TESS-Gaia Light Curve: a PSF-based TESS FFI light curve product

TL;DR

The paper tackles the challenge of extracting high-precision TESS light curves from full-frame images for millions of stars, including in crowded fields. It develops a Gaia-informed forward-modeling pipeline using a local, linear ePSF framework and sophisticated background modeling to decontaminate FFIs per epoch, producing PSF and aperture light curves (TGLC) down to TESS magnitude 16 and releasing them via MAST HLSP along with an open-source tglc package. The resulting photometry achieves photometric precision near pre-launch expectations (≈<2% for 16th mag) and enables robust exoplanet and variable-star science, as demonstrated by exoplanet case studies showing improved parameter estimates and cross-sector consistency. Overall, TGLC provides a scalable, reproducible approach to maximize the scientific return of TESS FFIs by leveraging Gaia priors for precise, decontaminated photometry across the sky.

Abstract

The Transiting Exoplanet Survey Satellite (TESS) is continuing its second extended mission after 55 sectors of observations. TESS publishes full-frame images (FFI) at a cadence of 1800, 600, or 200 seconds, allowing light curves to be extracted for stars beyond a limited number of pre-selected stars. Simulations show that thousands of exoplanets, eclipsing binaries, variable stars, and other astrophysical transients can be found in these FFI light curves. To obtain high-precision light curves, we forward model the FFI with the effective point spread function to remove contamination from nearby stars. We adopt star positions and magnitudes from Gaia DR3 as priors. The resulting light curves, called TESS-Gaia Light Curves (TGLC), show a photometric precision closely tracking the pre-launch prediction of the noise level. TGLC's photometric precision reaches <~2% at 16th TESS magnitude even in crowded fields. We publish TGLC Aperture and PSF light curves for stars down to 16th TESS magnitude through the Mikulski Archive for Space Telescopes (MAST) for all available sectors and will continue to deliver future light curves via DOI: 10.17909/610m-9474. The open-source package tglc is publicly available to enable any user to produce customized light curves.

Figures (13)

• Figure 1: Light curves of five TESS-discovered exoplanets from existing pipelines. Compared to the published work (similar to the SPOC 2-min light curves), the current FFI light curves perform poorer, especially for dimmer stars. The legend in each subplot indicates the sector(s) plotted. Light curves from extended missions are binned to a 30-minute cadence for compatible noise level with the primary mission. The periods and transit midpoints are adopted from each discovery literature. All lightcurves are detrended with wotanwotan and the phases are normalized to 1.
• Figure 2: Interpolation example of an ePSF model. The white squares and dots represent the pixels and their centers; the red dots represent the twice oversampled ePSF model; the dot sizes represent the ePSF values. The star represented by the red ePSF values is located at $(0, 0)$, the location of the largest red dot. Each pixel value (grayscale) is interpolated only by its nearest four values in the oversampled ePSF (red points) as shown in pixel (0, 1) by the white lines. This maintains the linearity (and therefore the computational efficiency) of the model we use to forward model a subimage of a full-frame TESS image from an ePSF.
• Figure 3: Example of background removal of TGLC. The left panel shows a FFI cutout with a background gradient and three vertical straps. We model the background as the middle panel. The residual image on the right shows a much cleaner field ready for an ePSF fit. Note that we preserve the resolution of the color map and only shift it down by 100 $e^-/s$ for the last panel.
• Figure 4: Normalized median absolute deviation (MAD) of residual images with different weighting power $l$ (Equation \ref{['eq:fullmodel_matrix_with_p']}). Larger values of $p$ weight lower count-rate pixels more heavily in deriving the ePSF, with $p=1$ weighting all pixels equally. These blue curves are generated from 196 different cutouts of Sector 7. The orange curve shows the MAD taken over all pixels from 196 residual images, which favors a weighting power $l \approx 1.4$
• Figure 5: $7 \times 14$ effective PSF models for half of Sector 1, camera 4, CCD 3. Each ePSF model is fitted in a $150 \times 150$ pixels FFI cutout, and $14 \times 14$ models will exactly cover the $2048 \times 2048$ pixels image with two-pixel-wide overlaps. Each CCD is divided in this manner to produce light curves published on MAST. We can observe the gradual spatial variation of ePSF: An obvious trend is that the ePSFs are narrower in the upper right corner, which is the closest to the center of the lens. One plausible reason is the optics of the telescope produce more compact pixel response functions (PRF) near boresights according to TESS Instrument Handbook v0.1. The gradual variation supports our assumption of constant ePSF in each cutout at the beginning of Section \ref{['sec:method']}.