Metallicities from High-Resolution TRES Spectra with The Payne and uberMS: Performance Benchmarks and Literature Comparison

As the field of exoplanetary astronomy has matured, there has been growing demand for precise stellar abundances to probe subtle correlations between stellar compositions and planetary demographics. However, drawing population-level conclusions from the disparate measurements in the literature is challenging, with various groups measuring metallicities using bespoke codes with differing line lists, radiative transfer models, and other assumptions. Here we use the neural-net framework uberMS to measure iron abundances and alpha enchriments from high-resolution optical spectra observed by the Tillinghast Reflector Echelle Spectrograph (TRES), a key resource used for the follow-up of candidate exoplanet hosts. To contextualize these measurements and benchmark the performance of our method, we compare to external constraints on metallicity using the Hyades cluster, wide binaries, and asteroids, to external constraints on $T_{\rm eff}$ and $\log g$ using stars with interferometric radii, and to the results of other abundance measurement methods using overlap samples with the APOGEE and SPOCS catalogs, as well as by applying the SPC method directly to TRES spectra. We find that TRES-uberMS provides parameter estimates with errors of roughly 100K in $T_{\rm eff}$, 0.09dex in $\log g$, and 0.04dex in [Fe/H] for nearby dwarf stars. However, [Fe/H] performance is significantly poorer for mid-to-late K dwarfs, with the bias worsening with decreasing $T_{\rm eff}$. Performance is also worse for evolved stars. Our [$α$/Fe] error may be as good as 0.03dex for dwarfs based on external benchmarks, although there are large systematic differences when comparing with specific alpha-element abundances from other catalogs.