VLA-ANGST: A HIGH-RESOLUTION H i SURVEY OF NEARBY DWARF GALAXIES

The parameters for our new observations and the subsequent data reduction strategy closely followed the design of the THINGS survey (Walter et al. 2008), with the goal of obtaining comparable sensitivity and spatial resolution. A significant difference between the two surveys, however, is the ∼5 times better velocity resolution of VLA-ANGST. Each of the VLA-ANGST galaxies was observed with the NRAO's VLA in the B (9 hr total observing time per galaxy), C (3 hr), and D (3 hr) array configurations (under project code AO215). The compact D-configuration exhibits the largest number of short baselines with antenna separations down to 35 m. The D-configuration is thus the most sensitive antenna configuration to image spatial scales of up to ∼16', which is the maximum for the VLA at 1.4 GHz. The addition of B-configuration observations yield spatial resolutions of ∼6'' or ∼100 pc for the nearby (D ∼ 3 Mpc) objects. This scale is necessary to compare star forming regions with their H i counterparts and matches well the resolution of surveys at other wavelengths such as LVL (Spitzer) and 11HUGS (GALEX). The southern sources NGC 3109, NGC 247, Antlia, and DDO 6 were observed mostly in the hybrid BnA-, CnB-, and DnC-array configurations that feature elongated placements of the antennas along the northern arm. The projected baselines of these antenna configurations produce a more circular beam toward southern targets.

At the time of the observations, 2007 October to 2008 August, the VLA was in the process of being upgraded as part of the EVLA project. This transition implied that the observations were taken with a mix of already converted EVLA and original VLA antennas. The conversion period mostly affected the signal distribution from the front-ends to the correlator. At the time, the old VLA correlator was still in use, and we configured it in the 2AC or 2AD modes to capture both RR and LL polarization products. Doppler tracking in the transition phase would have introduced phase jumps on baselines involving EVLA antennas. Consequently, we calculated and fixed the appropriate observing frequency for each observing run such that the H i emission line was well placed within a VLA bandpass at the start of an observation (Doppler setting). Line shifts during a single observation are minimal (<0.5 km s⁻¹) and are corrected in post-processing (Section 3.2).

The observational parameters for each galaxy were based on its H i linewidth as taken from single dish H i spectra, plus ∼20% additional line-free channels to enable a good continuum subtraction. To reach our goal of the best possible velocity resolution, we used a correlator mode with 0.78 MHz bandwidth and 256 channels for 15 galaxies in our sample. This corresponds to a channel width of ∼0.65 km s⁻¹ over a total velocity range of ∼120 km s⁻¹ after cropping about 20% edge channels, 10% on each side. Fourteen galaxies had wider H i linewidths and were therefore observed with a bandwidth of 1.56 MHz and 256 channels (∼300 km s⁻¹ velocity range and ∼1.3 km s⁻¹ channel width; see Table 3). The systemic velocity of KK 77 is unknown. To maximize on the range of velocities for this source, we used the 4IF mode, a correlator mode that enables observing with two simultaneous frequencies. NGC 247 was also observed in the 4IF mode because of its extreme H i linewidth of ∼200 km s⁻¹, by far the widest line of our sample. In the presence of extremely strong, narrow line features, the response of the VLA can include signs of the Gibbs phenomenon, a sinc function that oscillates between channels. This oscillation can be suppressed by "Hanning" smoothing the data with a triangular smoothing kernel. We decided not to use online Hanning smoothing as the lines are too weak to show any signs of the Gibbs phenomenon. Some of the archival observations that we added in, however, were observed with online Hanning smoothing turned on. We also applied a 25 MHz front-end filter to reduce the impact on radio frequency interference (RFI) in our data.

