Abstract Fluctuations in thermospheric neutral density affect the operational stability and lifetime of low Earth orbit (LEO) satellites. Solar activity and geomagnetic disturbances induce substantial density variations in the thermosphere, thereby impacting critical satellite operations such as orbit control, attitude maneuvers, and collision avoidance. However, existing empirical models fail to accurately capture these localized thermospheric density oscillations. To date, effective methods for the high‐precision prediction of LEO satellite orbital decay under varying geomagnetic conditions remain underdeveloped. This study proposes a machine learning‐enhanced method for predicting orbital decay at specified altitudes within the LEO region by making use of Gravity Recovery and Climate Experiment Level‐1B observations and integrating along‐track high‐precision thermospheric density, aerodynamic coefficients, and satellite mass parameters. During the 9–11 May 2024 storm event, along‐track thermospheric density surged, resulting in a 48‐hr semi‐major‐axis decay of approximately 168 m before stabilizing at around 83 m thereafter. For the 24 August 2005 interplanetary coronal mass ejection (ICME) case, the cumulative decay (−45.4 m) showed close alignment with the observed orbital data (−40.4 m). When independently tested across 113 ICME events, the random forest model accounted for 85% of the variance in orbital decay, achieving a test R 2 of 0.749 during all geomagnetically periods in 2005. The results demonstrate that our proposed approach delivers significantly improved prediction accuracy of satellite orbital decay across varying geomagnetic conditions compared with empirical models. This work provides novel insights into thermospheric disturbance impacts on satellite orbits and offers essential theoretical support for LEO mission planning and orbital management.
Zhou et al. (Sun,) studied this question.