5 Observational selection effects

It is certainly something of a surprise to find no obvious correlation between dust expelled from the main absorption layer and either HII regions in the disk or enhanced DIG brightness. Even for NGC 4013, which constitutes a possible exception to this statement, only a general association is evident, with chimneys tending to occur over regions of the disk where the DIG is most luminous. There is certainly no detailed correspondance apparent with, for example, individual dust filaments forming a boundary layer to H${\alpha }$ "bubbles'' expanding out of the disk. Observationally, there are severe obstacles in looking for a correlation between H${\alpha }$ emission and extraplanar dust. As far as the dust is concerned we are most sensitive to dust features on the near-side of the disk (given that we have selected on the basis of optical extinction). At the same time, H${\alpha }$ emission emanating from anywhere up to z-heights of 2 kpc is susceptible to extinction from these same features. Indeed, for a typical chimney optical depth of $\tau_{B}\sim1$, 6563 Å emission-line radiation emitted by the DIG will suffer up to 0.6 mag of extinction depending on whether the dust feature is situated in front of, or behind, the ionized gas. This value excludes attenuation by the more general dust lane although we expect this effect to be generally small (for example, with an edge-on, midplane optical depth A_B=10, $A(\lambda =6563~{\rm\AA}) \simeq$ 0.1 mag at 1 kpc or 4 scale-heights above the disk). Such considerations might argue that the recognition of extraplanar dust and the detectablility of DIG emission are mutually-exclusive. The problem is distinctly exacerbated by our necessary selection of edge-on disks. Choosing less inclined galaxies would, admittedly, mitigate the effects of extinction, but at the same time, we could never be sure that the structures we discover are extraplanar.

In the following sections, we try to assess how well our unsharp-masking technique reveals the true distribution of extraplanar grains. We do this in two different ways. Firstly, we use images of the (optically-thin) submillimeter emission recently detected from NGC 891. Secondly, we examine B-band images of the edge-on spiral NGC 55, which at a distance of only 1.6 Mpc offers superior spatial detail over galaxies comprising the R96 sample.

   
5.1 Submillimeter images of NGC 891

Alton et al. (1998a) obtained deep 450 and $850~\mu$m images of NGC 891 with a spatial resolution of 10'' and 16'' respectively (see also Israel et al. 1998). These submillimeter maps, obtained with the SCUBA array on the James Clerk Maxwell Telescope, trace primarily thermal emission from cold dust residing in the edge-on disk (temperature $\sim$17 K). Notably, however, fairly structured submm emission is detected up z=2 kpc in both these images (Fig. 13). Given a nominal FWHM beamwidth of 15'' (700 pc) and 9'' (400 pc) at 850 and $450~\mu$m, respectively, it is reasonable to ask whether this apparent high-z emission can identified with the extinction chimneys identified in our B-band image. Indeed, an initial inspection of the SCUBA data (Alton et al. 1999b) showed that the extended submm emission is much broader than the instrumental PSF, suggesting that it might well correspond to high-latitude dust. To make a more careful comparison between the extraplanar emission and the vertical extinction features, we use the results of the radiative transfer fit given in Table 3. Adopting the parameters derived for the distribution of optical depth, we produce an image of the standard dust layer which, after smoothing with the appropriate PSF, can be compared with the submm maps. Our objective here is to identify an excess in the submm maps which might be ascribed to the extraplanar extinction features. We utilize calibration images of Uranus, taken at regular intervals during the SCUBA observing schedule, to produce a signal-weighted map of the PSF at both 450 and 850 $\mu$m. Before convolving the model galaxy with the measured PSF we ensure that the chop direction is aligned for both the object and the beam images. This is important because the chop for NGC 891 was along the minor axis and the side lobes of the beam are known to be more pronounced along the chop direction. Having smoothed, in turn, the simulated disk with the respective 450 and $850~\mu$m PSF, profiles in z-height were created, for both wavelengths, at various locations along the major axis.