For flux and bandpass calibration purposes, we observed the VLA standard flux calibrators 3C286 and/or 3C48 (with fluxes of ∼15.0 Jy and ∼16.5 Jy, respectively) depending on their visibility at the time of the observations. Typical integration times were 12 minutes on the flux calibrator, split between the beginning and end of a track. We observed our target galaxies in 40 minute intervals alternating with 3 minutes on a nearby phase/complex gain calibrator. The complex gain calibrators were chosen to be nearby point sources with a minimum flux of ∼1 Jy. B-configuration observations were obtained in single 9 hr programs, or, in the case of southern galaxies, two 4.5 hr observations. In the C- and D-configurations, we combined sources into a few observations to reduce overhead and to obtain an improved uv-coverage.

During the VLA to EVLA transition time, the EVLA–EVLA baselines showed considerable aliasing and were largely unusable. Since the D-configuration observations had the largest number of EVLA antennas and therefore EVLA–EVLA baselines, we obtained an additional hour of observing time per object to reach our original sensitivity goals. Overall we obtained a total of 3 hr in D-configuration per galaxy. The aliasing is a monotonic function that affects the first ∼0.5 MHz (or ∼100 km s⁻¹ at the H i frequency), decreasing from the bandpass edge toward the center.¹⁰ In addition to the make-up time and using mostly baselines with old VLA antennas, we placed the H i lines away from the affected frequency ranges but could not fully eliminate the effects. As a result, the noise levels in our data products are not entirely uniform across all channels and increase toward lower frequencies where aliasing is strongest. Across the full width of our most narrow 0.78 MHz bandwidths, the noise can vary up to 35%. For the smaller width of the spectral line feature, the effect is smaller. In addition, for part of the observations, we tried to place the H i signal away from the most affected frequency ranges and for most galaxies the noise level changes could be constrained to an rms variation of ≲ 15 % over the width of the H i line. For the two galaxies that were observed in the 4IF mode, KK 77 and NGC 247, the full bandwidths were required to cover the requested velocity ranges. The aliased signal appears in each IF and, consequently, the full noise variations of up to 35% are visible across the combined spectrum. Overall, however, this will not have a huge impact on most of the data analysis; KK 77 is a non-detection, and NGC 247 is one of our brightest objects such that the signal-to-noise ratio is only moderately affected.

A few galaxies (marked "a" in Table 2) had H i emission whose velocity range overlapped with Galactic H i line features. We addressed this problem by observing the bandpass calibrator at two different frequencies, offset by ±4 MHz from the source frequency, in order to interpolate the calibration to the source frequency.

In addition to the new observations, in preparing for the program, we identified a few archival VLA observations which would be of use to the program. Most archival data, however, were observed at the relatively lower spectral resolution of 2.6 km s⁻¹ with online Hanning smoothing applied. To avoid interpolation, we kept the lower velocity resolution of the archival data and rebinned our new VLA-ANGST observations after calibration for the final cube.

Table 2 lists all of our observational setups. Column 1 lists the galaxy names followed by the array configurations of the observations and the project codes in Columns 2 and 3. AO215 is the genuine VLA-ANGST project, while other project codes refer to archival data. The observing dates are listed in Column 4. Asterisks denote observations that span across midnight, for which we refer to the start dates. The equatorial pointing positions are provided in Columns 5 and 6 and the appropriate equinox is tabulated in Column 7. Column 8 gives the phase calibrators that were used for the observations. The correlator setups are displayed in Columns 9 through 14 with Column 9 the correlator modes, Column 10 the bandwidths, Column 11 the number of channels, and Column 12 the channel widths in km s⁻¹ (for the rest frequency of H i at 1.420405752 GHz). VLA-ANGST data were taken at the fixed sky frequencies listed in Column 13, whereas many archival observations were Doppler tracked at the velocities listed in Column 14. Finally, Column 15 provides the number of converted EVLA antennas in the dataset with the remaining antennas being old, not yet converted, VLA antennas at the time.

Data calibration was performed using the AIPS¹¹ package and deviated from the "standard" VLA data reduction procedure due to the effects from the VLA to EVLA transition period. As mentioned above, EVLA–EVLA baselines showed considerable aliasing that affected primarily the narrow bandwidth observations. Our reduction scheme followed the following steps.

