Abstract
Cosmological observations must be corrected for Galactic dust extinction, yet existing maps often contain spatial correlations with the large-scale structure (LSS) of the Universe. These correlations introduce systematic biases into sensitive cosmological measurements. We present two new high-latitude integrated extinction maps based on stellar reddenings, utilizing a rigorous selection process to filter out galaxies and quasars. Our maps demonstrate significantly reduced spatial correlation with LSS compared to previous emission-based and stellar-reddening-based maps.
Introduction: The Problem of Contamination
To study the distant Universe, astronomers must account for the "cosmic fog" of the Milky Way—interstellar dust that causes extinction. While dust maps are essential for dereddening extragalactic objects, current maps are often "contaminated" by the very structures they aim to study. Emission-based maps (e.g., SFD) can pick up the cosmic infrared background from unresolved galaxies, while stellar-reddening maps can be biased by misclassified extragalactic objects in stellar catalogs. If dust corrections are correlated with the target galaxies, it introduces systematic errors in measurements of primordial non-Gaussianity ($f_{NL}$) and gravitational lensing.
Methodology: Purity Over Completeness
The authors developed two high-latitude extinction maps with resolutions of 6.1′ and 15′. The core innovation lies in a multi-stage "veto" process designed to ensure only true stars are used for reddening estimation.
Strict Filtering with SVMs
Using the Bayestar framework, the researchers applied Support Vector Machines (SVMs) trained on optical (SDSS) and infrared (WISE) colors to distinguish stars from quasars and galaxies. This "purity over completeness" approach reduced quasar contamination in test samples from 15% to approximately 0.1%.
2D Reconstruction
Unlike 3D maps that can be noisy when integrated, these maps are reconstructed in 2D. This approach provides a cleaner signal for cosmological applications by avoiding the discretization artifacts inherent in 3D models.
Data and Validation
The study integrated data from Pan-STARRS1 (PS1), 2MASS, and Gaia EDR3. To prove the maps are "clean," the authors measured the angular cross-correlation between their dust maps and known tracers of LSS at various redshifts.
The tracers included samples from the SDSS Main Galaxy Sample, BOSS (LOWZ and CMASS), and the DR14 Quasar catalog.
Key Results
The new maps show spatial correlations with extragalactic structures that are largely within the 1$\sigma$ noise level. In contrast, older maps like SFD or Bayestar19 show significant correlations at specific redshifts. Notably, the new maps successfully eliminated the "dip" in extinction previously found in stellar-based maps at redshifts where quasars are most dense ($z \approx 1.5$ to $2.5$). Despite the strict filtering, the maps remain consistent with the SFD map in low-extinction regions ($E(B-V) < 0.2$ mag).
Limitations and Future Outlook
The current maps are limited to the Pan-STARRS1 footprint ($\delta > -30^\circ$) and high Galactic latitudes ($|b| > 20^\circ$). Additionally, the strict filtering results in higher statistical noise in star-sparse regions. However, upcoming data from the Vera C. Rubin Observatory (LSST) will observe fainter stars, allowing for higher-resolution maps with lower noise while maintaining these high purity standards.
Reference: Mudur, N., Park, C. F., & Finkbeiner, D. P. (2023). Stellar-reddening-based Extinction Maps for Cosmological Applications. The Astrophysical Journal, 949:47. doi:10.3847/1538-4357/acc32c
Summary updated