Improving Ballistic Prediction Accuracy for Upper Stages Using Adaptive Kalman Filtering

Accurate ballistic trajectory prediction for upper stages on low Earth orbit is challenged by non-stationary atmospheric density driven by solar activity: as density varies, fixed noise covariance matrices in standard Kalman filters become mismatched with actual process statistics, causing degraded estimation accuracy. Non-stationary atmospheric density disrupts fixed-covariance Kalman filters in low Earth orbit, degrading trajectory prediction for upper stages. This paper characterises the quantitative position/velocity accuracy trade-off and performance boundaries of the Sage–Husa Adaptive EKF (AEKF), identifying an empirical sensor-noise approximate crossover region near σ* ≈ 100 m where the AEKF improves velocity accuracy by 38.9% at the cost of a 59% increase in position RMSE.