Presentation #412.02 in the session Evolution of Galaxies VII.
Self-Organizing Maps (SOM), an artificial neural network trained by unsupervised learning for dimensionality reduction, has shown to be a great tool to extract knowledge from photometric observations of galaxies. In this work, we demonstrate that SOMs can be a powerful tool to estimate the spectroscopic features of large samples of galaxies from their broadband photometry. We train SOMs with near ultraviolet to infrared photometry from the COSMOS2020 photometric catalogue and use measured spectroscopic properties for a subsample of galaxies from LEGA-C and zCOSMOS surveys to estimate spectroscopic features such as D4000 and line equivalent widths for the parent sample. Also, we use trained SOMs to identify regions of parameter space where we lack stellar abundance observations, to create the necessary complete sample to study chemical evolution of galaxies across cosmic time. This is crucial as new capabilities of future facilities like JWST can provide us with unprecedented spectra to observe such representative samples. This is especially important for the study of higher redshift galaxies whose stellar abundance patterns are only in vague agreement with chemical enrichment models informed by current low redshift observations. This technique will also benefit future massive photometric surveys with Euclid and Rubin where wide regions of sky will be observed with a limited number of filters.