Source-linked AI summary
eleanor: An open-source tool for extracting light curves from the TESS Full-Frame Images
Adina D. Feinstein, Benjamin T. Montet, Daniel Foreman-Mackey, Megan E. Bedell, Nicholas Saunders, Jacob L. Bean, Jessie L. Christiansen, Christina Hedges, Rodrigo Luger, Daniel Scolnic, Jose Vinicius de Miranda Cardoso
TL;DR
TESS Full-Frame Images offer a large archive for exoplanet searches but are difficult to exploit directly. This paper presents eleanor, an open-source pipeline for extracting and optimizing FFI light curves, and demonstrates its capabilities by recovering known planets and identifying new candidates and stellar variability.
Problem
TESS FFIs provide extensive time-series data but their format and computational requirements make them difficult for users to analyze.
Method
eleanor extracts FFI light curves using photometry, systematic-error correction, and CDPP optimization for planet searches.
Results
eleanor recovers known transiting planets and presents new planet candidates and stellar variability in early FFI science results.
Takeaways & Limitations
The paper provides publicly available eleanor light-curve products and software for analyzing TESS FFIs across a diverse range of scientific applications.
Takeaways & Limitations
Saturated stars are beyond eleanor’s intended scope, and PSF modeling produces poor results for them.
Abstract
from arXiv · showhide
During its two year prime mission the Transiting Exoplanet Survey Satellite (TESS) will perform a time-series photometric survey covering over 80% of the sky. This survey comprises observations of 26 24 x 96 degree sectors that are each monitored continuously for approximately 27 days. The main goal of TESS is to find transiting planets around 200,000 pre-selected stars for which fixed aperture photometry is recorded every two minutes. However, TESS is also recording and delivering Full-Frame Images (FFIs) of each detector at a 30 minute cadence. We have created an open-source tool, eleanor, to produce light curves for objects in the TESS FFIs. Here, we describe the methods used in eleanor to produce light curves that are optimized for planet searches. The tool performs background subtraction, aperture and PSF photometry, decorrelation of instrument systematics, and cotrending using principal component analysis. We recover known transiting exoplanets in the FFIs to validate the pipeline and perform a limited search for new planet candidates in Sector 1. Our tests indicate that eleanor produces light curves with significantly less scatter than other tools that have been used in the literature. Cadence-stacked images, and raw and detrended eleanor light curves for each analyzed star will be hosted on MAST, with planet candidates on ExoFOP-TESS as Community TESS Objects of Interest (CTOIs). This work confirms the promise that the TESS FFIs will enable the detection of thousands of new exoplanets and a broad range of time domain astrophysics.
1. INTRODUCTION
TESS FFIs offer a large, broad-cadence archive for exoplanet searches and diverse astrophysics, but extracting usable light curves requires substantial correction and computational resources. The paper introduces eleanor, an open-source pipeline and publicly available light-curve products for this purpose.
- TESS survey: TESS’s two-year prime mission surveys roughly 80% of the sky in approximately 27-day sectors, with selected targets observed at two-minute cadence.The four cameras cover 96° × 24° sectors, and approximately 20,000 stars are observed every sector.
- TESS FFIs: FFIs capture each sector at 30-minute cadence and contain roughly one million stars brighter than I=16 mag, enabling broad data mining and transit searches.The archive also supports asteroseismology, Solar System research, and studies of galactic and extragalactic sources.
- Processing challenges: Systematic effects can overwhelm astrophysical signals in FFIs, while global background correction fails to account for localized issues.These effects are especially problematic when TESS is near perigee.
- Processing challenges: Each FFI is approximately 35 MB, requiring roughly 45 GB for one target’s sector photometry and 1 TB for the entire Southern Hemisphere.These storage demands make the archive difficult to exploit without substantial computational resources.
- Contribution: The paper presents eleanor, an open-source pipeline for extracting light curves from TESS FFIs, together with publicly available light-curve data products.The paper describes the methods, demonstrates early science results, and explains product availability and software access.
2. CREATING LIGHT CURVES
eleanor extracts FFI light curves through a staged process that includes pointing modeling, quality-flag assignment, postcard creation, background subtraction, and TPF extraction. It tests multiple apertures to select light curves optimized for transiting-exoplanet searches.
- FFI processing: eleanor first creates a pointing model and assigns quality flags at the FFI level.These steps precede image cutout and light-curve extraction.
- Postcard creation: It produces time-stacked, background-subtracted postcards measuring 148 × 104 pixels.The postcards serve as intermediate cutouts for subsequent Target Pixel File extraction.
- Target extraction: From the postcards, eleanor extracts 13 × 13-pixel Target Pixel Files and tests multiple apertures for each target.Aperture selection is aimed at finding the best light curve for transiting-exoplanet searches.
2.1. Pointing Model
eleanor builds a detector pointing model because spacecraft motion can make the FFIs’ header WCS inaccurate for transforming pixel coordinates to sky positions. It selects suitable TIC stars and fits an affine transformation to align predicted and observed positions at each cadence.
