DebrisWatch II: Digging Deeper for Geosynchronous Debris Authors: James A Blake, Benjamin F Cooke, Cristina Paragini, William Feline, Christopher A Onken, Don Pollacco, Grant Privett, Toshifumi Yanagisawa, Robert Airey, Ioannis Apergis, Roberto Armellin, Lily Beesley, Paul Chote, Anna-Maria Cutolo, Stuart Eves, Tomoko Fujiwara, Daisuke Kuroda, Isobel S Lockley, Alexander MacManus, James McCormac, Morgan A Mitchell, Tokuhiro Nimura, Kota Nishiyama, Shin-ichiro Okumura, Thomas Schildknecht, Billy Shrive, Seitaro Urakawa, Dimitri Veras, Phineas Whitlock, Christian Wolf Published: 2026 ABSTRACT AbstractThe geosynchronous (GSO) debris environment is continually evolving. Regular monitoring of the region is consequently of great importance, though the trade-off between coverage and sensitivity makes this challenging for the population of optically faint debris, where collecting area becomes a pivotal factor. Surveys conducted with large-aperture telescopes have provided crucial insights into the nature of this largely uncharacterised population. In this paper, we revisit a survey conducted with the 2.54 m Isaac Newton Telescope (INT), presenting an overhaul of the astrometric calibration and object detection stages of the original analysis pipeline. We apply a blind stacking technique to boost target recovery, unearthing 25 tracklets previously missed by single-frame extraction methods, and pushing the sensitivity limit fainter by 1 magnitude. The same algorithm is applied to a contemporaneous dataset, captured with a 36 cm astrograph, enabling performance benchmarking through the attempted recovery of INT detections from commercial-off-the-shelf observations. We achieve sub-arcsecond astrometric accuracy through a combination of improved star trail centroiding and iterative distortion fitting, allowing short arc initial orbit solutions to be obtained. High-cadence light curves extracted for trailing detections indicate that faint fragments are proportionally more variable than bright derelicts, with many exhibiting photometric signatures of rapid tumbling, often straddling the image noise floor. Lastly, we present preliminary findings from a follow-up multi-national observation campaign, utilising telescopes in Australia, Japan and La Palma. As space traffic management concerns begin to extend beyond GSO altitudes, scientifically-driven surveys of high-altitude orbits have an important role to play in characterising the faint debris environment.Keywords: Orbital Debris, Geosynchronous Region, Blind Stacking, Short Arc Orbit Determination, Light Curves Introduction The Space Age is undoubtedly accelerating into a new era, characterised by the privatisation of space flight, increased access to space, and soaring launch rates to low Earth orbit, as large constellations of satellites begin to take shape [1–3]. Amidst this surge of activity close to home, it can be easy to overlook the higher-altitude orbits that have long supported a variety of broadcasting, surveillance, communications, and meteorological services. Nevertheless, the geosynchronous Earth orbit (GSO) region will continue to play a crucial role. Emerging business models are enhancing the affordability and flexibility of GSO operations, while a diversification of platforms and missions is fostering novel applications. An increasing number of multi-orbit solutions are leveraging the fixed-positioning and long-lifetime capabilities of geostationary satellites. Furthermore, the GSO environment has served as an effective testing ground for a variety of circular economy initiatives, most notably Northrop Grumman’s Mission Extension Vehicle demonstrations [5] and China’s tugging of a defunct navigation satellite from the geostationary belt to a supersynchronous graveyard orbit using SJ-21 [6]. Missions involving rendezvous and proximity operations, such as these, require high levels of situational awareness to maintain custody of the spacecraft involved and to prevent accidents from taking place. Debris generation is especially problematic at GSO altitudes. Contrary to low-altitude orbits, where atmospheric drag acts as a natural removal mechanism, no equivalent process exists to clear resident space objects (RSOs) from the GSO region. The rapid growth of the commercial space domain awareness sector has significantly improved the timeliness and persistence of GSO monitoring, enabling the swift detection and characterisation of anomalies and fragmentation events [7]. Most recently, the breakup of Intelsat 33E in October 2024 has added 36 fragments to the publicly available Space-Track catalogue, as of the time of writing [8, 9]. Many tens to hundreds of additional fragments continue to be tracked by responding surveillance networks [see, e.g., 10, 11], but maintaining custody of optically faint debris in high-altitude orbits remains a significant challenge. Furthermore, breakup events can generate many thousands of fragments that are too faint to be detected by ground-based