Julia Briden, Massachusetts Institute of Technology; Nicolette Clark, Massachusetts Institute of Technology; Peng Mun Siew, Massachusetts Institute of Technology; Richard Linares, Massachusetts Institute of Technology; Tzu-Wei Fang, National Oceanic and Atmospheric Administration
Keywords: Atmospheric Modeling, Orbit Prediction, Ballistic Coefficient, Machine Learning, Reduced-Order Model, Space Weather
Abstract:
Recently, fluctuations in solar and geomagnetic activity have resulted in increased Low Earth Orbit (LEO) satellite failure and Resident Space Object (RSO) loss of custody. With a single geomagnetic storms ability to inhibit the majority of LEO satellite tracking activity for several days, space weather activity poses a high conjunction risk for Earth-orbiting satellites. These failures occur because these fluctuations were not modeled accurately, resulting in inaccurate atmospheric density estimates used for computing the atmospheric drag. With the number of satellites in LEO growing by over 50% in the past year and the predicted solar cycle reaching maximum activity by 2025, accurate atmospheric density estimation for a range of solar and geomagnetic conditions is essential in deorbit event prediction, satellite orbit prediction, and collision avoidance.
Space weather activity in the Earths atmosphere occurs in two forms: solar activity and geomagnetic activity. Eruptions of plasma and magnetic field structures, known as coronal mass ejections (CMEs), occur in the suns atmosphere. When charged particles from these eruptions enter the Earths atmosphere, the atmospheric temperature rises due to Joule heating from electric fields in the ionosphere. The atmospheric heating from solar activity and the transient solar wind from Earth-directed CMEs determine the level of geomagnetic activity, sometimes resulting in geomagnetic storms. One of the largest solar storms of the Space Age occurred from mid-October to early November in 2003. The storm caused temporary to permanent communications losses for the ADEOS-2, CHIPS, and Department of Defense satellites and resulted in additional complications for Global Positioning System (GPS) applications and electrical power systems. Atmospheric density is influenced by time-dependent changes in solar and geomagnetic activity, as well as altitude-dependent changes in thermal activity. To capture and predict these fluctuations for LEO satellite orbit tracking, highly-accurate and efficient atmospheric density models must be employed.
In practice, atmospheric density models include empirical models, such as NRLMSISE-00, JB2008, and DTM-2013, and physics-based models, including the Global Ionosphere-Thermosphere Model (GITM), the coupled Whole Atmosphere Model-Ionosphere Plasmasphere Electrodynamics (WAM-IPE) model, and the Thermosphere-Ionosphere-Electrodynamics General Circulation Model (TIE-GCM). Empirical models are fast to evaluate, at the expense of limited accuracy, and exclude the dynamics of the atmosphere required for forecasting. Conversely, physics-based models solve the full Navier-Stokes equations for density, velocity, and temperature on a discretized grid, allowing for accurate forecasts at a high computational cost. Recent work in reduced-order modeling (ROM) has represented physics-based atmospheric density models with a small set of spatial modes to improve the computational efficiency of forecasting models. Additionally, data assimilation of two-line element (TLE) data and combined satellite radar and GPS tracking data into ROM has significantly increased the accuracy of these models. While performance comparisons of these atmospheric density models have previously been conducted, a full study of the effectiveness of these models in RSO orbit prediction for significant solar and geomagnetic space weather events on satellite reentry prediction has not been explored.
This work analyzes the accuracy of the NRLMSISE-00, JB2008, WAM-IPE, Proper Orthogonal Decomposition (POD) ROMs based on JB2008 and TIEGCM, and JB2008-based Machine Learning (ML) ROM atmospheric density models for LEO orbit propagation during deorbit events for Starlink, Rocket Labs Humanity Star, and the Freedom drag sail CubeSat. Drag is the largest perturbation for orbit prediction in LEO, requiring an accurate ballistic coefficient (BC) to be estimated. Often, the true values of the BCs for LEO RSOs are not readily available, requiring BCs to be estimated from TLE data; the BSTAR value in TLE files is only compatible with the SGP4 orbit propagator. The BC is estimated in this work by iteratively optimizing our estimate by comparing the change in semimajor axis based on TLE data to the semimajor axis change due to drag, as computed by orbit propagation. For the POD and ML ROMs, the reduced-order thermospheric density states are predicted using the Density Estimation Toolbox (DESTO). A sensitivity analysis is performed for each event in two parts: true solar index measurements are used to isolate the atmospheric density uncertainty from the space weather data and noise models are used to determine the sensitivity of atmospheric models to noise in the forecasted space weather indices.
By developing a method for evaluating the prediction ability and performance of atmospheric density models for deorbiting events, important preventative measures can be identified. With some of the most important space assets, along with the highest concentration of space debris residing in LEO, space weather modeling has the ability to predict and respond to future orbit determination hazards.
Date of Conference: September 27-20, 2022
Best Paper Award Winner 2022
Track: Atmospherics/Space Weather