1. INTRODUCTION

The analysis of nebular spectra constitutes the best, and in some cases the only one, method for the determination of chemical abundances in spiral and irregular galaxies, as well as in sites of recent star formation. The abundances of several elements like He, O, N and S can in principle be determined since strong emission lines of these elements, some of them in their dominant ionization states, are present in the optical region of the spectrum. This requires knowledge of the electron temperature which can be obtained by measuring appropriate line ratios like [OIII] [(4363) / (4959 + 5007)], [NII] [(5755) / (6548 + 6584)], [OII] [(7327) / (3727 + 3729)], or [SIII] [(6312) / (9069 + 9532)].

Unfortunately, these line ratios usually involve one intrisically weak line which can be detected and measured with confidence only for the brighter and hotter objects and in many cases - distant galaxies, low surface brightness objects, relatively low excitation regions - they become too faint to be observed.

In these cases, an empirical method based on the intensities of the easily observable optical lines is widely used. The method, originally proposed by Pagel et al. (1979) and Alloin et al. (1979), relies on the variation of these lines with oxygen abundance. Pagel et al. (1979) defined an abundance parameter R₂₃ = [([OII] 3727 + [OIII] 4959 + 5007) / (H)] which increases with increasing abundance for abundances lower than about 20% solar, and then reverses its behavior, decreasing with increasing abundance, since above this value a higher oxygen abundance leads to a more effective cooling, the electron temperature gets lower and the optical emission lines get weaker.

In principle, the calibration of the R₂₃ versus oxygen abundance relation can be done empirically in the low metallicity regime where electron temperatures can be derived directly, but requires the use of theoretical models for the so called high abundance branch. Several different calibrations have been made (Edmunds & Pagel 1984; McCall et al. 1985; Evans & Dopita 1986; Skillman 1989; McGaugh 1991) as more observational data and more improved models have become available. However, two problems that are difficult to solve still remain. The first one is related to the two-valued nature of the calibration, which can lead to important errors in the derived abundances. The second one concerns the dependence of R₂₃ on the degree of ionization of the nebula (see Skillman 1989). R₂₃ also depends on density, but this can be considered as a second order effect for low density regions (n_H 100 cm^-3) which constitute the majority of the extragalactic population. These two facts, taken together, produce a large dispersion of the data for values of logR₂₃ 0.8 and 12 + log(O/H) 8.0, with objects with the same value of log R₂₃ having actual abundances which differ by almost an order of magnitude. Unfortunately, a significant number of objects (about 40% of the observed HII galaxies; Díaz 1999) have logR₂₃ 0.8 for which the calibration is most uncertain, and this percentage is even higher for HII regions in normal spiral galaxies.

Here we present an alternative abundance calibration based on the intensities of the sulphur lines: [SII] 6716, 6731 and [SIII] 9069, 9532, through the use of the sulphur abundance parameter S₂₃ (Vílchez & Esteban 1996).

Spectroscopically, these lines are analogous to the optical oxygen lines defining R₂₃ but, due to their longer wavelengths, their contribution to the cooling of the nebula should become important at a somewhat lower temperature. Yet, the lower abundance of sulphur makes these lines less significant than the [OIII] lines as a contributor to cooling. On the other hand, the sulphur lines are less sensitive to temperature, therefore the reversal in the relation between their intensities and the average nebular abundance is expected to occur at a higher metallicity and the relation should remain single valued longer.

From the observational point of view they present two important advantages: first, the lines are easily detected both in high and low excitation ionized regions (Díaz et al. 1990) and second, they are less affected by reddening; moreover, they can be measured relative to nearby hydrogen recombination lines, H in the case of the [SII] lines and Paschen lines in the case of the [SIII] lines, which minimizes also any flux calibration uncertainties. These lines are accessible spectroscopically with CCD detectors up to a redshift of about 0.1.