Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact

arXiv:2611.10301 · astro-ph.IM · Submitted 2026-11-06 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics. The day in astrophysics: observations, models and what the data forces.

Vera: We're looking at the paper: "Satellite Streaks in HETDEX".

Jocelyn: Here's the gist of the paper. We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX), covering 6778 observations and ∼90 deg2 of sky from January 2017 to July 2024.

Vera: Contamination Statistics We detect 527 streaks across 492 observations, a contamination fraction of 7.3% that rises to monthly values above 20% by 2023–2024, although the streaks mask only 0.09% of the survey area.

Jocelyn: Of the 527 streaks, 475 (90.1%) are identified with a cataloged object using archival two-line elements: 258 (54%) lie in low or medium Earth orbit and 217 (46%) are geostationary or highly elliptical.

Vera: This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline.

Title and authors: Vera: So we're looking at this paper now titled "Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact". It’s basically taking a look at the data from the Hobby-Eberly Telescope Dark Energy Experiment. We’re seeing how much of that deep survey area is being contaminated by things that are actually satellites.

Jocelyn: Right, Vera. And it covers six thousand seven hundred seventy-eight observations spanning from January two thousand seventeen to July two thousand twenty-four covering about ninety degrees squared of sky <ref:2611.10301#pg1>. It’s a pretty big data set to analyze for streaks.

Subrahmanyan: I’m curious about the main takeaway here for cosmology. Is this just a nuisance we need to filter out, or does it hint at something bigger in how we interpret structure formation at redshift two point four?

Vera: Well, the paper shows that they detect five hundred twenty-seven streaks across those observations, which is a contamination fraction of about seven point three percent overall. But as the survey goes on, that contamination fraction actually gets higher, reaching values above twenty percent monthly by two thousand twenty-three and even July two thousand twenty-three <ref:2611.10301#pg1>.

Jocelyn: That rising trend sounds significant for time-domain astronomy, which is my area too. So what's the breakdown of those streaks? Are they all the same kind of object?

Subrahmanyan: The paper identifies that most of these streaks, about ninety point one percent, are matched to a cataloged object using archival two line elements. And when you look at that cataloged list, a big chunk—two hundred fifty-four objects—are payloads from space.

Vera: That’s interesting because it shows the sensitivity of HETDEX to faint objects and its long observing baseline from two thousand seventeen to two thousand twenty-four as the authors noted <ref:2611.10301#pg1>. The way they matched those streaks using archival data gives them a good constraint on the orbits.

Jocelyn: And when they look at the orbital identification, they find that for those matched streaks, ninety-nine point two percent are unambiguous in their orbit determination. That’s pretty solid work there, right?

Subrahmanyan: Solid work is good for checking our assumptions about how we model the universe. But I wonder if this is just a cataloging exercise or if it tells us anything about the density of objects in those specific orbital regimes over time.

Vera: The paper does look at that, Subrahmanyan. They found that the orbital identifications consist of two hundred fifty-four payloads, one hundred seventy-two rocket bodies, and forty-nine debris objects. That reflects what’s been launched across six decades of activity.

Title and authors: Jocelyn: It also gives some photometric data on these streaks themselves. They measured the SDSS g magnitudes ranging from eleven point five to eighteen point eight, with a median magnitude around sixteen point four. The observed lengths of these streaks are between zero point six arcminutes and seventeen point two arcminutes.

Subrahmanyan: Those magnitudes tell us something about the intrinsic brightness of these objects, or at least the apparent brightness they have when we observe them from Earth at those distances. How does that compare to what we expect from LEO versus geostationary orbits?

Vera: The paper gives some specifics on that. They found a median gAB value of five hundred kilometers is four point five five magnitudes, which varies from three point eight six for geostationary objects to nearly six point one two for low Earth orbit.

Jocelyn: That difference between the orbits is clear in the data. The instantaneous magnitudes they measured are even more spread out, spanning from three point nine to fourteen point six with a median of ten. That helps them separate the classes cleanly on a visual level, I guess.

Subrahmanyan: Separating them by apparent brightness is useful for classification, but does it constrain the actual physical properties of these payloads? The paper focuses more on the observational constraints than on modeling the physics of their trajectories.

