Astronomy & Astrophysics manuscript no.

aa54364-25 ©ESO 2025

October 27, 2025

Comparing the data reduction pipelines of FRIPON, DFN, WMPL,

and AMOS: Geminids Case Study
P.M. Shober1,⋆ , J. Vaubaillon1 , S. Anghel1, 2,⋆ , H.A.R. Devillepoix3, 4 , F. Hlobik5 , P. Matlovič5 , J. Tóth5 , D. Vida6 ,
E.K. Sansom4 , T. Jansen-Sturgeon7 , F. Colas1 , A. Malgoyre8 , L. Kornoš5 , F. Ďuriš5 , V. Pazderová5 , S. Bouley9 , B.
Zanda10 , and P. Vernazza11

1
LTE, Observatoire de Paris, Université PSL, Sorbonne Université, Université de Lille, LNE, CNRS, 61 Avenue de l’Observatoire,
Paris, 75014, France
arXiv:2510.21690v1 [[Link]] 24 Oct 2025

2
Astronomical Institute of the Romanian Academy, 5 - Cutitul de Argint Street, 040557 Bucharest, Romania
3
Space Science and Technology Centre, School of Earth and Planetary Sciences, Curtin University, Perth, WA 6845, Australia
4
International Centre for Radio Astronomy Research, Curtin University, Perth, WA 6845, Australia
5
Faculty of Mathematics, Physics and Informatics, Comenius University, Bratislava, Slovakia
6
Department of Physics and Astronomy, University of Western Ontario, London, Ontario, Canada
7
STELaRLab, Lockheed Martin Australia, Adelaide, South Australia, Australia
8
Service Informatique Pythéas (SIP) CNRS – OSU Institut Pythéas – UMS 3470, Marseille, France
9
Université Paris-Saclay, UMR CNRS 8148, GEOPS, Orsay, France
10
Institut de Minéralogie, Physique des Matériaux et Cosmochimie, Muséum National d’Histoire Naturelle, CNRS, Paris, 75005,
France
11
Laboratoire d’Astrophysique de Marseille, Aix-Marseille University, CNRS, CNES, LAM, Institut Origines, 38 rue Frederic Joliot
Curie, Marseille, 13388, France

Received March 4, 2025; accepted October 20, 2025

ABSTRACT

Context. The number of meteor observation networks has expanded rapidly due to declining hardware costs, enabling both profes-
sional and amateur groups to contribute substantial datasets. However, accurate data reduction remains challenging, as variations in
processing methodologies can significantly influence trajectory reconstructions and orbital interpretations.
Aims. Our goal is to thoroughly compare four professionally produced meteor data reduction pipelines (FRIPON, DFN, WMPL,
AMOS) by reprocessing FRIPON Geminid observations. Others can then use this analysis to compare their data reduction methods.
Methods. We processed a dataset of 584 Geminid fireballs observed by FRIPON between 2016-2023. The single-station astrometric
data is converted into the Global Fireball Exchange (GFE) standard format for uniform processing. We assess variations in trajectory,
velocity, radiant, and orbital element calculations across the pipelines and compare them to previously published Geminid measure-
ments.
Results. The radiant and velocity solutions provided by the four data reduction pipelines are all within the range of previously
published values. However, there are some nuances. Particularly, the radiants estimated by WMPL, DFN, and AMOS are nearly
identical. Whereas FRIPON reports a systematic shift in right ascension (- 0.3◦ ), caused by improper handling of the precession.
Additionally, the FRIPON data reduction pipeline also tends to overestimate the initial velocity (+0.3 km s−1 ) due to the deceleration
model used as the velocity solver. The FRIPON velocity method relies on having a well-constrained deceleration profile; however,
for the Geminids, many are low-deceleration events, leading to an overestimation of the initial velocity. On the other end of the
spectrum, the DFN tends to predict lower velocities, particularly for poorly observed events. However, this velocity shift vanishes for
the DFN when we only consider Geminids with at least three observations or more. The primary difference identified in the analysis
concerns the velocity uncertainties. Despite all four pipelines achieving similar residuals between their trajectories and observations,
their velocity uncertainties vary systematically, with WMPL outputting the smallest values, followed by AMOS, FRIPON, and DFN.
Conclusions. From this Geminid case study, we find that the default FRIPON data reduction methods, while adequate for meteorite-
dropping events, are not optimal for all cases. Specifically, FRIPON tends to overestimate velocities for low-deceleration events due
to the less constrained fit, and the nominal radiants are not correctly output in J2000. Meanwhile, the other data reduction pipelines
(DFN, WMPL, AMOS) produce consistent results, provided that the observational data is sufficiently robust, i.e., more than ∼50 data
points from at least three observers. A key takeaway is the need to re-evaluate how velocity uncertainties are estimated. Our results
show that uncertainty estimates vary systematically across different pipelines, despite generally similar goodness-of-fit statistics. The
increasing availability of impact observations from varying sources (radar, video, photo, seismic, infrasound, satellite, telescopic, etc.)
calls for greater collaboration and transparency in data reduction practices.
Key words. Methods: data analysis – Methods: observational – Meteorites, meteors, meteoroids

1. Introduction

Fireball observations have historically relied on film-based

⋆
corresponding authors techniques (Ceplecha 1961; McCrosky et al. 1971; Halliday
Article number, page 1 of 22
 A&A proofs: manuscript no. aa54364-25

et al. 1978), but the advent of charge-coupled devices (CCDs) (e.g., ablation, mass loss, fragmentation) requires careful the-
and, more recently, complementary metal–oxide–semiconductor oretical modelling. A meteoroid’s flight through Earth’s atmo-
(CMOS) sensors has revolutionised both the scale and precision sphere involves multiple interacting processes – shock wave
of meteor and fireball surveys (Molau 1999; Koten et al. 2019; generation, aerodynamic heating, ablation and mass loss, frag-
Arlt et al. 2022). mentation, ionisation, and the resulting luminous emission –
Photographic observations generally excel in spatial resolu- each influencing the meteor’s brightness, trajectory, and out-
tion and yield very precise astrometric solutions, making them come. Additionally, the unknown parameters of the meteoroids,
particularly well-suited for recovering meteorites. However, be- such as shape, spin, mass, and composition, further complicate
cause photographic techniques rely on relatively few discrete ex- this issue. Early models often adopted “single-body” ablation, in
posures (e.g., one exposure spanning minutes to hours), they pro- which the mass of a compact object with cross-section S evolves
vide limited time resolution; abrupt changes in meteor bright- under standard drag and ablation equations (Bronshtehn 1983;
ness, detailed velocity variations, or short-duration fragmenta- Ceplecha et al. 1998). This approach remains valuable in many
tion episodes can be difficult to capture. By contrast, video ob- automated or large-scale fireball networks, where a linear or
servations utilise continuous, frame-based recordings, typically straight-line fit to the geometric path is common (e.g., Borovicka
at 25–60 frames per second (or higher), providing near-real-time 1990; Gural 2012). Such simplifications can be sufficient un-
detection and automated data processing. This frame-by-frame der favourable conditions or for specific events. However, re-
approach provides excellent temporal resolution at the cost of cent work has underscored that fragmentation is pervasive and
lower spatial resolution. However, this shortcoming can be mit- can drastically alter a meteoroid’s velocity profile and final mass
igated by employing high-definition sensors or deploying mul- (Ceplecha et al. 1993; Boroviĉka & Kalenda 2003; Borovička
tiple narrow-field cameras to enhance positional accuracy. As et al. 2020, 2022b,a), necessitating more robust dynamical mod-
commercial CCDs and later CMOS sensors matured, sensitive els and careful uncertainty quantification. A primary challenge,
yet affordable camera hardware proliferated in both professional as emphasised by Egal et al. (2017), lies in accurately deriv-
and amateur networks (Jeanne et al. 2019; Koten et al. 2019; Arlt ing the velocity from photographic or video records. Many early
et al. 2022). methods that separate the geometric and velocity fits – such as
Film-based photographic observatories continued into the the straight-line least-squares approach (Borovicka 1990) – may
21st century for bright fireball observations, but as digital sen- underestimate uncertainties when the observations have complex
sors shrank in size, improved in sensitivity, and fell in cost, geometry or when the meteoroid undergoes multiple fragmenta-
analogue film gave way to digital imaging (Spurnỳ et al. 2006; tion episodes. In particular, Egal et al. (2017) highlights that even
Howie et al. 2017a). This digital revolution has spawned large- small systematic biases in frame timing or astrometric precision
scale photographic fireball observation networks that were pre- can lead to errors in the derived deceleration profile, with direct
viously infeasible due to geography or expense (Bland et al. consequences for orbit determination. This issue becomes espe-
2012; Koten et al. 2019; Devillepoix et al. 2020), leading to a cially critical in low-convergence-angle scenarios or when the
significant increase in data collection and meteorite recoveries. event has a protracted, curved path. Advanced pipelines increas-
In the four decades between 1959 and 1999, only five meteorite ingly employ dynamic trajectory solvers that integrate the mete-
falls (Přibram, Lost City, Innisfree, Benešov, Peekskill) were re- oroid’s three-dimensional equations of motion directly into the
covered based on instrumental observations. By comparison, in data-fitting step (Gural 2012; Vida et al. 2020b; Jansen-Sturgeon
2023 alone, meteorites from three distinct falls were recovered et al. 2020). Rather than first inferring a simple geometric path
on three consecutive days: 13 February in France (Zanda et al. and then fitting a velocity separately, these approaches can si-
2023), 14 February in Italy (Barghini et al. 2024), and 15 Febru- multaneously solve for position and velocity at each timestep, of-
ary in Texas1 . ten incorporating Monte Carlo or particle-filter methods to map
A diverse array of meteor and fireball networks now exists uncertainties more fully (Jansen-Sturgeon et al. 2020; Sansom
worldwide, ranging from higher-precision, photographic sys- et al. 2019b).
tems like the DFN (Howie et al. 2017a; Devillepoix et al. 2020) Along similar lines, Egal et al. (2017) stresses that reliably
and the EN (Borovička et al. 2022b), to more modular video computing velocity requires careful propagation of all measure-
approaches like FRIPON (Colas et al. 2020) and the Global Me- ment errors – most critically timing, pointing astrometry, and
teor Network (Vida et al. 2021). Additional examples include the any potential system offsets – to avoid unrealistically low or in-
Spanish Meteor Network (Trigo-Rodríguez et al. 2006), NASA’s consistent uncertainties. These modern methods reflect a broader
All-Sky Fireball Network (Cooke & Moser 2012), the Cameras push toward Bayesian or stochastic approaches that can capture
for all-sky Meteor Surveillance (CAMS) project (Jenniskens the inherent complexity of meteor flight and assign more realis-
et al. 2011), SonotaCo’s collaborative effort in Japan (Sono- tic error bars. While most authors now agree that velocity uncer-
taCo 2009), Polish Fireball Network (Wiśniewski et al. 2017), tainties below ∼ 0.1 km s−1 are necessary to reliably discriminate
and the AMOS initiative in Slovakia (Tóth et al. 2015), as well orbital source regions (Granvik et al. 2018), actual uncertain-
as a growing number of amateur-run networks (e.g., all-sky7; ties in automated or semi-automated networks sometimes exceed
Hankey et al. 2020). Although some focus on near-real-time, this figure due to sensor calibration, partial fragmentation mod-
frame-based video while others opt for longer-exposure, higher- elling, or constraints in processing time. Many networks still rely
resolution imaging, many of the calibration and data-reduction on simplified single-body or straight-line fits in their automated
methods are effectively the same (Koten et al. 2019). Each net- pipelines, principally for operational efficiency.
work’s hardware choices are shaped by its scientific goals and We are collecting more and more data on the meteoroids that
expense limitations. continuously impact our atmosphere (Koten et al. 2019). It is ex-
Building reliable links between observable parameters (e.g., cellent to have increasingly large meteor and fireball datasets.
light curves and velocity profiles) and the internal processes Albeit, we must be cautious. Despite the rapid amelioration of
hardware costs, enabling the development of many more profes-
1
[Link] sional networks and even amateur networks, analysing the col-
el-sauz-tx lected impact data is still complicated and difficult to reduce.
Article number, page 2 of 22
 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

