New AI Predicts Satellite Crashes Sooner to Keep GPS Safe

Earth's orbit is getting crowded. Thousands of working satellites now share space with dead rockets, old parts and broken spacecraft, and they regularly drift uncomfortably close to one another. When that happens, operators get standard alerts called Conjunction Data Messages — think of them as a weather warning for space. Each alert lists how close two objects will pass and the odds they will hit. The problem is those odds swing wildly from one alert to the next, so it's hard to know whether to act.

A team of researchers in Turkey built an AI system that reads those alerts and predicts what the next warning will say before it arrives. Their model combines two AI approaches: one that spots patterns over time, and another that weighs which details matter most. It studies the history of a single close call — distance, direction, timing, uncertainty — and estimates where the risk is heading next.

Why earlier warnings matter: many modern satellites use electric propulsion, an efficient but gentle engine that takes hours or even days to shift orbit. More lead time means a small, cheap nudge instead of a panicked, fuel-burning maneuver. It also cuts wasted maneuvers — false alarms that drain fuel and briefly interrupt phone, TV, internet and weather services.

The catch: this is early research, tested on alert data from one space agency, TÜBİTAK UZAY. The model predicts the next report, not the actual outcome — so if the early data is mistaken, the prediction inherits that error. It is a helpful head start for human operators, not an autopilot.