High-fidelity stellar extinction with Gaia and APOGEE -- I. The method and a new extinction curve

TL;DR

This work addresses the challenge of precise extinction corrections for Gaia-based stellar parameters by deriving star-by-star reddening and extinction for APOGEE DR19 stars using intrinsic colors predicted from APOGEE parameters with an XGBoost quantile model. A robust anchor, $A_{BP}/A_{RP}=1.694\pm0.004$, ties reddening to absolute extinctions, and a new empirical extinction curve is built for 39 broadband filters across 10 systems, achieving an Av precision of about $0.03$ mag. The study also maps $R_V$ across the APOGEE DR19 footprint, finding an average $R_V\approx3.15$ with significant spatial structure, and provides a publicly available catalog of $A_{\zeta}/A_V$ and $A_{\zeta}/E(B-V)$ for broad application. These results enable improved stellar parameter inference from Gaia parallaxes, calibrate differential reddening in clusters, and offer a path toward unifying diverse dust maps into a coherent all-sky extinction framework, with broad implications for Galactic structure and stellar population studies.

Abstract

The scarcity of high-fidelity extinction measurements remains a bottleneck in deriving accurate stellar properties from Gaia parallaxes. In this work, we aim to derive precision extinction estimates for APOGEE DR19 stars, establishing a new benchmark for Galactic stellar population studies. We first determine reddening by comparing observed colorsr, etrieved from photometric surveys or standardized synthetic magnitudes from Gaia BP/RP spectra, to intrinsic colors predicted via an XGBoost model. The model is trained on minimally reddened stars to infer intrinsic colors and their associated uncertainties, using APOGEE stellar parameters (Teff, logg, [Fe/H], and [alpha/Fe]). The derived reddening values are then converted into extinctions using an anchor ratio of A_BP / A_RP = 1.694 +/- 0.004, derived from red-clump-like stars. Here, we provide extinction measurements in 39 filters across 10 photometric systems and introduce a new empirical extinction curve optimized for broadband passbands. Our extinction estimates (Av) outperform existing results (Bayestar19, StarHorse, SEDEX), achieving a typical precision of 0.03 mag in Av. Notably, we identify systematic deviations of up to 30% between monochromatic and passband-integrated extinction ratios at wavelengths greater than 700 nm. This result highlights the necessity of adopting passband-specific coefficients when correcting extinction to derive stellar parameters. As the foundation for a forthcoming series of papers, these benchmark measurements will be used to (1) revise asteroseismic scaling relations, (2) calibrate differential reddening in open clusters, and (3) reconcile heterogeneous dust maps into a unified, all-sky extinction scheme.

Figures (11)

• Figure 1: Schematic overview of the methodology used to derive extinction ratios $A_\zeta/A_V$. Intrinsic colours are inferred from APOGEE stellar parameters using XGBoost quantile regression. Reddening is measured by comparing observed and predicted intrinsic colours, while red-clump-like stars with precise Gaia distances provide an anchor extinction ratio $A_{BP}/A_{RP}$, enabling extinction estimates for individual filters and final extinction ratios.
• Figure 2: Comparison of observed and predicted intrinsic colors for 9126 stars in the validation set with negligible reddening ($E(B-V)_{\rm SFD} < 0.02$). The top panel shows $(BP - RP)$, while the bottom panel displays $(BP - J)$. Intrinsic colors are predicted using XGBoost regression models trained on the low-reddening stellar sample. The models utilize $T_{\rm eff}$, $\log g$, [Fe/H], and [$\alpha$/Fe] from APOGEE DR19 as input features, with the observed colors of these stars serving as target labels. $BP$ and $RP$ magnitudes are retrieved from Gaia DR3, and $J$ magnitudes are from 2MASS. Red points represent the median predicted values, with error bars indicating the $1\sigma$ uncertainty derived from the 16th and 84th percentiles.
• Figure 3: Comparison of reddening between $E(BP - RP)$ and $E(BP - \zeta)$, where $\zeta$ corresponds to various bands. The first row shows 2MASS bands $(J, H, K_S)$, and the second row shows WISE bands $(W1, W2)$. The black dashed line represents a quadratic fit to the binned averages of the data (green squares; associated standard errors are smaller than the symbol size), while the red dashed line indicates a linear fit to the average values. The density of points is depicted using a color map, with yellow regions indicating higher densities and purple regions representing lower densities.
• Figure 4: Extinction $A_{BP}$ versus color excess $E(BP - RP)$ for 8,426 red clump candidate stars (see the text for their selection), color-coded by star number density on a logarithmic scale. The orange dashed line represents the best linear fit (with $A_{BP}$ as the independent variable) to all the red clump candidate stars, while the red dashed line shows the best linear fit to the mean values binned in color excess $E(BP - RP)$ with a bin size of 0.12. In both regression models, $A_{BP}$ was treated as the independent variable and $E(BP - RP)$ as the dependent variable. The slope and intercept of each best-fit line are annotated in the legend. The $3\sigma$ outliers w.r.t the orange dashed line, comprising 1.7% of the full sample, are excluded and not shown here.
• Figure 5: Precision and accuracy of our extinction measurements in $A_V$. Upper panel: Comparison of $A_V$ values derived in this work with those from Bayestar19 (black), StarHorse (green), and SEDEX (purple). The red dashed line represents the 1:1 relationship. Lower panel: Distributions of $A_V$ derived in this work (blue) compared with those from SEDEX (purple) and StarHorse (green) for nearby stars with distances less than 70 pc. These stars are expected to experience low extinction, making the dispersion of these distributions indicative of the internal $A_V$ precision. The mean ($\mu$) and 1$\sigma$ uncertainty (calculated from the 16$^{\rm th}$ and 84$^{\rm th}$ percentiles) are shown in the plot for each method, with results for this work indicated by vertical dashed lines. The 1$\sigma$ dispersion is 0.042 for this work, compared to $0.096$ from SEDEX and $0.367$ from StarHorse. Negative $A_V$ values in both panels arise due to measurement uncertainty.