We are using this data to better understand the meteoroid en- estimated to weigh 500 grams or more, by modelling the im-
vironment in near-Earth space, the source regions of meteorites, pact via photometric and dynamic methods (Anghel et al. 2021;
and the physical and orbital evolution of small-body populations. Jeanne et al. 2019). To date, the global expansion of the consor-
Thus, we must be transparent about our data-reduction methods, tium has aided in the recovery of five meteorites, marking signif-
as slight variations or systematic errors in data reduction can sig- icant strides in the field of meteorite retrieval and interplanetary
nificantly alter how we interpret the results and their implications study (Gardiol et al. 2021; McMullan et al. 2023; Antier 2023;
for our understanding of small bodies in the solar system. This Barghini et al. 2024; Egal et al. 2025). The data reduction pro-
is especially important as many networks, particularly video- cess within FRIPON is described in detail in Jeanne et al. (2019),
based, are increasingly fully automated in their data acquisition, Jeanne (2020), and Colas et al. (2020). The processed data col-
processing, and reduction (Molau 2001; Gural & Segon 2009; lected by the FRIPON project is publicly available2 .
Brown et al. 2010a; Gural 2011; Weryk et al. 2013; Suk & Šim- FRIPON estimates the internal error of each camera
berová 2017; Koten et al. 2019; Colas et al. 2020; Peña-Asensio (σi ≃ 0.75 arcmin; Colas et al. 2020) by fitting a reference plane
et al. 2021; Vida et al. 2021). through all picks. A local comparison (100 px box) between
In today’s modern research environment, we must be in- measured and HIPPARCOS positions gives the systematic term
creasingly adept at sharing our code so that others may test and si (Bessell 2000). The straight–line trajectory T is obtained by
scrutinise it. The first clear example was the observation and re- minimising
covery of the Winchcombe meteorite fall in the UK in Febru-
ary 2021. This carbonaceous meteorite fall was observed by five
ncam X
ni
different camera systems and six networks in total. Due to the X ϵi j (T )2
extraordinary nature of the fall and observations, it was neces- S (T ) = , (1)
i=1 j=1
σ2i + ni s2i
sary for the first time to combine and coordinate the observa-
tions and pipelines of several fireball networks. This collabo-
ration led to the development of the Global Fireball Exchange where ϵi j (T ) is the residual between the j-th measure taken by
(GFE) format, a universal method of sharing the detailed single- the i-th camera and the candidate trajectory T , si is its system-
station observations of a fireball event (Rowe 2021). Networks atic error, σi is the internal random error of the i-th camera, and
then implemented this format so that they could all communicate ni is the number of images acquired by that camera. FRIPON
and process the data in coordination, leading to the recovery of uses this method to characterise the systematic errors of our cam-
the meteorite and its complete characterisation (King et al. 2022; eras (e.g., a misaligned lens); however, this approach doesn’t ad-
McMullan et al. 2023). dress errors such as the camera’s location. To address these er-
This work aims to assess these cross-network variations rors, FRIPON computes an initial estimate of the trajectory and
and promote best practices for ensuring consistent, reliable re- compares the residuals with the expected random and systematic
sults across different pipelines. We compare four established, errors. If they are larger than expected for a specific camera, they
professionally developed fireball data-reduction pipelines, all iteratively decrease its weight during the calculation of the tra-
tested against actual FRIPON network observations. Our cross- jectory. The final systematic error is usually on the order of 0.3
evaluation highlights differences in calibration strategies, trajec- arcmin, which ends the iterative process.
tory modelling approaches, and orbital solutions, offering in- The entry speed is recovered based on the single–body
sights for more convergent standards in future multi-network drag/ablation model of Bronshtehn (1983) and Stulov et al.
meteor research. (1995). Methods such as Gritsevich (2009) reformulate these
The paper is organised as follows: Section 2 introduces the without assuming any parameters, enabling a dimensionless co-
four contributing networks and their pipelines. Section 3 outlines efficient approach. The FRIPON pipeline employs the method
the data sets and processing methods, while Section 4 compares described by Jeanne et al. (2019). Although similar to Gritse-
the derived trajectories and orbital elements. We discuss implica- vich (2009), it uses a different set of equations that do not make
tions for multi-network validation in Section 5, and finally, Sec- use of the dimensionless parameters:
tion 6 explores strategies for improving meteor data reduction
and future collaborative efforts.
dV h i
= − 21 A ρatm V 2 exp B(1−µ)
2A (V 2
− V 2
e ) ,
dt (2)
2. Networks and software used h i
m = exp 2A B
(V 2 − Ve2 ) .
2.1. FRIPON
Initiated in 2015, FRIPON is a pioneering French-led project with a deceleration parameter A = cd S e /Me and an ablation pa-
dedicated to the detailed observation of fireballs and the recov- rameter B = ch S e /(HMe ), where cd is the drag coefficient, ch
ery of meteorites. This collaborative effort now spans 15 coun- the heat–transfer coefficient, H the enthalpy of destruction, ρatm
tries on four continents, incorporating over 220 cameras cover- the atmospheric gas density (NRLMSISE-00 model; Lyytinen &
ing approximately 2 million km 2 . Using all-sky CCD cameras Gritsevich 2016), m the normalised meteoroid mass, Me the pre-
that record at 30 frames per second, FRIPON achieves high- entry mass, s the normalised cross-sectional area, S e the pre-
time-resolution data capture. The network has a detection limit entry cross-sectional area, and µ the shape-change coefficient.
of about zero magnitude, making it tuned for fireball observation Fitting the projected picks on the straight line (Jeanne et al.
and meteorite recovery. Extended exposures are taken every 10 2019) yields the parameters Ve , A, and B and their joint confi-
minutes to enhance astrometry and photometry, achieving effec- dence region. The estimated Ve value represents the velocity at
tive signal-to-noise ratios up to magnitude 6 in optimal condi- the top of the atmosphere, within FRIPON this is arbitrarily de-
tions(Anghel et al. 2019; Jeanne et al. 2019; Jeanne 2020; Colas fined as 10 km above the first measured point.
et al. 2020). The automated system of FRIPON enables swift
2
mobilisation for meteorite recovery efforts, focusing on objects [Link]

Article number, page 3 of 22

 A&A proofs: manuscript no. aa54364-25

Table 1: Comparison of trajectory, velocity solvers and outlier removal approaches.

Pipeline Designed For Trajectory Solver Velocity Solver Outlier Removal? Sources
FRIPON Lower- Straight-line least squares Fits physically based Yes †
resolution (Borovicka 1990) single-body differential
video, fireballs equations that model
meteor deceleration and
ablation, rewritten into
two independent equations
by Jeanne et al. (2019)
following the method of
Turchak & Gritsevich
(2014); Gritsevich (2009).
DFN Higher- Straight-line least squares Meteoroid initial states No ‡
resolution (Borovicka 1990) (Me and Ve ) are esti-
photo, fireballs mated using an Extended
Kalman Filter on straight
line triangulated positions
(measurements). This
updates estimates of po-
sition, mass, and velocity
throughout the trajectory,
while modelling uncer-
tainties.
WMPL Higher- Monte Carlo trajectory A linear “lag” fit is Outlier weighting is §
resolution solver: initial geometry performed on the early applied on a per-
video, meteors via IP/LoS, followed by 25–80% of the meteor’s measurement basis
noise injection (based path (where deceleration (with lower weights for
on measured angular is minimal). The best observations with poor
residuals) and global fit, i.e., the one with the perspective)
minimisation of a timing smallest standard devia-
cost function tion, provides the initial
velocity estimate.
AMOS Lower- Intersecting planes and Exponential fit of decel- Yes ¶
resolution Straight-line least squares erated meteors along the
video, meteors, methods (Ceplecha 1987; length of trajectory, other-
spectroscopy Borovicka 1990) wise linear fit.
Notes. “Higher-resolution” implies pixel scale ∼1 arcminute; “lower-resolution” implies pixel scale ∼10 arcminutes.
†
Jeanne et al. (2019), Jeanne (2020), Colas et al. (2020)
‡
Howie et al. (2017a), Sansom et al. (2015), Devillepoix et al. (2020)
§
Vida et al. (2018), Vida et al. (2020b)
¶
Duris et al. (2018), Tóth (in prep.)

