Accurate orbit prediction (OP) of space debris is vital in Space Situation Awareness (SSA) related tasks, like space collision warnings. However, due to the sparse and low precision observations, unknown geometrical and physical features of debris, and effects of incomplete force models, OP based on the orbital mechanics theory or physics-based OP of space debris suffers from rapid error growth over a long duration, limiting the period of validity of debris OP for precise space applications. Considering that the tracking arcs of a debris object over a single station often share a similar temporal and spatial distribution in the inertial space, the resultant OP errors possibly have a coherent relationship with the temporal and spatial distribution of tracking arcs. This paper proposes a machine learning (ML)-based approach to model the underlying pattern of debris OP errors from historical observations and apply it to modify the future physics-based OP results. The approach includes three steps: constructing a historical OP error set, training a ML model to fit the historical OP error set, and correcting the future physics-based OP with ML-predicted orbital errors. The ensemble learning algorithm of boosting tree (BT) is studied as the primary ML method for the error modeling and predicting process. Experiments with three low-Earth-orbit (LEO) objects, tracked by a single radar station, demonstrate that the trained ML models can capture more than 80% of the underlying pattern of the historical OP errors. More importantly, the errors of physics-based OP over the future 7 days reduce from thousands of meters to hundreds or even tens of meters through the error correction with the learned error pattern, achieving at least 50% accuracy improvement. Such dramatic OP improvements show the promising potential of ML for enhanced SSA capability.

With the undergoing and planned implementations of mega constellations of thousands of Low Earth Orbiting (LEO) satellites, space will become even more congested for satellite operations. The enduring effects on the long-term space environment have been investigated by various researchers using debris environment models. This paper is focused on the imminent short-term effects of LEO mega constellations on the space operation environment concerned by satellite owners and operators. The effects are measured in terms of the Close Approaches (CAs) and overall collision probability. Instead of using debris environment models, the CAs are determined from integrated orbit positions, and the collision probability is computed for each CA considering the sizes and position covariance of the involving objects. The obtained results thus present a clearer picture of the space operation safety environment when LEO mega constellations are deployed. Many mega constellations are simulated, including a Starlink-like constellation of 1584 satellites, four possible generic constellations at altitudes between 1110 km and 1325 km, and three constellations of 1584 satellites each at the altitudes of 650 km, 800 km, and 950 km, respectively, where the Resident Space Object (RSO) spatial density is the highest. The increases in the number of CAs and overall collision probability caused by them are really alarming. The results suggest that highly frequent orbital maneuvers are required to avoid collisions between existing RSOs and constellation satellites, and between satellites from two constellations at a close altitude, as such the constellation operation burden would be very heavy. The study is not only useful for satellite operators but a powerful signal for various stakeholders to pay serious attention to the development of LEO mega constellations.

Concerns for the collision risk involving Starlink satellites have motivated the interest in obtaining their accurate orbit knowledge. However, accurate orbit determination (OD) and prediction (OP) of Starlink satellites confront two main challenges: mismatching or missed matching of sparse tracklets to maneuvering satellites, and unknown or unmodeled orbit maneuvers. How to exactly associate a tracklet to the right satellite is the primary issue, since a maneuvering satellite does not follow the naturally evolving orbit during the maneuvering, while more tracklets are needed for developing an accurate orbit maneuver model. If these two challenges are not well addressed, it may lead to catalog maintenance failure or even loss of objects. This paper proposes a method to correctly match tracklets to the climbing Starlink satellites. It is based on the recursive OD and OP, in which the orbit maneuver is modeled and the thrust is estimated, such that the subsequent OP accuracy guarantees the correct match of tracklets shortly after the OD time. Experiments with climbing Starlink satellites demonstrate that the tracklets within three days of the last TLE (two-line element) are all correctly matched to the right satellites. With the matched tracklets, the thrust accelerations of climbing Starlink satellites can be precisely estimated through an orbit control approach, and the position prediction accuracy over 48 hours is at the level of a few kilometers, providing accurate orbit knowledge for reliable collision warning involving Starlink satellites.