Vera: The paper doesn't get into that deep physics modeling, but they do link it back to how fast things are moving. They say that each exposure is about three hundred and sixty seconds long, which is much longer than the time a satellite spends crossing our field of view.

Jocelyn: So the instantaneous magnitude measurement is effectively a lower limit because the exposure time covers the transit. That's an important caveat for anyone trying to use these magnitudes to estimate distance or speed.

Subrahmanyan: And what about the overall impact on our survey area? The paper calculates that summing up all those five hundred twenty-seven streaks over all observations flags a total area of one point zero three times ten six arcsec squared, which is about zero point zero eight degrees squared.

Vera: That’s just zero point nine percent of the effective survey area, which is only about one tenth of a degree square. So while there are many streaks, they don't completely blind the telescope.

Title and authors: Jocelyn: But that contamination fraction is what keeps going up monthly past twenty percent by two thousand twenty-three and four point two arcminutes after twilight in the first hour after astronomical twilight <ref:2611.10301#pg1>. That’s where the real story is for time-domain astronomy, I think.

Subrahmanyan: The paper also points out that this rising contamination fraction broadly tracks the growth of payloads in orbit from what we call GCAT. It suggests that we aren't dealing with just one new constellation, but rather the cumulative effect of decades of launch activity.

Vera: Exactly. They are pointing out that the dominance of older, spent objects isn't proof that modern constellations pose a lesser threat to astronomy. The brighter LEO population is demonstrably increasing over time in their sample.

Jocelyn: It sounds like the conclusion is that these streaks trace six decades of accumulated launch activity rather than any one constellation, reflecting HETDEX’s baseline observational coverage and sensitivity, not a reprieve from constellation growth.

Subrahmanyan: I agree with that assessment. And the paper notes a specific selection effect where nearly half of those identified objects are in geostationary or highly elliptical orbits, which is more about the telescope's reach than anything else.

Vera: It’s also worth mentioning that even fully compliant satellites can register as contamination in deep surveys because of how bright they appear when you look at them from a large aperture like HET and the moderately long exposure times.

Jocelyn: And they show that the per-streak brightness doesn't change significantly over time, so the rising impact is driven by the growing frequency of contamination rather than any change in the streaks themselves.

Subrahmanyan: So to wrap up this paper on "Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact", it confirms that as long as we have these deep surveys running, we’ll keep seeing this accumulation of objects from the past.

Vera: It really does. And it sets a baseline for how we need to account for this kind of contamination in future wide-field imaging projects.

Jocelyn: Right, and it gives us a clearer picture of what to expect as we look at time-domain data going forward, especially with the increasing number of LEO satellites.

Subrahmanyan: And it reminds us that even in deep observations, the constraints we get have their own limitations due to the observational setup itself.

The paper's summary: Vera: So, to wrap up this discussion on HETDEX satellite streaks, what we've seen is that these aren't just random noise; they track the history of launches over sixty years.

Jocelyn: Right. The paper breaks down how they found them—using AI classification and checking those archival two line elements from SpaceTrack to link the streaks to actual space objects.

Vera: And it gives us some hard numbers on what’s happening in terms of contamination, showing that even though only about one point zero three times ten six arcsec squared of sky is masked, the fraction of observations affected goes up significantly over time.

Jocelyn: That monthly contamination fraction hitting above twenty percent by twenty twenty-three is a real marker for how quickly the LEO population is growing.

Vera: It’s telling us that we aren't just dealing with one new batch of satellites; we're seeing this slow accumulation of launches spanning from nineteen sixty three up to two thousand twenty-four.

Jocelyn: And the paper highlights something interesting about the orbital types they find—nearly half of those identified objects are in geostationary or highly elliptical orbits, which is really more about what the telescope can see than where the satellites are actually going.

Vera: That's a key distinction. It’s a selection effect of the instrument, because the HET and its exposures bring those slow-moving GEO and highly elliptical objects into view at almost the same brightness as closer LEO things.

Jocelyn: And it shows that even fully compliant satellites can register as contamination in these deep surveys because of how bright they look when you look at them from a large aperture like HET.