2.2. DFN coupled with an extended Kalman smoother, ensuring accurate

trajectory and velocity data while accounting for observational
The Desert Fireball Network (DFN), established to cover the ex- and fitting uncertainties (Borovicka 1990; Sansom et al. 2015).
pansive Australian outback, surveys over 2.5 million km2 . This
equates to over a third of the Australian continent, underscoring Expanding beyond its Australian origins, the DFN now
the DFN’s extensive observational reach since its digital transi- forms the core of the Global Fireball Observatory (GFO)3 , in-
tion in 2013-2015 (Bland et al. 2012; Howie et al. 2017a). The tegrating ten partner networks and 18 collaborating institutions
deployment of 50 digital fireball observatories, equipped with across nine countries, all employing DFN-developed observato-
high-resolution DSLR cameras and all-sky fisheye lenses, en- ries (Devillepoix et al. 2020). The GFO has led to the retrieval of
ables the network to achieve unprecedented capture rates of me- 18 orbital meteorites in total4 , representing about 30% of all me-
teoroids, thereby enhancing the precision of meteorite fall po- teorites recovered with known orbits to date (Bland et al. 2009;
sition determinations through GNSS-synchronised liquid crystal Dyl et al. 2016; Devillepoix et al. 2018; Jenniskens et al. 2019;
shutters (Howie et al. 2017b; Devillepoix et al. 2019). The pre- Sansom et al. 2020; King et al. 2022; Shober et al. 2022; Dev-
cision and manual picking of fireball observations have resulted illepoix et al. 2022; Anderson et al. 2022; Brown et al. 2023).
in a vast, accurate fireball dataset that helps us better understand
the dynamics of centimeter- to meter-sized debris in the inner so-
lar system (Devillepoix et al. 2019; Sansom et al. 2019a; Shober
et al. 2020, 2021, 2025). The DFN’s automated methodology
for determining atmospheric trajectories and extracting veloc- 3
[Link]
4
ity profiles utilizes a refined straight-line least squares approach [Link]

Article number, page 4 of 22

 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

2.3. AMOS In addition to WMPL, the Raspberry Pi Meteor Station

5 (RMS) library7 provides universal tools for reducing optical me-
Based in Slovakia, the All-Sky Meteor Orbit System (AMOS) teor data (both photographic and video), implementing the lat-
is an advanced global meteor observation network that utilises est astrometric and photometric calibration methods (Vida et al.
automated intensified all-sky video cameras to track and anal- 2021). The measurements made by RMS tools are fully com-
yse meteor events, and wide-field spectral cameras to study patible with the international standards (Rowe 2021) and can be
meteoroid composition. Established in 2009 by the Faculty of transparently used in WMPL.
Mathematics, Physics, and Informatics at Comenius University
Bratislava, AMOS aims to provide coverage over Central Europe
with a network currently spanning six stations in Slovakia and, 3. Methods
in total, 17 stations around the globe. Other network locations
include the Canary Islands, Chile, Hawaii, Australia, and South 3.1. FRIPON Observations
Africa. Each station has a digital CCD camera capable of record-
ing high-resolution video (1600x1200) at up to 20 frames per When integrating and comparing meteor data across different
second, covering a wide range of meteor magnitudes, typically networks, it is crucial to account for variations in hardware, ob-
between +4.0 and -3.0 in stellar magnitude, including fireballs serving cadence, exposure lengths, network density, and calibra-
as well Tóth et al. (2015). AMOS’s primary goal is to calcu- tion and reduction methods. These factors can significantly influ-
late precise trajectories and orbits of meteors and correlate them ence the accuracy of the trajectory and origin estimations and the
with their spectral properties studied by the AMOS-Spec cam- overall interpretation of the meteor data. These factors – ranging
eras (Matlovič et al. 2020) to improve the knowledge of weak from sensor resolution and frame rate to the specific observa-
meteor showers and their parent body characterisation, or to tional strategies employed – directly influence the accuracy and
provide comprehensive studies of meteoroid dynamics, physi- precision of derived trajectories, orbital elements, and their cor-
cal properties and composition (Matlovič et al. 2019; Matlovič responding uncertainties.
et al. 2022). The network has successfully integrated cutting- The FRIPON observatories use a CCD Sony ICX445 chip
edge software algorithms to determine the atmospheric paths, with 1296 × 964 pixels and a pixel size of 3.75 × 3.75 µm. The
radiant points, and orbital parameters of meteors, utilising meth- lens is a fish-eye with a 1.25 mm focal length opened to f/2, giv-
ods like the Least Squares Monte Carlo fitting technique (Duris ing a pixel scale of 10 arcmin (Colas et al. 2020). These obser-
et al. 2018) for enhanced accuracy in trajectory reconstruction vatories are lower resolution than those used by the EN or DFN
(Tóth et al. 2019). These cameras are part of an integrated sys- (∼1-2 arcmin, i.e., 5-10× the resolution), which both utilise au-
tem designed for advanced meteor detection and analysis, con- tomated fireball observatories with high-resolution DSLR cam-
tributing to research on meteor trajectories, origins, and potential eras (Howie et al. 2017a; Borovička et al. 2022b). Despite the
recovery of meteorites. lower optical resolution of the FRIPON sensors, the network
operates and aims to recover meteorites through its higher net-
work density and increased frame rate. The CCD observations
2.4. WMPL and RMS made by FRIPON observatories are captured at 30 frames per
second. Additionally, FRIPON cameras have a target spacing
The Western Meteor Python Library (WMPL), developed by
of 80 km compared to the 100-150 km of the higher-resolution
the Meteor Physics Group at the University of Western Ontario
DSLR systems (Howie et al. 2017a; Colas et al. 2020). Re-
in Canada, is a comprehensive open source6 toolset written in
lying on the law of large numbers, this tighter station spac-
Python designed to process, analyze, and interpret the observa-
ing and increased sampling rate were found to adequately re-
tional data obtained from various meteor detection systems, fa-
duce the uncertainties on the trajectories for FRIPON (Jeanne
cilitating detailed analysis of meteor trajectories. The library in-
et al. 2019). FRIPON’s unique approach, which utilises a dense
cludes a variety of commonly used methods in the meteor field
network of lower-resolution cameras capturing at higher frame
(Vida et al. 2018), a novel meteor trajectory solver (Vida et al.
rates, enables effective triangulation and photometric measure-
2020b), a novel method for meteor shower mass index estima-
ments across a wide field of view, despite the lower resolution of
tion (Vida et al. 2020a), and implementation of the Borovička
individual cameras.
et al. (2007) faint meteor model (Vida et al. 2024), and imple-
mentation of the Borovička et al. (2013) semi-empirical fireball
fragmentation model (Vida et al. 2023). 3.2. Data processing
Applicable to this work, WMPL’s meteor trajectory solver
uses both the geometric and dynamical data to constrain me- This study focused on Geminid meteor shower FRIPON obser-
teor trajectories and applies rigorous statistical methods, includ- vations during 2016-2023 (584 fireballs). The goal is to check
ing Monte Carlo simulations, to fully propagate the uncertain- the differences in results between the data reduction pipelines of
ties in all observed trajectory parameters and derived orbital el- FRIPON, DFN, AMOS, and WMPL (as summarised in Table 1).
ements. The transparent nature of WMPL represents a signif- The FRIPON pipeline was initially used to process the observa-
icant step towards standardising the methodology used in the tions and make the astrometric reductions. For a description of
field. WMPL has been adopted as the main trajectory analysis the astrometric reduction methodology of FRIPON, please refer
toolset for several major meteor networks, including the originat- to Jeanne et al. (2019). Afterwards, the single-station observa-
ing Global Meteor Network (Vida et al. 2021), all-sky7 (Hankey tions were converted to the GFE8 standard format for processing
et al. 2020), the UK Meteor Network (Campbell-Burns & Kac- by the three other data processing pipelines (Rowe 2021). Astro-
erek 2014), and the NASA fireball network (Cooke & Moser metric calibration is highly network–specific and cannot realis-
2012). tically be rerun in four independent codes for this comparison.
5 7
[Link] RMS library on GitHub: [Link]
6 CroatianMeteorNetwork/RMS (accessed July 10, 2024)
WMPL library on GitHub: [Link]
8
WesternMeteorPyLib (accessed July 10, 2024) [Link]

