Seismological Reconstruction of Meteoroids: Trajectories, Physical Parameters, and Strewn Fields in Wind-Perturbed Atmospheres

Every day, the Earth's atmosphere is struck by meteoroids, and the recovery of fresh meteorite material is the main way of improving our understanding of how the Solar System and the Earth formed. However, meteorite recovery is often limited because not all meteorite-producing meteoroids are observed. Meteoroid detection is currently dominated by optical fireball networks, whose individual cameras are affected by clouds, daylight and spatial coverage gaps. Dense, globally distributed seismic networks offer a time- and weather-independent alternative. Since meteoroids continuously generate acoustic waves during their supersonic flight, the resulting air-to-ground coupled seismic waves can be used to reconstruct their atmospheric path. This provides a distinct physical method of constraining meteoroid parameters. However, exploiting this approach requires a new framework to systematically detect meteoroid signals in seismic data and reconstruct trajectories in wind-perturbed atmospheres. Only then can the seismic method reliably complement the current approaches to estimating meteoroid parameters, trajectories, and meteorite ground locations.
... mehr

In this work, we outline the development and application of this end-to-end framework, which covers meteoroid detection to meteorite strewn-field reconstruction. First, we compile the Seismo-Acoustic Meteoroid Observation Recording Database (SMORD), which contains 307 meteoroid events and 3,295 manually labelled arrivals across a global set of stations. We then use this database to train a PhaseNet-based deep-learning phase picker in SeisBench. The model achieves 91% precision and 94% recall (AUC 0.89), with a median absolute timing error of 0.02 s. We demonstrate how this model can automatically detect meteoroid onsets and reconstruct the trajectory of a meteoroid over the Adriatic Sea.
We then improve the trajectory reconstruction method by using the uncontrolled re-entry of the Starlink-2382 satellite as a meteoroid equivalent. By incorporating a 3D wind model and a lift-like parameter into a ray-based direct-search optimisation, we significantly improve the travel-time fit to recording data from 226 stations. This allows us to recover an entry velocity of $8.9 \pm 0.7$ km s$^{-1}$ and an effective ablation coefficient consistent with near-complete ablation of the ${\sim}100$ kg main satellite fragment. Using collocated acoustic and seismic sensors, we found that the acoustic-to-seismic ground coupling coefficients vary by up to three orders of magnitude (from $4.31 \times 10^{-10}$ m s$^{-1}$ Pa$^{-1}$ to $5.86 \times 10^{-7}$ m s$^{-1}$ Pa$^{-1}$), mainly due to differences in near-surface rock stiffness.

Finally, we use probabilistic Markov chain Monte Carlo ensemble sampling to perform a statistically robust reconstruction of the 8 March 2026 Koblenz meteoroid trajectory, which was recorded at 95 stations. Replacing direct ray computation with a precomputed lookup table reduces new ray calculations to simple interpolation, enabling this full Bayesian inversion of the meteoroid trajectory. With this trajectory and the associated uncertainties that can be propagated, we recover the heliocentric orbit and initial mass of 305 +180/-90 kg of the meteoroid. Additionally, we identify five distinct fragmentation events along the atmospheric trajectory and calculate a strewn field, which is validated against fragments that struck a building in Koblenz-Güls. Together, these results demonstrate the successful application of the end-to-end acoustic-seismic framework which can shape future meteoroid trajectory reconstructions and enables the rapid, automated calculation of meteorite strewn fields.