- The pipeline downloads all FFIs for a sector and builds a pointing model to determine each star’s true detector position.
- Spacecraft motion can make the WCS recorded in FFI headers inaccurate for transforming pixel space to sky position.
- Target stars are selected from TIC version 7.0 using 7.5 ≤ T mag ≤ 12.5 and 0 ≤ contamination ≤ 5 × 10^-3.
- At each cadence, an affine transformation minimizes squared differences between predicted and observed detector positions for the selected stars.The transformation captures spacecraft rotation or translation and apparent stellar-position changes from differential velocity aberration.
2.2. From FFIs to Postcards
The pipeline converts TESS FFIs into time-stacked, background-subtracted postcards for efficient single-target analysis. These localized products support background subtraction, TPF and light-curve extraction, systematics modeling, and preservation of WCS and quality flags.
- Postcard construction: Postcards are 148 × 104 pixel FFI cutouts with a 50 pixel overlap, time-stacked across available cadences and background-subtracted.The overlap reduces edge effects for individual stars and makes postcards more efficient for single-target analysis.
- Background subtraction: A constant postcard-level background is calculated for each cadence with photutils’ MMMBackground to address structured, spatially varying FFI backgrounds.The localized postcard scale provides a sufficient region for initial background subtraction.
- Data products: TPFs and light curves are extracted from the background-subtracted postcards.The pipeline also models a two-dimensional background using low-flux pixels and principal-component analysis.
- Background systematics: The two-dimensional background model uses the bottom 40% of postcard pixels, approximately 6300 pixels, five principal modes, and shifts of up to 15 cadences.Frames without quality flags are stacked before low-flux pixels are selected and common systematics are identified with PCA.
- Metadata and quality control: Postcards conserve each FFI’s WCS and include quality flags that identify potentially corrupted cadences.The paper notes that quality-flag assignment follows a two-step process described in the section.
2.3. Quality Flags
eleanor propagates eight TESS quality flags applicable to Full-Frame Images and adds a pointing-model flag for cadences with poorly modeled spacecraft motion. The custom flag is assigned per orbit after iterative 2σ clipping and is combined with existing short-cadence quality flags.
- Mission quality flags: Eight of TESS’s twelve quality flags apply to Full-Frame Images, covering spacecraft pointing, brightening, desaturation, cosmic rays, stray light, and processing exclusions.The listed issues include attitude tweaks, coarse or Earth point, argabrightening, reaction-wheel desaturation, collateral-pixel cosmic rays, Earth or Moon stray light, and manual exclusions.
- Pointing-model flag: eleanor introduces a pointing-model quality flag by fitting measured x and y pixel coordinates and iteratively sigma-clipping at 2σ.Bad-pointing-model cadences are identified from these fits, performed independently for each orbit.
- Flag propagation: The custom pointing-model flag has value 4096 and is applied to cadences that already carry short-cadence quality flags.The short-cadence flags are copied into postcards by identifying short-cadence targets located on each camera.
2.4. From TPFs to Light curves
eleanor extracts 13 × 13-pixel TPFs, tests diverse aperture masks and background treatments, and selects the light curve with minimum one-hour CDPP for transit detection. It also provides PCA detrending and optional PSF photometry, which substantially improves precision for WASP-100.
- TPF extraction: 13 × 13-pixel TPFs are extracted from the postcard containing each target closest to its center, and photometry is performed at the TPF level.Targets can appear on multiple postcards because adjacent postcards overlap by 50 pixels.
- Aperture photometry: eleanor tests apertures spanning rectangular, L-shaped, circular, square, rotated, binary, and weighted designs, while allowing users to define custom masks.All tested apertures and their extracted light curves are retained in the data product.
- Flux extraction and correction: RAW FLUX is the background-subtracted aperture–TPF pixel sum, while CORR FLUX applies orbit-by-orbit correction for instrumental systematics.Background subtraction is performed on the postcard, with additional TPF-level background treatments tested for precision.
- Light-curve selection: The ideal aperture and background treatment minimize one-hour CDPP, preserving short-timescale transit features and recording the selected configuration in the data-product header.CDPP is measured in parts per million and assesses sensitivity to weak terrestrial planet transits.
- Systematics detrending: PCA removes additional systematics shared by nearby detector stars, and PCA FLUX subtracts the first 3 CBVs while allowing users to vary the component count.All SPOC-generated CBVs are available for further analyses.
- PSF photometry: 161 ppm versus 279 ppm: for WASP-100, PSF time-series photometry and PSF-regressed flux outperform standard corrected flux in one-hour CDPP.The PSF model fits a uniform background and Gaussians for stars, with positions constrained and the Gaussian network allowed to vary by frame.
2.5. The eleanor Data Product
The eleanor data product is a FITS file containing cadence-stacked pixel data, aperture masks, and raw and corrected fluxes for light-curve analysis. Community users can generate light curves for fainter or extragalactic objects and optimize analyses of individual targets.