Article number, page 5 of 22

 A&A proofs: manuscript no. aa54364-25

DFN solves a single long–exposure DSLR frame that contains tioned studies, providing a comprehensive comparison across
∼ 103 reference stars (Howie et al. 2017a); WMPL/RMS derives multiple methodologies. This table displays a general consis-
solutions from 88◦ ×48◦ HD video with typically 50–200 stars tency of the radiant and velocity estimates obtained from the dif-
per stack (Vida et al. 2021); whereas low-resolution FRIPON ferent pipelines with prior recorded data. Although the pipeline
cameras must integrate a dedicated 5-s exposure and merge a results are generally congruent with the established records,
few dozen detections from many epochs to reach a ≳ 100-star there are discernible, albeit minor, systematic variations at-
solution (Jeanne et al. 2019). AMOS performs all-sky video as- tributed to each pipeline. To visualise all pairwise differences
trometry based on the procedure of Borovicka et al. (1995) with between pipelines, we provide compact ∆–matrix histograms
typically 100–300 stars. Porting these four very different astrom- in Appendix A for radiants (Figs. A.1, A.2), geocentric speed
etry strategies into a single cross-network astrometric pipeline (Fig. A.3), and orbital elements (Figs. A.4–A.8).
would be a major software project and is beyond the scope of
the present work; instead, we rely on the published FRIPON so-
lutions (median ≃ 1 arc-min residuals; Jeanne et al. 2019) as a 4.1. Velocities
common, quality-controlled baseline for the trajectory and orbit
Previous velocity estimates for the Geminids meteor shower,
comparison that follows.
as shown in the top six rows of Table 2, report a 1.5 km s−1
Several methods have been developed to identify meteor range of nominal median values for vg . Typically, the video ob-
showers and assess orbital similarity. The earliest was the D-
criterion introduced by Southworth & Hawkins (1963), which servations report the slowest speeds (33.5-33.8 km s−1 ; Sono-
calculates orbital similarity based on the orientation (i, ω, Ω), taCo 2009; Jenniskens et al. 2016), linked to previous find-
perihelion (q), and shape (e) of the orbits. Improvements to ings that video observations tend to underestimate Geminids ve-
this approach have been proposed by Drummond (1981), Jopek locities (Hajduková Jr et al. 2017). Photographic observations
(1993), and Jopek et al. (2008). Jenniskens (2008), on the other of the Geminids tend to be slightly faster (34.6 km s−1 ; Jopek
hand, proposed a criterion rooted in dynamical arguments, which et al. 2003); however, the IAU Meteor Data Center (MDC)
evaluates the likelihood of a meteoroid stream’s association with only lists one study that uses 51 older, legacy Harvard photo-
a parent body by considering how orbital evolution due to secu- graphic plate observations (1940s-1950s), along with five Me-
lar perturbations aligns the orbits over time. teorite Observation and Recovery Project (MORP) observations
(1971-1985), and 223 photographic observations by the Dutch
In this study, we applied a distance function, DN , based on
Meteor Society (1972-96). More recent photographic EN ob-
four geocentric quantities directly linked to observational data,
servations of Geminid fireballs report slightly lower speeds
as proposed by Valsecchi et al. (1999). Unlike conventional or-
(∼34.1 km s−1 ;Borovička et al. 2022b; Henych et al. 2024). Fi-
bital similarity criteria, which rely on comparing derived orbital
nally, radar observations of the Geminids have the highest veloc-
elements, the DN method operates in a multidimensional space
ities (34.5-35.0 km s−1 ; Jopek et al. 2003; Brown et al. 2010b).
defined by the independently measured physical quantities. It in-
corporates components of the geocentric velocity at the Earth en- The four tested pipelines provide an estimate of the velocity
counter and introduces two near-invariant variables that account in the lower range of previously published values. The maximum
for the effects of secular perturbations on meteoroid orbits. This velocity difference is 1.4 km s−1 (35.0 km s−1 from Brown et al.
focus on directly observable quantities offers a more robust ap- (2008) and 33.6 km s−1 from DFN - this work). However, the
proach to identifying meteor showers, minimising reliance on lowest nominal velocity of the shower is by SonotaCo (2009),
derived parameters that can be subject to larger uncertainties. with a given velocity of 33.5 km s−1 . The AMOS and WMPL
To ensure reliability, a threshold of DN < 0.1 was chosen, as data reduction pipelines yield very similar median geocentric ve-
Shober & Vaubaillon (2024) demonstrated that this value main- locities (∼33.8 km s−1 ), while the DFN’s median value is slightly
tains a false-positive rate < 2.5% for EN meteor showers. Con- less (33.6 km s−1 ) and FRIPON is more (34.0 km s−1 ). While
sequently, assuming similar statistics, more than 97.5% of the relatively minor, such differences could be crucial for accurate
meteors identified here should be confidently classified as Gem- stream modelling and impact the interpretation of expected ac-
inids. tivity. As seen in Fig. 1, where pre-luminous deceleration correc-
tions are applied to all v∞ estimates, the deviations between the
estimated v∞ values of the four pipelines persist. In other words,
4. Results the velocity deviations are not explained by pre-luminous decel-
eration and likely stem from the triangulation or velocity solver
The Geminid meteor shower, known for its robust and consistent used (Table 1). Assuming an asteroidal composition observed by
activity in mid-December originating from 3200 Phaethon, has a FRIPON-like system, the median estimated pre-luminous de-
been extensively monitored by the FRIPON network from 2016 celeration is only 25 m s−1 (Fig. 2). The corresponding pairwise
to 2023. Here, we have compiled and analysed observations of distributions of ∆vg are provided in Appendix A (Fig. A.3).
584 FRIPON-observed Geminid fireballs. This dataset provides
a substantial basis for evaluating the efficacy and consistency of
four distinct meteor data reduction pipelines (FRIPON, DFN, 4.2. Radiants
AMOS, and WMPL) while also providing a new analysis of
the Geminid meteor shower. Furthermore, the results can also Fig. 3 showcases the calculated radiant distributions, which align
be compared against historically established values for the radi- closely with those recorded in earlier studies, such as those by
ant, velocity, and orbital elements of the Geminids estimated by Jopek et al. (2003) (MORP, Dutch Meteor Society, Harvard) and
five separate studies completed over the last 20 years and docu- Jenniskens et al. (2016). Very slight deviations are observed, par-
mented by the IAU Meteor Data Center9 . ticularly in the right ascension values. The nominal right ascen-
Table 2 summarises the resulting median values from the sion values reported by the four other previous studies (A, C, D,
four pipelines alongside established values from the aforemen- E in Table 2) are notably smaller than any of the distributions
produced here by the four different pipelines, providing strong
9
[Link] evidence that this difference is real for the FRIPON Geminid
Article number, page 6 of 22
 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

Table 2: Summary table for the nominal median measurements of the Geminid meteor shower.

Activity RA σRA Dec σDec vg σvg a q e Peri Node Inc N Ref Obs
annual 112.9 — 32.3 — 34.6 — 1.38 0.141 0.898 324.2 261.6 23.5 51 A photo.
annual 113.8 — 32.3 — 34.6 — 1.40 0.14 0.897 324.4 262.2 23.9 279 B photo.
2002-2006 112.8 — 32.1 — 35.0 — 1.416 0.136 0.904 324.6 261.3 24.0 4384 C radar
2007-2008 112.8 — 32.3 — 33.5 — — — — — — — 2510 D video
2002-2008 112.5 — 32.1 — 34.5 — 1.35 0.1373 0.898 324.95 261.0 23.2 10381 E radar
2010-2013 113.5 — 32.3 — 33.8 — 1.31 0.145 0.889 324.3 261.7 22.9 5103 F video
2016-2023 113.6 0.73 32.3 0.44 34.0 0.41 1.32 0.144 0.891 324.3 262.3 23.2 584 FRIPON video
2016-2023 113.9 0.74 32.3 0.48 33.6 0.60 1.29 0.147 0.886 324.2 262.3 22.6 584 DFN video
2016-2023 113.9 0.75 32.3 0.51 33.8 0.56 1.30 0.145 0.888 324.3 262.3 22.9 584 AMOS video
2016-2023 113.9 0.75 32.3 0.42 33.8 0.47 1.30 0.145 0.888 324.3 262.3 22.8 584 WMPL video

Notes. The table compares values from previous studies and the 2016-2023 FRIPON observations processed through four separate data reduction
pipelines. Radiant and orbital information is provided using the same nomenclature as the IAU-MDC, and units are given in au, degrees, and
km s−1 . The letter next to the reference indicates the radiant in Fig. 3. The last four rows are based on the same 584 Geminid fireballs observed by
the FRIPON network, where the σ values represent the estimated standard deviation for the distribution of Geminids, rather than an uncertainty
of an individual measurement.
A D
(Jopek et al. 2003) (Harvard meteors) (SonotaCo 2009)
B E
(Jopek et al. 2003) (MORP, Dutch Meteor Society, Harvard) (Brown et al. 2010b)
C F
(Brown et al. 2008) (Jenniskens et al. 2016)

Fig. 2: Pre-luminous deceleration estimated for the 584 Geminid

FRIPON fireballs using the model of Vida et al. (2018), showing
that >98% experienced less than 100 m/s of deceleration before
detection. The dashed vertical line indicates the median value.

dataset. As illustrated in Fig. 4, there is also a discernible varia-

tion in the computed radiants when comparing the results from
different pipelines relative to those calculated by the FRIPON
automated system. The DFN, AMOS, and WMPL radiants are
all almost identical, displaying only slightly lower declinations
(-0.1◦ ) but much larger right ascensions (+0.3◦ ) relative to the
radiants calculated by FRIPON’s pipeline. Full pairwise com-
parisons for ∆RA and ∆Dec across all pipelines are shown in
Appendix A (Figs. A.1 and A.2).
To comprehensively analyse the distribution of radiants, we
also quantified both the spread and skew of the geocentric ra-
diant measurements. The spread was characterised using Prin-
cipal Component Analysis (PCA) of the two-dimensional (RA,
Dec) distribution, allowing us to identify the semi-major and
semi-minor axes of the distribution ellipse. These axes corre-
spond to the square root of the eigenvalues of the covariance ma-
Fig. 1: Distribution of v∞ corrected for pre-luminous decelera- trix, scaled to represent a 3 σ dispersion along the directions of
tion using the model of Vida et al. (2018). The dashed vertical greatest and least variance. Skewness was computed separately
line and corresponding value correspond to the median v∞ . for RA and Dec to quantify any asymmetry in the distribution
Article number, page 7 of 22
 A&A proofs: manuscript no. aa54364-25

Fig. 3: Calculated geocentric radiant distribution for the Geminid meteor showers based on FRIPON observations between 2016-
2023. The six black points with crosses indicate radiants identified from previous studies of the Geminids (A: Jopek et al. 2003, B:
Brown et al. 2008, C: SonotaCo 2009, D: Brown et al. 2010b, E: Jenniskens et al. 2016). Nominal values for the four pipelines and
previous results can also be found in Table 2.

