Saikat Majumder, Digantara; Archana Trivedi, Digantara; Bhanu Chandana, Digantara; Subramanian Arumugam, Digantara; Thamim Ansari, Digantara; Tanveer Ahmed, Digantara
Keywords: Space Weather, Uncertainty Quantification, Machine Learning, Satellite Drag, Orbit Prediction, Conjunction Assessment
Abstract:
Accurate orbit prediction and collision risk assessment for low-Earth orbit (LEO) satellites are highly dependent on precise thermospheric density modeling, which is strongly influenced by space weather drivers such as solar radio flux (F10.7) and the planetary geomagnetic index (Kp). In this work, we develop and evaluate machine learning (ML) forecasts of these drivers using CelesTrak and OMNI datasets, employing a multivariate Variational Mode Decomposition (VMD) and Long Short-Term Memory (LSTM) architecture, referred to as the VL model, to effectively capture nonlinear and dynamic space weather variations. These forecasts are ingested into the NRLMSISE-00 density model within Digantara’s proprietary high fidelity orbit propagator (EPIC) to drive orbit prediction for synthetic LEO satellites spanning low to high altitudes and having configurations from 1U to 16U. In the simulation framework, only the atmospheric density input is varied, while all other force and drag parameters are held constant. Forecast-driven orbits are compared against truth-driven references under space weather conditions ranging from quiet to extreme, quantifying prediction sensitivities with altitude and object size. By propagating forecasted states and uncertainties, we further map model-driven density errors into collision probability (Pc) estimation during representative conjunction scenarios.
Our results show that objects propagated using VL model outputs demonstrate improved predictive accuracy as compared to current operational space weather driver forecasts. For a historical conjunction event, VL models closely reproduce Pc values computed using observed space weather parameters at shorter lead times, thereby improving the timeliness and reliability of conjunction assessments. These results highlight the operational value of integrating ML-driven space weather forecasting with operational neutral density modeling to enhance orbit determination and improve conjunction analysis capabilities.
Date of Conference: September 16-19, 2025
Track: Atmospherics/Space Weather