EXTINCTION AND DUST GEOMETRY IN M83 H ii REGIONS: AN HUBBLE SPACE TELESCOPE/WFC3 STUDY

Interstellar dust attenuates and reddens the light from stars, by both absorbing and scattering it, producing what is termed "dust extinction." When dealing with individual stars, one can usually assume that the dust (absorber) is located exclusively in front of the star (emitter) and is well separated from it, thus simplifying the treatment of dust extinction. The analysis of the light-attenuating effects of dust quickly becomes extremely complicated when extended sources, such as stellar populations, are the emitters, and absorbers and emitters can distribute to produce complex geometries. This is generally the case of external galaxies, where individual stars are unresolved; in this case we term "dust attenuation" the combined effects of extinction and geometry. The assumption that the absorbers are located in front of the emitters and well separated from them is widely applied in astronomy even for external galaxies, but is far from accurate.

Probing the actual geometry of the dust distribution in an external galaxy is a challenging task that requires multiple-band measurements across a wide range of the electromagnetic spectrum with very high spatial resolution. The hydrogen recombination emission lines provide a convenient tool, commonly used for deriving both the dust extinction and the dust geometry. The intrinsic ratios of these lines are easily calculated theoretically, and show little variation for a wide range of physical and chemical conditions. For case B recombination, changes in the electron density of an H ii region between 10² and 10⁴ cm⁻³ and in the temperature between 5 × 10³ and 10⁴ K, the Hα/Hβ line ratio only changes between 3.04 and 2.85 (∼7%), and Hα/Paβ between 16.5 and 17.6 (also ∼7%, Osterbrock & Ferland 2006; Dopita & Sutherland 2003). Case B recombination and the adopted density and temperature ranges are appropriate for most extragalactic nebulae. Observing two hydrogen lines yields a crude measurement of the foreground extinction (similar to what is done for individual stars), but at least three or more lines, ideally widely spaced in wavelength, are needed to constrain both extinction and geometry.

The starburst galaxy M83 has been imaged in multiple narrow bands targeting hydrogen recombination lines (Hβ and Hα in the optical, and Paβ in the near-infrared; IR), as part of the Early Release Science (ERS) observations made by the Hubble Space Telescope/WFC3 (HST/WFC3) Scientific Oversight Committee (SOC), which enables an unprecedented detailed study of the dust extinction and geometry in the nebular gas of a nearby galaxy.

M83 (NGC 5236, a.k.a. the "Southern Pinwheel Galaxy"), is classified as an SAB(s)c (de Vaucouleurs et al. 1991), is one of the closest grand design-spirals (4.56 Mpc), is virtually face-on, and has a starburst nucleus, a marked bar, and prominent spiral arms. These properties make it an ideal target for detailed investigations of dust extinction in extragalactic H ii regions.

M83 was mapped with HST/WFC3 in 2009 August, with a series of narrow and broad band filters in the UVIS and IR channels, as part of the ERS program (GO-11360, PI: Robert O'Connell; see Chandar et al. 2010 for further details of the program and the full list of filters used to observe M83). The narrow-band filters included those in the Hβ (λ4861 Å, F487N), Hα+[N ii] (λ6563 Å + λλ6548, 6584 Å, F657N), and Paβ (λ12818 Å, F128N) emission lines. The observations cover two pointings: the southern pointing targets the nuclear region of M83 and about half of its northeastern arm, and the northern pointing targets an adjacent field of the same area where the southwestern arm winds over. The field of view (FoV) for each pointing subtends 3.6 × 3.6 kpc² for Hα and Hβ, and 3.1 × 2.7 kpc² for Paβ.

The raw data were first processed with the MultiDrizzle software (Fruchter 2009) to accomplish both basic and high-level processing, including flat-fielding, cosmic-ray cleaning, combination, mosaicing of the two pointings, and registration of all mosaics onto a common grid (Mutchler 2010).

We proceed at matching the point-spread functions (PSFs) of the UVIS and IR channels. Direct measurements of point sources in the images yield PSFs with 007, 007 and 022 FWHM for Hβ, Hα, and Paβ, respectively. We thus degrade the UVIS images (007 resolution) to match the resolution of IR data (022) by convolving the former to a Gaussian kernel. Finally, we resample the images from both channels (original pixel sizes are 004 for UVIS and 013 for IR) to a 7 × 7 grid of the original UVIS data, obtaining a final pixel size of 028 (6.1 pc).