around its mean. Positive skew indicates a longer tail to the right mean radiant for each pipeline. These offsets are 0.45◦ , 0.47◦ ,
(higher values), while negative skew indicates a longer tail to 0.45◦ , and 0.48◦ for FRIPON, DFN, WMPL, and AMOS, re-
the left (lower values). Symmetric distributions exhibit skewness spectively. For reference, Moorhead et al. (2021) found a me-
near zero. dian offset of 0.38◦ ) for 1279 Geminids observed by GMN and
All four pipelines produce similar radiant distributions and Kresák & Porubcan (1970) found a 0.49◦ ) value based on 82
ranges, with standard deviations for RA all concentrating around observations.
0.74◦ and Dec ranging from 0.43◦ to 0.51◦ . The 3 σ PCA spread All four pipelines deliver similar radiant precision; how-
is found to be generally around (2.2◦ × 1.3◦ ). Regarding skew- ever, the DFN pipeline’s 95% confidence region is the small-
ness, all pipelines show similar moderate negative skew in RA est (Table 3). The median σRA and σDec cluster tightly around
and Dec, indicating a slight asymmetry with a bias toward lower 0.02–0.11 ◦ . Fig. 5 compares 1-σ radiant uncertainties against
RA and Dec values. The RA skew values range between −1.0 to the cross-track RMS. DFN is the only pipeline whose uncertain-
−0.8, and Dec skew with a larger diversity from −3.03 to −0.20. ties scale strongly with the RMS (Spearman ρ ≈0.7), meanwhile,
Dec skewness is more pronounced for AMOS (−3.0), primarily the other pipelines show only weak correlations (ρ <0.2). This
due to a small number of outliers. occurs because, for this FRIPON comparison only, we supplied
Additionally, in accordance with several previous works, we the DFN pipeline with a single, fixed measurement noise value
have also calculated the median absolute angular offset from the for every pick; when the residuals exceed this assumption, the
Article number, page 8 of 22
 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

Fig. 4: Variation in the FRIPON-observed Geminid fireball radi-

ants calculated by the DFN, AMOS, and WMPL data reduction
pipelines relative to those calculated by the FRIPON automated
pipeline.

Table 3: One–sigma single radiant measurement uncertainties

for the FRIPON Geminids sample.

σRA (deg) σDec (deg)

Pipeline
Median 95 % range Median 95 % range
FRIPON 0.117 0.030–0.372 0.117 0.030–0.372
WMPL 0.067 0.015–0.462 0.053 0.011–0.461
DFN 0.030 0.013–0.083 0.021 0.011–0.064 Fig. 5: Median 1σ radiant uncertainties in (top) right ascen-
AMOS 0.073 0.024–0.763 0.062 0.023–0.550 sion and (bottom) declination as a function of the trajectory-fit
Notes. The “95 % range” column encloses the central 95 % of the distri- cross-track RMS residuals. Values are binned logarithmically in
bution (low–high). This range is distinct from the larger range of nomi- RMS; the solid line in each colour-coded curve is the median
nal radiant values (as seen in Fig. 3). of all events in the bin, while the shaded region encloses the 5 –
95 % percentile range (central 90 %) and illustrates the event-
to-event spread. FRIPON, WMPL, and AMOS pipelines (nearly
EKF inflates the model covariance, resulting in larger radiant σ flat curves) report uncertainties that are largely independent of
values. In contrast, WMPL, AMOS, and FRIPON weight astro- fit quality. In contrast, the DFN solution shows a clear positive
metric noise on a per-pick basis, so their σ-to-RMS correlations trend, indicating that its formal errors scale with the residuals.
remain weak (ρ <0.2). Nevertheless, the vast majority of individ-
ual uncertainties are well below the observed radiant dispersion
of the Geminid stream (Fig. 3), thus confirming the capability to tial for reliable orbital characterisation, this inconsistency af-
discern substructure from FRIPON observations. fects subsequent interpretations of meteoroid stream dynamics.
Therefore, discussing uncertainty quantification across pipelines
is crucial for improving the robustness and comparability of me-
4.3. Velocity Uncertainties teor observations and derived orbital parameters.
A systematic difference emerges in the reported velocity un-
certainties across the networks. This variation in uncertainties 5. Discussion
is clearly illustrated in Fig. 6, where one-sigma velocity un-
certainties are plotted against the spatial RMS of the cross- Our comparative analysis across the four pipelines – FRIPON,
track residuals of the trajectories fitted by each pipeline. Al- DFN, AMOS, and WMPL – indicates that while the nominal
though all pipelines exhibit comparable RMS values, indicat- radiant and velocity estimates derived from DFN, AMOS, and
ing similar goodness-of-fit in trajectory modelling, the derived WMPL are largely consistent, the FRIPON pipeline shows sub-
velocity uncertainties differ markedly. In particular, the WMPL tle but notable systematic deviations. Specifically, the FRIPON
pipeline consistently reports the lowest uncertainties, followed radiant exhibits a right ascension approximately 0.3◦ lower than
by AMOS, then FRIPON, with the DFN pipeline exhibiting the the median 113.9◦ reported by DFN, AMOS, and WMPL, ac-
highest uncertainties despite similar, nearly identical RMS dis- companied by a marginally larger declination distribution. Addi-
tributions. tionally, FRIPON tends to produce geocentric velocities that are
These findings underscore a critical challenge in meteor tra- 300–500 m s−1 higher than the medians provided by the other
jectory analysis: while multiple pipelines can achieve similar pipelines. Although the spread and skew of the velocity distri-
trajectory fits, their approaches to estimating velocity uncertain- butions among DFN, AMOS, and WMPL are similar, the ve-
ties diverge significantly. Since accurate uncertainties are essen- locity uncertainty estimates vary systematically across all net-
Article number, page 9 of 22
 A&A proofs: manuscript no. aa54364-25

works, with DFN reporting the largest uncertainties and WMPL

the smallest. These small systematic offsets propagate into the
orbital elements; the pairwise ∆a, ∆e, ∆i, ∆ω, and ∆Ω distribu-
tions are summarised in Appendix A (Figs. A.4–A.8).

5.1. Outliers

As shown in Table 1, the FRIPON network employs a fully au-

tomated data reduction pipeline designed to minimise meteorite
recovery times. Candidate events are detected, matched across
stations, and the trajectory is solved without human intervention
(Jeanne et al. 2019; Colas et al. 2020; Anghel et al. 2023). The
survey is optimised for bright, multi-station fireballs (peak mag-
nitude Mp < 0), thus, each event is typically recorded by sev-
eral cameras. This redundancy, together with built-in pixel-level
vetoes for slow artefacts (e.g., the Moon, street lamps, aircraft,
etc.), means that most spurious picks are either never linked into
an event or are down-weighted during the least-squares fit of
Eq. (1). In practice, only a few percent of frames have to be
rejected, and the residual false detections have a negligible in-
fluence on the final bright-flight line.
In contrast, the DFN does not automate the fireball pick-
ing step, leading to precise trajectories and no outliers. Further
manual refinement is also done for potential meteorite-dropping
events. Thus, the DFN pipeline does not include outlier detec-
tion by design. As a result, when initially fed FRIPON data
containing false detections and large residuals, the DFN solver
often fails, since it is designed to raise errors when it encoun-
ters large residuals during trajectory determination. Similarly,
the AMOS pipeline encountered difficulties with FRIPON data.
Many FRIPON observations required manual cleaning before
AMOS could process them. When bad data were not removed,
the resulting trajectories and orbital solutions deviated signif-
icantly, underscoring the importance of effective outlier rejec-
tion. These issues highlight the need for rigorous preprocessing
of fireball observation files as well as the tighter radiant and ve-
locity distribution output by the FRIPON pipeline.
Fig. 6: Variation in the RMS of the cross-track spatial residu-
5.2. Radiant and Reference Frames
als (m) for the trajectory fitting versus the velocity uncertainty
estimated by each pipeline. Vertical dashed lines indicated the
FRIPON’s astrometric pipeline is anchored to an inertial celes- median value. All pipelines have relatively equal RMS ranges;
tial frame. It uses the ICRF2/J2000 celestial reference frame for however, the velocity uncertainties differ systematically.
star calibration (Colas et al. 2020). Stars from the Hipparcos
catalogue (an ICRF/J2000 reference) are matched in each cam-
era’s field to determine orientation. In practice, FRIPON fits the ordinates – leading to a systematic RA offset if not corrected for
camera model by converting star coordinates from the celestial precession.
frame to the Earth-fixed frame (ITRF) at the time of observation
(Jeanne et al. 2019). This implies that FRIPON does account for 5.3. Velocity and uncertainties
Earth’s rotation and precession when mapping star positions –
essentially using star coordinates precessed to the observation The FRIPON network’s tendency to report higher meteoroid ve-
epoch during calibration. The intent is to produce meteor direc- locities likely results from its trajectory-fitting approach and de-
tions in a J2000-equatorial frame for consistency (Colas et al. celeration model (Fig. 7). FRIPON employs a dynamic model of
2020). However, if the final radiant coordinates are not explic- deceleration and mass loss, as described in Jeanne et al. (2019)
itly transformed back to the J2000 epoch, a slight epoch mis- (Eq. 2), which describes the deceleration and ablation of a mete-
match could remain. A ∼ 20-year difference between the obser- oroid in the atmosphere. In practice, the velocity solver overes-
vation date and J2000 can shift right ascension on the order of timates Ve when the observed track shows insufficient resolved
0.3° due to precession. In other words, if FRIPON assumes its deceleration to constrain the deceleration model.
results are “in J2000” without actually rotating from the date-of- Jeanne (2020) (c.f. pages 111-119 within) showed that in the
observation equinox, the RA would appear about 0.2◦ –0.3◦ off weak-deceleration regime, the separation of Ve from A becomes
(drifting 50.3 arcseconds per year), exactly as observed. Thus, ill-conditioned, biasing slightly high Ve and inflating its formal
FRIPON’s radiant output is effectively in the epoch of the event uncertainty. In other words, many (Ve , A) pairs fit the luminous
(or a recent epoch), whereas the other networks report J2000 co- segment equally well. Also, if an object ablates away before any
Article number, page 10 of 22
 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

