Angthopo, J., Granett, B. R., La Barbera, F., Longhetti, M., Iovino, A., Fossati, M., Ditrani, F. R., Costantin, L., Zibetti, S., Gallazzi, A., Sánchez-Blázquez, P., Tortora, C., Spiniello, C., Poggianti, B., Vazdekis, A., Balcells, M., Bardelli, S., Benn, C. R., Bianconi, M., Bolzonella, M., Busarello, G., Cassarà, L. P., Corsini, E. M., Cucciati, O., Dalton, G., Ferré-Mateu, A., García-Benito, R., Delgado, R. M. González, Gafton, E., Gullieuszik, M., Haines, C. P., Iodice, E., Ikhsanova, A., Jin, S., Knapen, J. H., McGee, S., Mercurio, A., Merluzzi, P., Morelli, L., Moretti, A., Murphy, D. N. A., Pizzella, A., Pozzetti, L., Ragusa, R., Trager, S. C., Vergani, D., Vulcani, B., Talia, M., and Zucca, E.
The WHT Enhanced Area Velocity Explorer (WEAVE) is a new, massively multiplexing spectrograph. This new instrument will be exploited to obtain high S/N spectra of $\sim$25000 galaxies at intermediate redshifts for the WEAVE Stellar Population Survey (WEAVE-StePS). We test machine learning methods for retrieving the key physical parameters of galaxies from WEAVE-StePS-like spectra using both photometric and spectroscopic information at various S/Ns and redshifts. We simulated $\sim$105000 galaxy spectra assuming SFH with an exponentially declining star formation rate, covering a wide range of ages, stellar metallicities, sSFRs, and dust extinctions. We then evaluated the ability of the random forest and KNN algorithms to correctly predict such parameters assuming no measurement errors. We checked how much the predictive ability deteriorates for different S/Ns and redshifts, finding that both algorithms still accurately estimate the ages and metallicities with low bias. The dispersion varies from 0.08-0.16 dex for ages and 0.11-0.25 dex for metallicity, depending on the redshift and S/N. For dust attenuation, we find a similarly low bias and dispersion. For the sSFR, we find a very good constraining power for star-forming galaxies, log sSFR$\gtrsim$ -11, where the bias is $\sim$ 0.01 dex and the dispersion is $\sim$ 0.10 dex. For more quiescent galaxies, with log sSFR$\lesssim$ -11, we find a higher bias, 0.61-0.86 dex, and a higher dispersion, $\sim$ 0.4 dex, for different S/Ns and redshifts. Generally, we find that the RF outperforms the KNN. Finally, the retrieved sSFR was used to successfully classify galaxies as part of the blue cloud, green valley, or red sequence. We demonstrate that machine learning algorithms can accurately estimate the physical parameters of simulated galaxies even at relatively low S/N=10 per angstrom spectra with available ancillary photometric information., Comment: 19 pages, 10 + 2 figures, 4 tables, accepted in A&A