line contamination in each galaxy by calibrating them to the MUSE spectroscopic observations. In both cases, we allowed the slope and intercept to be free parameters in the fits; however, we only applied the offset in the former case (background correction) and the slope in the latter ([N ii] correction). In this section, we explore the impact on the derived correction factors of fixing the slope to unity or the offset to zero, following the same procedures outlined in Sections 4 and 4.4.

Columns (2), (3), and (4) of Table 6 report the background level fits (the y-intercept) obtained when the slope is fixed to unity rather than treated as a free parameter. Column (5) of Table 6 lists the correction factors for the [N ii] emission lines when the y-intercept is fixed to zero. We find that these values, derived under fixed-slope or fixed-intercept conditions, are broadly consistent with those obtained when both slope and offset are allowed to vary freely.

In Figure 12, for NGC 628 c we showed the ratio of HST to MUSE Hα fluxes versus the MUSE Hα flux, MUSE Hα velocity, galactocentric radius, and the Hα equivalent width measured directly from the MUSE spectra. There is some range in the results for the galaxies in our sample, so in Figure 16 we present the exact same measurements for NGC 1566 (top panels) and NGC 4303 (bottom panels). These plots allow a deeper understanding of any systematic variations in the HST-to-MUSE flux ratios, as well as the reliability of cross-calibration for Hα measurements across different galactic conditions.

Zoom In Zoom Out Reset image size

Unlike the results shown in the main text for NGC 628 (Figure 11), here we observe a trend of decreasing HST-to-MUSE flux ratios at higher velocities (>50 km s⁻¹), which is particularly pronounced for NGC 4303 (see also >100 km s⁻¹ for NGC 1566). This effect is likely due to the position of NGC 4303 at the edge of the F658N filter response curve (see Figure 3), which could lead to a reduction in the effective Hα transmission at higher velocities relative to MUSE. In contrast, NGC 628 does not show this trend in Figure 11, supporting the hypothesis that proximity to the filter band edge influences flux measurements in cases with high nebular velocities. To correct for this, velocity-dependent flux correction will be included in a future version of the pipeline.

The F555W filter was used (along with F814W) to estimate and subtract the continuum level in the narrowband images. This filter may suffer contamination from strong emission lines such as [O iii] and potentially introduce residuals in continuum subtraction that affect the derived flux maps when compared with the narrower F550M or F547M filters. We tested the impact of using the F550M or F547M filters rather than the F555W filter by using data available for NGC 1672 and NGC 3351. We ran the full pipeline, which includes the background subtraction and [N ii] correction steps (using the F550M or F547M filter transmission curves applied to the MUSE cubes), as well as using the filters for the continuum subtraction.

Figure 17 shows the resulting maps and a comparison of the per-pixel fluxes. We observe that in NGC 1672 the F555W filter appears to have some minor residual continuum in the image when compared to the F550M filter subtraction, which appears predominantly toward the center of the galaxy. This is consistent with expectations that [O iii] emission might contaminate the F555W filter more significantly in regions with strong ionizing sources. Nonetheless, overall we see a very close one-to-one match of the fluxes across the galaxy (see the top right panel of Figure 17). In NGC 3351 we see that there is a slight shift to lower values for the F547M filter subtraction when compared to the F555W subtraction (see dotted line in Figure 17), though this appears to be an effect of poor background subtraction in the F547M filter rather than due to line contamination, which would predominantly affect smaller scales. Overall, we found that switching to the F547M or F550M filters makes little difference to the final flux-calibrated, continuum-subtracted images, indicating that line contamination does not significantly propagate into the final products for these galaxies.

Zoom In Zoom Out Reset image size