Figure 11: Extraplanar dust mass plotted against H${\alpha }$ luminosity in the disk. We have integrated the emission line flux and dust mass over 1 kpc intervals along the major axis. The graphs are arranged in the following order: top-left (NGC 891); top-right (NGC 4013); bottom-left (NGC 4302); and bottom-right (UGC 4278)

Figure 12: Extraplanar dust mass plotted against H${\alpha }$ luminosity of the diffuse ionized gas (DIG). We have integrated the emission line flux and dust mass over 1 kpc intervals along the major axis. The graphs are arranged in the following order: top-left (NGC 891); top-right (NGC 4013); bottom-left (NGC 4302); and bottom-right (UGC 4278)

Figure 13: Submillimeter emission from the disk of NGC 891. We show the central part of the galaxy at $850~\mu$m (left) and $450~\mu$m (right) with contours beginning at $3\sigma$(10 mJy/16'' beam and 40 mJy/10'' beam respectively). In each image, the beamsize is represented by a circle at the bottom left and adjacent isophotes are separated by a factor $\surd$2 (0.4 mag). The galaxy is orientated with North at the top and East to the left

Figure 14: Profiles at $850~\mu$m (top figure) and $450~\mu$m (bottom plot) perpendicular to the major axis of NGC 891. In each case, the observed data are denoted by the solid circles and the exponential dust model (e-height = 0.21 kpc) is represented by a dashed curve. The PSF of the beam, as measured from calibration maps of Uranus, is given by the dotted line

Figure 15: Submillimeter emission from NGC 891. $F_{450~{\mu}{\rm m}}$ and $F_{850~{\mu}{\rm m}}$represent, respectively, the fraction of 450 and $850~\mu$m flux detected at heights greater than 1 kpc above the stellar disk. The data are sampled at 500 pc intervals along the major axis

It was found that the widths of the simulated profiles were considerably larger than the observed data. After several trials, we established that a modified exponential z-height of 0.22 kpc gave a far more consistent fit between model and observation (at both 450 and $850~\mu$m and for all locations along the disk). This value is somewhat smaller than the original scale-height of 0.31 kpc but it seems likely that B-band radiation transfer modelling of NGC 891 will over-estimate the height of the general dust layer by incorporating high-z extinction features into the photometric fit (they are particularly conspicuous in a blue filter). Figure 14 shows the simulated profiles, with a 0.22 kpc scale-height, alongside measured profiles of regions of the disk exhibiting high-z structure in the submm image. Our conclusion from this figure is that the detected submm emission even at 30-60'' (1.5-3.0 kpc) is attributable to the main dust layer. The wings of the PSF appear to amplify the tail of the exponential distribution so that spurious high-z "features'' are produced which extend up to 3 beamwidths from the midplane. To clarify the situation we also ascertained the fraction of emission occuring at z>1 kpc in both the 450 and $850~\mu$m SCUBA images. This quantity can be expected to vary at different locations along the disk if it is not attributable to the wings of the PSF. Moreover, the high-z flux should correlate at both 450 and $850~\mu$m if it has a physical, rather than an instrumental, origin. Figure 15 demonstrates quite clearly that a constant fraction of 450 and 850 $\mu$m flux is recorded at z>1 kpc severely undermining our precursory impression that the submm filaments might represent high-latitude dust. The detectability of high-z extinction features in thermal emission is, in fact, doubtful if one considers that a chimney of intrinsic $\tau_{B}\sim1$ will be smoothed out parallel to the major axis by a 16'' beam in the $850~\mu$m image. The $\tau_{B}$ that is then effectively measured is $\sim 0.05$ or 0.5 mJy/beam for cold (17 K) dust (Alton et al. 1999b). The anticipated emission level, therefore, is an order of magnitude smaller than that recorded at z=1 kpc from the (convolved) main dust layer (Fig. 14).

Although it is disappointing not to be able to use submm images to infer the true distribution of high-z dust, we are still in a position to place an upper limit on the total amount of material in the extraplanar layer. Refering once again to Fig. 14, the observed submm emission seldom deviates more than $2\sigma$ from the simulated profiles. We therefore attribute, as an upper limit, $2\sigma$ of the flux to grains located outside the standard exponential layer (5 mJy/16'' beam and 16 mJy/10'' beam at 850 and $450~\mu$m respectively). Integrating over a z-height of 1-2 kpc we find that <5% of both the 450 and $850~\mu$m flux can originate from high-latitude dust. Thus, with the proviso that high-z grains are not much colder than 17 K, the SCUBA observations indicate that less than 5% of galactic dust exists outside the notional absorption layer. For lower grain temperatures this upper limit must be relaxed somewhat - dust temperatures of 10 K would allow up to 9% of galactic dust to reside outside the disk. Considered together, however, our optical and submm images for NGC 891 tend to constrain the fraction of dust outside the main extinction lane to $\sim1-5$%. If dust chimneys possess solar-type gas-to-dust ratios, our results imply that only a few percent of the neutral gas present in spiral galaxies reside above the main disk.

