THE CRUNCH
Scientists led by Northwestern's Rodrigo Ferrer-Chavez have devised a method that strips starlight out of telescope images, and in trials on data from the James Webb Space Telescope it made faint exoplanets two to 15 times easier to detect than standard processing.
Their framework, Physics-based Learning for Imaging Planets (FLIP), pairs a physics model of how light travels through the telescope with optimisation methods drawn from machine learning. It accounts for flaws in the optics and works out the glare that needs removing. The problem it tackles is that photographing a planet directly is hard: the host star outshines it, and small shifts in telescope optics leave leftover glare that current subtraction methods fail to catch.
When run on data from Webb's NIRCam instrument, FLIP managed what standard processing could not: it captured the first image of the already-known giant planet 51 Eridani b at the 4.3 micrometer infrared wavelength. The team's Nature paper states that, as far as they know, no comparable method had previously shown such a large sensitivity gain on genuine on-sky coronagraphic imaging.
According to the researchers, the method should also suit other observatories now operating or yet to launch, in space or on the ground, and could ease the extreme stability demands those instruments face. Work still to do includes trimming the computing cost and coping with the far bigger datasets ground-based telescopes generate, in the hope of one day imaging Earth-like worlds directly without tiny instabilities getting in the way.