To avoid low signal-to-noise solutions during calibration, we started by eliminating edge channels. About 20% of the entire bandwidths were cropped, ∼10% at the upper and ∼10% at the lower frequency ends (task UVCOP). After correcting the absolute antenna positions (VLANT), the absolute flux scale of the primary flux calibrators 3C286 and 3C48 (see Section 3.1) were calculated by the task SETJY using NRAO-provided models and "Perley-Taylor 99" flux scales. We manually inspected all calibrator data in each array configuration for bad visibilities due to RFI or cross-talk between antennas (AIPS tasks TVFLAG, SPFLG, UVFLG, WIPER). If calibrators showed higher than expected signal on short baselines due to solar interference, we excluded baselines with a uv range between 0 and 1 kλ in the calibration. Solar RFI was usually well removed by continuum subtraction in the source visibilities, so we performed no additional source flagging due to solar interference.

The data were then bandpass-corrected by deriving a channel-based, normalized gain solution from the flux calibrator data via BPASS. We produced a new broadband "channel zero" (CH0) map from these bandpass-corrected data (AVSPC), and utilized the new CH0 map in all subsequent calibration steps. Next, we calculated the complex antenna gain as a function of time for all calibrators (CALIB). The complex gain is a solution for both gain and phase, and for all solutions we assumed that the complex gain calibrators are point-like. With CLCAL we linearly interpolated all phase/gain solutions across all time intervals. GETJY transferred the absolute flux calibration to the complex gain calibrator. At this stage, we manually inspected the quality of the calibration and repeated the above procedure if RFI corrupted the solution and further flagging was required. Finally, we applied all calibration solutions to the target galaxies. We estimate the calibration flux uncertainties to be ∼5%.

For galaxies whose H i emission is within the velocity range of the foreground Galactic emission (marked "a" in Table 2), the flux and bandpass calibrators were observed at offsets of ±4 MHz relative to the frequency of the targets. The calibration of these data followed the same steps as above with each observed frequency calibrated separately. We then obtained an interpolated flux and bandpass solution calibration from a linear interpolation across these offset frequencies. The EVLA uses different internal local oscillator settings from the VLA. For our observations during the VLA to EVLA transition phase, this difference had the unfortunate effect that, for some observations, the 25 MHz front-end filter overlapped in frequency with the upper frequency offset bandpass observations. This resulted in an extreme phase gradient across that particular offset frequency and the affected observations were unusable. In these cases, we calibrated the source data by extrapolating the calibration from the single offset frequency that most closely matched the bandpass derived from the phase calibrator. This method provided bandpasses that are accurate to a few percent. For observations where the complex gain calibrator was contaminated by Galactic H i, we simply flagged the affected channels before calculating a new CH0 to be used for the gain and phase solutions.

After applying all calibration tables to the source, the galaxy data were separated (SPLIT) from the flux and phase calibrator data for further processing. Both NGC 247 and KK 77 were observed in the 4IF mode. We calibrated each of the two IF windows separately as described above, and "stitched" the data together, averaging overlapping velocity ranges in the process (UJOIN) prior to continuum subtraction.

To determine the continuum level, we fit a linear function to the line-free channels, typically 20 channels on each side of the spectrum. The fit was then subtracted from the complex uv-data (task UVLSF). Since the frequencies for each observing setup were fixed (see Section 3.1), we regridded the H i data to a common heliocentric, optical velocity system (CVEL). Finally, data from all observations and array configurations were combined (DBCON) into a single data set for imaging. We produced dirty images with IMAGR and inspected the cubes for artifacts. If further flagging was required, we went back to the original source data, flagged, and reapplied the calibration.

