The TEQUILA catalog of variables in TESS full-frame images: Differential photometry light curves from the first two years of observations

TL;DR

The TEQUILA catalog provides a comprehensive set of over six million light curves from TESS FFIs, enabling studies of stellar variability and transients with improved processing techniques and machine learning classification.

Contribution

This work introduces a new pipeline for extracting and analyzing light curves from TESS FFIs, producing a large, accessible catalog with automated noise and artifact identification.

Findings

Over six million light curves produced from TESS FFIs

Median differential variability noise achieved between 10^-3 and 10^0

Convolutional neural network effectively identifies instrumental noise in light curves

Abstract

Stellar variability and transient events provide critical insights into astrophysics, accelerated by missions like CoRoT, Kepler, and K2. NASA's Transiting Exoplanet Survey Satellite (TESS) adds a unique combination of long baseline and all-sky coverage, though extracting light curves from full-frame images (FFIs) is challenging due to scattered light and blending. We processed TESS FFIs to produce TEQUILA (TESS quick-look and light curve analysis), a comprehensive catalog of variable point sources from the prime mission, enabling diverse studies without requiring raw data processing. We used difference image analysis, constructing reference images from quality-filtered FFIs for each CCD across sectors 1-26. Iterative subtraction mitigated systematics, and light curves were created using aperture photometry for sources varying in residual images. The pipeline yields over six million light curves, including stellar variables, transients, systematics, and moving objects. Approximately $6 \times 10^5$ span multiple sectors, with roughly $10^3$ from continuous viewing zones. We achieve median differential variability noise of $10^{-3}$ to $10^0$ for sources between 5.0 and 16.0 Tmag, while typical photometric RMS variability ranges from $10^{-2}$ to $10^1$. A convolutional neural network identifies light curves caused by instrumental noise, assigning a confidence score to each classification. To avoid confusion with astrophysical variables, we also flag light curves prompted by known Solar System objects (SSOs). All light curves are accessible via MAST as a High-Level Science Product. This catalog serves as a discovery tool for new variables; future work will refine methods and extend coverage to the TESS extended mission.