To produce pure hydrogen emission line images, we subtract a rescaled F110W image from the F128N image to remove the stellar continuum. The stellar continuum images for Hβ and Hα lines are determined by linearly interpolating between the F438W and F555W and between the F555W and F814W images, respectively. For each individual pointing in each line, we subtract the stellar continuum by matching the fluxes of a sample of stars, and then measure and remove the global background. Filter throughput corrections for the shifted lines at the 513 km s⁻¹ heliocentric recession velocity of M83 (Koribalski et al. 2004) amount to only 1–3%. Each emission line image is also corrected for foreground Galactic extinction using the color excess value E(B − V) = 0.066 mag, as reported in NED.¹⁰

The F657N filter allows [N ii]λλ6548, 6584 Å doublet, which needs to be removed to obtain the Hα emission. To account for spatial variations in the [N ii]/Hα ratio, we use the measurements in Bresolin & Kennicutt (2002), who have obtained spectroscopic data for eleven H ii regions in the disk and five regions in the core of this galaxy. Our FoV covers five of the disk regions and all the nuclear regions. We assign the spectroscopic [N ii]/Hα ratio determinations to areas surrounding each H ii region from the Bresolin & Kennicutt (2002) paper. Figure 1 shows the distribution of the adopted ratios, color coded in terms of the total ([N ii]λ6548 Å + [N ii]λ6584 Å)/Hα ratio. For the areas outside the immediate surroundings of each H ii region and for the entire northern pointing (which does not contain any H ii region studied by Bresolin & Kennicutt 2002), we assume the value of Region 11 in their paper, the one closest to the average of the four regions located on the northeastern spiral arm.

Zoom In Zoom Out Reset image size

For our galaxy with super-solar metallicity (12 + log [O/H] = 8.94, Bresolin & Kennicutt 2002; Hong et al. 2013), we adopt an electron temperature T_e = 7500 K and a number density n_e = 10³ cm⁻¹ as the fiducial environment in the whole galaxy, implying intrinsic ratios I_Hα/I_Hβ = 2.92 and I_Hα/I_Paβ = 17.1 (Dopita & Sutherland 2003) (note that our results are not sensitive to the specific choice of the electron density or temperature, see Section 1). We adopt the extinction curve presented by Cardelli et al. (1989), with k(Hα) = 2.535, k(Hβ) = 3.609 and k(Paβ) = 0.840, in the expression I_obs/I_intr = 10^{−0.4k(λ)E(B − V)}. Two extinction maps are then created, one derived from the ratio of Hα to Hβ and the other from that of Hα to Paβ, as shown in Figure 2, using only pixels above the 5σ detection level in all three maps. Our main limitation is the depth of the Hβ line, which is the most dust attenuated one among the three.

Zoom In Zoom Out Reset image size

3.1. Extinction and Dust Content

Figure 3 shows the histograms of A_V derived from the two line ratios. A significant number of pixels in the Hα/Hβ ratio map have negative values of A_V, due to the large error bars (typically ∼1 mag). In general, A_V from Hα/Hβ shows a narrow distribution peaked at ∼0.5 mag, while A_V from Hα/Paβ shows a double peak at ∼2 and ∼5.5 mag. Rebinning our maps with a coarser grid hardly changes the shapes and the peak locations of the histograms (see the color lines in the figure). We also show these histograms in a series of galactic annuli, which clearly demonstrate that A_V (Hα/Paβ) peaks at a larger value than A_V(Hα/Hβ) in every annulus (also see Table 1), and the second peak of A_V(Hα/Paβ) mentioned above is largely contributed by the nucleus, where many pixels with large and patchy dust attenuation are present (e.g., Thatte et al. 2000).

Zoom In Zoom Out Reset image size

Table 1. Derived Dust Properties in M83

Region	$A_V{\rm (\frac{H\alpha _{tot}}{H\beta _{tot}})}$	$A_V{\rm (\frac{H\alpha _{tot}}{Pa\beta _{tot}})}$	$\langle A_V{\rm (\frac{H\alpha }{H\beta })}\rangle$	$\langle A_V{\rm (\frac{H\alpha }{Pa\beta })}\rangle$	log  〈Nd〉
(1)	(2)	(3)	(4)	(5)	(6)
All	1.48	2.65	0.61	4.04	10.07
0–1 kpc	1.75	3.40	1.15	4.65	10.13
1–2 kpc	0.82	2.65	0.23	1.58	9.67
2–3 kpc	1.37	1.46	0.72	1.88	9.74
3–4 kpc	1.33	0.68	0.50	1.14	9.53
4–5 kpc	1.42	0.02	0.31	1.09	9.50

Notes. Column 1: region for calculation. Columns 2 and 3: visual extinction derived from the ratio of the total line emission from the considered region (mag). Columns 4 and 5: median value of the A_V maps (i.e., Figure 2) within this region (mag). Column 6: dust column density converted from Column 5 (cm⁻²). Note that the method for Columns 4–6 is for high resolution studies, while the calculation for Columns 2 and 3 is relevant for analyses where only coarse resolution images (e.g., those of distant galaxies) are available.