After satisfactory calibration and source editing, we used the AIPS task IMAGR to generate data cubes and final data products. We followed the THINGS protocol when possible so that the two data sets could be easily compared. For each VLA-ANGST galaxy, we imaged the visibilities with two different weighting schemes: one using natural weighting and one using the "robust" weighting (originally described by Briggs 1995, with small modifications as described in the AIPS help files). Natural weighting yields high sensitivity at moderate resolution (typically ∼6''–12'' for our galaxies, or about ∼90–170 pc for a distance of 3 Mpc), while robust weighting decreases the size of the synthesized beam at the cost of reduced surface brightness sensitivity. We applied a robust parameter of 0.5, which was found to be a good compromise between resolution and sensitivity and matches the maps generated by THINGS. When compared to the naturally weighted cubes, the noise levels of the robust-weighted cubes is typically ∼20% higher and the beam sizes ∼40% smaller. Depending on the angular extent of each galaxy's H i emission, the cubes were imaged with either 1024² pixels at 15 per pixel or 2048² pixels at 10 per pixel. The cubes were deconvolved using the Clark Clean deconvolution algorithm (Clark 1980), stopping at a residual flux threshold of 2.5 times the noise level as measured in the cubes. Finally, we produced primary beam corrected data cubes that were later used in the moment map analysis (Section 4).

The properties of all data cubes are listed in Table 3 where Column 1 lists the galaxy names followed by Columns 2, 3, 4, and 5 that contain the weighting algorithms and the resulting beam major and minor axes sizes as well as the position angles of the deconvolved data. The average root mean squared noise values per channel are shown in Column 6, the channel widths in Column 7, the number of pixels in each plane in Column 8 followed by the pixel sizes in Column 9.

The most important difference between the VLA-ANGST THINGS data cubes is the higher velocity resolution in the majority of the VLA-ANGST data cubes which was possible because of the overall smaller range in velocity of detectable H i in the VLA-ANGST sample of galaxies. The sensitive and spatial resolution of THINGS and VLA-ANGST are very similar. The spatial resolution of LITTLE-THINGS and SHIELD are similar to VLA-ANGST, too, since all surveys share the same VLA array configuration setups. Given the wider lines of the somewhat more massive galaxies in LITTLE-THINGS (see Figure 2), the bandwidth had to be slightly increased for this survey and their velocity resolution hovers around the upper limit of ours at ∼2 km s⁻¹. On the other hand, LITTLE-THINGS has about 1/3 longer integration times than VLA-ANGST. The two Westerbork surveys HALOGAS and WHISP exhibit two to five times lower spatial resolution than the VLA surveys at which their surface brightness sensitivity limits become comparable. They are observed at a lower velocity resolution of ∼5 km s⁻¹. FIGGS has a velocity resolution of ∼1.5 km s⁻¹ and a beam two to eight times larger than VLA-ANGST.

To ensure that our data cubes would be as directly comparable to the THINGS data cubes as possible, we reduced THINGS observations using our calibration, mapping, and deconvolution protocols. Comparisons of our reductions of THINGS observations with the publicly available THINGS data cubes showed no significant differences.

To suppress noise for the production of moment maps, we generated image cube masks that defined regions containing detectable H i emission from the galaxies. To do so, we convolved the natural-weighted images to twice the original beam major axis and applied spectral Hanning smoothing with a three channel wide kernel. New "mask" cubes containing predominantly H i signal were constructed by keeping all regions corresponding to H i emission above the 1σ noise level and blanking all other regions. To remove the effects of sidelobes, noise spikes, or other spurious signals from the masks, any individual region with an area smaller than the beam size was automatically removed. We then eliminated any remaining non-emission regions by eye inspection.

The result is a single mask cube per galaxy that we applied prior to the generation of the integrated spectra and moment maps. The same mask is used for both the natural and robust data cubes. New cubes containing predominantly H i signal were constructed by keeping all H i emission corresponding to unblanked regions in the mask cube. For our data, this method discriminated well between significant, low-level emission and pure noise. Note that this process produces a lower limit to the total H i emission. Some very low surface brightness H i may be have been eliminated from these data cubes.

Note that mask generation is not entirely automatic, and, therefore, our mask generation cannot be said to be strictly following THINGS protocols. This is exacerbated by the differences in typical channel widths, which leads to differences between VLA-ANGST and THINGS in average noise levels in the individual channels. Nonetheless, we have followed the intentions of the THINGS project to use masking to suppress the noise and to provide optimal moment maps. However, one should be aware that small differences in masks are possible and that the resulting moment maps have a small inherent systematic uncertainty.