instrumentation, as demonstrated by laboratory experiments [12] and reflected in the space debris environment models that draw from them [13]. An overview of recent GSO anomalies and fragmentation events is provided in Table 1, collectively accounting for large parts of the steep increase in the number of trackable high-altitude RSOs that can be seen in Fig. 1. It is important to note that the rise in object count also reflects improvements in surveillance capabilities that have come about over the past decade, and many of the fragments discovered during this period will have originated from events that took place long before. Table 1Notable (well-reported) anomalies and fragmentation events in high-altitude orbits that have taken place in the past decade. Information sourced from a mixture of academic/conference literature and newsletters, as indicated in the final column, alongside the History of On-orbit Satellite Fragmentations database [17]. Two counts are provided for the number of trackable fragments generated by each event: the number of associated entries that appear in the publicly available Space-Track catalogue [8]; and, in parentheses, the number of fragments claimed to have been initially identified by responding survey campaigns. Note that these inventories are limited by the sensitivities of the instrumentation employed, thus they will severely underestimate the true number of fragments generated across the full range of fragment sizesNameINTLDESEvent dateApo/Perigee [km]Inclination []FragmentsReferenceAMC-92003-024A17-Jun-201735798/357740.01740 (13+)[14]Telkom 11999-042A25-Aug-201735793/357810.01120 (10+)[14]Titan 3C Tra.1969-013B28-Feb-201837257/358866.218 (150+)[15]Atlas 5 Cent.2014-055B30-Aug-201835090/811222.3109 (590+)[15]Atlas 5 Cent.2009-047B24-Mar-201934565/681023.31 (250+)[15]Atlas 5 Cent.2018-079B6-Apr-201935092/852612.2214 (250+)[15]Intelsat 29e2016-004A7-Apr-201935707/356750.092 (10+)[16]Atlas 5 Cent.2018-022B6-Sep-202434949/76229.30 (10+)[9]Intelsat 33e2016-053B19-Oct-202435797/357730.03736 (500+)[11] Notable (well-reported) anomalies and fragmentation events in high-altitude orbits that have taken place in the past decade. Information sourced from a mixture of academic/conference literature and newsletters, as indicated in the final column, alongside the History of On-orbit Satellite Fragmentations database [17]. Two counts are provided for the number of trackable fragments generated by each event: the number of associated entries that appear in the publicly available Space-Track catalogue [8]; and, in parentheses, the number of fragments claimed to have been initially identified by responding survey campaigns. Note that these inventories are limited by the sensitivities of the instrumentation employed, thus they will severely underestimate the true number of fragments generated across the full range of fragment sizes Recent evolution of the GSO region, sourced from the latest edition of the European Space Agency’s Annual Space Environment Report [4]. The main panel shows the number of RSOs penetrating the GSO Protected Region, defined by the ranges h∈35586,35986 km and δ∈-15,15∘, where h is altitude and δ is declination. The RSOs are categorised according to their orbit type: ‘GEO’ (geostationary) with inclination i∈0,25∘, perigee height hp∈35586,35986 km, and apogee height ha∈35586,35986 km; ‘EGO’ (extended geostationary) with semi-major axis a∈37948,46380 km, eccentricity e∈0.00,0.25, and i∈0,25∘; and ‘Other’, encompassing all other types, including GEO transfer and highly eccentric orbits. Inset, GEO launch traffic is provided, with payloads categorised according to their mass m: ‘S’ (small), with 101000 kg Observations of the GSO region are typically carried out using optical telescopes, as their sensitivity drops off more gradually than that of alternative active sensing techniques, such as radar. Early attempts to probe the population of small GSO debris were made by NASA using the Charge-Coupled Device (CCD) Debris Telescope (CDT), a portable 32 cm Schmidt telescope initially based in Maui, Hawaii, and subsequently at Cloudcroft, New Mexico [18]. Despite the instrument’s small format and large pixel scale, the CDT was able to detect many ∼1 m RSOs that were not previously catalogued. These efforts were followed by two long-lasting and highly influential programmes: the NASA-funded Michigan Orbital DEbris Survey Telescope (MODEST) [19]; and ESA’s 1 m telescope campaigns [20]. MODEST is a 0.6 m Schmidt telescope located at the Cerro Tololo Inter-American Observatory in Chile. From 2001 to 2014, MODEST was used to carry out extensive observations of the GSO region, providing coverage of most orbital slots over the continental US. The system achieved a relatively wide field of view (FOV) of 1.3∘×1.3∘ and a limiting magnitude of roughly 17.5 with a broad R filter. Its successor, the 1.3 m Eugene Stansbery-Meter Class Autonomous Telescope, has recently been deployed on Ascension Island [21, 22]. The ESA 1 m telescope is installed at the Optical Ground Station at Teide Observatory in Tenerife, Canary Islands. Since 2001, the system has been used by the Astronomical Institute of the University of Bern (AIUB) to conduct regular surveys of the GSO region and other high-altitude orbital regimes, often supported by instruments at AIUB’s Zimmerwald Observatory in Switzerland [see, e.g., 23, 24]. Compared to MODEST, the ESA 1 m has a smaller FOV of 0.7∘×0.7∘, but its wider collecting area enables the detection of sources fainter than 19th magnitude, corresponding to RSOs 10-20 cm in size, assuming a Lambertian sphere with an albedo of 0.1 [25]. Both the MODEST and ESA 1 m campaigns served to highlight a bimodal brightness distribution, with known (catalogued) satellites and rocket bodies clustered around 12th magnitude and an abundance of uncatalogued RSOs fainter than 15th magnitude. The bright-end drop-off and the dearth of objects between the two peaks are characteristics of the GSO population, while the faint-end drop-off is due to the sensitivity limit of the instrument. It is also important to note that the brightness of RSOs is strongly dependent on the illumination geometry between the Sun, the target, and the observer at the time of observation [see, e.g., 26]. As surveys typically report an average brightness across an observation window (or multiple windows), the characteristics of the brightness distribution will vary according to the observational strategy employed. Surveys attempting to detect faint GSO debris usually track the telescope and detector at the same rate as geostationary satellites. However, target RSOs are moving relative to this rate, as uncontrolled debris will drift under the influence of the natural perturbative forces that dominate in the GSO region [27]. Increasing the exposure time to boost the signal-to-noise ratio (SNR) and push the limiting magnitude fainter is consequently not an option, as longer exposure times will spread the target signal across a trail of pixels, while the background noise accumulates proportionally to the trail area. It proves necessary, therefore, to appeal to larger-aperture telescopes for their superior collecting areas. Large-aperture nodes of the International Scientific Optical Network (ISON), a non-government project led by the Keldysh Institute of Applied Mathematics, Russia, have been used to detect and continuously track faint debris in high-altitude orbits [28], most notably the narrow-field 1.25 m and 2.6 m telescopes at the Crimean Astrophysical Observatory in Nauchny, with limiting magnitudes of 19 and 20, respectively. Working in collaboration with AIUB, ISON telescopes conducted follow-up observations of high-altitude RSOs with high area-to-mass ratios, extending coverage and enabling a catalogue of these objects to be built and maintained [29]. The ISON project continues to conduct research into a variety of orbital regimes with a geographically diverse network of over 50 optical telescopes [30]. Bolden et al. [31] carried out observations of faint GSO debris using the 1.8 m Pan-STARRS PS1 telescope on Haleakala, Maui, achieving a survey depth of V∼21. Interestingly, the Pan-STARRS survey observed a sharp increase in the number of very faint RSOs (fainter than V∼19.5) before the instrumental drop-off in sensitivity. This is unsurprising, as the instrument is well-situated to probe the surroundings of the geopotential well at 105∘ W, where many small RSOs originating from fragmentation events will be oscillating about the well centre if captured, and such events will typically generate many more small fragments than large ones [32]. That said, the authors exercise caution, highlighting partial streak detection as a possible source of brightness underestimation. The 6.5 m Magellan Telescope, ‘Walter Baade’, at the Las Campanas Observatory, Chile, has been used to conduct a number of ‘spot’ surveys targeting known fragmentation events [33]. Despite its relatively narrow 0.2 square degree FOV, the telescope detected a significant number of RSOs with brightness 1819 is slowly varying in brightness over timescales longer than the INT observation window. The outlier was still in view of the RASA during a subsequent pointing, where its brightness had increased sufficiently to be detected by the blind stacking pipeline. We move on to explore the guided stacking of RASA frames in panel (g) of Fig. 11, leveraging the extracted motions for the INT detections to predict their positions within the RASA frames and inform a median stack. Inherent uncertainty stems from the INT trail centroiding, especially for cases where the RSO brightness straddles the noise floor (see Section 6). That said, the guided nature of the stacking meant that extraction thresholds and false positive vetting could be eased relative to the blind stacking pipeline. We find that the guided approach is capable of recovering INT detections with V>18, highlighting the potential