New Software Reveals a Giant Planet in Webb Data

Astronomers have recovered a giant planet’s faint signal from James Webb Space Telescope observations by developing a more precise way to subtract its star’s glare. The signal had remained buried in leftover starlight after conventional processing.

Astronomers first discovered the planet, 51 Eridani b, in 2015. The new work gives them their first image of it at an infrared wavelength of 4.3 microns, where carbon dioxide absorbs light and makes the planet particularly difficult to see.

The technique, described in Nature on November 4, improved sensitivity to faint planets by up to 15 times in the observations tested. At the location of 51 Eridani b, the improvement was about fourfold. The researchers achieved those gains by reprocessing existing observations, without changing Webb’s hardware.

That newly recovered light could help researchers investigate the giant planet’s atmosphere and formation history. It also gives astronomers a reason to revisit other observations in which residual starlight may have concealed useful planetary signals.

A planet inside the glare

The team, led by Northwestern University astronomy doctoral student Rodrigo Ferrer-Chavez, tested its method on Webb observations of three stars. Two, 51 Eridani and HIP 65426, have known planets. The third, TYC 5899, provided a test of how well the software could recover artificial planets inserted into real telescope data.

For 51 Eridani b, the telling comparison used the same observation processed two ways. The conventional approach left the planet indistinguishable from the star’s remaining glare. The new method recovered a signal at the planet’s expected location, with a reported signal-to-noise ratio of 6.3.

“We showed that we can actually recover a planet that people previously couldn’t see,” co-author Jason Wang said in Northwestern’s announcement.

Webb uses a coronagraph to block much of the star’s light, but some still reaches the detector. Small optical imperfections and alignment changes alter the pattern of that leftover light.

Astronomers commonly estimate the unwanted pattern from reference images, then subtract it. The difficulty is that the telescope can change slightly between observations. Even a good reference can leave residual structure strong enough to overwhelm a faint planet.

Modeling how the telescope behaves

The new method is called Physics-based Learning for Imaging Planets, or FLIP. It builds a computational model of how starlight travels through Webb’s optics and reaches the Near-Infrared Camera, NIRCam. The researchers can adjust this model to reproduce the glare in a particular observation.

FLIP first uses a reference star to calibrate its model, bringing the simulation close to the instrument’s real behavior. It then adjusts selected parameters against the target observation to account for the differences that ordinary reference subtraction struggles to remove.

Those parameters describe the star’s position and brightness, distortions in the light passing through the instrument, and the response of individual detector pixels. The software repeatedly compares the modeled image with the observed one and changes the model to reduce the mismatch.

Making that process practical requires automatic differentiation, a computational technique also used to train machine-learning models. It efficiently calculates how parameter changes affect the mismatch, guiding the next adjustments. FLIP applies those tools to a physical model of the instrument; its job is to reconstruct the unwanted starlight.

There is a risk in fitting a model this flexible: it could start removing a real planet along with the glare. The researchers inserted artificial planets and followed how well those signals survived as the optimization proceeded. They stopped when the signals reached their best strength or leveled off, and used the tests to calibrate signal loss and detection sensitivity.

The largest reported gain, roughly 15 times better sensitivity, occurred in the TYC 5899 data at a separation of 0.5 arcseconds from the star. That number describes how faint a planetary signal the processing could recover under those conditions. The gain varied with the observation and distance from the star. The TYC 5899 test used artificial planetary signals inserted into the observations.

More science from observations already taken

The recovered 4.3-micron image is particularly useful because Earth’s atmosphere interferes with ground-based observations in that wavelength range. Webb can observe it from space, but separating a faint planet from its host star remains essential.

Carbon dioxide absorbs strongly in this part of the spectrum. A reliable measurement of the planet’s brightness there, combined with other observations and atmospheric models, could help constrain its composition. The authors identify that detailed atmospheric analysis as future work. The FLIP paper demonstrates the recovery of the signal.

The immediate opportunity is to apply the method more widely to Webb data, potentially bringing colder, older or lower-mass giant planets within reach. The researchers have made their code and example data available, giving other teams a way to test the approach.

The study covered four NIRCam datasets across three stars, and none reached the fundamental photon-noise limit, the uncertainty associated with collecting a finite number of photons. The authors identify improvements to detector modeling and computational efficiency among the next steps.

Farther ahead, the approach could inform NASA’s planned Habitable Worlds Observatory, which aims to directly image Earth-like worlds. Such planets demand far more extreme suppression of their stars’ light. The paper suggests that sufficiently accurate modeling might ease some telescope-stability requirements if the technique can be extended to those much harder conditions.

Astronomers can now study a measurable 4.3-micron signal from a known planet in observations where it had previously been hidden.

Sources

Ferrer-Chavez R et al. Imaging exoplanets below the telescope stability limit. Nature 659, 58–62. Published November 4, 2026.

Northwestern University. Hidden planet emerges from the glare. November 4, 2026. Source of the Jason Wang quotation.

NASA Science. 51 Eridani b. Discovery history.

NASA Science. Habitable Worlds Observatory. Mission objectives.

The Nature paper is licensed under Creative Commons Attribution 4.0. Its findings are paraphrased here in an original news article.