Recovery of the total H i flux from each channel and the resulting H i spectra is more complicated than simply summing up the total emission. Jörsäter & van Moorsel (1995) have shown that standard Clean maps do not in actuality yield correct flux measurements. Maps in AIPS are created by summing the Clean components, convolved with the Clean beam, to the signal that is still present in the residuals. While both maps are purportedly measured in Jy (beam area)⁻¹, the relevant beam is different in each map: the convolved Clean component map is measured in Jy (Clean beam area)⁻¹ while the residual map has units of Jy (dirty beam area)⁻¹. Because the areas of the Clean beam and the dirty beam are different, the flux in the Clean components and in the residuals must be corrected to obtain the proper H i flux. A full discussion of the following correction technique is given in Jörsäter & van Moorsel (1995). The corrected flux of a channel is given by

$$\begin{equation} G = C + \epsilon \times R, \end{equation} \tag{ 1 }$$

where G is the corrected flux, C is the flux in the convolved Clean components (with units of Jy (Clean beam)⁻¹), R is the flux in the residual map (with units of Jy (dirty beam)⁻¹), and is the correction factor that takes into account the ratio of the dirty beam area to the Clean beam area. IMAGR provides a method to automatically apply this correction and, following the THINGS protocol, we determined within the inner 50 × 50 pixels of the dirty and Clean beams. This produces a set of two new natural and robust weighted cubes with the above correction applied in addition to our standard cubes. When the residuals are scaled by , the noise in the corrected cubes is artificially suppressed. We thus produced two sets of data cubes for different analyses.

• 1.  
  "Standard" cubes: the standard output from our pipeline, with uncorrected H i fluxes but uniform noise properties. No primary beam correction is applied to these data. These cubes should be used for any analysis that requires uniform noise properties or uses selection based on noise (e.g., fitting of individual profiles in order to construct velocity fields or measure profile shapes).
• 2.  
  "Rescaled" cubes: the cubes with the flux correction applied, to be used only in regions with genuine H i emission. The flux values in this cube are correct, and therefore any analysis that requires selection based on H i fluxes should use these cubes (e.g., mass and column density measurements). The "rescaled" cubes are corrected for the attenuation from the primary beam.

The data products that we make available follow this recipe; all global H i spectra (Section 3.6) and moment maps (Section 4) were derived using the masked, "rescaled cubes." The online data cubes themselves are the "standard" cubes, without primary beam attenuation or flux corrections.

Global H i spectra are derived from the masked, rescaled data cubes. The spectra are used to derive velocity widths at 20% (w₂₀) and 50% (w₅₀) of the peak. The central H i velocity of each galaxy is calculated by taking the midpoint of the w₂₀ boundaries.

We also use the integrated H i spectra to calculate the total H i masses of our galaxies using the following equation:

$$\begin{equation} M_{{\rm H\,\mathsc{\sc i}}}\left[ M_{\odot }\right] = 2.36 \times 10^5 \; D^2 \times \sum _i S_i \Delta v, \end{equation} \tag{ 2 }$$

where D is the distance to the galaxy in Mpc (as given in Table 1) and S_iΔv is the total flux of a single channel in Jy km s⁻¹ (e.g., Rohlfs & Wilson 2004). This formula assumes that the H i emission is optically thin, an assumption that is valid over a large flux range and may begin to fail at very extreme column densities of ≳ 10²² cm⁻² (e.g., Allen et al. 2012; Braun 2012). At our spatial resolution, we do not observe column densities of this magnitude.

In Table 4, we present all of the derived H i parameters starting with the galaxy names in Column 1, followed by the integrated H i flux densities $S_{{\rm H\,\mathsc{i}}}$ and the derived H i masses in Columns 2 and 3. For comparison, we compiled single dish fluxes $S_{{\rm H\,\mathsc{i}}}^{\rm SD}$ from the literature and list them in Column 4. The w₂₀ and w₅₀ values as well as the central velocities are shown in Columns 5, 6, and 7, respectively, followed by the peak H i column densities (Section 4.3.1) taken from the natural-weighted maps in Column 8. To derive upper limits for the non-detections, we assume an H i disk the same size as the optical diameter D₂₅ and a hypothetical linewidth of 20 km s⁻¹. Galaxies typically exhibit H i dispersions of 5–10 km s⁻¹, and a linewidth of 20 km s⁻¹ thus implies little rotation or face-on orientation.

