[Paper Review] Gaia Data Release 1: The variability processing & analysis and its application to the south ecliptic pole region

L. Eyer, N. Mowlavï+71 more|arXiv (Cornell University)|Feb 10, 2017

Astronomy and Astrophysical Research19 citations

TL;DR

This paper presents the variability processing pipeline for Gaia Data Release 1, focusing on the South Ecliptic Pole region. Using multi-epoch G-band photometry from the first 14 months of Gaia operations, the method detects and classifies variable stars—particularly Cepheids and RR Lyrae—through statistical and machine learning techniques, identifying 3,194 such stars with 67% and 58% completeness for Cepheids and RR Lyrae, respectively.

ABSTRACT

The ESA Gaia mission provides a unique time-domain survey for more than one billion sources brighter than G=20.7 mag. Gaia offers the unprecedented opportunity to study variability phenomena in the Universe thanks to multi-epoch G-magnitude photometry in addition to astrometry, blue and red spectro-photometry, and spectroscopy. Within the Gaia Consortium, Coordination Unit 7 has the responsibility to detect variable objects, classify them, derive characteristic parameters for specific variability classes, and provide global descriptions of variable phenomena. We describe the variability processing and analysis that we plan to apply to the successive data releases, and we present its application to the G-band photometry results of the first 14 months of Gaia operations that comprises 28 days of Ecliptic Pole Scanning Law and 13 months of Nominal Scanning Law. Out of the 694 million, all-sky, sources that have calibrated G-band photometry in this first stage of the mission, about 2.3 million sources that have at least 20 observations are located within 38 degrees from the South Ecliptic Pole. We detect about 14% of them as variable candidates, among which the automated classification identified 9347 Cepheid and RR Lyrae candidates. Additional visual inspections and selection criteria led to the publication of 3194 Cepheid and RR Lyrae stars, described in Clementini et al. (2016). Under the restrictive conditions for DR1, the completenesses of Cepheids and RR Lyrae stars are estimated at 67% and 58%, respectively, numbers that will significantly increase with subsequent Gaia data releases. Data processing within the Gaia Consortium is iterative, the quality of the data and the results being improved at each iteration. The results presented in this article show a glimpse of the exceptional harvest that is to be expected from the Gaia mission for variability phenomena. [abridged]

Motivation & Objective

• To develop and validate a robust pipeline for detecting and classifying variable stars in the Gaia time-domain survey.
• To assess the performance of machine learning classifiers on early Gaia photometric data, particularly for Cepheids and RR Lyrae stars.
• To provide a high-fidelity, homogeneous, all-sky variability catalog with characteristic parameters and time-series data for key stellar populations.
• To evaluate the completeness and reliability of variable star detection under the initial data quality and sampling conditions of Gaia DR1.
• To demonstrate the feasibility and scientific potential of Gaia’s time-domain survey for stellar variability studies.

Proposed method

• Employed classical statistical methods, data mining, and time series analysis techniques tailored to Gaia’s unique photometric sampling and noise characteristics.
• Developed custom software tools for data handling, visualization, and validation of variability detection results.
• Applied three machine learning classifiers—Random Forest, Boosted Bayesian Networks, and Gaussian Mixture models—to classify variable stars based on light curve features.
• Used Kullback-Leibler divergence to compare classifier outputs with reference data from OGLE surveys, validating classification accuracy.
• Applied membership probability thresholds (p > 0.5 to p > 0.9) to refine classifier outputs and reduce false positives.
• Conducted visual inspections and applied selection criteria to finalize the final catalog of 3,194 confirmed Cepheids and RR Lyrae stars.

Experimental results

Research questions

• RQ1 How effective are machine learning classifiers in detecting and classifying Cepheids and RR Lyrae stars in early Gaia photometric data?
• RQ2 What is the completeness of Gaia DR1 in detecting known Cepheid and RR Lyrae populations in the South Ecliptic Pole region?
• RQ3 How do the classifier outputs compare with established literature data from OGLE surveys in terms of statistical consistency?
• RQ4 To what extent does the Gaia sampling law and photometric quality affect the reliability of variability detection and classification?
• RQ5 What is the impact of iterative data processing on the quality and completeness of variability results in successive Gaia data releases?

Key findings

• Out of 694 million sources with calibrated G-band photometry, 2.3 million had at least 20 observations within 38 degrees of the South Ecliptic Pole.
• Approximately 14% of these 2.3 million sources were detected as variable candidates, with 9,347 identified as Cepheid or RR Lyrae candidates via automated classification.
• After visual inspection and selection, 3,194 Cepheids and RR Lyrae stars were published, including 443 new RR Lyrae and 43 new Cepheids.
• The completeness for Cepheids was estimated at 67%, and for RR Lyrae stars at 58%, under the restrictive conditions of Gaia DR1.
• Classifier performance was validated using Kullback-Leibler divergence, showing strong agreement with OGLE reference data sets.
• The results demonstrate that Gaia’s time-domain survey, though in early stages, already provides high-quality, homogeneous, and scientifically valuable variability data for the stellar population around the South Ecliptic Pole.