5 Conclusions

We have reanalyzed the emission line properties of three large samples of H  II galaxies taken from the literature (Terlevich et al. 1991; Izotov and collaborators 1994-2000; Popescu & Hopp 2000) with the aim of studying the temporal evolution of these objects during the lifetime of the ionizing stars ( 10 Myr). We have constructed a series of diagrams using observed emission line ratios and equivalent widths, and found significant trends.

We interpreted these trends with photoionization models for integrated stellar populations. This study extends the work of Stasinska & Leitherer (1996), by including objects with no direct metallicity determination and using updated evolutionary tracks and atmospheric models.

The changes brought about by the inclusion of the latest non-rotating Geneva stellar evolution models and the CoStar non-LTE line blanketed atmosphere models including stellar winds for O stars have been summarized in Sect. 3.1. (cf. Fig. 5). These models have been successfully compared to observations in a large variety of studies related to massive stars (see e.g. Maeder & Meynet 1994; Stasinska & Schaerer 1997; Oey et al. 2000; Schaerer 2000). Despite this, the adopted models have also some known weaknesses, e.g. the neglect of stellar rotation leading to modifications of the stellar tracks (cf. Maeder & Meynet 2000), a possible overestimate of the hardness of the ionizing flux above $\sim$40 eV (Oey et al. 2000), or the lack of line blanketing in WR atmospheres (see Sect. 3.1). Essentially, the main dependence of the photoionization models on the stellar input physics enters through 1) the evolutionary timescales and the temperatures of main sequence O stars, and 2) the hardness of the He⁰/H ionizing spectrum. It is important to note that, while detailed outputs from photoionization models are sensitive to the exact stellar input physics, our main conclusions summarized below are quite robust.

The underlying principle of the comparison between models and observations is that the metallicity distribution is age-independent for sufficiently large and homogeneous samples, despite the fact that the metallicities of individual objects may be a priori unknown.

We have shown that H  II galaxies selected by objective-prism surveys do not correspond to the simplistic view of instantaneous starbursts surrounded by constant density, ionization bounded H  II regions. The observed relations between emission line ratios and H$\beta$ equivalent width can be understood if most H  II galaxies contain older stellar populations contributing, sometimes rather significantly, to the observed optical continuum. This is in line with the recent findings of Raimann et al. (2000a,b) who, from a spectroscopic analysis on the stellar features in very high signal-to-noise templates of H  II galaxies, showed that populations as old as 100 Myr and up to a few Gyr are detectable in the spectra and significantly affect the observed H$\beta$ equivalent widths. This is also consistent with the model of a long lasting low level of star formation in I Zw 18, as suggested by Legrand (1999). Two additional mechanisms may play a role. First, stars seem to suffer a smaller dust obscuration in the visible light than the ionized gas. Second, some Lyman continuum radiation is probably leaking out of most of the nebulae encompassed by the observing slits.

The combination of these effects reduces $EW({\rm H}\beta )$ with respect to the value predicted for an instantaneous starburst surrounded by an ionization bounded nebula. Therefore objective-prism selected samples of H  II galaxies are unlikely to contain significant numbers of starbursts older than about 5 Myr. Older stages in the evolution of starbursts must be selected from photometric surveys based on broad- or narrow-band colors.

An interesting consequence of this selection effect is that the strong line methods for deriving oxygen abundances work rather well in metal poor H  II galaxies because there is no large mean effective temperature spread. Under these conditions, a strong line method which is calibrated on objects with electron temperature based oxygen abundances determinations like that proposed by Pilyugin (2000), is the most valid approach to derive oxygen abundances in low signal-to-noise spectra of H  II galaxies. One must, however, keep in mind the statistical nature of this method and the fact that the unknown stellar absorption in H$\beta$ provides a further source of uncertainty.

We have shown that [O  I]/H$\beta$ uniformly increases with decreasing $EW({\rm H}\beta )$. This behavior suggests that this line ratio results from dynamical effects that shape the nebula and whose importance increases with time. These dynamical effects could also be responsible for the small range in ionization parameters that account for the observed emission line trends.

The classical emission line ratio diagnostic diagrams such as [O  III]/H$\beta$ vs. [O  II]/H$\beta$ imply a sequence in oxygen abundance and ionization parameter. This suggestion was made earlier on the basis of single-star photoionization models, but a physical interpretation remains elusive. Extra heating sources in addition to the Lyman continuum radiation from massive stars seem necessary in order to explain the largest observed [O  II]/H$\beta$ ratios.

The [N  II]/[O  II] ratio is shown to increase as $EW({\rm H}\beta )$ decreases. The observed trend is even stronger when considering the sample of high signal-to-noise H  II galaxy templates of Raimann et al. (2000a,b) after correction for the old stellar population. We conclude that either the relation between N/O and O/H is steeper than that adopted in our models (N/O $\propto$ O/H^0.5) and the underlying stellar population is stronger at higher metallicities, or the N/O ratio increases with time on a time scale of $\sim$5 Myr. This last option would support a scenario of self-pollution of giant H  II regions by nitrogen produced in situ. High signal-to-noise observations to directly uncover the old stellar populations and the H$\beta$ contamination in individual objects, especially those with the highest [N  II]/[O  II] ratios, are required to settle this important issue. The possible dependence of the above findings on the various main unknowns (underlying populations, escape of Lyman continuum photons, non stellar heating sources, etc.) affecting our analysis have been discussed in the respective sections. Luckily, not all the problems raised in this study are interconnected, and they may be attacked from various angles. E.g. our claim of underlying populations being responsible for an effective "age bias'' against burst events with ages 5 Myr and the puzzling behavior of [N  II]/[O  II] can be tested by additional and more detailed stellar population studies of H  II galaxies along the lines of Raimann et al. (2000a,b), using high signal-to-noise spectroscopy. A purely "stellar'' solution seems now clearly excluded for the problem of [O  III]/H$\beta$ vs. [O  II]/H$\beta$ as well as [S  II]/H$\beta$ (Sect. 4.3 and Stasinska & Schaerer 1999). Hydrodynamical models of the interaction of the interstellar medium with clusters of hot stars (in the vein of Cantó et al. 2000 or Franco et al. 2000) combined with photoionization calculations, should allow one to propose quantitative solutions. Such an approach may at the same time explain the behavior of [O  I]/H$\beta$ as a function of $EW({\rm H}\beta )$ and provide some quantitative estimate of the leakage of ionizing photons from giant H  II regions. Multi-wavelength high spectral resolution observations should provide important constraints to the models in this respect. The dust obscuration issue is perhaps trickier, since dust effects are extremely dependent on geometry. Here, systematic high resolution imaging of giant H  II regions, such as performed by Calzetti et al. (1997) or Johnson et al. (2000), should allow one to clarify the dust location and its effect on integrated H  II galaxies spectra. The insight gained from this study and the proposed directions for further theoretical and observational investigations will considerably increase our understanding of the physics of H  II regions and the complex interactions between its stellar and interstellar components.

Acknowledgements

We thank E. Telles for sending us the list of objects in the Terlevich sample that proved to be giant H  II regions in spiral galaxies. We thank D. Raimann for sending us unpublished data that were required to construct Fig. 13.