One million data centers in orbit would erase up to 80% of Rubin Observatory images some nights

A simulation of a constellation of about one million computing satellites shows that it would contaminate on average 20% of LSST poses over a year, and would multiply by more than ten the brightness of the sky at dusk.

As the Vera C. Rubin Observatory begins its Great Sky Survey (LSST), several operators are preparing constellations of orbital data centers to run AI in low Earth orbit. Researchers simulated a fleet of about one million of these satellites, which were recently filed in two low orbits and two sun-synchronous orbits, and then compared it with a year of Rubin observations.

Streaks in the majority of images at dusk

Their six-step calculation chain — orbit propagation, cross-checking with the field of view, observation geometry, brightness model, surface luminance and estimation of lost pixels — leads to a stark conclusion: depending on the season, up to 80% of a night's pointing contains at least one illuminated satellite trail, and about 20% on average annually. Sun-synchronous orbits weigh more than their share, because their geometry keeps them in the sun more often. The specular reflections of solar panels could reach a magnitude of 2 to 4.

Even with only 10% deployment, the constellation would leave trails visible year-round. The background sky would become more than ten times brighter near dusk without the Moon, and nighttime pixel losses would remain 5 to 10% on average at these times, even with multiple mitigation strategies. The authors conclude that such projects would seriously harm the science of the LSST.