The spectra were reduced using a modified version of the MULTIRED package, which runs within IRAF, using the following procedure: (i) for each starplate the location and dimensions of each slit spectrum are determined; (ii) using these parameters, subsections of the image are trimmed to produce individual 2D spectra corresponding to each slitlet; (iii) identical trimming sections are applied to the corresponding bias, flat-field, and wavelength calibration frames; (iv) the bias and flat-field corrections are pipelined in the standard way for the whole set of stellar and arc spectra. This produced 231 individual slit spectra averaging 25 1700 pixels each.
Wavelength calibration was performed on the 2-D images. This is essential in order to correct for possible distortions along the slit, which can prevent a proper nebular background subtraction. He-Ar lines were identified in a selected column of each slit, the central one in most cases, and were then traced along the direction of the slit. Unfortunately, problems with the arc lamps that occurred during the observing run imply that in some calibration frames the argon lines are not well exposed and the intense HeI 5875 Å line is saturated. Therefore, in order to avoid systematic effects within the whole sample, we used a 5th order Legendre polynomial for all the pixel-wavelength transformations, and we used the same sub-set of He-Ar lines for all spectra. This procedure is not critical for spectral classifications, but is crucial for stellar radial velocities which will be discussed in a forthcoming paper. The mean rms error of the wavelength calibration for the 13 lines in common to all spectra is 0.06 Å.
Notice that, since the multislit spectrograph yields different wavelength ranges for different slit positions, the reddest line available for wavelength calibration varies from star to star. This, together with the saturation problems of HeI 5875 Å, introduces different limits to the reliable wavelength interval for each object. These limits, however, are beyond the wavelength range used for spectral classification
EMMI spectra are slightly curved in the spatial direction forming an arc with up to 5 pixels transversal deviation from centre to edge. On top of that, crowding sometimes causes overlapping of information in adjacent spectra. These two effects often left only a few pixels available for background subtraction, so the usual procedure of selecting background strips in the image for subtracting the sky and nebular contamination could not be performed. On the other hand, the strength of the nebular emission lines from the HII region itself makes the background subtraction crucial for the final quality of the reduced data. Therefore, in order to minimise residuals in the nebular subtraction due to the curvature of the spectral lines, we used the IRAF APEXTRACT task to trace the nebular background as close as possible to the stellar spectra. (In order to use this procedure it is essential to eliminate distortions in the slit direction by performing the wavelength calibration in 2-D, as described above.)
Due to the large variations in the intensity of the nebular HeI emission lines within the slit length, we used - whenever possible - two background windows situated symmetrically on both sides of the stellar spectra. Each window was 3 pix-wide and 6 pix (1.7'') away from the aperture centre, and linear interpolation between the two windows was used to determine the background spectrum.
We have devised a method to improve the background subtraction which uses the [OIII]4959, 5007 Å lines to minimise the residual contamination by the nebular lines.
Since the [OIII] lines are not emitted by the stars and are the strongest lines in the nebular spectrum, the presence of residual emission (or "absorption'') features of [OIII] in the final stellar spectra is a sensitive indication of under (or over) subtraction of the nebular background. Although, due to the clumpiness of the ionised gas, the intensity of the nebular spectrum can change significantly on very small distance scales, the intensity ratios of the relevant nebular lines are relatively insensitive to density variations and thus remain constant within the slitlets. We have checked for the stability of the line ratios ([OIII]/H, [OIII]/HeI 4471) for the slits with the strongest [OIII] nebular lines. In all cases, the observed ratios changed less than 20%. Therefore, our estimation (and correction where necessary) using the [OIII] residuals also applies to other nebular lines, and in particular to HeI 4471 Å which is critical for spectral classification of O stars.
The correction was estimated as follows:
As shown in Fig. 3, this procedure allowed us to achieve excellent subtraction of the nebular lines in most cases. Where this was not possible, the procedure gave us a good estimation of the amount of contamination present. The critical diagnostic for O-type stars is the ratio of HeI 4471 to HeII 4542. As an indicator of the accuracy of the background subtraction for the former line, we used the ratio of the equivalent widths of HeI 4471 to [OIII]5007 in the nebular spectra. For the whole sample, the average ratio is , which indicates that only 0.4% of the residual in [OIII]5007 present in the stellar spectrum will be affecting the flux of the HeI 4471 line.
Both panels show, at different wavelengths, the reduction of the
nebular residual with the procedure described in the text, for the
case of star 977. Within each
panel both spectra, the corrected one being plotted below, are drawn
at the same scale. The correction introduced from
the [OIII] lines is shown in the right panel. The effects of improved
background subtraction can be seen in the left panel, as the observed
intensity of the HeI 4471 Å absorption changes|
Figure 4: Spectral classification for early O-type dwarf stars. HeII and HeI lines have been identified as a reference. The range suitable for classification is shown in this figure. The y axis scale is in arbitrary units, each spectrum being normalised to its continuum
Copyright The European Southern Observatory (ESO)