2 Sample selection

2.1 The database

The prerequisites for our observational data base are the following. We require a large sample of objects with uniform observation parameters. The sample must be large since there are several factors determining the line spectra, the most important being metallicity and starburst age. We wish to use information from the He  I $\lambda$5876 line, which is directly related to the mean effective temperature of the ionizing stars (cf. Stasinska 1996). Deep spectroscopy is required for this generally weak line. In addition, we want to make use of a direct estimator of the starburst age, such as provided by the equivalent width of H$\beta$ (hereafter $EW({\rm H}\beta )$). While this estimator has its well-known drawbacks (effects of dust, incomplete absorption of the stellar ionizing photons by the observed nebula), it is relatively independent of the assumed nebular properties, as compared with emission line ratios such as [S  II]  $\lambda\lambda$6717+30/H$\beta$ or [S  II] $\lambda\lambda$6717+30/[S  III] $\lambda$9069 proposed by García-Vargas et al. (1995). However, $EW({\rm H}\beta )$ is difficult to use for giant H  II regions with a strong underlying old stellar component, such as in nuclear starbursts or giant H  II regions in evolved galaxies. Our sample is thus restricted to giant H  II regions in irregular or blue compact galaxies, where the old stellar component is expected to be weaker.

The largest, homogeneous sample meeting these requirements is the spectrophotometric catalogue of H  II galaxies by Terlevich et al. (1991). It contains spectra of 425 emission line galaxies, among which about 80% are classified as genuine H  II galaxies by the authors (as opposed to active galactic nuclei and nuclear starbursts). Further imaging of these objects identified some of these H  II galaxies as giant H  II regions in spiral galaxies (Telles, private communication). Those objects were removed to produce the final sample, hereafter referred to as the Terlevich sample, which consists of 305 objects. Most of the observations for the Terlevich catalogue were made with 2-m class telescopes, and the typical signal-to-noise in the continuum is 5. Note, however, that second-order contamination could affect some of the line ratios for $\lambda >$ 6000 Å (Terlevich et al. 1991).

As a control sample, we consider a much more restricted collection of isolated extragalactic H  II regions with very high quality observational data. These are the blue compact galaxies observed by Izotov and coworkers (Izotov et al. 1994, 1997; Thuan et al. 1995; Izotov & Thuan 1998; Guseva et al. 2000) on 2-m and 4-m class telescopes. Their signal-to-noise in the continuum is typically 20-40. The latter sample, referred to as the Izotov sample, comprises 69 objects.

Recently, another survey of emission line galaxies meeting our requirements has been published. This is the catalogue of Popescu & Hopp (2000) which contains 90 objects, out of which 70 are classified as H  II galaxies. The quality of the data is not as good as for the Izotov sample, but the sample is more homogeneous, as explained below.

2.2 Selection effects

Objective-prism surveys of emission line objects tend to underestimate the true proportion of objects with weak emission line equivalent widths. On the other hand, genuine, isolated giant extragalactic H  II regions will always be discovered in such surveys if they are bright enough. The Terlevich catalogue comprises 75% (25%) of the non-quasar emission line objects identified in the Michigan (Tololo) objective-prism survey, and additional biases are not expected to be important. Therefore, one expects a roughly uniform distribution of ages of the most recent starbursts, and the same distribution of metallicities at each age bin above a certain threshold in H$\beta$ equivalent width in the Terlevich sample.

In the case of the Izotov sample, the objects are mainly blue compact galaxies from the First and Second Byurakan objective-prism surveys. The sample is obtained by merging two sub-samples. One is composed of objects which had been selected for their low metallicities ( $Z < {\rm Z}_\odot/10$) as estimated from low signal-to-noise 6-m telescope spectra. The other subsample (the 39 objects studied by Guseva et al. 2000) is composed of objects selected on the basis of broad Wolf-Rayet features seen in the 6-m telescope spectra. The metallicities in the latter subsample range between $Z_\odot$/40 and $Z_\odot$. The H$\beta$ equivalent widths in the Guseva subsample are, on average, lower than those in the other Izotov et al. subsample, and contamination from old stellar populations is likely present at least for some objects known as nuclear starbursts. Compared to the Terlevich sample, the Izotov sample is affected by strong biases which must be kept in mind when interpreting the diagnostic diagrams shown below.

