Deep Learning Model Sharpens Forecasts for When Dead Satellites Fall to Earth

Researchers have combined a neural network with the physics of orbital motion to predict more precisely when uncontrolled satellites and spent rocket stages reenter the atmosphere. Tested against three real reentries, the method produced fall-time predictions within minutes of the observed events, a margin that matters because a one-minute error shifts the predicted impact zone by hundreds of kilometers.

The work, published in Acta Astronautica, responds to a growing population of objects in low Earth orbit. By the end of 2025, low Earth orbit held 25,060 cataloged objects, nearly twice as many as a decade earlier, according to the European Space Agency. Many old satellites and spent rocket stages will eventually fall back without control and may come down anywhere on Earth.

The study was led by M. Khawariz Andaristiyan of the Research Center for Satellite Technology at Indonesia's National Research and Innovation Agency, known as BRIN, with Affiani Machmudah and Wahyudi Hasbi of the same center and Ridanto Eko Poetro of the Institut Teknologi Bandung.

The team tested its approach against three uncontrolled falls, making predictions after the fact from tracking data gathered up to shortly before each reentry. For Cosmos-482, a Soviet lander built for Venus and stranded in orbit since 1972, the model predicted 06:22 UTC, give or take 11.4 minutes, against a reported 06:24. For NASA's NEOWISE telescope, it predicted 00:54 UTC, give or take 7.8 minutes, against 00:49. For a Long March 4B rocket stage, it predicted 22:45 UTC, give or take 13.2 minutes, against debris seen over Spain from 22:33 to 22:36.

Cosmos-482 was launched in 1972 and stranded by a failed upper stage of its rocket, circling Earth for 53 years before coming home on May 10, 2025. Roscosmos later reported that it came down in the Indian Ocean, west of Jakarta. In its final days, several space agencies and research groups forecasted when it would fall, and therefore where.

The researchers explain why timing is the core of the problem. An object near the end of its life in orbit still flies at about 7.8 kilometers per second. An error of one minute in the predicted time of a fall shifts the place where it comes down by about 460 kilometers along its path. An error of one hour shifts it by about 28,000 kilometers, two-thirds of the way around Earth. To say where an object will fall, the team notes, one must first say when.

What drives these objects down is drag, the braking caused by the thin air of the upper atmosphere. When the sun is active, it heats the upper atmosphere, which swells, so the air at a given height grows denser and low orbits decay faster. The sun's activity rises and falls in a cycle of about 11 years, and the three falls studied all came in its recent active years, from 2023 to 2025.

The authors present the work as a way to improve both when and, by extension, where uncontrolled objects reenter, a question that grows more pressing as the number of cataloged objects in low Earth orbit continues to climb.

The method has been demonstrated on three reentries drawn from the sun's recent active years. The source does not name further planned tests or operational deployments.