The field of galaxy evolution will make a great leap forward in the next decade as a consequence of the huge effort by the scientific community in multi-object spectroscopic facilities. To maximise the impact of such incoming data, the analysis methods must also step up, extracting reliable information from the available spectra. In this paper, we aim to investigate the limits and the reliability of different spectral synthesis methods in the estimation of the mean stellar age and metallicity. The main question this work aims to address is which signal-to-noise ratios (S/N) are needed to reliably determine the mean stellar age and metallicity from a galaxy spectrum and how this depends on the tool used to model the spectra. To address this question we built a set of realistic simulated spectra containing stellar and nebular emission, reproducing the evolution of a galaxy in two limiting cases: a constant star formation rate and an exponentially declining star formation. We degraded the synthetic spectra built from these two star formation histories (SFHs) to different S/N and analysed them with three widely used spectral synthesis codes, namely FADO, STECKMAP, and STARLIGHT. For S/N < 5 all three tools show a large diversity in the results. The FADO and STARLIGHT tools find median differences in the light-weighted mean stellar age of ~0.1 dex, while STECKMAP shows a higher value of ~0.2 dex. Detailed investigations of the best-fit spectrum for galaxies with overestimated mass-weighted quantities point towards the inability of purely stellar models to fit the observed spectra around the Balmer jump. Our results imply that when a galaxy enters a phase of high specific star formation rate the neglect of the nebular continuum emission in the fitting process has a strong impact on the estimation of its SFH when purely stellar fitting codes are used, even in presence of high S/N spectra.
Download