Fig. 8: FRIPON-derived entry speed ventry versus the FRIPON

Fig. 7: Comparison of the velocity and velocity uncertainty deceleration parameter A (Jeanne et al. 2019; Colas et al. 2020).
of Geminid meteor shower FRIPON events processed by the The solid line is an ordinary-least-squares fit; the shaded band
DFN, AMOS, WMPL pipelines versus the automatic reduction is the 95% confidence interval. A clear positive trend is evident:
pipeline of FRIPON. The x-axis is the difference between the meteoroids with larger A, i.e., events that display little measur-
σvg values processed by the DFN (a), AMOS (b), or WMPL (c) able deceleration, yield systematically higher fitted velocities,
and those calculated by FRIPON’s pipeline. The y-axis similarly indicating that the default FRIPON single-body solver tends to
shows the differences for the nominal vg values. over-estimate the true pre-atmospheric speed when deceleration
is weak.

deceleration occurs, this can also result in an artificially high A

value. So in practice, for short, fast meteors that show minimal tion, and initial velocity (Vida et al. 2018). However, this de-
deceleration, there’s a fundamental ambiguity in the fit that can celeration is not influential within this particular observational
overestimate velocities and can be identified as a positive corre- dataset, as over 98% of the meteoroids experienced less than
lation between parameters A and Ve (See Jeanne (2020) pg. 117). 100 m s−1 of pre-luminous deceleration (Fig. 2), with a median
Fig. 8 shows a clear correlation between these parameters by the value of 25 m s−1 . When this extra initial deceleration is taken
FRIPON pipeline (Spearman ρ = 0.497, p = 1.5 × 10−36 ). into account by the DFN, AMOS, and WMPL (Fig. 1), the v∞
Consequently, velocities for meteorite-dropping fireballs values barely change.
(which decelerate strongly) are expected to be more accurate, Even after pre-luminous deceleration is accounted for, all
whereas the method is not optimal for low-deceleration impacts four data reduction pipelines converge to a lower vg value com-
such as the Geminids examined here. pared to nearly all other reported values by the IAU MDC (top
On the other hand, the lower velocities reported by the DFN of Table 2). Systematic underestimation of Geminid speeds has
pipeline (Fig. 7) are linked to lower-quality observations. As the been reported for several earlier video surveys (Hajduková Jr
number of observational data points goes below 50 and the num- et al. 2017), mainly because those studies (i) centroided faint
ber of cameras is limited to 2, the DFN pipeline tends to produce (mpeak ≳ +2) meteors and (ii) used straight–line fits that
slightly lower velocities relative to WMPL and AMOS. This is did not correct for deceleration. The four pipelines used here
partially true for all the pipelines; however, the DFN was de- solve for the full dynamical state, and we have accounted for
signed for high-precision photographic observations, making it pre-luminous corrections based on the model of Vida et al.
the least equipped to handle noisier observations. This is exem- (2018). The higher photographic speed value tabulated in the
plified by the fact that most of the Geminids analyzed here (321 IAU-MDC derives almost exclusively from a smaller photo-
events), were only observed by two FRIPON cameras. If we graphic set of Jopek et al. (2003), based on 1940–1950s Har-
only considered events observed by 3+ cameras, the vg estima- vard plates, a handful of MORP observations (1971-1985), and
tions start to converge (DFN: 33.7 km s−1 , WMPL: 33.7 km s−1 , 223 DMS observations (1972-96); producing a vg estimate that is
FRIPON: 34.0 km s−1 , AMOS: 33.8 km s−1 ). The number of ob- ∼ 800 m s−1 faster than our results. However, more recent mod-
servations significantly influences the data quality of FRIPON ern photographic fireball observations for equally large Gemi-
events, as the network was not designed with 2-station detec- nids recorded by the European Fireball Network produce a me-
tions in mind, but rather meteorite-dropping events detected by dian vg = 34.1 km s−1 (Borovička et al. 2022b; Henych et al.
multiple stations. 2024). A Kolmogorov–Smirnov test also confirms the difference
Pre-luminous atmospheric deceleration before the first ob- with EN is significant (D = 0.38, p = 4.5 × 10−5 ). Thus, there
servation can also significantly influence the velocity measure- is still a 250-300 m s−1 underestimate likely due to instrumen-
ments of meteoroids before their visible ablation phase. Vida tal effects. Principally, as pointed out in Hajduková Jr et al.
et al. (2018) demonstrated that the velocity measured at the start (2017), the automation of point-picking smeared-out motion-
of the luminous trajectory is systematically lower than the true blurred meteor streaks can systematically influence the velocity
pre-atmospheric velocity. This underestimation is particularly estimates. In particular, this causes the speed to be underesti-
significant for smaller or slower meteoroids, where deceleration mated when taking the centre of the streak, or overestimated if
due to drag and ablation can result in discrepancies ranging from picking the leading edge (Hajduková Jr et al. 2017). The 30-
100 to 750 m s−1 , depending on the meteoroid’s mass, composi- fps frame integration of FRIPON (33 ms) smears a Geminid
Article number, page 11 of 22
 A&A proofs: manuscript no. aa54364-25

by ∼1 km, and centroiding that streak shifts each pick slightly and not systematic. Vida et al. (2020b) demonstrated that sys-
backwards along the path when picking the middle. This should tematic biases are also present due to the limitations of measure-
account for the observed 0.25 km s−1 shift relative to EN. In ments (e.g., estimating the initial speed on a segment of the tra-
contrast, the much larger MDC discrepancy most likely reflects jectory after some deceleration has already occurred); thus, the
legacy calibration issues in the Harvard plates rather than a gen- problem cannot be collapsed to a one-dimensional consideration
uine physical difference. of simply inflating the uncertainties. The same effect was also
More notably, as illustrated in Fig. 6, although all four data recently noticed by Barghini et al. (2025), who defined an esti-
reduction pipelines provide generally similar spatial residual mator R (with R ≃ 1 for realistic uncertainties and R > 1 when
root-mean-square (RMS) values for trajectory fits, there is a clear uncertainties are underestimated) and measured R = 1.19 ± 0.05
systematic trend in their estimated velocity uncertainties. The for FRIPON but R = 3.10 ± 0.02 for GMN and CAMS, indi-
variation in velocity uncertainty outcomes is striking, with the cating that the latter two databases publish velocity/radiant er-
WMPL pipeline yielding the lowest uncertainties, followed by rors that are too small near the parabolic limit. This follows the
AMOS, FRIPON, and DFN, which reports the highest uncer- long-noted link between surplus hyperbolic orbits and underes-
tainty estimates. This discrepancy is not due to poor model fits, timated errors (e.g., Hajdukova et al. 2024), and implies that,
as the RMS values for spatial residuals are comparable across while not the smallest, FRIPON’s uncertainties are the most rep-
all four pipelines at a population scale. Instead, the differences resentative of the actual scatter in these observations. Between
arise from varying methodologies in how each pipeline propa- the four pipelines studied here, the results of Barghini et al.
gates and estimates uncertainties. (2024) do not necessarily imply that the FRIPON pipeline’s un-
A factor contributing to differences in reported velocity un- certainties are the most representative; however, they are less un-
certainties is the treatment of astrometric precision in each derestimated than the WMPL estimates, according to this recent
pipeline. The DFN pipeline is the only method that takes work.
astrometric uncertainties into account and propagates them This highlights a broader issue in the field: a clear disconnect
through to velocity models. These uncertainties are calculated between networks regarding how velocity uncertainties should
for DFN images within DFN image reduction. FRIPON does be assessed. Achieving consensus on uncertainty propagation
not output astrometric uncertainties, though Jeanne et al. (2019) and how various factors should be accounted for, such as system-
notes that FRIPON’s lens calibration routinely achieves ∼ 2 atics, timing, astrometric precision, and deceleration modelling,
arcmin (1σ) residuals on astrometric measurements. As the is essential for the standardised and reliable understanding of the
DFN pipeline explicitly requires them (it requires columns meteoroid populations in near-Earth space.
err_minus_altitude, err_plus_altitude, , err_minus_azimuth,
err_plus_azimuth defined as optional in the astrometric stan-
5.4. Advice for future meteor data creation and
dard of [Link] to run FRIPON
recommendations
data through the DFN pipeline, we had to set default uncer-
tainties for all observation points. We manually added values Based on our work, we have provided a handful of recommen-
of 2 arcmin in azimuth and 4 arcmin in elevation to all obser- dations:
vations. Including astrometric uncertainties has contributed to
the systematically higher velocity uncertainties for the DFN re- 1. Be able to output/input observations in the GFE10 format so
sults (Figure 6). However, a more likely factor is the way the that data can easily be shared, tested, and combined.
EKS method is set up to handle the uncertainties while using a 2. Include per-event quality metadata in GFE (e.g., station in-
single-body model. As fragmentation is not explicitly modelled formation, picks, convergence angle, min range, cross-track
by the DFN pipeline, any abrupt deceleration manifests as extra RMS, long-track RMS, frame rate, exposure, lens model,
process noise and is absorbed into the covariance matrix. Af- clock source, etc.).
terwards, the filter steps down the main-body mass and velocity 3. Share the single-station GFE files via Zenodo or some other
(Sansom et al. 2015), resulting in increased overall uncertainties. file-sharing service.
It is designed to ‘trust’ the observations over the model. When 4. All data processing pipelines should be publicly available on
combined with the noisier data of FRIPON, an artificially in- GitHub, Gitlab, etc., with clear documentation.
flated covariance matrix is triggered, explaining the apparent bi- 5. Each network should clearly document the assumed celes-
modal distribution in Figure (6). The lower σv cloud represents tial reference frame (e.g., “J2000 equatorial coordinates” or
those in ‘single body’ mode, while the higher ones are using the “no-atmosphere topocentric horizontal coordinates”), detail-
‘fragmenting’ covariance matrix. The same two clusters appear ing the reference epoch and refraction correction; all calcu-
within DFN data, confirming that the uncertainty distribution is lations (or final outputs) should be transformed to that frame
an intrinsic consequence of the EKS. Given these two factors, the within arcsecond precision.
larger uncertainties presented here and the large cluster around 6. Confirm that each observatory’s time stamps, particularly
300-500 m s−1 are a result of the low spatial resolution of the data shutter-opening times or frame mid-exposure times, are ac-
and the high sampling rate of FRIPON observations, which the curately and consistently converted to UTC, with verified
EKS is not optimised to handle. While similarly high velocity leap seconds and known latencies/delays.
uncertainties are estimated by the DFN pipeline for DFN obser- 7. Propagate per-pick astrometry and timing errors through the
vations, they typically constitute a minority subset. trajectory and velocity fits, and report per-event 1σ uncer-
The accuracy of uncertainties has also been studied in Vida tainties and the propagation method (especially for auto-
et al. (2020b), where it was found that a factor of 2-4 under- mated point-picking).
estimation occurs in the radiant uncertainty when 1-σ errors 8. Specify the spectral sensitivity of the instrument and docu-
are used during the Monte Carlo procedure. This suggests that ment the bandpass of the star catalog used for photometric
higher measurement errors should be introduced in the Monte calibration and measurements.
Carlo procedure to ensure accurate uncertainty estimation. How-
10
ever, this analysis assumed that all sources of errors are random [Link]

