Enhancing Space Surveillance and Orbit Propagation of Space Objects in Very Low Earth Orbit through Wide-Angle Passive Optical Observations

Kurzfassung

The German government’s recently published strategy paper dated November 4, 2025 emphasizes the paramount importance of a national space strategy in light of the current geopolitical situation. As a result, space has become the focus of society’s attention not only as a physical space in terms of its potential for research and technical applications, but also in terms of its significance for security policy. For the vast majority of orbital objects, publicly available two-Line element (TLE) data concerning position predictions are available. However, this data is subjected to individual update cycles, so it cannot be ruled out that objects may deviate significantly from their predictions. First, the absolute positional difference between the individual data points of the respective TLE predictions and thus the effect of the influence of temporal epoch differences was examined, yielding the resulting angular deviation. Subsequently, a split into radial-. tangential, normal-, i. e., RTN-coordinates was conducted. When considering Falcon 9 R/B with an average orbital altitude of 180–230 km, the absolute position differences yield the following result: After 1.5 days of epoch difference, ≈ 60% of all data points are captured within a field of view (FOV) of 10°, while ≈ 80% of the data points are included within a FOV of 20°, meaning that the satellite position prediction could be successfully detected. This result was significantly improved by the decomposition, as only the cross-track differences were included in the analysis: After 2.5 days of epoch difference, ≈ 85% of all data points are below 10° and ≈ 90% are below 20°. This underscores the comprehensive possibility of detecting objects in a VLEO using the wide-angle feature, even if the position forecast has an epoch difference of up to 2.5 days. Correlation serves to link RTN components (linear & non-linear) with the physical orbit parameters. The resultant drivers are intended to act as indicators, attributing patterns and structures within the position differences of the prediction data to the respective physical influences of orbital mechanics. One of the most important findings is that the position errors in the radial and tangential directions are primarily influenced by three Keplerian elements (∆Orbital period, ∆Mean anomaly, ∆Mean motion), i.e., they are dominated by orbital dynamics. The position error of the normal component, in contrast, is dominated by a single orbital element (∆RAAN) and can therefore be interpreted as a different orbital orientation. As a further application, targeted training of ML algorithms (GBDT) based on tracking data generated by APPARILLiO was utilized to improve trajectory propagation. The mean angular deviation, the deviation between the TLE prediction + correction term from the ML model and the TLE prediction, was improved by ≈ 50% for the vector multi-output model and by ≈ 80% for the scalar output model. In addition, the position of the laser ranging satellite Terra SAR-X was taken as a reference position, which improved the accuracy of the corrected propagation by 235 m in total magnitude compared to pure TLE propagation.