Modern large-scale, ground-based stellar spectroscopic surveys produce large datasets of stellar atmospheric parameters, abundances and ages. These datasets are crucial for testing current chemo-dynamical models for the Milky Way and for unraveling its formation history. Two different approaches are usually adopted for deriving stellar parameters and chemistry from stellar spectra: “physical” spectroscopic pipelines that fit a synthetic spectrum to the observed one, and machine learning methods that derive stellar labels after being trained on spectra of well known stars.
How can we ensure the accuracy, precision, and homogeneity of the parameters provided by surveys? Presently, spectroscopic surveys base their homogenization on only thirty-six benchmark stars and, when machine learning is in use, they adopt their own training sample (often with incomplete parameter coverage, especially in the metal-poor regime). As a consequence, abundance zero-points and trends vary from one survey to another, thereby introducing erratic biases when used for characterizing our Galaxy.
Our proposed International Team will remedy this situation lastingly. Using targets with available asteroseismology and targets in clusters, we will build a reference catalogue of ~10⁵ stars. The catalogue will include reliable atmospheric parameters, abundances, and ages that span the required range for Milky Way investigations. Our reference catalogue will have a significant importance for present and future spectroscopic surveys such as e.g. 4MOST and MSE.