Article number, page 12 of 22

 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

9. Document any idiosyncrasies of the instrument that will help then DFN. These inconsistencies primarily reflect contrasting
understand and reproduce the results. treatments of astrometric weighting, timing precision, and sys-
10. If a group is starting a meteor or fireball network, it is highly tematic effects rather than deficiencies in any single solver. Re-
recommended to use WMPL11 to process your data. It is well liable orbital uncertainties are crucial for modelling meteoroid
written, publicly available, and well-documented. streams and linking meteors to their parent bodies; achieving
a consensus on uncertainty propagation is key for the commu-
nity to address. Future work should focus on standardised ap-
6. How to test your pipeline vs ours? proaches to specifying astrometric uncertainty, modelling pre-
All FRIPON single-station observation files used in this study, luminous deceleration, and automating outlier handling, thereby
along with a summary table of the triangulation results for each improving consistency and interpretability across diverse meteor
data reduction pipeline, are available on Zenodo12 . networks.
For access to pipelines compared in this paper, WMPL As meteor and fireball networks become more numerous due
is open-source software13 . The DFN, FRIPON, and AMOS to the decreasing cost of precise sensors, along with the in-
pipelines are currently being refactored to remove network- creased ubiquity of other observations from seismic, infrasound,
specific dependencies and improve documentation, with the aim radar, weather radar, telescopic, and satellite observations, we
of making them publicly available. In the interim, please contact believe that this study represents the first step toward improv-
the authors. ing collaboration between different networks and data sources.
In this comparison, it is also vital to note that the per–station By combining more sensors and integrating increasingly diverse
astrometry in the shared GFE files was only computed by the data reduction methods, we anticipate that future studies will
FRIPON pipeline. These GFE files, with FRIPON reduced as- enable a more comprehensive understanding of meteoroid pop-
trometry, were then used by the DFN, WMPL, and AMOS. Our ulations and their dynamics. In future work, we plan to con-
cross-pipeline differences, therefore, reflect trajectory/velocity duct similar analyses for meteorite-dropping fireballs and pre-
modelling and uncertainty propagation rather than astromet- impact telescopic detections, such as the 2023 CX1 data, as hav-
ric/timing calibration. ing an external and independent estimate of the orbit would be
extremely beneficial to the work.
Acknowledgements. This project received funding from the European Union’s
7. Conclusions Horizon 2020 research and innovation programme under the Marie Skłodowska-
Curie grant agreement No945298 ParisRegionFP (P.M.S.), and the grant agree-
By reprocessing 584 FRIPON-observed Geminid fireballs with ment No. 101150536 (S.A.) The Global Fireball Observatory and data pipeline
four independent data reduction pipelines (FRIPON, DFN, is enabled by the support of the Australian Research Council (DP230100301,
WMPL, and AMOS), we found that all achieve results simi- LE170100106). FRIPON was initiated by funding from ANR (grant N.13-BS05-
0009-03), carried by the Paris Observatory, Muséum National d’Histoire Na-
lar to previous studies; however, with some differences in ve- turelle, Paris-Saclay University and Institut Pythéas (LAM-CEREGE). VigieCiel
locity and radiant coordinates. The per–station astrometric solu- was part of the 65 Millions d’Observateurs project, carried by the Muséum Na-
tions produced by FRIPON were used by the DFN, WMPL, and tional d’Histoire Naturelle and funded by the French Investissements d’Avenir
AMOS; therefore, the cross-pipeline differences we discuss do program. FRIPON data are hosted and processed at Institut Pythéas SIP (Service
not arise from astrometric processing. FRIPON retrieved veloci- Informatique Pythéas), and a mirror is hosted at LTE (Le Laboratoire Temps
Espace / Paris Observatory). This project was partially funded by the Excellent
ties around 0.2-0.4 km s−1 faster than DFN, WMPL, and AMOS. Grant of Comenius University no. UK/3055/2024 (Flux of primitive and differ-
This overestimation is due to the deceleration-based model entiated material to Earth). This work was also supported by the Slovak Grant
(Bronshtehn 1983) it uses to estimate velocities, which are not Agency for Science (grant VEGA 1/0218/22), and by the Slovak Research and
accurate when deceleration is minimal. The DFN, WMPL, and Development Agency grants APVV-23-0323 and VV-MVP-24-0232. The au-
thors would also like to thank Simon Jeanne for his communication and help in
AMOS each employ more advanced velocity solvers, which pro- interpreting some of the results. Also, we would like to thank the anonymous
duce similar results as long as the observations do not contain reviewer whose thorough analysis and recommendations greatly improved the
significant noise. Otherwise, when only 2 cameras have observed manuscript.
the fireball, or there are fewer than 50 data points, the DFN
tends to estimate slightly lower velocities relative to WMPL and
AMOS. Concerning the radiant, FRIPON solutions are systemat- References
ically shifted by about −0.3◦ in the right ascension relative to the
other pipelines due to an epoch misalignment (no final preces- Anderson, S. L., Towner, M. C., Fairweather, J., et al. 2022, ApJL, 930, L25
Anghel, S., Birlan, M., Nedelcu, D.-A., & Boaca, I. 2019, RoAJ, 29, 191
sion to J2000), and possibly compounded by FRIPON using an Anghel, S., Drolshagen, E., Ott, T., et al. 2021, MNRAS, 508, 5716
older ITRF93 model. This error is currently being remedied, and Anghel, S., Nedelcu, D. A., Birlan, M., & Boaca, I. 2023, MNRAS, 518, 2810
the FRIPON pipeline will, in the future, output correctly shifted Antier, K. 2023, Bolide du 10 septembre, 00H13
J2000 radiants. All pipelines converge on orbital solutions that Arlt, R., Asher, D. J., Brown, P. G., et al. 2022, Handbook for Meteor Observers,
ed. J. Rendtel (International Meteor Organization)
remain consistent with the known Geminid stream parameters, Barghini, D., Carbognani, A., Gardiol, D., et al. 2024, in XIX Congresso
with over 99% of fireballs retaining a DN < 0.1 relative to pub- Nazionale di Scienze Planetarie, Società Italiana di Scienze Planetarie, 0–0
lished Geminid observations. Barghini, D., Ďurišová, S., Koten, P., et al. 2025, A&A, 701, A135
A more pressing discrepancy emerges in the formal veloc- Bessell, M. S. 2000, Publ. Astron. Soc. Pac., 112, 961
Bland, P., Spurný, P., Bevan, A., et al. 2012, Aust. J. Earth Sci., 59, 177
ity uncertainties. Even though each pipeline generally fits the Bland, P. A., Spurnỳ, P., Towner, M. C., et al. 2009, Science, 325, 1525
raw observations to the same degree (i.e., the RMS of the spa- Borovicka, J. 1990, Bull. astr. Inst. Czechosl., 41, 391
tial residuals has a similar magnitude), the reported 1-σ veloc- Boroviĉka, J. & Kalenda, P. 2003, Meteorit. Planet. Sci., 38, 1023
ity uncertainties differ systematically, with WMPL reporting the Borovicka, J., Spurny, P., & Keclikova, J. 1995, A&AS, 112, 173
Borovička, J., Spurný, P., & Koten, P. 2007, A&A, 473, 661
lowest velocity uncertainties, followed by AMOS, FRIPON, and Borovička, J., Spurnỳ, P., & Shrbenỳ, L. 2020, AJ, 160, 42
11 Borovička, J., Spurnỳ, P., & Shrbenỳ, L. 2022a, A&A, 667, A158
[Link] Borovička, J., Spurnỳ, P., Shrbenỳ, L., et al. 2022b, A&A, 667, A157
12
[Link] Borovička, J., Tóth, J., Igaz, A., et al. 2013, Meteorit. Planet. Sci., 48, 1757
13
[Link] Bronshtehn, V. A. 1983, Physics of meteor phenomena.

