🔭 astrophysics

Predicting Atmospheric Re-entries Using Segmented M/A Calibration of Public TLE Data

This paper presents a reproducible method for predicting atmospheric re-entries with approximately one-hour accuracy using only public Two-Line Elements (TLEs) by applying a segmented calibration of the effective mass-to-area ratio to filter out catalogue noise and isolate stable drag dynamics.

Original authors: Joseph Remis

Published 2026-07-23

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Joseph Remis

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Sky's Trash Can: Why We Need to Know When Space Junk Falls

Imagine the sky above us is a busy highway, but instead of cars, it's filled with thousands of satellites and chunks of metal from old rockets. Most of these are working hard, but some are broken, dead, or just running out of fuel. When they run out of energy, Earth's gravity gently pulls them down. As they fall, they hit thicker air, heat up, and usually burn up like a shooting star. This is called "atmospheric re-entry."

The tricky part is knowing exactly when and where a piece of space junk will fall. Scientists use a special list called "Two-Line Elements" (or TLEs) to track where these objects are. Think of TLEs like a series of snapshots taken by a camera that isn't perfectly steady; sometimes the picture is a little blurry or jumps around a bit. Because the air gets thicker and the objects tumble and spin, predicting the final fall is like trying to guess when a leaf will hit the ground in a storm. It's important to get this right because, while most junk burns up, some pieces can survive to hit the ground, and we want to know if they might land on a house or a city.

The Paper's Big Idea: A New Way to Guess the Fall

In this paper, Joseph Remis from France proposes a clever, low-tech way to predict these falls using only the public snapshots (TLEs) that anyone can see. Instead of trying to build a super-complex computer model that guesses how the wind blows or how the metal heats up, Remis uses a method he calls "segmented M/A calibration."

Here is how it works, using a simple analogy: Imagine you are trying to guess how fast a runner is slowing down as they get tired. If you look at their position every second, your eyes might be tricked by them wobbling or stumbling (the "noise"). But if you look at their position every 12 hours, you can clearly see the steady trend of them slowing down.

Remis does exactly this with space junk. He breaks the timeline into chunks of at least 12 hours. For each chunk, he calculates a "magic number" called the Mass-to-Area ratio (M/A). You can think of this as the object's "drag personality." A heavy, compact object (like a solid brick) has a high M/A and cuts through the air easily. A light, flat object (like a sheet of paper) has a low M/A and gets slowed down quickly. By adjusting this number for every 12-hour chunk, he can match the computer's prediction to the actual "blurry" snapshots from the TLE list.

What They Found: It Works, But with Limits

The paper tested this method on many real-life examples, including old rocket parts and broken satellites. Here is what they discovered:

It gets better at the end: The method isn't perfect days in advance, but in the final hours before an object falls, it becomes very accurate. When the object gets low enough, the drop in its orbit becomes so huge that it swamps the "blurry" noise in the data.
The Accuracy: In the final stretch, the method can predict the fall time within about ±1 hour. For example, if a rocket body was predicted to fall at 10:00 AM, the method would likely say it falls between 9:00 AM and 11:00 AM.
The "Magic" Number: By looking at how the "drag personality" (M/A) changes, the method can tell if an object is just falling naturally or if it is being pushed by a rocket engine.
- Natural Fall: The number stays steady.
- Rocket Push: The number jumps wildly. The paper notes that if a satellite suddenly fires its engines to fall faster (like the Starlink-34110 case), the method can't predict the exact moment while it's happening, but it can figure out what happened afterwards.

What This Method is NOT

It is important to know what this paper says it doesn't do. The author explicitly states that this method does not rely on complex guesses about solar flares or the exact shape of the tumbling metal. Those details are too chaotic to model perfectly. Instead, the paper argues that the main problem isn't the atmosphere; the problem is the "noise" in the public data list itself.

The paper also rules out the idea that you can get perfect, minute-by-minute predictions days in advance. The limit is the quality of the public snapshots, not the computer's brain. If the snapshots are fuzzy, the prediction will be fuzzy.

Real-World Proof

The author didn't just run simulations; they checked real events.

The ISS View: A rocket part fell in April 2026, and astronauts on the International Space Station saw it. The time and location matched the prediction.
The Sri Lanka Sightings: In May 2026, people in Sri Lanka saw a bright fireball. The prediction matched the time and place.
The Meteor Society: In June 2026, 91 people reported seeing a falling object. The prediction lined up with their photos and videos.

The Bottom Line

Joseph Remis's paper suggests that you don't need a supercomputer or secret government data to predict when space junk will fall. By taking the public data, chopping it into 12-hour blocks, and adjusting a single "drag number" for each block, you can get a surprisingly good guess—usually within an hour—of when the final fall will happen. It's a simple, transparent tool that turns a noisy, confusing list of numbers into a clear picture of the sky's final moments.

