带有光谱红移和机器学习测光红移的盖亚DR3星系候选体增强星表
An Enhanced Catalog of Gaia DR3 Galaxy Candidates with Spectroscopic and Machine-Learning Photometric Redshifts
浏览论文内容
中文总结 AI 辅助
该研究通过交叉匹配光谱红移星表、用MBRNN模型计算测光红移，更新了盖亚DR3星系候选体星表，提升了其科学价值，为河外及宇宙学研究提供了重要资源。
中文摘要 AI 辅助
我们发布了基于盖亚（Gaia）数据发布3（DR3）星系候选体表的更新星系星表，新增了光谱红移和基于机器学习的测光红移估计值。尽管盖亚DR3星系候选体的原始星表提供了前所未有的全天河外天体样本，但由于缺少红移信息和源纯度相对较低，其在河外及宇宙学研究中的科学效用受到限制。为解决这些局限，我们将盖亚DR3星系候选体星表与现有光谱红移星表（包括DESI、SDSS、2MRS和NED）进行交叉匹配。对于无光谱测量值的源，我们使用多区间回归神经网络（Multiple-Bin Regression Neural Network，MBRNN）机器学习模型计算测光红移。该模型的输入特征来自新构建的多波段星表，该星表结合了盖亚DR3测量值、2MASS和unWISE测光数据，以及形状参数、类星体（QSO）分类标志和消光值。模型利用光谱子样本进行训练和验证。最终，24.5%的星表源被分配了光谱红移，其余天体则提供了测光红移。与仅基于盖亚数据的已发表测光红移估计值相比，MBRNN模型实现了更高的红移精度和准确度。除红移值外，我们还提供了量化给定源为统计异常值可能性的评分。生成的星表显著提升了盖亚DR3星系候选体样本的科学价值，为大规模河外及宇宙学研究提供了宝贵资源。
英文摘要
We present an updated galaxy catalog based on the Gaia Data Release 3 (DR3) Galaxy Candidates table, augmented with both spectroscopic and machine-learning based photometric redshift estimates. While the original catalog of Gaia DR3 galaxy candidates provides an unprecedented all-sky sample of extragalactic sources, its scientific utility for extragalactic and cosmological studies has been limited by the absence of redshift information and by its relatively low source purity. To address these limitations, we cross-match the catalog of Gaia DR3 galaxy candidates with existing spectroscopic redshift catalogs including DESI, SDSS, 2MRS, and NED. For sources without spectroscopic measurements, we compute photometric redshifts using a Multiple-Bin Regression Neural Network (MBRNN) machine-learning model. The input features for this model are drawn from a newly constructed multiwavelength catalog that combines Gaia DR3 measurements with 2MASS and unWISE photometry, as well as shape parameters, QSO classification flags, and extinction values. The model is trained and validated using the spectroscopic subsample. As a result, spectroscopic redshifts are assigned to 24.5\% of the catalog sources, while photometric redshifts are provided for the remaining objects. The MBRNN model achieves improved redshift accuracy and precision compared to previously published photometric redshift estimates based on Gaia data alone. In addition to redshift values, we provide a score quantifying the likelihood that a given source represents a statistical outlier. The resulting catalog significantly enhances the scientific value of the Gaia DR3 galaxy candidates sample and provides a valuable resource for large-scale extragalactic and cosmological studies.