Article number, page 13 of 22

 A&A proofs: manuscript no. aa54364-25

Brown, P., Weryk, R., Kohut, S., Edwards, W., & Krzeminski, Z. 2010a, WGN, Sansom, E. K., Gritsevich, M., Devillepoix, H. A., et al. 2019a, ApJ, 885, 115
vol. 38, no. 1, p. 25-30, 38, 25 Sansom, E. K., Jansen-Sturgeon, T., Rutten, M. G., et al. 2019b, Icarus, 321, 388
Brown, P., Weryk, R., Wong, D., & Jones, J. 2008, Icarus, 195, 317 Shober, P. M., Devillepoix, H. A., Sansom, E. K., et al. 2022, Meteorit. Planet.
Brown, P., Wong, D., Weryk, R., & Wiegert, P. 2010b, Icarus, 207, 66 Sci., 57, 1146
Brown, P. G., McCausland, P., Hildebrand, A., et al. 2023, Meteorit. Planet. Sci., Shober, P. M., Devillepoix, H. A., Vaubaillon, J., et al. 2025, Nat. Astron., 1
58, 1773 Shober, P. M., Jansen-Sturgeon, T., Sansom, E. K., et al. 2020, AJ, 159, 191
Campbell-Burns, P. & Kacerek, R. 2014, WGN, 42, 139 Shober, P. M., Sansom, E. K., Bland, P. A., et al. 2021, PSJ, 2, 98
Ceplecha, Z. 1961, Bull. astr. Inst. Czechosl., 12, 21 Shober, P. M. & Vaubaillon, J. 2024, A&A, 686, A130
Ceplecha, Z. 1987, Bull. astr. Inst. Czechosl., 38, 222 SonotaCo, A. 2009, WGN, 37, 55
Ceplecha, Z., Borovička, J., Elford, W. G., et al. 1998, Space Sci. Rev., 84, 327 Southworth, R. & Hawkins, G. 1963, Smithsonian Contr. Astrophys., 7, 261
Ceplecha, Z., Spurny, P., Borovicka, J., & Keclikova, J. 1993, A&A, 279, 615 Spurnỳ, P., Borovička, J., & Shrbenỳ, L. 2006, Proc. Int. Astron. Union, 2, 121
Colas, F., Zanda, B., Bouley, S., et al. 2020, A&A, 644, A53 Stulov, V., Mirskii, V., & Vislyi, A. 1995, Moscow: Science. Fizmatlit
Cooke, W. J. & Moser, D. E. 2012, in Proceedings of the International Meteor Suk, T. & Šimberová, S. 2017, Earth Moon Planets, 120, 189
Conference, 30th IMC, Sibiu, Romania, 2011, 9–12 Tóth, J., Kornoš, L., Zigo, P., et al. 2015, Planet. Space Sci., 118, 102
Devillepoix, H., Cupak, M., Bland, P., et al. 2020, Planet. Space Sci., 191, Tóth, J., Šilha, J., Matlovič, P., et al. 2019, in 1st NEO and Debris Detection
105036 Conference_ESA2019, 63
Devillepoix, H. A., Bland, P. A., Sansom, E. K., et al. 2019, MNRAS, 483, 5166 Trigo-Rodríguez, J. M., Llorca, J., Castro-Tirado, A. J., et al. 2006, A&G, 47, 6
Devillepoix, H. A., Sansom, E. K., Bland, P. A., et al. 2018, Meteorit. Planet. Turchak, L. I. & Gritsevich, M. I. 2014, J. Theor. Appl. Mech., 44, 15
Sci., 53, 2212 Valsecchi, G., Jopek, T., & Froeschlé, C. 1999, MNRAS, 304, 743
Devillepoix, H. A. R., Sansom, E. K., Shober, P., et al. 2022, Meteorit. Planet. Vida, D., Brown, P. G., & Campbell-Brown, M. 2018, MNRAS, 479, 4307
Sci., 57, 1328 Vida, D., Brown, P. G., Campbell-Brown, M., & Egal, A. 2024, Icarus, 408,
Drummond, J. D. 1981, Icarus, 45, 545
115842
Duris, F., Kornos, L., & Toth, J. 2018, in Proceedings of the International Meteor
Vida, D., Brown, P. G., Devillepoix, H. A., et al. 2023, Nat. Astron., 7, 318
Conference, 37th IMC, Pezinok-Modra, Slovakia, 2018, ed. R. Rudawska,
Vida, D., Campbell-Brown, M., Brown, P. G., Egal, A., & Mazur, M. J. 2020a,
J. Rendtel, C. Powell, R. Lunsford, C. Verbeeck, & A. Knofel, 127–128
A&A, 635, A153
Dyl, K. A., Benedix, G. K., Bland, P. A., et al. 2016, Meteorit. Planet. Sci., 51,
Vida, D., Gural, P. S., Brown, P. G., Campbell-Brown, M., & Wiegert, P. 2020b,
596
MNRAS, 491, 2688
Egal, A., Gural, P., Vaubaillon, J., Colas, F., & Thuillot, W. 2017, Icarus, 294, 43
Egal, A., Vida, D., Colas, F., et al. 2025, Nat. Astron. Vida, D., Šegon, D., Gural, P. S., et al. 2021, MNRAS, 506, 5046
Gardiol, D., Barghini, D., Buzzoni, A., et al. 2021, MNRAS, 501, 1215 Weryk, R., Campbell-Brown, M., Wiegert, P., et al. 2013, Icarus, 225, 614
Granvik, M., Morbidelli, A., Jedicke, R., et al. 2018, Icarus, 312, 181 Wiśniewski, M., Żoładek,
˛ P., Olech, A., et al. 2017, Planet. Space Sci., 143, 12
Gritsevich, M. 2009, Adv. Space Res., 44, 323 Zanda, B., Égal, A., Steinhausser, A., et al. 2023, in 86th Annual Meeting of the
Gural, P. & Segon, D. 2009, WGN, vol. 37, no. 1, p. 28-32, 37, 28 Meteoritical Society
Gural, P. S. 2011, in Proceedings of the International Meteor Conference, 29th
IMC, Armagh, Northern Ireland, 2010, 28–31
Gural, P. S. 2012, Meteorit. Planet. Sci., 47, 1405
Hajdukova, M., Stober, G., Barghini, D., et al. 2024, A&A, 691, A8
Hajduková Jr, M., Koten, P., Kornoš, L., & Tóth, J. 2017, Planet. Space Sci., 143,
89
Halliday, I., Blackwell, A. T., & Griffin, A. A. 1978, JRASC, 72, 15
Hankey, M., Perlerin, V., & Meisel, D. 2020, Planet. Space Sci., 190, 105005
Henych, T., Borovička, J., Vojáček, V., & Spurnỳ, P. 2024, A&A, 683, A229
Howie, R. M., Paxman, J., Bland, P. A., et al. 2017a, Exp. Astron., 43, 237
Howie, R. M., Paxman, J., Bland, P. A., et al. 2017b, Meteorit. Planet. Sci., 52,
1669
Jansen-Sturgeon, T., Sansom, E. K., Devillepoix, H. A., et al. 2020, AJ, 160, 190
Jeanne, S. 2020, PhD thesis, Université Paris Sciences et Lettres
Jeanne, S., Colas, F., Zanda, B., et al. 2019, A&A, 627, A78
Jenniskens, P. 2008, Icarus, 194, 13
Jenniskens, P., Gural, P., Dynneson, L., et al. 2011, Icarus, 216, 40
Jenniskens, P., Nénon, Q., Albers, J., et al. 2016, Icarus, 266, 331
Jenniskens, P., Utas, J., Yin, Q.-Z., et al. 2019, Meteorit. Planet. Sci., 54, 699
Jopek, T., Valsecchi, G., & Froeschlé, C. 2003, MNRAS, 344, 665
Jopek, T. J. 1993, Icarus, 106, 603
Jopek, T. J., Rudawska, R., & Bartczak, P. 2008, Advances in Meteoroid and
Meteor Science, 73
King, A. J., Daly, L., Rowe, J., et al. 2022, Sci. Adv., 8, eabq3925
Koten, P., Rendtel, J., Shrbený, L., et al. 2019, Meteoroids: Sources of Meteors
on Earth and Beyond, 90
Kresák, L. & Porubcan, V. 1970, Bull. astr. Inst. Czechosl., 21, 153
Lyytinen, E. & Gritsevich, M. 2016, Planet. Space Sci., 120, 35
Matlovič, P., Tóth, J., Rudawska, R., Kornoš, L., & Pisarčíková, A. 2019, A&A,
629, A71
Matlovič, P., Kornoš, L., Kováčová, M., Tóth, J., & Licandro, J. 2020, A&A,
636, A122
Matlovič, P., Pisarčíková, A., Tóth, J., et al. 2022, MNRAS, 513, 3982
McCrosky, R. E., Posen, A., Schwartz, G., & Shao, C.-Y. 1971, J. Geophys. Res.,
76, 4090
McMullan, S., Vida, D., Devillepoix, H. A., et al. 2023, Meteorit. Planet. Sci.
Molau, S. 1999, in Proceedings of the International Meteor Conference, 17th
IMC, Stara Lesna, Slovakia, 1998, 9–16
Molau, S. 2001, in Meteoroids 2001 Conference, Vol. 495, 315–318
Moorhead, A. V., Clements, T., & Vida, D. 2021, MNRAS, 508, 326
Peña-Asensio, E., Trigo-Rodríguez, J. M., Gritsevich, M., & Rimola, A. 2021,
MNRAS, 504, 4829
Rowe, J. 2021, WGN, 49, 211
Sansom, E. K., Bland, P., Paxman, J., & Towner, M. 2015, Meteorit. Planet. Sci.,
50, 1423
Sansom, E. K., Bland, P. A., Towner, M. C., et al. 2020, Meteorit. Planet. Sci.,
55, 2157

Article number, page 14 of 22

 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

Appendix A: Pipeline-comparison matrix plots

Fig. A.1: ∆RA histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 15 of 22

 A&A proofs: manuscript no. aa54364-25

Fig. A.2: ∆Dec histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 16 of 22

 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

Fig. A.3: ∆vg histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 17 of 22

 A&A proofs: manuscript no. aa54364-25

Fig. A.4: ∆a histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 18 of 22

 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

Fig. A.5: ∆e histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 19 of 22

 A&A proofs: manuscript no. aa54364-25

Fig. A.6: ∆i histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 20 of 22

 P.M. Shober et al.: Comparing the data reduction pipelines of FRIPON, DFN, WMPL, and AMOS: Geminids Case Study

Fig. A.7: ∆ω histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 21 of 22

 A&A proofs: manuscript no. aa54364-25

Fig. A.8: ∆Ω histograms (column – row) for all pairwise combinations of FRIPON, WMPL, DFN, and AMOS. The solid vertical
line denotes zero, the dashed vertical line indicates the median, and the grey region encompasses the 95% confidence region. ∆ is
the median value, and the ± indicates half of the 95% region.

Article number, page 22 of 22