Estimating the Atmospheric Parameters of Early-type Stars from the Chinese Space Station Telescope (CSST) Slitless Spectra Survey

TL;DR

This study demonstrates that machine learning can accurately estimate atmospheric parameters of early-type stars from simulated and real spectra expected from the upcoming CSST survey, enabling efficient analysis of large stellar datasets.

Contribution

The paper introduces a machine learning approach for estimating stellar parameters from CSST slitless spectra, validated with synthetic and observed data, highlighting its accuracy and robustness.

Findings

Average Teff deviation less than 5% with NGSL data

Teff and log g estimations are robust against spectral shifts

Machine learning achieves high accuracy in parameter estimation for early-type stars

Abstract

The measurement of atmospheric parameters is fundamental for scientific research using stellar spectra. The Chinese Space Station Telescope (CSST), scheduled to be launched in 2024, will provide researchers with hundreds of millions of slitless spectra for stars during a 10 yr survey. And machine learning has unparalleled efficiency in processing large amounts of data compared to manual processing. Here we studied the stellar parameters of early-type stars (effective temperature Teff more than 15,000 K) based on the design indicators of the CSST slitless spectrum and the machine learning algorithm, Stellar LAbel Machine. We used the Potsdam Wolf-Rayet (POWR) synthetic spectra library for cross validation. Then we tested the reliability of machine learning results by using the Next Generation Spectrum Library (NGSL) from Hubble Space Telescope observation data. We use the spectra with the impact of interstellar extinction (AV = 0, 0.5, 1, 1.5, 2 mag) and radial velocity (RV = -50, -30, 0, 30, 50 km s-1) from the POWR library as the test set. When RV = 0 km s-1 and AV = 0 mag, the average value and standard deviation for 3 wavelength ranges (2550-4050 Ang (R = 287); 4050-6300 Ang (R = 232); 6300-10000 Ang (R = 207)) are -66 K, 550 K, and 356 K for Teff, and 0.004 c.g.s, -0.024 c.g.s, and 0.01 c.g.s for log g. When using the observed data from NGSL as the testing samples, the deviation of Teff is less than 5%, and the deviation of log g is less than 11%. In addition, we also test the influence of shifting of spectra on the parameters accuracy. The deviation of Teff for the case with a shift of 5 Ang and 10 Ang are 3.6% and 4.3%, respectively; the deviation of log g are 4.2% and 5.1%. These results demonstrate that we can obtain relatively accurate stellar parameters of a population of early-type stars with the CSST slitless spectra and a machine-learning method.