Unified Deep Learning Approach for Estimating the Metallicities of RR Lyrae Stars Using light curves from Gaia Data Release 3

TL;DR

The paper presents a unified GRU-based deep learning framework to estimate photometric metallicities $[\mathrm{Fe/H}]$ of RR Lyrae stars (RRab and RRc) from Gaia DR3 $G$-band light curves. By framing metallicity estimation as time-series extrinsic regression (TSER) and applying a rigorous preprocessing pipeline (phase folding, smoothing, standardization) plus density-weighted loss, the model achieves high predictive accuracy on held-out data ($R^2$ ≈ 0.94–0.96; wRMSE ≈ 0.07–0.08 dex). A single model handles both pulsation modes, aided by 3 GRU blocks with dropout and a final dense layer, optimized via Hyperband with weighted MSE and regularization under repeated stratified K-fold cross-validation. Validation reveals strong correlations with literature photometric metallicities and a dependence of prediction quality on the number of $G$-band epochs, underscoring the benefit of denser sampling. Applying the model directly to Gaia DR3 light curves increases the number of stars with metallicities by substantial factors and enables large-scale chemical mapping of the Milky Way, with Gaia DR4 and LSST expected to further boost precision and reach.

Abstract

RR Lyrae stars (RRLs) are old pulsating variables widely used as metallicity tracers due to the correlation between their metal abundances and light curve morphology. With ESA Gaia DR3 providing light curves for about 270,000 RRLs, there is a pressing need for scalable methods to estimate their metallicities from photometric data. We introduce a unified deep learning framework that estimates metallicities for both fundamental-mode (RRab) and first-overtone (RRc) RRLs using Gaia G-band light curves. This approach extends our previous work on RRab stars to include RRc stars, aiming for high predictive accuracy and broad generalization across both pulsation types. The model is based on a Gated Recurrent Unit (GRU) neural network optimized for time-series extrinsic regression. Our pipeline includes preprocessing steps such as phase folding, smoothing, and sample weighting, and uses photometric metallicities from the literature as training targets. The architecture is designed to handle morphological differences between RRab and RRc light curves without requiring separate models. On held-out validation sets, our GRU model achieves strong performance: for RRab stars, MAE = 0.0565 dex, RMSE = 0.0765 dex, R^2 = 0.9401; for RRc stars, MAE = 0.0505 dex, RMSE = 0.0720 dex, R^2 = 0.9625. These results show the effectiveness of deep learning for large-scale photometric metallicity estimation and support its application to studies of stellar populations and Galactic structure.

Figures (8)

• Figure 1: Distributions of RRab and RRc stars from our selected development datasets on the amplitude in the $G$ band versus period diagram, color-coded by metallicity.
• Figure 2: Normalized splined $G$-band light curves of 6002 RRab stars (a) and 6613 RRc stars (b) from our development datasets.
• Figure 3: Photometric metallicity distributions and corresponding sample weights (black lines) for RRab ( left panel) and RRc ( right panel) stars. Regions with lower data density are assigned higher weights to ensure a balanced contribution during model training.
• Figure 4: Schematic overview of the GRU-based neural network architecture used for predicting stellar metallicity ([Fe/H]) from pre-processed light curves. The model comprises an input layer, a sequence of GRU layers with Tanh activations, interleaved with dropout layers to prevent overfitting, followed by a dense linear layer producing the final regression output.
• Figure 5: Training (red) and validation (green) loss curves across epochs for each of the five cross-validation folds.The left panel (a) corresponds to RRab stars, while the right panel (b) represents RRc stars. A steady decrease in both losses indicates effective model learning, while the close alignment between training and validation loss across folds suggests good generalization and minimal overfitting. Darker colors denote greater consistency between folds.