Vera: The paper also points out that the per-streak brightness itself doesn't really change much over this period, which means the rising impact is purely due to the frequency of these objects appearing rather than any physical change in the satellites themselves.

Jocelyn: So, even though it’s a nuisance for pure cosmology data, this research gives us a clearer picture of how to model and filter that noise as we push these wide-field surveys further.

Vera: It sets a baseline for how much we need to account for this kind of contamination in future wide-field imaging projects.

Jocelyn: And it makes me wonder what those implications are for time-domain astronomy, especially when the LEO population keeps growing, and whether we can use these streaks as a tracer rather than just something to subtract.

The paper's improvements: Vera: So, the paper doesn't just stop at identifying these streaks; they suggest several ways to make their analysis even better for future work.

Jocelyn: Right. They point out that we could implement a more precise contamination model using orbital dynamics to predict how much noise we’ll get in future surveys.

Vera: That means instead of just looking at the data, we could use that model to forecast the expected contamination fraction based on what orbits are launching next, which is helpful for planning telescope time.

Jocelyn: And they also suggest integrating those instantaneous magnitude corrections into the AI systems so they can better estimate what’s actually happening on the sky when you account for transit time.

Vera: That way, the AI gets a better sense of the physical quantity we care about, like how bright an object truly is rather than just its apparent brightness during a brief transit.

Jocelyn: Plus, they mention refining the satellite identification pipelines by checking them against independent services like SatChecker to boost that accuracy and cross-validation.

Vera: That cross-checking is important because it shows the matching pipeline works well when you use different ways of looking at those archival two line elements, which is a big win for robustness.

Jocelyn: And they suggest training the AI systems to recognize the spectral signature of reflected sunlight, which should help them distinguish these streaks from other kinds of noise or artifacts.

Vera: That’s interesting because they noted that all the detected stacks display clear solar spectrum features, so teaching the AI to look for those specific patterns should be a solid step toward cleaner data.

Jocelyn: So, what's the big picture here? It seems like these improvements are moving us from just cataloging streaks to creating a more reliable tool for filtering out this background noise in deep surveys.

Vera: Exactly. If we can get better at modeling and classifying this contamination, it means our cosmological measurements won't be as noisy when we’re looking at the faint signals from the early universe.

Jocelyn: It’s about making sure that when we look for those faint structures, we aren't accidentally picking up a bright satellite instead of a real astronomical feature.

Conclusion: Vera: So, to wrap up this paper on "Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact," we’ve seen that these streaks are not just noise; they map out sixty years of launch activity.

Jocelyn: That's right. They show how those streaks tie directly into real space objects by using archival data from SpaceTrack to check the orbits.

Subrahmanyan: I think it’s a good exercise in separating the local observational noise from the actual cosmological signal we’re trying to measure at redshift two point four.

Vera: The main implication is that we have a clearer picture of how much contamination to expect when we run these deep, wide-field surveys over long baselines.

Jocelyn: It changes how we treat the data by showing that the rising contamination fraction isn't just random; it’s tracking the growth of LEO constellations in orbit.

Subrahmanyan: And for structure formation studies, this means we need to be careful about what we subtract from our images when trying to map dark matter or galaxy clustering.

Vera: It also highlights that even the brightest objects, like geostationary ones, can still be contamination because of how bright they appear from a large telescope setup.

Jocelyn: So, while it’s a nuisance for pure cosmology data, this research gives us a clearer way to filter out that background noise in future wide-field surveys.

Subrahmanyan: And it reminds us that the constraints we get from these deep observations have limitations due to the observational setup itself.

Vera: Exactly. We’re using this kind of data to build better tools for analysis, which is what really matters for the next generation of astronomy projects.

Jocelyn: It makes me wonder what follow-up observations we need to do now, given how much they’ve taught us about these streaks and their orbital behavior.