- File contents: Each FITS file contains cadence-stacked, background-subtracted flux pixels and corresponding flux-error pixels for a 13 × 13 region centered on the source.The region is stored at the cadence-stacked level for both flux and flux-error data.
- File contents: The files include all 21 aperture masks tested during light-curve extraction, along with raw and corrected fluxes.These products preserve intermediate and corrected photometric outputs across the tested apertures.
- Community use: Community members can use eleanor to create light curves for fainter or extragalactic objects or optimize analyses of individual objects.The package supports applications beyond the primary analyzed targets.
3. RESULTS
eleanor produces low-scatter FFI light curves, recovers known transiting planets and transient events, and identifies new periodic-signal candidates. Performance varies with detector systematics, neighboring-star contamination, moving objects, saturation, and crowded fields.
- CDPP performance: Camera 4 CCD 4 experiences more systematics than other CCDs, increasing overall CDPP, while stable diffuse LMC light does not significantly alter CDPP.CDPP remains fairly consistent across CCDs within a camera except for Camera 4 CCD 4.
- CDPP performance: 2,200 stars with Tmag > 12 fell below the general CDPP trend, and 90% had brighter neighbors within 50”, indicating aperture contamination.The neighboring stars are roughly two TESS pixels away and decrease overall CDPP for these faint stars.
- Pipeline comparison: For TOI-172, eleanor achieved CDPP = 325 ppm versus 376 ppm for QLP and 579 ppm for Oelkers & Stassun (2019).The comparison used a Tmag = 10.71 host star and uniform eleanor quality flags.
- Pipeline comparison: Approximately 12,000 of 17,000 crossmatched Sector 1 stars had OS19/eleanor CDPP > 1, suggesting less scatter in eleanor light curves.The comparison used light curves for the same TIC IDs and calculated CDPP identically.
- Astrophysical validation: eleanor recovered known transiting planets and known supernovae, including Type Ia and Type II events, from TESS FFI observations.The analysis derived planet parameters from known transits and recovered events across Sectors 1 and 2.
- Candidate search: By-eye vetting retained periodic-signal candidates from a limited Sector 1 search, but further vetting is required and candidates will be hosted as CTOIs.The reported candidate list is incomplete, with a full list planned for future work.
4. DATA AVAILABILITY AND SOFTWARE TOOLS
eleanor provides community-accessible light-curve products and an open-source package for generating TPFs and light curves from user-specified sources or detector positions. Users can customize extraction and background-subtraction settings, while new candidates and other astrophysical events will be publicly released.
- Data products: eleanor data products will be released for each Southern Hemisphere sector and later reprocessed into a uniform library.Reprocessing may improve earlier light curves as methods for background removal, spacecraft pointing, and crowded regions develop.
- Software access: The open-source eleanor package is downloadable through GitHub or the Python Package Index and creates TPFs and light curves from TIC IDs, Gaia IDs, or coordinates.Users can request any source or detector position, after which eleanor performs the described processing steps.
- User customization: Users can customize TPF dimensions, enable PSF-modeled light curves, and select the region used for background subtraction.TPF height and width must be odd, and publicly available products use a 13 × 13 pixel background region by default.
- Community releases: After each sector, the team will search FFIs for new planet candidates and publicly release candidate light curves through ExoFOP-TESS.The planned catalog will also include eclipsing binaries, RR Lyrae stars, and other interesting astrophysical events.
- Community releases: The open-source eleanor products will support a diverse set of scientific discoveries achievable with TESS FFIs.This extends the intended community use beyond the planet-candidate search.
APPENDIX A. eleanor SOFTWARE DEMONSTRATION
The demonstration shows how to generate eleanor light curves for a TIC target or coordinate set, including sector selection, quality filtering, PCA correction, and optional PSF modeling. It also lists the resulting raw, corrected, PCA, and PSF flux products and their availability conditions.
- Target-based extraction: A known TIC ID can be passed to eleanor.Source with a chosen sector, including when the target was observed in multiple sectors.The example uses TIC 38846515 in Sector 1.
- Target-based extraction: PSF-modeled flux must be explicitly enabled in eleanor.TargetData, while PCA correction is enabled through do_pca=True.The demonstration sets do_psf=True and do_pca=True; PSF modeling otherwise defaults to off.
- Target-based extraction: Quality filtering selects cadences with data.quality == 0 before extracting raw_flux, corr_flux, pca_flux, and psf_flux.These arrays represent the available flux options after applying the quality mask.
- Coordinate-based extraction: Coordinate-based extraction supports targets observed in multiple sectors through eleanor.multi_sectors, which returns a list of Source objects.The example requests sectors [1, 2] for specified right ascension and declination coordinates.
- Coordinate-based extraction: Available flux products depend on processing: data0 lacks PSF-modeled flux, while data1 lacks PCA-corrected flux and provides raw_flux, corr_flux, and psf_flux.The demonstration also calls PSF light-curve generation with a Gaussian model and Poisson likelihood.