Table 4. Galaxy H i Properties

(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Galaxy	$S_{{\rm H\,\mathsc{i}}}$	$M_{{\rm H\,\mathsc{i}}}$	$S_{{\rm H\,\mathsc{i}}}^{\rm SD}$	w20	w50	vcen	Peak $N_{{\rm H\,\mathsc{i}}}$
	(Jy km s−1)	(106 M☉)	(Jy km s−1)	(km s−1)	(km s−1)	(km s−1)	(1021 cm−2)
NGC 247	382.6	1106.2	608a	201.3	193.9	163.7	5.4
DDO 6	1.2	3.2	3.7b	20.9	13.7	292.5	0.9
NGC 404c	66.7	146.4	76d	80.5	63.2	−54.0	0.5
KKH 37e	<3.4	<0.8	1.8f,g	...	...	...	<0.08
UGC 4483	12.0	32.8	13.6h	51.2	34.3	153.9	3.2
KK 77e	<4.4	<2.3	<5.5i	...	...	...	<0.06
BK3N	6.3	22.0	<0.75j	44.4	20.0	−42.5	0.7
AO 0952+69k	61.3	206.6	<0.59j	56.0	45.6	112.8	1.3
Sextans B	91.0	41.5	72.9h	58.1	40.6	302.2	2.6
NGC 3109	720.9	270.1	1148a	127.7	116.0	405.1	6.6
Antlia	1.4	0.5	1.7l	23.4	13.4	363.4	0.3
KDG 63e	<4.2	<1.1	<0.2m	...	...	...	<0.06
Sextans A	138.1	62.1	169n	59.8	46.2	324.8	6.1
HS 117e	<1.7	<0.6	...	...	...	...	<0.03
DDO 82	0.8	2.8	<0.62j	35.8	26.7	56.2	0.9
KDG 73	0.1	0.5	1.0h	9.2	8.5	116.3	0.1
NGC 3741	32.8	81.1	44.6o	85.4	70.6	229.1	3.4
DDO 99	29.7	46.9	47.1o	51.6	28.6	242.1	2.6
NGC 4163	4.8	9.3	9.6p	33.5	22.7	161.6	2.1
NGC 4190	15.5	44.8	23.2p	73.2	52.8	227.0	3.5
DDO 113e	<1.6	<0.4	23.6p	...	...	...	<0.04
MCG +09-20-131	3.1	11.9	5.2q	39.0	26.1	157.6	3.3
DDO 125	18.3	28.7	21.8h	39.7	27.0	196.1	2.1
UGCA 292	12.9	40.0	14.3h	37.1	25.2	308.3	4.2
GR 8	5.8	5.9	7.8h	30.7	21.4	213.7	1.7
UGC 8508	12.3	19.3	14.8p	62.7	48.1	59.8	2.9
DDO 181	10.5	24.4	12.5o	52.1	40.8	201.4	1.7
DDO 183	8.2	20.1	9.6p	42.2	26.4	191.2	2.2
KKH 86	0.1	0.1	0.5h	7.7	6.9	285.5	0.2
UGC 8833	5.9	13.1	6.0h	41.0	29.4	225.9	2.2
KK 230	0.8	0.7	2.6h	17.4	11.5	60.6	0.6
DDO 187	10.1	11.6	12.0h	46.0	31.8	152.2	3.2
DDO 190	22.5	41.3	8.5o	62.3	45.2	148.8	3.6
KKR 25e	<1.0	<0.1	2.2g,r	...	...	...	<0.03
KKH 98	2.2	3.3	4.1h	25.5	17.0	−137.8	0.8

