New SVD-Based Method Predicts Rocket Debris Footprints in Real Time

When a launch vehicle breaks apart in flight, the consequences on the ground depend on one deceptively simple question: where will the pieces land? Answering that question quickly enough to matter has long been a computational headache for range safety engineers. A team of researchers in South Korea now reports a new analytical technique that can map the potential debris dispersion area of a failing rocket with an average error of less than 0.15 percent compared with established statistical methods, while running dramatically faster as the number of potential fragments grows. The study, published in the International Journal of Aeronautical and Space Sciences, was validated against real flight data from the second launch of Korea’s KSLV-II rocket and points toward a future in which debris hazard zones could be computed onboard, in real time, rather than on the ground after the fact.

The work was carried out by Bum-Yong Park and Dong-Hyun Cho of the Department of Aerospace Engineering at Pusan National University, together with Tae-Hun Kim of Korea Aerospace University and Ha-Ryong Song of the Flight Safety Technology Division at the Korea Aerospace Research Institute. Their collaboration reflects the practical stakes of the problem: KARI, Korea’s national space agency, provided flight data and materials for the study, grounding the mathematics in the realities of an actual launch campaign. The research was supported in part by a 2-Year Research Grant from Pusan National University.

To understand why the new method matters, it helps to consider how debris dispersion has traditionally been estimated. The gold standard is the Monte Carlo simulation, in which thousands or millions of virtual debris fragments are generated, each assigned slightly different initial conditions, and each propagated through the atmosphere under gravity and drag until impact. The resulting scatter of impact points defines the hazard area that range safety officers must keep clear of people, ships, and aircraft. Monte Carlo methods are robust and statistically well understood, but their cost scales directly with the number of samples and fragments. When a vehicle breakup produces dozens or hundreds of candidate fragments, each requiring its own trajectory integration, the computation can become too slow for the time-critical decisions that follow an anomaly.

An alternative that has gained traction in guidance and estimation circles is the Unscented Transform, a technique drawn from unscented Kalman filtering. Rather than sampling randomly, the Unscented Transform deterministically selects a small set of sigma points around the nominal state, propagates each one through the nonlinear dynamics, and reconstructs the mean and covariance of the resulting distribution. It is far cheaper than Monte Carlo for modest state dimensions, but it still requires explicit propagation of multiple representative points, and its accuracy can degrade for strongly nonlinear problems or when the number of uncertain parameters grows. Both approaches share a common burden: they must actually generate and fly candidate fragments, whether sampled or deterministic, before any statement about the dispersion area can be made.

The Korean team’s contribution, which they call the F&G SVD method, sidesteps that burden entirely. Instead of simulating individual fragments, the method characterizes the spread of debris analytically. The F&G approach, combined with projection techniques, expresses how variations in the initial conditions of a fragment translate into variations in its eventual impact point. Singular value decomposition then enters as the mathematical engine of the method: by decomposing the matrix that maps initial uncertainties to impact-point uncertainties, the technique extracts the principal directions and magnitudes of the dispersion ellipse directly. The singular values quantify how strongly each mode of initial variation amplifies into ground-level scatter, and the corresponding singular vectors define the orientation of the hazard footprint. In effect, the method delivers a conservative envelope of the debris dispersion area without ever flying a single virtual fragment.

This is a meaningful conceptual shift. Conventional statistical approaches treat the dispersion area as an emergent property of many simulated trajectories; the F&G SVD method treats it as a geometric property of the mapping from launch conditions to impact points. Because singular value decomposition is a well-established tool of numerical linear algebra with decades of stable, efficient implementations behind it, the resulting algorithm is lightweight and predictable. The authors emphasize that their technique provides a rapid and conservative estimate, which is precisely what range safety demands: a hazard zone that errs on the side of caution can be computed in the moments available after an anomaly, whereas a perfectly refined but late answer is useless.

