"Spatially resolved stellar populations in galaxies from the galaxy survey SDSS-IV/MaNGA "

Evolutionary population synthesis models have been instrumental to our
understanding of galaxy formation and evolution. For the successful application of these models, they must be underpinned by a comprehensive
catalogue of theoretical or empirical stellar spectra. In the latter case, spectra must be described by certain atmospheric parameters: effective temperature (Teff), surface gravity (log g) and chemical abundance. Thus, to accurately model the spectrophotometric properties of
distant galaxies, we must first study the stars around us. In this thesis we
utilise the MaNGA Stellar Library (MaStar), an SDSS-IV survey of 80,592
spectra representing 27,945 unique stars, to create the basis for the nextgeneration of stellar population models for extragalactic astronomy.
We have developed a stellar parameter pipeline that fits theoretical spectra from a combination of model atmosphere libraries to each MaStar observation using the penalised pixel-fitting method (pPXF). We match the
MaStar observations to Gaia photometry and use this to set reliable, flat
priors for Teff and log g. The stellar atmospheric parameters and their uncertainties are then estimated with a Bayesian approach in combination with
the Markov Chain Monte Carlo (MCMC) algorithm. Initially, we calculated stellar parameters using solar-scaled models, which was then
expanded to include as a fourth parameter. Our method is corroborated by its application to well known stars, such as the Sun and Vega, by
comparing our results to several other literature sources of stellar parameters
and by the parallel development and testing of stellar population models on
viii
star clusters.
We further leverage the vastness of MaStar by identifying carbon- and
oxygen-rich, thermally-pulsing asymptotic giant branch (TP-AGB) stars for
their inclusion in stellar population models. This research is done using the
SDSS photometric bands (u, g, r, i, z ) due to time efficiency and as a proof of
application to future large photometric surveys, such as the Legacy Survey
of Space and Time (LSST). We apply an unsupervised machine learning
technique called t-distributed stochastic neighbour embedding to test its
ability in TP-AGB identification. This method is able to isolate such stars
and correlates well with the atmospheric parameters. We then define a
novel colour selection of (g - r) > 2 and (g - i) < 1.55(g - r) -0.07, based
solely on SDSS photometry. This is compared to literature colour cuts and a
supervised learning, support vector machine (SVM) which we have trained.
We find the SVM method most efficient, recovering an average F1-score,
the chosen performance metric, of 0.82 over 200 test cases. Through such
methods we are able to identify 69 carbon-type and 118 oxygen-type spectra
in MaStar. Lastly, we determine the spectral types of MaStar TP-AGB stars
using empirical templates and a discrete version of our stellar parameter
fitting routine.
The application of MaStar spectra has marked a new era in galaxy modelling, allowing for the study of complex stellar populations over a wide
wavelength domain. The latest MaStar-based stellar population models, using the atmospheric parameters presented here, have already been applied to
the SDSS MaNGA data analysis pipeline and can be used for future surveys
by probing the optical and near-infrared window.