2 Observations and reduction of the data

The Fabry-Perot (FP) observations were taken on the night of May 10th 1997, with the 4.2 m William Herschel Telescope on La Palma. The night was photometric, but with a seeing of 1.3 $^{\prime \prime }$. The galaxy was observed in H$\alpha$ using the TAURUS II FP instrument at the Cassegrain focus. The etalon used had a spacing of 125 $\mu$m, and the camera focal ratio was f/2.11; the detector was a TEK CCD. We windowed the camera to a size of 540 $\times$ 540 pixels (to avoid vignetting by the filter wheel), at a pixel size of 0.56 $^{\prime \prime }$$\times$0.56 $^{\prime \prime }$, yielding a field size of 5 arcmin square. The observations consisted of exposures with the spacing varied in steps, to scan the full wavelength range of the H$\alpha$line as emitted by the whole galaxy. A narrow band filter, centred on the wavelength of H$\alpha$ appropriately red-shifted: $\lambda_{\rm c}$ = 6589 Å and $\Delta \lambda$ = 15 Å, was used for order-sorting. The full spectral range, 17.228 Å, was scanned in 55 steps, so that the wavelength interval between consecutive planes was 0.34 Å at the centre of the field. The full effective exposure time was 140 s per plane. We took a calibration cube with a CuNe lamp at the beginning of the night, and performed the calibration in phase and wavelength using calibration rings taken before and after the data cube. Both types of calibration were carried out using the TAUCAL package in the FIGARO environment, converting the raw data, in which the surfaces of constant wavelength are paraboloids, into a cartesian cube.

The sky background emission was subtracted separately from each plane of the cube, and all the planes were put into common spatial coordinates using field star images. Positional astrometry was performed by comparing the positions of these field stars and of the brightest H II regions in the TAURUS field with their positions on a very well resolved H$\alpha$ image of the galaxy (Fig. 1 from Paper I), yielding maximum uncertainties in position of 0.5 $^{\prime \prime }$.

The resulting data cube was used as the input for the program MOMENTS in the GIPSY programme suite, to calculate moment maps. While the channel maps in Fig. 1 give global information about the way the emission changes with velocity across the face of the galaxy, a more detailed description of the spatial distribution of intensities and velocities is provided by the moment maps. The first gives the integrated intensity of the emission at each point, specified in R.A. and dec.; MOMENTS then allows us to calculate the zeroth, first and second order moments at each pixel, which yield the peak intensity of the emission line, the velocity (centre of the emission line) and the dispersion in intensity from the profile respectively. Only signals coming from the same positional point in three adjacent planes were considered true emission, and used to compute the moment maps.

A careful inspection of the moment maps obtained showed that there were zones where the integration over all velocities gave place to quite strong noise features. This made necessary the revision of the initial data cube to eliminate all the noisy areas which were distorted. To do this, we produced cubes at lower spatial resolution, by smoothing the original cube in order to derive intensity and velocity maps at different resolutions, following traditional radio-astronomical practice. In this way, from an initial resolution of 1.5 $^{\prime \prime }$ we obtained cubes at 6, 8, 10 and 16 $^{\prime \prime }$; for the first two we used an unchanged pixel size, while for the others we binned into 2 $\times$ 2 pixel bins. Although we did not use all the low resolution data cubes to obtain the results shown in this paper, they were a necessary intermediate step to obtain the final data cube. Then we subtracted the continuum emission from each cube, using those planes in the original cube from which H$\alpha$ emission was absent. A linear interpolation to these latter gave us the appropriate continuum level to subtract from each cube plane to leave only the H$\alpha$.

Figure 1: Planes of the H$\alpha$ high resolution (1.5 $^{\prime \prime }$) data cube for NGC 3359 (before "cleaning'' the data cube; see text for details) at a series of velocities about the central systemic value of 1008 km s-1

Figure 1: continued

Figure 2: Intensity map (zeroth moment) of the H$\alpha$ emission in NGC 3359 obtained with the high resolution (1.5 $^{\prime \prime }$) H$\alpha$ data cube. The dynamical centre is marked with an asterisk

We now briefly describe the procedure followed for the full resolution data set. The first step was to produce a conditionally transferred data cube, in which values were retained only at positions where the intensity in the smoothed data cube at 6 $^{\prime \prime }$ was larger than 2.5 times the rms noise of the smoothed maps. Pixel values at all other positions were set undefined. Then, noise peaks outside the area where H$\alpha$ emission is expected were removed by setting pixel values at those positions to undefined too. This was done interactively by inspecting the (high resolution) channel maps one by one, continually comparing with the same channel and referring to adjacent channels in the smoothed cube (6 $^{\prime \prime }$). Previously we had followed an analogous routine for "cleaning" the cubes at lower resolution, using the 8 $^{\prime \prime }$ cube as a reference to clean the 6 $^{\prime \prime }$ cube, and the 16 $^{\prime \prime }$ cube to clean that at 10 $^{\prime \prime }$ resolution.

The emission in H$\alpha$ as a function of wavelength, i.e. of velocity, for the high resolution cube is shown in Fig. 1 in the form of a set of channel maps, again following radio-astronomical usage.

Finally the obtained data cubes were used to construct the moment maps (intensity, velocity and velocity dispersion) in the way explained above. The high resolution (1.5 $^{\prime \prime }$) intensity and velocity maps are shown in Figs. 2 and 3 respectively. These maps could be obtained not only by calculating moments, but also fitting each spectrum to a Gaussian function. The velocity maps obtained with both methods are essentially the same, but considerable differences could appear in the intensity maps and especially in the velocity dispersion maps. This happens because when moments are calculated, it is possible to lose information about the wings of the lines specially if these are low intensity and high velocity dispersion lines (van der Kruit & Shostak 1982). This fact is due to the selected threshold (normally a fixed multiple of the measured rms noise) bellow which the data is neglected. For Gaussian profiles, the threshold is applied to the line amplitude and not as a fixed value over the noise. This implies that the fit can take into account all the spectrum, but this method is susceptible to error (in the low signal points) because we are considering noisy areas as signal. In this work, we have chosen the moment method. We are aware of the possible effect on the results that we present here, and the analysis of the velocity dispersion map constructed using both methods, a study of all the velocity components of the emission line, and the relation between the principal velocity components with the H II region luminosity, will be presented in Zurita et al. (2000).

Figure 3: High resolution (1.5 $^{\prime \prime }$) velocity map for NGC 3359. The darkest zone is the receding side (positive redshift) and the brightest part is the approaching side of the galaxy. The dynamical centre is marked with a asterisk