Tapley (1975): Unmodeled Drag Accelerations

Source

Tapley, B.D., “Orbit Determination in the Presence of Unmodeled Accelerations,” AFOSR-TR-75-1073, Department of Aerospace Engineering and Engineering Mechanics, The University of Texas at Austin, January 1975. DTIC Accession Number: ADA013842. Grant: AFOSR 72-2233.

Download PDF · DTIC Citation

Why This Paper Matters for SGP4

SGP4 encodes atmospheric drag as a single number: the $B^*$ drag term in the TLE. This is a time-averaged ballistic coefficient that assumes constant atmospheric density, constant drag coefficient, and no solar activity variations between TLE updates.

Tapley’s paper, from the same AFOSR/14th Aerospace Force ecosystem that produced the Spacetrack Reports, quantifies why this breaks down:

• Atmospheric density at satellite altitudes varies by factors of 2—10 over a single solar rotation (~27 days)
• Density model error is the dominant source of orbit prediction error for LEO satellites
• A single ballistic coefficient cannot capture time-varying density fluctuations

This is the fundamental reason that Kelso (2007) measured SGP4/TLE accuracy degrading at ~1—3 km/day.

The Lane & Cranford Citation

Lineage Paper Now Acquired

Reference 2 in this paper is Lane, M.H. and Cranford, K.H., “An Improved Analytical Drag Theory for the Artificial Satellite Problem,” AIAA Paper 69-925, August 1969 — the key link in the SGP4 lineage between Brouwer/Lyddane and the Spacetrack Reports. Now part of this archive: see Lane & Cranford (1969). Tapley’s citation confirms the paper number and date, and his use of it establishes that Lane’s drag theory was the standard atmospheric model in the AFOSR/NORAD orbit prediction community by the mid-1970s.

The B* Parameter in Context

SGP4’s drag term collapses the entire atmospheric drag problem into a single scalar:

$B^* = \frac{\rho_0}{2} \cdot \frac{C_D A}{m}$

where $\rho_0$ is a reference atmospheric density, $C_D$ is the drag coefficient, $A$ is the cross-sectional area, and $m$ is the satellite mass. This value is fitted at the TLE epoch from radar fence observations.

What Tapley’s work reveals is why this is both necessary (limited tracking data quality) and fundamentally limiting (density varies faster than TLEs are updated):

Factor	SGP4 Assumption	Reality
Atmospheric density	Lane power-law, constant $\rho_0$	Varies by factors of 2—10 with solar activity
Drag coefficient	Constant $C_D$	Changes with attitude and flow regime
Cross-section	Constant $A$	Changes with satellite orientation
Solar flux (F10.7)	Not modeled	Drives density variations at all altitudes
Geomagnetic storms	Not modeled	Can increase density by 10x in hours

The Dynamic Model Compensation Method

Tapley’s alternative treats unmodeled accelerations as stochastic processes rather than constants. The satellite state vector is augmented with acceleration components modeled as first-order Gauss-Markov processes:

$\dot{a}_j + \beta_j a_j = u_j(t)$

where $a_j$ are acceleration components, $\beta_j$ are inverse correlation times, and $u_j(t)$ is white noise. An extended Kalman filter simultaneously estimates the orbit state and the unmodeled accelerations.

From the estimated drag acceleration, local atmospheric density can be recovered:

$\rho(r) = \frac{-2m \cdot a_{\text{drag}}}{C_D A \cdot v^2}$

Key Results

1. Density recovery to ~$10^{-15}$ g/cm$^3$ at 200—400 km altitude from ground-based tracking alone
2. Tracking precision required: ~1 m in range and ~0.1 arcsec in angle — far beyond the radar fence observations that generate TLEs
3. Active satellite participation (on-board accelerometers) dramatically improves estimation, but the typical SGP4/TLE scenario uses only passive radar tracking

Atmospheric Models Compared

Model	Type	Inputs	SGP4 Equivalent
Lane power-law	Analytical	Perigee height only	Direct ($B^*$ parameter)
Modified Harris-Priester	Semi-empirical	Solar position, F10.7	None
Analytic Jacchia-Roberts	Semi-empirical	F10.7, Ap, diurnal bulge	None

SGP4 uses only the Lane power-law model. The more sophisticated models incorporate solar UV flux (F10.7), geomagnetic activity indices (Ap/Kp), diurnal density bulge, and seasonal-latitudinal variations — none of which SGP4 accounts for.

The Accuracy Gap: 1975 to Today

The fundamental problem Tapley identified remains the limiting factor for TLE-based orbit prediction:

Era	Approach	Drag Data	Useful Prediction
1975	DMC + Kalman filter	High-quality tracking	Hours to days
1980	SGP4 + $B^*$	Radar fence	~1—3 days
2007	SGP4 + $B^*$	Radar fence + optical	~1 km at epoch, 1—3 km/day
2020s	Precision OD	GRACE-FO accelerometers	Days to weeks
2020s	Operational TLE	Still $B^*$	Still ~1—3 days

For Craft/Astrolock’s purposes, the operational limit remains the TLE/$B^*$ approach. The ~1 km/day accuracy degradation Kelso measured in 2007 is a direct consequence of the limitation Tapley analyzed 32 years earlier.

Orbit Decay Prediction Limits

If you’re using SGP4/TLEs to predict when a satellite will reenter, understand that the accuracy degrades rapidly with prediction horizon. The $B^*$ value becomes stale as solar conditions change. During solar maximum (high F10.7), atmospheric density at LEO altitudes can be 10x higher than during solar minimum, making reentry predictions especially unreliable. Fresh TLEs help, but cannot overcome the single-parameter limitation of $B^*$.

References of Note

The paper’s bibliography connects several threads in the orbit prediction community:

Ref	Author(s)	Year	Relevance
1	King-Hele	1964	Foundational book: satellite orbit decay in atmosphere
2	Lane & Cranford	1969	The SGP4 drag theory foundation (AIAA 69-925)
3	Jacchia	1971	Revised static atmospheric models (basis for J-R model)
5	Rauch	1965	Optimum estimation with random drag fluctuations
6	Ingram	1971	Sequential estimation of atmospheric parameters

---

Kelso (2007): SGP4 Validation Empirical measurement of the accuracy limits Tapley predicted

WGS-72: SGP4 Constants The geodetic constants that SGP4 uses alongside B*

STR#2: Near-Earth Theory Lane & Hoots' formalization of the drag theory Tapley references

SGP4 Theory Overview Full document lineage and cross-references