Computer model uncovers exoplanet concealed by starlight in Webb data
The researchers reported two to 15 times greater sensitivity in tests on known planets, using software to account for small optical imperfections.
A team of researchers has detected a previously obscured exoplanet in observations from the James Webb Space Telescope using a computer simulation to remove stellar glare. The work appeared online in Nature on November 4.
The technique, Physics-based Learning for Imaging Planets (FLIP), simulates Webb’s Near Infrared Camera to work out how optical imperfections affect the recorded light. The team reported two to 15 times greater sensitivity than conventional approaches when testing the method on known planets.
FLIP uses optimization methods developed for machine learning to adjust millions of variables until the simulation closely matches telescope observations. The resulting calculation of the star’s appearance can then be subtracted from an image, exposing much fainter objects nearby.
An earlier attempt to train machine learning systems to predict stellar glare did not succeed. Lead author Rodrigo Ferrer-Chavez, now a Northwestern University graduate student, and his collaborators instead turned those computational tools toward refining a physical model of the instrument.
Ferrer-Chavez said archived Webb observations could contain other planets that earlier searches failed to detect. The team intends to refine FLIP and investigate applications to additional observatories, including the Roman Space Telescope and instruments on Earth.