Download table as:  ASCIITypeset image

Although the size of dust grains spans a wide range, the extinction in V-band (λ = 5500 Å) is dominated by the grains with a radius a ∼ λ/2π ∼ 0.1 μm, thus we can derive the dust column density using A_V = 1.086 τ_V = 1.086 πa²N_d, and further calculate the dust mass by assuming the solid density of the grain material to be ρ_d = 3 g cm⁻³, a compromise between graphite and crystalline olivine (Draine 2011). Using A_V(Hα/Paβ) which is likely closer to the actual extinction than A_V (Hα/Hβ), we find the dust mass in the whole FoV to be 7 × 10⁵ M_☉. The entire galaxy has ∼24 times of this amount (∼2 × 10⁷ M_☉), as our FoV covers ∼8% of the disk (129 × 115, NED), and a factor of two is to account for the dust in the excluded signal-to-noise ratio (S/N) < 5 regions. This estimation, although an upper limit (these faint regions likely do not contribute that much, and the galaxy center, the most dusty region, is covered by our FoV), is consistent with previous studies using IR and (sub)millimeter data (Galametz et al. 2011 find 8.5 × 10⁶ M_☉; Foyle et al. 2012 find 4 × 10⁷ M_☉), and is similar to the dust content of other metal-rich spiral galaxies (e.g., Liu et al. 2010; Draine et al. 2007).

3.2. Dust Geometry

The availability of three hydrogen emission lines also enables a crude analysis of the dust geometry in the H ii regions, by comparing the observationally derived extinction to models of the emitters/absorbers geometry. The simplest geometry, the foreground screen, in which the dust grains form a homogeneous, non-scattering screen foreground to the emitting light source, is such that any pair of hydrogen lines will yield the same A_V value. This is clearly not the case in M83. When the A_V values derived from Hα/Hβ and from Hα/Paβ are compared to each other, as shown in Figure 4, they do not follow a relation with a slope of unity, as expected in the case of the foreground screen. Instead, the data points are broadly distributed in the upper half of the diagram, delimited by the one-to-one line and the vertical A_V = 0 line.

Zoom In Zoom Out Reset image size

In order to describe the behavior observed in Figure 4, we follow Calzetti et al. (1994) and Natta & Panagia (1984) and calculate the predictions of five simple models with plane-parallel absorber/emitter configurations (for a schematic representation of the models, see Calzetti et al. 1994).

• 1.  
  The foreground screen model, as described above (no clumpiness and scattering is present, as the screen is uniform and physically distant from the light emitter).
• 2.  
  The clumpy dust screen model. Like Natta & Panagia (1984) and Calzetti et al. (1994), we assume all the clumps to have the same optical depth and to be Poisson distributed, with an average of N clumps along the line of sight.
• 3.  
  The uniform scattering slab model. In this case the dust is located close to the emitter and the scattering by the dust grains into the line of sight has an important positive contribution. We follow the calculation procedure of Calzetti et al. (1994), but with the physical parameters of the dust from Witt & Gordon (2000).
• 4.  
  The clumpy scattering slab model. The properties of this model is a combination of models 2 and 3. Same dust parameters are used as in model 3. The same value of N as model 2 has been adopted.
• 5.  
  The uniform mixture model. The dust and emitters are homogeneously mixed together, and the scattering into the line of sight is taken into account.

The relations between the visual extinction derived from these hydrogen lines predicted by the above models are shown in Figure 4. The average number of clumps in models 2 and 4 are set to be N = 6 in the figure. Model 5 has the interesting property that the tip of the line is the limit when the optical depth becomes infinitely large, in which case neither of the three emission lines can explore deep into the H ii regions. None of the five models can, alone, account for the distribution of all data points, even when taking the fairly large error bars into account. The spread in extinction values is significant and spans the range covered by all five models. A significant number (30.4%) of regions with A_V(Hα/Hβ) < 2 mag have extinction values A_V > 2 mag when the latter is derived from Hα/Paβ. For these regions, the traditional use of the Hα/Hβ line ratio to derive extinction values produces a significant underestimate of the intrinsic values, by factors >2 (up to ∼10). This is in line with the conclusions of Israel & Kennicutt (1980) and Skillman & Israel (1988) that deriving dust extinction in H ii regions at longer wavelengths results in larger values than at shorter wavelengths.