of the blind stacking technique if rate constraints were not in place. Short Arc Orbit Determination The observational strategy detailed in Section 2 was designed to mitigate the narrow FOV of the INT, promote sky coverage, and boost sensitivity for the survey. In doing so, however, it placed severe limitations on the time available for observing any individual RSO, resulting in very short orbital arcs. A preliminary initial orbit determination (IOD) assessment was conducted using a least-squares minimisation of normalised right ascension and declination residuals. A common initial guess was first adopted, with a semi-major axis equal to 42,000 km, zero eccentricity, and zero inclination. For cases that failed to converge, an alternative initial guess obtained from a surrogate optimisation of the residuals was used. The analysis considered observational arcs ranging from roughly 45 s (two exposures) to 220 s (seven exposures). Only arcs spanning multiple exposures were retained to ensure basic observability. Convergence was assessed using a combination of a root mean square error threshold and visual inspection. The IOD solutions are summarised in the top panels of Fig. 12. We see that the inclinations i and right ascensions of the ascending node (RAANs) Ω follow the expected evolutionary trend induced by the perturbative forces acting on RSOs in the GSO region [27, 61]. The Earth’s oblateness and its gravitational interactions with the Moon and the Sun cause the orbital planes to precess with a 53-year cycle from 0∘-15∘-0∘ about the stable Laplacian plane. The majority of bright detections are found along the zero inclination line, associated with operational geostationary satellites that are N-S station-kept to counter the natural increases in inclination. On the other hand, the uncontrolled debris have progressed through part (possibly all) of the 53-year cycle, gradually evolving from their assumed origin near the geostationary belt with low inclination and Ω∼100∘, then decreasing in RAAN and increasing in inclination to the maximum i∼15∘ at the halfway point of the cycle. Many of the faint detections have surpassed this stage, now decreasing in both RAAN and inclination, approaching i∼0∘ once more, where the oldest fragments of debris will have already restarted the cycle. There are tentative signs of clusters in Ω-i space, typically indicative of groups of fragments with a common explosive origin, though the numbers are too small here to draw any statistically significant conclusions. We note that, for many of the apparent clusters, the constituent members vary widely in orbital eccentricity, perhaps due to different area-to-mass ratios governing their evolution under solar radiation pressure. Initial orbit determination results for the INT dataset. (Top left) Right ascension of the ascending node Ω and inclination i for the solved arcs. Marker colours indicate the median brightness of the target, as mapped by the colour bar above. Markers are scaled according to the root mean square of the normalised residuals in right ascension–declination space. For clarity, grey lines highlight Ω=0∘ and equatorial orbits, i=0∘. (Top right) Semi-major axis a and eccentricity e for the solved arcs. Marker colours indicate the rate of the target during the observation window, as mapped by the colour bar above. Markers are scaled according to arc length. For clarity, grey lines highlight circular orbits, e=0, and the geostationary semi-major axis, a=42164 km. (Bottom) Histograms of median brightness measurements (left) and temporal lengths (right) for the sample of orbital arcs extracted from the INT survey frames (grey), showing the number of arcs with initial orbit solutions (blue). Alongside, we plot the absolute rates in hour angle and declination for the sample (grey), with solved arcs highlighted (blue) Some degree of spread is to be expected; however, detections straying significantly from the evolutionary path are likely in eccentric orbits. The IOD solutions span a wide range of semi-major axis and eccentricity values, with arc length playing a key role in solution quality. In particular, solutions yielding semi-major axes larger than 50,000 km are likely to be spurious. For short observation windows, RSOs in eccentric orbits with their apogees near the GSO region, such as geostationary transfer orbits, can resemble objects in higher circular orbits, thus misleading the solver with degenerate solutions. These cases are also consistently associated with very short arcs, typically shorter than ∼2 minutes, highlighting the intrinsic ambiguity of orbit determination from limited angular data. The results show a strong dependence of convergence on arc length, with the majority of solved arcs in excess of 200 s, as illustrated in the bottom panels of Fig. 12. Using the nominal initial guess, the overall convergence rate is approximately 18%, increasing to 38% when the initial condition is instead taken from the surrogate