Better Starlight Removal Lets Webb See Fainter Planets Without New Hardware

A team led from Northwestern University reports a new way of subtracting starlight from James Webb Space Telescope images that reaches 2 to 15 times the sensitivity of standard processing, with no change to the telescope. Run on archived Webb data, it detected the giant planet 51 Eridani b at a wavelength of 4.3 microns, where the original analysis of the same observations found nothing. The work was published Nov. 4, 2026, in Nature.

The team writes in the paper that modeling starlight more accurately than the telescope's own steadiness allows would put more planets within reach of direct imaging, and would be a step toward imaging Earth-like planets with future observatories. Rodrigo Ferrer-Chavez of Northwestern led the work, with Jason J. Wang of Northwestern and Katherine L. Bouman of the California Institute of Technology as joint senior authors.

Imaging a planet means modeling the glare of its star and subtracting it. Standard methods build that model out of reference pictures of other stars, or of the same star at other times; pointing errors and thermal drift leave residual starlight that buries faint planets. The new method, which the authors call FLIP, instead fits a physical model of the instrument, solving for the telescope's optical distortions and the detector's response until the simulated glare matches what Webb recorded.

The paper puts the fit at about 35 minutes on a single graphics processor, and reports tests on published observations of three stars behind Webb's coronagraph, the mask that blocks a star's light. On one, close in to the star, FLIP reached roughly 15 times the sensitivity of the standard method: the top of its range.

The 4.3-micron band holds a carbon dioxide absorption feature and cannot be seen from the ground, because Earth's atmosphere blocks it. The authors say future work will measure the planet's brightness there. They describe the method as the first, to their knowledge, to improve sensitivity substantially on real coronagraph data, and say it could be extended to the Nancy Grace Roman Space Telescope and the planned Habitable Worlds Observatory. The FLIP code is posted publicly.