Charles Constant, University College London; Shaylah Mutschler, Space Environment Technologies; Santosh Bhattarai, University College London; Marcin Pilinski, University of Colorado at Boulder / Laboratory for Atmospheric and Space Physics
Keywords: Atmospheric Density Inversion, POD, Orbit Propagation, Data Assimilation
Abstract:
Accurate, timely knowledge of thermospheric density in Low Earth Orbit (LEO) is increasingly critical for Space Situational Awareness (SSA) and satellite operations as large constellations proliferate. Existing approaches fall short of the community’s needs on several counts (e.g. accuracy, latency, licensing restrictions). The growing number of LEO spacecraft carries with it the potential to enable a step change in the quality and quantity of data available to drive the next generation of assimilative thermospheric density models.
We quantify how four key factors influence Precise Orbit Determination (POD) density retrievals using the Energy Dissipation Rate (EDR) method: POD noise level, drag acceleration magnitude, averaging interval (“fit-span”), and the number of concurrent spacecraft used to generate observations. We propagate Starlink-like trajectories over the month of April 2023, using the High Accuracy Satellite Drag Model (HASDM) as our density model. Onto these trajectories we overlay a realistic LEO POD noise model, synthesizing three representative noise scenarios informed by published characterizations. We then modify the semi-major axis of these trajectories to sample ten drag levels (altitudes ~300-700 km), vary fit-span from one to five orbits, and aggregate observations from up to 18 equally spaced co-orbital satellites.
Three rules of thumb emerge:
Fit-span rule: Doubling the observation window roughly halves the RMS percentage error (scales ~ T-1/2).
Satellite-count rule: For satellites distributed along a single orbital track, error decreases with count following a power law: N-0.6 for averages of direct density retrievals and N-0.25 for averages of individually assimilated estimates.
Drag rule: Relative RMS percentage errors show a logistic-like response to drag acceleration. For weak signals (< 1e-7 m/s-2) improvements are negligible; in a transition band (1e-7-1e-6 m/s-2) each ~3-fold increase in drag yields ~50% error reduction; beyond ~1e-6 m/s-2 returns diminish. This pattern is consistent across noise levels. When observations are sufficiently numerous or low-noise, a simple mean of direct density retrievals can match or even exceed the average of individually assimilated estimates, consistent with background-model bias leakage. Assimilation remains essential for provision of thermospheric density values across the full operational domain; however, local estimates may benefit more from simpler, computationally cheaper data fusion methods under these conditions. These results aim to enable design-to-requirement mappings for mission designers, model developers, and SSA practitioners (e.g., required POD accuracy, optimal fit-span, satellite count, and processing strategy) to support development of the next generation of thermospheric density models. Date of Conference: September 16-19, 2025
Track: Atmospherics/Space Weather