   
5.2 The nearby edge-on galaxy NGC 55

At a distance of 1.6 Mpc (Puche et al. 1991), NGC 55 is ten times closer than the galaxies comprising the Rand (1996) sample (Table 1). Under normal conditions of optical seeing, this object provides a remarkable opportunity to resolve regions down to a physical size of 13 pc - a region much smaller than the anticipated width of dust chimneys (30 pc). NGC 55 is a late-type galaxy classified as SB(s)m (de Vaucouleurs et al. 1991). With a D₂₅ size of 18 kpc, it is physically smaller than most of the spirals in our sample (cf. NGC 891 with a D₂₅ corresponding to 37 kpc) and its overall optical appearance is more irregular than a typical $L_{\star}$ disk. In emission-line radiation, NGC 55 possesses bright, knotted structure in the disk and a spectacular network of curvi-linear filaments extending up to 2.6 kpc away from the midplane (Ferguson et al. 1996; Bomans & Grebel 1993). The DIG is most conspicuous within the inner 3-4 kpc of the galaxy where the extraplanar H${\alpha }$ structures correspond well to fragmented shells of gas presumably swept up by superbubbles expanding from recent enhanced star-formation in the central disk.

To compare with the H${\alpha }$ image taken by Ferguson et al. (1996), we obtained archival CCD data of NGC 55 in a blue filter. Since this object is located in the Sculptor Group (Dec = $-39^{\circ}$) the only image that could be retrieved was a 120 second exposure of the central $9' \times 9'$ region, which had been taken on the 4-m Anglo-Australian Telescope. These archival frames were reduced in the standard manner and an unsharp-mask B-band image created using the same method adopted for the R96 sample. Figure 16 compares the extinction evident in the unsharp-mask image with the pronounced network of H${\alpha }$ filaments located at the centre of the galaxy. We have superimposed on the disk a box to denote the apparent centroid of the most prominent H${\alpha }$ shells. The dust does not extend to such large distances from the midplane as the diffuse emission-line gas (only up to z=700-1000 kpc in fact). However, there is a suggestion that, perhaps, at either side of the active centre, the dust clouds have been prised away from the midplane by the superbubble's impact on the main gas layer. If this is the case, the situation would be reminiscient of M 82, where the central starburst cavity levers ambient dust and gas away from a rather chaotic-looking absorption layer (Ichikawa et al. 1994).

We should point out that the dust lane in NGC 55 assumes an equally non-uniform, distorted shape in the outlying parts of the disk as it does close to the central star-forming activity. This can be inferred from the extensive R-band image of Ferguson et al. which was used in the subtraction of the continuum from their H${\alpha }$ filter. Indeed, the maximum z-height at which dust manifests itself through R-band absorption varies little along the major axis and is typically no more than $\sim 1$ kpc. Parallel to the disk, the horizontal width of the "vertical'' absorption structures is generally $\geq$40 pc which is consistent with the width of dust chimneys in NGC 891, as measured by the Hubble Space Telescope (HS97). The fact that vertically-extending dust structures do not appear to be concentrated where the DIG is brightest might be considered indicative that disruption in the disk and halo occur on different timescales. A chaotic dust lane may be the remnant of a more widespread epoch of star-formation whilst the brightest H${\alpha }$ filaments above the central disk represent a more recent episode of stellar creation. We elaborate these ideas in the next section.

Although we obtain significantly more spatial detail in using NGC 55, our reliance on unsharp-masking to identify concentrations of high-latitude dust means that we are always sensitive to the relative geometry between stars and dust. The proximity of NGC 55, however, makes it an ideal candidate for future submm/mm imaging where sampling of the optically-thin thermal emission from high-z grains can be carried out.

Figure 16: Images of the inner disk of NGC 55 ( $6' \times 6'$ or 2.8 kpc $\times$ 2.8 kpc). At the top, we show part of the deep emission-line exposure (H${\alpha }$ +NII) obtained by Ferguson et al. (1996). The same area is shown as an unsharp-mask B-band image at the bottom. In both cases, a box marks the centroid of the most extensive, high-z H${\alpha }$ filaments