Deep learning-driven atmospheric parameter prediction for hot subdwarf stars with synthetic and observed spectra

TL;DR

This work presents a CNN with channel and spatial attention to predict hot subdwarf atmospheric parameters ($T_{\rm eff}$, ${\log g}$, $\log(n{\rm He}/n{\rm H})$) from spectra, trained on a synthetic grid across nine SNR levels and augmented with 945 observed spectra. Nine SNR-specific models achieve high $R^{2}$ ($>0.96$) on test data, with $T_{\rm eff}$ and $\log(n{\rm He}/n{\rm H})$ reaching $R^{2}>0.98$ and ${\log g}$ around $0.96$–$0.98$, and MAEs of $730\,K$, $0.09$ dex, and $0.03$ dex when compared to spectral-fitting references. Applying the models to LAMOST DR12 data yields 1512 confirmed hot subdwarfs, including 291 new identifications, demonstrating both accuracy and speed suitable for large-scale spectroscopic surveys. The approach holds promise for rapid atmospheric parameter estimation in upcoming surveys (e.g., Gaia XP, DESI, CSST) and for accelerating the discovery and classification of hot subdwarfs in massive datasets.

Abstract

We design a convolutional neural network (CNN) incorporating channel attention and spatial attention mechanisms to predict atmospheric parameters of hot subdwarfs. The experimental dataset comprises spectra at nine distinct signal-to-noise ratio (SNR) levels, with each SNR level containing 11 396 synthetic spectra and 945 observed spectra. The trained deep learning models achieves mean absolute errors (AME) in predicting hot subdwarf atmospheric parameters of 730 K for effective temperature (Teff ), 0.09 dex for surface gravity (log g), and 0.03 dex for helium abundance (log(nHe/nH)), respectively, which reaches the accuracy of traditional spectral fitting methods. Utilizing the trained deep learning models and low-resolution spectra from LAMOST DR12, we confirm 1512 hot subdwarfs from the catalog of hot subdwarf candidates, of which 291 are newly identified. Our results demonstrate that the deep learning model not only achieves accuracy comparable to traditional methods in obtaining hot subdwarf atmospheric parameters, but also far exceeds them in speed and efficiency, making it particularly suitable for the analysis of large datasets of hot subdwarf spectra.

Figures (6)

• Figure 1: A synthetic spectrum of ${\rm log}\ g$ = 6.3 (blue-dashed curve) obtained by extrapolation from the synthetic spectrum with ${\rm log}\ g$ = 6.2 (gray-solid curve) and 6.1 (red dot-dashed curve). The three synthetic spectra shown in the figure have same surface temperature ($T_{\rm eff}$ = 36 000 K) and He abundance ($\log(n{\rm He}/n{\rm H})$ = -3.40). The inset plot magnifies the region surrounding the $H_\alpha$ line to demonstrate the subtle variations among the three spectra with little different ${\rm log}\ g$ values. See the context for details.
• Figure 2: Top panel: a synthetic spectrum without noise (black-solid curve) and with noise (blue-dashed curve) of SNRu=20.0. Bottom panel: the same synthetic spectrum without noise (black-solid curve) and with noise (blue-dashed curve) of SNRu=80.0.
• Figure 3: Comparisons between CNN model predictions and the true parameter values for the test set. From left to right, it gives the comparison of $T_{\rm eff}$, ${\rm log}\ g$ and $\log(n{\rm He}/n{\rm H})$, respectively. Top panels show the results of SNRu = 25 model, while bottom panels show the results of SNRu = 80 model. Horizontal axis presents parameters from spectral fitting method (training labels), while vertical axis presents CNN model predictions. The coefficient of determination $R^{2}$ is shown in each panel (see text for details).
• Figure 4: Comparison between CNN model predictions and the parameter values for the known hot subdwarfs in 2022AA...662A..40C. The mean absolute errors (MAE) are 730 K for $T_{\rm eff}$ (top left panel), 0.09 dex for ${\rm log}\ g$ (top right panel) and 0.03 dex for $\log(n{\rm He}/n{\rm H})$ (bottom middle panel).
• Figure 5: The relationships among atmospheric parameters for the 291 newly identified hot subdwarf stars. The markers and number counts for different types of hot subdwarfs are shown in each panel. Top left panel: $T_{\rm eff}$ vs $\log{g}$ plane. The ZAHB and TAHB sequences with [Fe/H]= -1.48 from 1993ApJ...419..596D are denoted by dashed lines. The He main-sequence from 1971AcA....21....1P is marked by black solid line. Three evolution tracks for hot HB stars from 1993ApJ...419..596D are presented by brown dotted curves, for which the masses from top to bottom are 0.495, 0.490, and 0.488 $M_{\odot}$, respectively. Top right panel: $T_{\rm eff}$-$\log(n{\rm He}/n{\rm H})$ plane. The black dotted line and dot-dashed line are the linear regression lines fitted for the two He sequences by 2003AA...400..939E and 2012MNRAS.427.2180N, respectively. Bottom middle panel: $\log{g}$-$\log(n{\rm He}/n{\rm H})$. The red horizontal dashed line denotes the solar value of He abundance (e.g., $\log(n{\rm He}/n{\rm H})$ = -1)