Notes. ^aHuchtmeier & Richter (1989); ^bMeyer et al. (2004); ^cNGC 404 is contaminated by foreground Milky Way H i emission; ^dBaars & Wendker (1976); ^eVLA-ANGST non-detection. Limits based on a width of 20 km s⁻¹ and the optical diameter D₂₅; ^fKarachentsev et al. (2001); ^gMight be Galactic H i; ^hHuchtmeier et al. (2003); ⁱHuchtmeier et al. (2000b); ^jvan Driel et al. (1998); ^kAO 0952+62 is contaminated by M81 tidal H i emission; ^lBarnes & de Blok (2001); ^mSchneider et al. (1990); ⁿBarnes & de Blok (2004); ^oSpringob et al. (2005); ^pKoribalski et al. (2004); ^qPustilnik & Martin (2007); ^rHuchtmeier et al. (2000a).

Download table as:  ASCIITypeset image

In ∼70% of all cases the single dish fluxes are somewhat larger than the interferometric VLA flux measurements. This difference is expected to some level given that the VLA can only image structures with an extent of up to ∼16' in the D-configuration at 1.4 GHz (limited by the minimum distance between two antennas). Missing flux may therefore only be a significant issue for the most extended objects in our sample. Some single dish flux measurements deviate substantially from the trend of being slightly larger than the VLA fluxes. The deviations can be either way: galaxies like DDO 6, UGC 4483, DDO 113, and KK 230 have much larger single dish measurements whereas others like BK 3N, AO 0952+69, Sextans B, DDO 82, and DDO 190 have smaller single dish fluxes. Such discrepancies may be explained by difficulties in single dish baseline subtraction or by the larger single dish beam that tends to pick up larger fractions of Galactic H i emission as well as flux from nearby objects.

We used the naturally weighted data cubes to derive H i spectra (Section 3.6) for our galaxies given their higher surface brightness sensitivity compared to the robust-weighted data. This approach captures as much low-level, extended emission as is possible with interferometric VLA data. All fluxes are calculated from the rescaled cubes described in Section 3.5 and are presented in Figure 3. In the case of NGC 247, velocities around 108 km s⁻¹ were strongly contaminated with RFI; to estimate the H i flux in this channel, we interpolated the emission from the adjacent channels.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Since our sample is composed primarily of dwarf galaxies, the galaxy spectra typically show narrow, singly peaked profiles. Extreme cases like KDG 73 and KKH 86 exhibit linewidths of <10 km s⁻¹, which implies very little velocity dispersion, maybe due to low signal-to-noise ratio in the line wings. On the other end of the mass spectrum, a few galaxies (NGC 404 NGC 3741, DDO 190, Sextans A, DDO 181, NGC 3109, NGC 247) exhibit hints of the more familiar double-horned profile expected from larger spiral disks. The maximum linewidth is observed in NGC 247 with w₂₀ ∼ 200 km s⁻¹.

Channel maps of the galaxies are presented in Figures 4–32 (natural weighting). Given the high spectral resolution of our data, there is typically only little flux in each velocity bin. Some galaxies, mainly the more massive ones such as NGC 247 or NGC 3109 show the classic "butterfly" pattern of a spiral galaxy, a tell-tale sign for a flat rotation curve. The bulk of galaxies exhibit rotation despite the fact that the dispersion adds a stochastic component to the velocity structure. NGC 247 also features a H i absorption feature along the line of sight to the background quasar NVSS J004713-205114 at R.A. (J2000) = $00^{{\rm h}}{\rm :}47^{{\rm m}}{\rm :}13\mbox{$.\!\!^{\mathrm s}$}6$ and decl. (J2000) = −20°:51':15'' (e.g., Dickey et al. 1992). Some data cubes are contaminated by Galactic foreground emission, but only for NGC 404 is the Galactic H i close to the systemic velocity of the source. Other data cubes, such as that for AO 0952+69, contain emission from a nearby object. AO 0952+69, in fact, is likely not a real galaxy but might be a feature within a spatially coincident spiral arm that belongs to the massive M81 galaxy.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We used the AIPS task XMOM to generate moment maps from the masked, flux-corrected cubes. For all calculations, we require that each pixel in a moment map is calculated from at least four unmasked channels; pixels with fewer channels are masked in all moment maps.

