Distance and stellar parameter estimations of solar-like stars from the LAMOST spectroscopic survey

Context . The Gaia mission has opened up a new era for the precise astrometry of stars, thus revolutionizing our understanding of the Milky Way. However, beyond a few kiloparseconds from the Sun, parallax measurements become less reliable, and even within 2 kpc, there still exist stars with large uncertainties. Aims . Our aim was to determine the distance and stellar parameters of 521 424 solar-like stars from LAMOST DR9; these stars lacked precise distance measurements (uncertainties higher than 20% or even without any distance estimations) when checked with Gaia. Methods . We proposed a convolutional neural network (CNN) model to predict the absolute magnitudes, colors, and stellar parameters ( T eff , log ɡ , and [Fe/H]) directly from low-resolution spectra. For spectra with signal-to-noise ratios at ɡ band (S/N g ) greater than 10, the model achieves a precision of 85 K for T eff , 0.07 dex for log ɡ , 0.06 dex for [Fe/H], 0.25 mag for M G , and 0.03 mag for (BP – RP) 0 . The estimated distances have a median fractional error of 4% with a standard deviation of 8%. Results . We applied the trained CNN model to 521 424 solar-like stars to derive the distance and stellar parameters. Compared with other distance estimation studies and spectroscopic surveys, the results show good consistency. Additionally, we investigated the metallicity gradients of the Milky Way from a subsample, and find a radial gradient ranging from −0.05 < Δ[Fe/H]/Δ R < 0.0 dex kpc −1 and a vertical gradient ranging from −0.26 < Δ[Fe/H]/ΔZ < −0.07 dex kpc −1 . Conclusions . We conclude that our method is effective in estimating distances and stellar parameters for solar-like stars with limited astrometric data. Our measurements are reliable for Galactic structure studies and hopefully will be useful for exoplanet researches.