Determination of metallicities of red giant stars using machine learning techniques applied to the narrow and broadband photometry of the S-PLUS survey

Aims. The aim of this study is to obtain metallicities of red giant stars from the Southern Photometric Local Universe Survey (S-PLUS) and to classify giant and dwarf stars using artificial neural networks applied to the S-PLUS photometry. Methods. We combined the five broadband and seven narrow-band filters of S-PLUS – especially centred on prominent stellar spectral features – to train machine learning algorithms. The training catalogue was made by cross-matching the S-PLUS and Apache Point Observatory Galactic Evolution Experiment 2 (APOGEE-2) survey catalogues. The classification neural network uses the colours (J0378 - u), (J0395 - g), (J0410 - g), (J0515 - g), (J0660 - r), (g - z) and (r - i) as input features, whereas the network for metallicities uses the colours (J0378 - u), (J0395 - g), (J0410 - g), (J0515 - g), (J0660 - r), (u - g) and (r - z) as input features. Results. The resulting network is capable of identifying ~99% of the giants in the test set. The network for determining the photometric metallicities of giant stars estimates metallicities in the test set a with a standard deviation of σ_giants ~ 0.07 dex with respect to the spectroscopic values. Finally, we used the trained artificial neural networks to generate a publicly available catalogue of 523 426 stars classified as red giant stars from S-PLUS, which we used to explore metallicity gradients in the Milky Way.