4.3.1. Integrated H i Maps

Integrated H i column density maps are created from the masked, rescaled data cube by integrating along the velocity axis to generate the moment 0 map:

$$\begin{equation} I_{{\rm H\,\mathsc{\sc i}}}= \sum _i S_i \times \Delta v, \end{equation} \tag{ 3 }$$

where i is the channel, S_i is the flux density in the ith channel in Jy beam⁻¹, and Δv is the channel spacing in km s⁻¹. We then convert the moment maps to column density with

$$\begin{equation} N_{{\rm H\,\mathsc{\sc i}}}= 1.104 \times 10^{24} \frac{1}{b_{{\rm maj}} \,\,b_{{\rm min}}} \sum _i S_i \Delta v, \end{equation} \tag{ 4 }$$

where b_maj and b_min are the beam major and minor axes in arcseconds and ∑_iS_iΔv is the value of moment 0 map at each pixel in units of Jy beam⁻¹ km s⁻¹. In Figures 4–32, we show column density maps for all galaxies with detected H i (upper left panels of Section (b) of each figure). We also show column density contours overlaid on optical images for each galaxy. On these maps (upper right), we placed the footprints of the HST observations that are available through ANGST.

The maps exhibit resolved H i structures comparable to their beam sizes. Some galaxies, like KK 230, or NGC 404 have low columns with peak values of a few times 10²⁰ cm⁻². Other galaxies like DDO 190, or UGCA 292 reach columns of a few times 10²¹ cm⁻² (or ∼10 M_☉ pc⁻²), which is a canonical threshold for SF (e.g., Skillman 1987; Kennicutt 1989; Bigiel et al. 2008; Leroy et al. 2008).

4.3.2. Intensity-weighted Velocity Field Maps

The H i intensity-weighted velocity fields (moment 1) maps are calculated using

$$\begin{equation} \langle v \rangle = \frac{\sum _i S_i \times v_i}{\sum _i S_i}. \end{equation} \tag{ 5 }$$

For well-behaved disks, this equation gives a good indication of the average velocity of gas in a given pixel. However, bulk motions, outflows, and other non-circular motions can shift the derived velocities to unexpected values. Therefore, profile fitting in order to determine the velocity of the peak of the emission is a more reliable method for finding the average rotational velocity of the gas at a given location in the galaxy. While the velocity fields of lower mass dwarfs are less ordered than those of their larger disky counterparts, most still show velocity gradients across their disks that are indicative of rotation, which is typical for H i in dwarf galaxies (Begum et al. 2008; Walter et al. 2008). The H i intensity-weighted velocity maps of each galaxy are shown in the lower right panels of Section (b) in Figures 4–32.

4.3.3. Second Moment Maps

The linewidth of H i emission can be characterized by the intensity-weighted second moment as given by

$$\begin{equation} \sigma = \sqrt{ \frac{ \sum _i S_i \times \left(v_i - \langle v \rangle \right)^2 }{ \sum _i S_i } }, \end{equation} \tag{ 6 }$$

where 〈v〉 indicates the intensity-weighted velocity derived in the first moment map. While the second moment can be indicative of the turbulence of the ISM, it also reflects the influence of large-scale gas flows such as expanding shells or tidal material. At lower resolution, the velocity dispersion can be artificially inflated by beam smearing over the gradient in the velocity field, especially toward the centers of the galaxies, where this gradient is the steepest. Overall, the velocity dispersion values fall in a relatively narrow range of 5–15 km s⁻¹, as seen in the lower right panels of Section (b) of Figures 4–32).

Pixels which yield first velocity moments outside the velocity ranges of the data cubes are blanked in all moment maps. The first and second moment maps generated from the robust data cubes are noisier than those from natural-weighted cubes and occasionally have pixels with unrealistic values in low column density regions. To counter this problem, we blanked all pixels with column densities $N_{{\rm H\,\mathsc{i}}} <3 \times 10^{19}$ cm⁻² in the robust moment maps.