Erin Mentuch Cooper, Karl Gebhardt, Dustin Davis, Daniel J. Farrow, Barbara G. Castanheira, Gavin Dalton, Gary J. Hill, Stephen C. Hummel, 你大 Liu (Liu Chenxu), 世 Shiro Mukae, 北 Shun Saito Donald P. Schneider Sarah Tuttle

Department of Astronomy, The University of Texas at Austin · McDonald Observatory, The University of Texas at Austin · E. A. Milne Centre for Astrophysics, University of Hull · Centre of Excellence for Data Science, Artificial Intelligence & Modelling (DAIM), University of Hull · Department of Physics and Astronomy, Baylor University · Department of Physics, Denys Wilkinson Building, Keble Road, Oxford · South-Western Institute for Astronomy Research, Key Laboratory of Survey Science of Yunnan Province, Yunnan University · Department of Physics, School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University · Institute for Multi-messenger Astrophysics and Cosmology, Department of Physics, Missouri University of Science and Technology · Kavli Institute for the Physics and Mathematics of the Universe (WPI), The University of Tokyo Institutes for Advanced Study (UTIAS) · Department of Astronomy & Astrophysics, The Pennsylvania State University · Institute for Gravitation and the Cosmos, The Pennsylvania State University · Department of Astronomy, University of Washington

astro-ph.IM

Submitted: 2026-11-06

Updated: 2026-11-06

Comments: 15 pages, 5 figures, 1 table. Published at Mentuch Cooper et al., 2026, AJ, 172, 303. Data available at https://doi.org/10.5281/zenodo.22962479. Full workflow at https://github.com/eriniocentric/hetdex_sats

Journal ref: Mentuch Cooper et al., 2026, AJ, 172, 303

DOI: 10.3847/1538-3881/aeb0a7

Code: https://github.com/eriniocentric/hetdex_sats

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 90/100

The gist: We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX)…

Summary

We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX), covering 6778 observations and ∼90 deg2 of sky from January 2017 to July 2024. p. 0

Where the paper says it

Milne Centre for Astrophysics University of Hull, Cottingham Road, Hull, HU6 7RX, UK 4Centre of Excellence for Data Science, Artificial Intelligence & Modelling (DAIM), University of Hull, Cottingham Road, Hull, HU6 7RX, UK 5Department of Physics and Astronomy, Baylor University, Waco, TX 76798-7316 6Department of Physics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, UK 7South-Western Institute for Astronomy Research, Key Laboratory of Survey Science of Yunnan Province, Yunnan University, Kunming, Yunnan 650500, People’s Republic of China 8Department of Physics, School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo 169-8555, Japan 9 Institute for Multi-messenger Astrophysics and Cosmology, Department of Physics, Missouri University of Science and Technology, 1315 N. , Rolla MO 65409, USA 10Kavli Institute for the Physics and Mathematics of the Universe (WPI), The University of Tokyo Institutes for Advanced Study (UTIAS), The University of Tokyo, Chiba 277-8583, Japan 11Department of Astronomy & Astrophysics, The Pennsylvania State University, University Park, PA 16802, USA 12Institute for Gravitation and the Cosmos, The Pennsylvania State University, University Park, PA 16802, USA 13Department of Astronomy, University of Washington, Seattle, 3910 15th Ave NE, Room C319, Seattle WA 98195-0002 ABSTRACT We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX), covering 6778 observations and ∼90 deg2 of sky from January 2017 to July 2024.

Contamination Statistics

We detect 527 streaks across 492 observations, a contamination fraction of 7.3% that rises to monthly values above 20% by 2023–2024, although the streaks mask only 0. p. 0

Where the paper says it

We detect 527 streaks across 492 observations, a contamination fraction of 7. 3% that rises to monthly values above 20% by 2023–2024, although the streaks mask only 0.

09% of the survey area. p. 0
Where the paper says it

3% that rises to monthly values above 20% by 2023–2024, although the streaks mask only 0. 09% of the survey area.

Of the 527 streaks, 475 (90. p. 0
Where the paper says it

09% of the survey area. Of the 527 streaks, 475 (90.

1%) are identified with a cataloged object using archival two-line elements: 258 (54%) lie in low or medium Earth orbit and 217 (46%) are geostationary or highly elliptical. p. 0
Where the paper says it

Of the 527 streaks, 475 (90. 1%) are identified with a cataloged object using archival two-line elements: 258 (54%) lie in low or medium Earth orbit and 217 (46%) are geostationary or highly elliptical.