The quantitative comparison with Monte Carlo simulation and the Unscented Transform is the heart of the paper’s case. Across the evaluated scenarios, the proposed method achieved accuracy comparable to the conventional techniques, with an average error within 0.15 percent in the estimated dispersion characteristics. That level of agreement indicates that the analytical characterization captures the same physics that brute-force sampling reveals, at least for the launch vehicle configurations tested. More striking is the computational behavior. While the cost of Monte Carlo and Unscented Transform calculations grows as the number of potential fragments increases, the F&G SVD method maintains a nearly constant computation time. In other words, whether a breakup scenario involves a handful of plausible fragments or a large population of them, the analytical method finishes in essentially the same amount of time, and its advantage over the conventional methods widens as the fragment count climbs.

The validation step distinguishes this work from purely theoretical proposals. The team applied the method to flight data from the second launch of the Korea Space Launch Vehicle II, known as KSLV-II or Nuri, the country’s indigenous three-stage orbital rocket. Using real telemetry and flight conditions rather than synthetic scenarios, the researchers demonstrated that the algorithm produces dispersion estimates consistent with the operational picture of the mission. The authors conclude that the method shows strong potential as a lightweight and reliable algorithm for real-time onboard debris dispersion prediction, a phrase that deserves unpacking. Onboard prediction means the flight computer itself, or a range safety processor riding with the vehicle, could continuously update the hazard footprint during ascent, without downlinking data to ground stations and waiting for ground-based simulations to catch up. For launches over ocean or remote ranges, and especially for missions where the response window after a breakup is measured in seconds, that capability could reshape how flight safety systems are designed.

The broader context reinforces the significance. Range safety has long relied on instantaneous impact point calculations, a lineage of methods that traces back decades in sounding rocket programs and has been progressively refined to include atmospheric drag and more sophisticated decision rules. Researchers have explored response-surface methods, deep neural networks, and integrated filtering approaches to speed up impact prediction, and hazard analyses for uncontrolled reentries have grappled with the same trade-off between statistical fidelity and computational speed. The F&G SVD method adds a distinct option to that toolbox: an approach whose cost does not scale with the size of the debris population. It also connects to a wider trend in aerospace engineering, in which linear algebra techniques such as singular value decomposition are used to compress, characterize, and control complex systems, from structural dynamics to trajectory optimization.

Certain caveats remain, and the authors are transparent about the scope of their work. The method is designed to be conservative, meaning it deliberately bounds the hazard area rather than pinpointing the exact distribution of impacts, which is the appropriate posture for safety but means the technique complements rather than replaces detailed post-flight analysis. The underlying flight data used in the study contains sensitive information and is not publicly available for security reasons, with limited access possible after a security review. And while the validation on KSLV-II is compelling, the method’s performance across other vehicle classes, breakup modes, and atmospheric conditions will be the test of its generality. Still, the combination of sub-0.15-percent average error, fragment-count-independent runtime, and demonstrated performance on a real orbital launch makes a strong case that the next generation of flight safety systems could carry their own debris forecasters, computing the shape of danger in the sky before the debris ever begins to fall.

Subject of Research: Analytical estimation of launch vehicle debris dispersion areas using singular value decomposition for real-time flight safety

Article Title: Estimation of Launch Vehicle Debris Dispersion Area Using the F&G SVD Method

Article References: Park, B.-Y., Cho, D.-H., Kim, T.-H., & Song, H.-R. (2026). Estimation of Launch Vehicle Debris Dispersion Area Using the F&G SVD Method. International Journal of Aeronautical and Space Sciences. https://doi.org/10.1007/s42405-026-01248-x

Image Credits: AI Generated

DOI: 10.1007/s42405-026-01248-x

Keywords: launch vehicle, debris dispersion, singular value decomposition, range safety, Monte Carlo simulation, Unscented Transform, KSLV-II, flight safety, aerospace engineering, impact prediction, Korea Aerospace Research Institute, real-time computation