optimisation, with failures in both cases dominated by very short arcs. We see that the majority of bright tracklets solve; an unsurprising result, given that the bright, point-like nature of geostationary satellites makes them far less likely to drop out due to blending, streaking, or sensitivity. Interestingly, we observe a slight upward trend in convergence towards the faintest detections, likely due to the fact that the average angular rate of detections decreases as we approach the sensitivity limit of the instrument (a systematic bias stemming from the surface brightness considerations discussed in Section 4), thus reducing the chance of blending and increasing the likelihood that the targets remain within the FOV for the duration of the seven-exposure pointing. Overall, while further work using improved data processing (e.g., outlier rejection) and more robust approaches [see, e.g., 62] is needed for a comprehensive analysis, this assessment demonstrates the feasibility of IOD and provides a promising basis for future refinement and extension. Brightness Variability An important finding from the original survey analyses [34] was that many of the faint, uncatalogued RSOs detected by the INT showed significant brightness variability throughout the observation window. In the following section, we explore this variability more quantitatively. The reflected light from an orbiting body carries signatures of its shape and attitude, convolved with sensor characteristics, atmospheric effects, and viewing geometry. Isolating these influences is challenging, and light curve characterisation for space domain awareness remains an active area of research [see, e.g., 63]. Numerous studies have made use of photometric light curves to characterise [64–66] and classify [67–69] objects in near-Earth orbit. The trailing nature of near-geostationary detections in the INT frames enabled the extraction of high-cadence light curves via the placement of apertures along the trails. Discrete pixel-wide rectangular apertures were used to avoid correlated noise injection (see [34] for further details). The light curve extraction method resulted in 10 s segments (spanning the exposure time) of closely sampled brightness measurements, separated by 25 s gaps due to the readout of the CCD mosaic, as illustrated in Fig. 13. To enable a shape-based comparison between light curves that differ widely in mean brightness and variability amplitude, we rescale each brightness measurement by subtracting the mid-range and dividing by the maximum absolute deviation from the mid-range, yielding a dimensionless series Δt∈-1,1. Specifically, we plot a normalised mid-range deviation, Δt=ft-M/Δmax, where ft is the time-dependent, background-subtracted flux extracted via the aperture photometry, M=fmax+fmin/2 is the flux mid-range, and Δmax=fmax-fmin/2 is the maximum deviation from the flux mid-range, where fmax and fmin are the maximum and minimum flux, respectively. Through this interpretation, we can obtain a clearer picture of the variety of variability structures present within the overall dataset while losing information pertaining to the relative amplitudes of variability. In Fig. 13, we see evidence of brightness variability on timescales shorter (see, e.g., q) and longer (see, e.g., m) than the exposure time, and indeed longer than the observing window entirely (see, e.g., t), with the nature of the variability ranging from smooth periodic oscillations (see, e.g., f) to sharp sporadic glints (see, e.g., u), often straddling the noise floor of the survey frames. Examples of temporal structure exhibited by light curves extracted from the INT survey frames. The normalised mid-range deviation, Δ (see main text), is plotted as a function of time relative to the start of the exposure sequence. Dark colours indicate regions of the light curves that are faint relative to the mid-range, while light colours highlight bright regions. Grey regions are either inter-exposure gaps due to readout, or intra-exposure gaps arising from blending or bad pixels. Labels specify the NORAD ID for light curves that correlate with a known RSO in the Space-Track catalogue, or ‘UCT’ for uncorrelated detections Temporal resolution within each light curve segment is governed by the target’s angular motion across the CCD chip. A target moving through the INT field at 1′′s-1 will have traversed roughly 15 px during the 10 s exposure, resulting in a 0.66 s cadence for the corresponding segment. On the other hand, an angular motion of 5′′s-1 would enable a much higher cadence of 0.13 s. This is demonstrated nicely in Fig. 14, where we compare light curves extracted from the INT and RASA frames for RSOs observed contemporaneously. In general, we find excellent agreement between the extracted INT and RASA photometry after accounting for a systematic offset of ∼0.5 magnitudes, attributed to the different bandpasses of the two instruments. Although the RASA was scheduled to survey the same fields as