The Popescu & Hopp sample is a complete subsample of emission line galaxies found on the objective-prism plates for the Hamburg Quasar Survey. The single selection criterion was the location of the galaxies with respect to voids. Therefore, as for the Terlevich sample, we expect a roughly uniform distribution of ages, and the same metallicity distribution at each age bin. The sample is however much smaller than the Terlevich sample; therefore statistical fluctuations may not be negligible.

2.3 Reddening corrections

Izotov and coworkers used an iterative procedure to deredden their sample where the reddening constant and stellar absorption at H$\beta$ are determined simultaneously. Such a procedure was feasible due to the high signal-to-noise of the data. They assumed the Whitford (1958) extinction law and took for the intrinsic H$\alpha$/H$\beta$ ratio the recombination value corresponding to the electron temperature derived from [O  III] 4363/5007, when available. In most cases, the contribution of the stellar absorption in H$\beta$ turns out to be only a few percent, except in some objects from the Guseva sample which includes a few nuclear starbursts (see Guseva et al. 2000; Schaerer et al. 2000).

The line intensities in the Popescu & Hopp (2000) sample are published corrected for reddening. For the galaxies with a strong underlying continuum relative to the H$\beta$ emission ( $EW({\rm H}\beta )$ < 20 Å), they assumed a constant underlying absorption at H$\beta$ of 2 Å. The reddening was computed from H$\alpha$/H$\beta$ assuming an intrinsic ratio of 2.87 and using the Howarth (1983) extinction law.

In the case of the Terlevich sample, the data are published without reddening correction. We have dereddened them adopting the Seaton (1979) reddening law and an intrinsic H$\alpha$/H$\beta$ ratio of 2.87. We assumed a constant underlying stellar absorption of 3 Å for all the spectra in the H$\alpha$ and H$\beta$ lines, which is predicted by models for synthetic absorption line spectra in starburst galaxies (Olofsson 1995a; González Delgado et al. 2000). Unfortunately, spectra in the H$\alpha$ region have second order contamination in many objects (cf. Terlevich et al. 1991), affecting the reddening correction. Alternatively, one could use H$\gamma$/H$\beta$ instead of H$\alpha$/H$\beta$ to determine the reddening correction, but this procedure is plagued by the fact that H$\gamma$ is a much weaker line than H$\alpha$, introducing substantial errors due to the reduced signal-to-noise, and by the lack of published H$\gamma$ equivalent widths. In any event, for objects without published H$\alpha$ data, we used H$\gamma$/H$\beta$ to derive the reddening, assuming that the stellar (negative) contribution to the observed H$\gamma$ line is three times as large as for the H$\beta$ line.

The reddening correction introduces some uncertainties into each sample. Even in the case of the Izotov sample, where dereddening was performed in a completely self-consistent way, the reddening correction may not be perfect. First, departures from the standard reddening law exist (e.g. Mathis & Cardelli 1988 or Stasinska et al. 1992). Second, the intrinsic H$\alpha$/H$\beta$ ratio in a nebula is not necessarily the recombination value. In objects with high electron temperatures, collisional excitation increases H$\alpha$/H$\beta$ with respect to the recombination value, in a proportion that depends on the quantity of neutral hydrogen in the emitting regions (Davidson & Kinman 1985; Stasinska & Schaerer 1999). Adopting the same underlying absorption of 3 or 2 Å at H$\beta$ for all the objects of the Terlevich and Popescu & Hopp samples may not be justified if stellar populations from previous star-forming events contribute significantly to the continuum (see below). Among the line ratios discussed in this paper, the one most affected by reddening is [O  II]/H$\beta$. It is prudent to assume that the [O  II]/H$\beta$ ratios may well be uncertain by up to about 30%, and even more for some objects in the Terlevich sample.

Figure 1: The Terlevich sample of H  II galaxies. On the upper right of each panel is given the total number of the objects appearing in the diagram

Figure 2: Same as Fig. 1 for the Izotov sample of H  II galaxies

Figure 3: Same as Fig. 1 for the Popescu & Hopp sample of H  II galaxies

2.4 Observed emission line trends

The nine panels of Figs. 1, 2, and 3 are various observational diagrams for the Terlevich, Izotov, and Popescu & Hopp samples, respectively. Panels a through f in these figures show the behavior of various emission line ratios as a function of $EW({\rm H}\beta )$. In spite of some (expected) dispersion, striking line trends are seen. These trends are very similar in the three samples, and are shown for the first time.