The extinction values are almost evenly distributed between the two extreme lines marked by models 1 and 5 (and even beyond those models), suggesting a large variation in the dust geometry among different H ii regions. At the same time, these models clearly encompass the majority of the data, when the data uncertainties are taken into account. The envelope represented by model 1 and model 5 suggests that for most H ii regions/nebulae, the uniform mixture and the foreground screen models bracket the range of absorber/emitter geometric configurations.

Although clearly an over-simplification of the complex dust geometry, model 1 represents 46% of the area of H ii regions well (within 1/2 mag of extinction). The most deviant points are located in the nuclear regions of this galaxy.

Moreover, rebinning our maps onto a very coarse (≳ 20 × 20) grid significantly suppresses the necessity to introduce model 5. For instance, at a sampling scale of ∼180 pc (30 × 30) all data points agree with a dust screen configuration (Figure 4). This is consistent with Kennicutt et al. (2009): when averaged over large sub-galactic regions, the mean extinction becomes consistent with the foreground dust screen model, suggesting that other geometries tend to be more localized.

In spite of the significant uncertainty of the derived A_V, the data points in Figure 4 appear to form two branches: one of which approximately follows the locus of model 5 (the "uniform mixture branch"), while the other favors models 1–4 whose degeneracy between each other is difficult to break and therefore can be represented by model 1 (the "foreground screen branch"). Projecting these two branches to the M83 images will provide an empirical prescription for extinction correction in spatially resolved galaxies.

These two branches can be reasonably divided by the straight line that bisects all the data points A_V(Hα/Paβ) = 1.428 A_V(Hα/Hβ) (the blue line in Figure 4). To investigate the spatial distribution of the data points separated by this line, we calculate the ratio of the area occupied by the points from each branch within a series of annuli with a fixed width of 0.5 kpc starting from the galaxy center. Interestingly, we find this ratio to persist at roughly a constant on either side of a radius of ∼2 kpc, but abruptly switch their roles at this radius (Figure 4). Specifically, in terms of the mean values and standard deviations, we find

$$\begin{equation} \left<\frac{{\rm Area_{uniform \; mixture}}}{{\rm Area_{foreground\; screen}}}\right> = \left\lbrace \begin{array}{@{}l l}1.57\pm 0.34\quad {\rm ({\lesssim }2\, kpc)}; \\ 0.51\pm 0.18\quad {\rm ({>}2 \,kpc)}. \\ \end{array} \right. \end{equation} \tag{ 1 }$$

Hence, we conclude that in the central region of M83 within a galactocentric distance of ∼2 kpc, 61% of the nebular-emitting area follows the dust extinction properties of model 5 (uniform mixture). Beyond this radius, 66% of the nebular-emitting area follows the foreground screen model. This result is particularly useful for pixel-based statistical analyses of spiral galaxies with properties similar to M83, for which we suggest that models 5 (1) be used with a weight of ∼1/3 (2/3) for the center and ∼2/3 (1/3) for the outer disk. Even a more crude prescription—emitters and absorbers are assumed to be uniformly mixed in the center (≲ 2 kpc) and the foreground dust screen is employed for the outer disk—will significantly improve the accuracy of dust extinction correction.

As part of the ERS observations made by the HST/WFC3 SOC, the starburst galaxy M83 has been imaged in multiple narrow bands targeting three hydrogen recombination lines (Hβ, Hα and Paβ). These data enable us to scrutinize the extinction values and geometry in the H ii regions by deriving the A_V map using two combinations of hydrogen lines: Hα/Hβ and Hα/Paβ. The pixel-by-pixel comparison between the two extinction maps shows that larger optical depths are probed by the longer wavelength line emission, yielding A_V values that are larger by ≳1 mag than those derived from shorter-wavelength hydrogen lines. This produces a factor ≳2 discrepancy in the intrinsic Hα luminosity when using the different line ratios for the extinction correction. By comparing these observations to a series of simple models, we conclude that the data require a large diversity of absorber/emitter geometric configurations, but when averaged over large (≳100–200 pc) sub-galactic regions, the mean extinction becomes consistent with the foreground dust screen model, suggesting that other geometries tend to be restricted to more local scales. Moreover, we can provide a simple prescription for improving the extinction corrections in spatially resolved analyses, by expressing the correction in terms of two extreme geometrical dust configurations: (1) foreground non-scattering dust screen, and (2) uniform mixture of emitters and absorbers. These two configurations can be combined in the following proportions: ∼2/3 (1/3) for uniform mixture (foreground screen) in the central area (≲ 2 kpc) and ∼1/3 (2/3) in the outer disk.

This paper is based on Early Release Science observations made by the WFC3 Science Oversight Committee. Support for program GO–11360 was provided by NASA through a grant from Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. We are grateful to the Director of STScI for awarding Directors Discretionary Time for this program.

Facility: HST (WFC3) - Hubble Space Telescope satellite