Technical Summary: Predicting Atmospheric Re-entries Using Segmented M/A Calibration of Public TLE Data

Problem Statement
Predicting the precise timing of atmospheric re-entries for low-altitude objects is hindered by the intrinsic noise and cadence of publicly available Two-Line Element (TLE) data. Traditional long-arc drag fitting often diverges because it accumulates short-period catalogue noise, obscuring the true dynamical evolution of the object. Furthermore, the physical state of a re-entering object (attitude, tumbling, thermal conditions, and ballistic coefficient) can vary on minute timescales, rendering detailed ab-initio physical modeling obsolete before it can be completed. The challenge lies in extracting a stable dynamical signal from noisy, discrete TLE observations without requiring detailed knowledge of the object's geometry or aerodynamic state.

Methodology: Segmented M/A Calibration
The paper proposes an empirical, reproducible method that reconstructs the terminal decay path by calibrating the effective mass-to-area ratio ( $M/A$ ) over successive short propagation arcs. The core principles and steps are:

Segmented Approach: Instead of a single long-arc fit, the method uses pairs of successive TLEs to define independent segments. The earlier TLE is propagated to the epoch of the later TLE.
Parameter Calibration: The effective drag area ( $A$ ) is adjusted so that the simulated semi-major-axis decay ( $\Delta a_{sim}$ ) matches the observed decay ( $\Delta a_{obs}$ ) between the two TLEs. The optimal area is determined via linear interpolation, yielding an effective $M/A$ ratio for that specific interval.
Noise Suppression: To isolate the secular decay from catalogue noise, a minimum separation of 12 hours between TLEs is enforced for standard calibration. In the terminal phase, where the semi-major axis drop becomes large enough to dominate fluctuations, segments are accepted if the semi-major axis decreases by at least 10 km, even with closer spacing.
Force Model: Propagations are performed using NASA's GMAT with a fixed force model (EGM96 gravity, NRLMSISE-00 density, solar radiation pressure, and Earth orientation parameters). The only adjusted parameter is the effective area $A$ , which acts as a proxy for the net drag behavior, absorbing variations in geometry, attitude, and thermal state.
Data Source: The method relies exclusively on public data: successive TLEs and final Tracking and Impact Prediction (TIP) messages from Space-Track.

Key Results
The method was applied to recent natural re-entries of rocket bodies, debris, and non-manoeuvring satellites, as well as specific Starlink cases:

Accuracy: In the terminal phase, the method consistently reproduces published TIP times with an accuracy of approximately $\pm 1$ hour. The deviation relative to final TIPs often reduces by a factor of two to four in the final 6–12 hours of orbital life compared to earlier estimates.
Final Orbit Identification: The approach reliably identifies the final complete orbit prior to atmospheric entry, allowing for the exclusion of all regions outside the final ground track with certainty.
Case Studies:
- Natural Decay: For objects like NORAD 43802 and 61440, the method converged smoothly to the final TIP.
- Starlink Variations: The method successfully tracked natural decay (STARLINK-2171) and thrusting events (STARLINK-1363). However, for rapid deorbit maneuvers (STARLINK-34110), real-time prediction was impossible during the thrusting phase; the calibrated $M/A$ values spiked (indicating propulsion rather than drag), and accurate reconstruction was only possible retrospectively once the maneuver concluded.
Independent Validation: Several predictions were confirmed by independent observations, including an ISS sighting of the PROGRESS MS-34 re-entry, witness reports in Sri Lanka for ICEYE-X26, and 91 reports from the American Meteor Society for CLARITY-1.

Limitations
The paper explicitly defines the operational envelope and constraints of the method:

Catalogue Constraints: The primary limiting factor is the intrinsic noise, quantization, and publication cadence of the TLE catalogue, not the atmospheric or solar-flux modeling.
Maneuvers: Controlled maneuvers or thrusting episodes temporarily invalidate short-term predictions because the semi-major axis evolution reflects propulsion rather than drag. Predictions only regain reliability once the object returns to a purely ballistic regime.
Precision Ceiling: Even under favorable conditions, the final decay time can typically be recovered with an accuracy of about $\pm 1$ hour. This is a fundamental limit imposed by the information content of the TLE data.

Significance and Claims
The paper claims that segmented $M/A$ calibration offers a transparent, lightweight, and operationally robust alternative to long-arc drag fitting when only public tracking data are available. Its significance lies in:

Empirical Robustness: By anchoring each segment to a new catalogue measurement, the method prevents long-arc divergence and isolates catalogue noise as the dominant uncertainty.
Practicality: It does not require detailed knowledge of the object's aerodynamic state or geometry, making it applicable to a wide range of objects (debris, rocket bodies, satellites) where such data is unknown.
Reproducibility: The method provides a reproducible framework for event identification and rapid decay reconstruction, suitable for independent verification of publicly reported re-entries.

The author concludes that while the method cannot overcome the intrinsic limitations of the TLE catalogue, it effectively extracts the maximum possible dynamical signal from public data, providing hour-scale accuracy in the critical terminal phase of re-entry.