Some differences among the samples are likely due to the quality of the data. The smaller dispersion in He  I $\lambda$5876/H$\beta$ in the Izotov sample, for example, is probably due to the better signal-to-noise. We also note that the Terlevich and Popescu & Hopp samples contain some objects with rather large [O  II]/H$\beta$: $\sim$ 20% of the Popescu & Hopp sample and 3% of the Terlevich sample have [O  II]/H$\beta$ $\ge$ 5, while no object of the Izotov sample has such large values for this ratio. However, if we retain only the 50 objects from the Terlevich sample with measurements in the four lines [O  II] $\lambda$3727, [O  III] $\lambda$5007, He  I $\lambda$5876 and [N  II] $\lambda$6584, the diagrams become very similar to those corresponding to the Izotov sample. Therefore, we will ignore the objects with [O  II]/H$\beta$ > 5 since such high ratios may be attributed to poor signal-to-noise and/or to inadequate reddening corrections.

We generally note a gradual increase in the dispersion and a decrease of the average value of [O  III]/H$\beta$ as $EW({\rm H}\beta )$ decreases, while the He  I $\lambda$5876/H$\beta$ ratio remains remarkably constant. The [O  I]/H$\beta$ ratio, on the other hand, steadily increases with decreasing $EW({\rm H}\beta )$, extending the trend already noted by SL96 to lower $EW({\rm H}\beta )$. The [O  II]/H$\beta$ increases and tends to level off at $EW({\rm H}\beta )$ about 30 Å. The ([O  II]+ [O  III])/H$\beta$ ratio is rather constant at the largest $EW({\rm H}\beta )$, and becomes more dispersed with a tendency to decrease as $EW({\rm H}\beta )$ decreases. The [N  II]/[O  II] ratio shows a clear tendency to increase as $EW({\rm H}\beta )$ decreases, although the dispersion is larger than in the remaining diagrams. Overall, all the trends of emission line ratios with H$\beta$ equivalent width are significant and require an explanation.

Finally, panels g, h, and i in Figs. 1-3 show the standard emission line ratio diagrams ([O  III]/H$\beta$ vs. [O  II]/H$\beta$, [O  III]/H$\beta$ vs. [S  II]/H$\beta$ and [N  II]/[O  II] vs. ([O  II]+[O  III])/H$\beta$), similar to those that have been widely used in the literature to distinguish H  II regions from active galaxies (Baldwin et al. 1981; Veilleux & Osterbrock 1987; van Zee et al. 1998; Martin & Friedli 1999) and for studies of abundance ratios in H  II regions (McGaugh 1994; Ryder 1995; Kennicutt & Garnett 1996; van Zee et al. 1998; Bresolin et al. 1999). For comparison, we have plotted in Fig. 4 the same three emission line ratio diagrams for the sample of giant H  II regions in spiral galaxies from McCall et al. (1985).

Figure 4: Emission line ratio diagrams for the sample of giant H  II regions in spiral galaxies from Mc Call et al. (1985)

The H  II galaxies of our samples clearly appear like an extension of the giant H  II region sequence of McCall et al. toward the high [O  III]/H$\beta$ end and towards the low [N  II]/[O  II] end. Some of the H  II galaxies (mainly in the Terlevich sample) reach values of [N  II]/[O  II] which are as high as those observed in the giant H  II regions of the central parts of spiral galaxies. The true scatter in [O  II]/H$\beta$ is probably smaller than it appears in our figures: as discussed in Sect. 2.3, [O  II]/H$\beta$ is strongly affected by uncertainties in the dereddening procedure. Note the pronounced bending of the [O  III]/H$\beta$ vs. [O  II]/H$\beta$ and [O  III]/H$\beta$ vs. [S  II]/H$\beta$ sequences toward the left at high [O  III]/H$\beta$ in the three H  II galaxies samples. The relation between [N  II]/[O  II] and ([O  II]+[O  III])/H$\beta$ is extremely tight in all the samples we consider. This was also seen in diagrams originally studied by the authors mentioned above. Considering that line ratios in nebulae are determined by several factors (abundance ratios, intensity and spectral distribution of the ionizing radiation field, density distribution of the nebular gas), the question arises as to why the observed sequences are so narrow.

We will try to understand the observational data in the light of the photoionization models described in the next section.