the INT, the exposure sequences during each pointing were not synchronised. Consequently, the RASA’s superior duty cycle served to fill the large readout-induced gaps in the INT light curves, enabling targets to be monitored over a larger fraction of the observing window. That said, the light curves also serve to highlight some limitations of the RASA observations. The temporal resolution is lower for the COTS instrument, owing to its larger pixel scale. In the top panels of Fig. 14, we see that the RASA’s lower temporal resolution acts to suppress the amplitude of variability. Furthermore, for the fainter detection in the bottom panel of Fig. 14, the brightness measurements extracted from the RASA frames show a high level of scatter, reflecting the lower SNR of sources in the RASA frames, compared to equivalent sources observed by the INT. In the following analyses, we consider solely the trailed detections from the INT dataset. For point-like geostationary targets, it was only possible to extract a single photometric measurement per exposure, yielding much lower-cadence light curves with comparatively limited information over the observing window. Examples of contemporaneous light curves extracted from the INT (black) and RASA (cyan) survey frames. We apply a 0.5 magnitude correction to align the two sets of measurements, a systematic offset arising from the different bandpasses of the two instruments. The light curves shown in the top and middle panels are associated with SL-12 rocket bodies NORAD 16797 and NORAD 15581, respectively, as labelled. The bottom panel shows the light curve extracted for a tracklet that failed to correlate with the publicly available Space-Track catalogue (UCT). Adapted from [38] With high-cadence measurements, it is possible to resolve finer details in the target’s light curve, such as sharp glints and/or rapid periodicity. A lower sampling rate acts to suppress this information, smoothing over sharp features and painting a less distinct picture of the underlying variability. However, an inherent trade-off arises from the light curve extraction method between cadence and signal-to-noise, as reflected light from faster-moving targets will spread across more pixels, reducing surface brightness. To boost signal-to-noise for such cases, we bin the light curves to sample at a maximum cadence of 0.25 s. Care is taken to bin the light curves in linear flux space, before converting to calibrated magnitudes with the updated photometry from Section 3. In Fig. 15, we explore the amplitude of variability for the resampled INT light curves. The left-hand panel displays a ridgeline plot that shows the spread of calibrated brightness measurements for each light curve, sorted by mean brightness. We see that the faint end of the sampled population exhibits significant brightness variability, with some examples spanning several magnitudes. It is also interesting to note that the very faint end (V>20) is dominated by slow-moving targets, as expected from the surface brightness considerations discussed above. Variability exhibited by the photometric light curves extracted for trailed detections in the INT survey frames. (Left) A ridgeline plot showing the spread of calibrated brightness measurements for the overall sample. Each row corresponds to a separate light curve, and the rows are sorted by mean calibrated magnitude. Detections satisfying the ‘GEO’ rate conditions are black, while other detections are shown in orange. (Bottom right) For each light curve, we plot the difference in brightness ΔV (delta magnitudes) between each brightness measurement and the mean brightness across the observation window, with maximal uncertainty subtracted. The differences are plotted as a function of the mean brightness for each case, such that each spike represents a separate light curve. For reference, the photometric noise floor is provided in the middle right panel. (Top right) The light curves are categorised by the amplitude of their brightness variability, as presented in the bottom right panel. The number of light curves comprising each 0.5 magnitude bin is labelled above the corresponding bar Of course, the level of photometric uncertainty increases as we move to fainter signals, and is particularly impactful for targets straddling the noise floor of the observations. To account for this, we plot the absolute deviation from the mean brightness for the measurements comprising each light curve, with maximal uncertainty subtracted in linear flux space, in the bottom-right panel of Fig. 15. This metric gives a sense of the minimum level of variability that cannot be attributed to photometric uncertainty. Despite the uncertainty suppression, we find that the faint end remains proportionally more variable than the bright end of the distribution. Indeed, the normalised histogram in the top-right panel of Fig. 15 shows that the highest levels of variability are exhibited by detections with a mean brightness in the range 18