2 The observations

Observations were made with the HST STIS, using the GL140 grating which provides a resolution from 310 to 210 km s^-1 in the wavelength range from 1150 to 1700 Å. We used a 02 wide slit, which is recommended for optimizing the spectral purity. Table 1 gives our list of objects and other details of the observations.

There are two spectra for each star. We merged these two spectra after checking their relative displacement by cross-correlating them.

Merged stellar spectra are corrected for the relative velocity between stars and observer. This is done by measuring the displacement from rest wavelength of metal lines that are purely photospheric. There are a few photospheric lines that could in principle be used in the observed spectral range (Prinja 1990). They are difficult to identify in all spectra. Thus, after inspection of the spectra, we decided to use the Si III doublet at $\lambda \lambda$ 1500.24, 1501.19 Å, because it is the only one that can be clearly identified in all spectra. The spectral resolution does not allow us to resolve both components of the Si III doublet. We use as rest wavelength that of a composed line centered at $\lambda$ 1500.72 Å.  Because of the small signal to noise ratio of the spectra, we decided to correct one of the spectra of higher quality and then refer the others to this one. We selected the spectrum of M 33-0900 for that purpose. Once the M 33-0900 spectrum had been shifted so that the Si III line was at rest wavelentgh, we checked this correction with the position of other strong interstellar lines (see f.e. Prinja 1990), as these are probably produced by the interstellar medium close to M 33-0900. We can see in Fig. 1 that the shifted interstellar lines are at their expected rest wavelengths.

To correct the other spectra we cross-correlated them with the corrected spectrum of M 33-0900. To avoid possible biases, strong P-Cygni profiles have been masked before correlations (see Howarth et al. 1997). For M 33-1137 and M 33-B-526, the stars with lowest SN spectrum, the cross-correlation function does not present a well defined maximum. Checking the strong interstellar lines present in the spectra (Si II $\lambda \lambda$ 1260.40, 1526.70) we found good concordance between all spectra, excluding M 33-1137 and M 33-B-526. We decided to shift both spectra until the position of their interstellar lines agree with those of other spectra. Special care was taken in the case of M 33-B-526, as the unshifted spectrum shows several lines in the vicinity of the rest wavelength of the intestellar lines.

After having shifted the spectra we rectified the continuum by tracing a polynomial through a number of selected continuum points chosen iteratively, in the same way as described in Herrero et al. (2001).