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. p. 0
Where the paper says it

Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft. This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline.

Data Acquisition and Detection

HETDEX is a wide-field integral-field spectroscopic survey designed to map the large-scale structure of the Universe at z ∼ 2.4. HETDEX obtains low-resolving-power (R ∼ 800–950) optical spectroscopy over 3500–5500 ˚A using the Visible Integral-field Replicable Unit Spectrograph (VIRUS, G. p. 1

Where the paper says it

Gebhardt et al. HETDEX obtains low-resolving-power (R ∼ 800–950) optical spectroscopy over 3500–5500 ˚A using the Visible Integral-field Replicable Unit Spectrograph (VIRUS, G.

J. Hill et al. p. 0
Where the paper says it

Castanheira , 5 Gavin Dalton , 6 Gary J. Hill , 2, 1 Stephen C.

2021). Object detection is performed without imaging preselection using complementary searches for emission-line and continuum sources across more than 6 × 108 spectra. p. 2
Where the paper says it

′′5 in diameter, and a detections database generated directly from these data. Object detection is performed without imaging preselection using complementary searches for emission-line and continuum sources across more than 6 × 108 spectra.

Satellite streaks are detected from spectrally collapsed image cutouts classified using the proprietary AI-classification software Rapid Automatic Image Categorization (RAIC; RAIC Labs 2025). p. 3
Where the paper says it

In addition to astrophysical sources, the detection database contains a significant number of artifacts, including those arising from satellite streaks, which are identified and masked as part of the data reduction pipeline. Satellite Detection Satellite streaks are detected from spectrally collapsed image cutouts classified using the proprietary AI-classification software Rapid Automatic Image Categorization (RAIC; RAIC Labs 2025).

A satellite streak is confirmed when a group of detections in an observation displays an alignment of point-like detections tracing a linear path across the field. p. 3
Where the paper says it

From these labels, satellite streaks are identified by searching observations that contain at least five continuum detections assigned a “satellite” label by RAIC. A satellite streak is confirmed when a group of detections in an observation displays an alignment of point-like detections tracing a linear path across the field.

Orbital Identification and Constraints

For each streak we search for a catalogued object whose propagated orbit reproduces the observed track, using archival two-line elements (TLEs) from the SpaceTrack gp history service, 16 which is the only source spanning the full 2017–2024 PDR1 baseline. The primary discriminant is the perpendicular offset of the streak midpoint from the great circle locally tangent to the propagated track, together with the position-angle difference. p. 4

Where the paper says it

space-track. org onds to ∼1 ′ , the primary discriminant is the perpendicular offset of the streak midpoint from the great circle locally tangent to the propagated track, together with the position-angle difference.

For 475 matched streaks, 99. p. 5
Where the paper says it

Of the 527 streaks, 475 (90. 1%) are matched to a catalogued object via the SGP4 identification pipeline described in Section 2.

2% are unambiguous. p. 4
Where the paper says it

◦5)2 , and the best plus four runners-up are retained. A match is unambiguous when no runner-up exists or the next-best score exceeds three times the best plus one.

The orbital identifications consist of 254 payloads (53. p. 8
Where the paper says it

5281/zenodo. p. 0

Where the paper says it

The catalog, spectra, and code are archived at 10. 5281/zenodo.

Orbital Properties By object type, the orbital identifications consist of 254 payloads (53.

5%), 172 rocket bodies (36. p. 8
Where the paper says it

Orbital Properties By object type, the orbital identifications consist of 254 payloads (53. 5%), 172 rocket bodies (36.

2%), and 49 debris objects (10. p. 0
Where the paper says it

Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft. This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline.

3%).

Photometric Findings

The measured streak SDSS-g magnitudes span 11.5–18.8 mag (median 16. p. 5

Where the paper says it

41), set by the gaps between IFUs along a streak’s path; without it tcross is overestimated and ginst too faint by ∼1. 4 mag at the median.

4 mag) and observed lengths range from 0. p. 4
Where the paper says it

5 grated flux divided by the projected on-sky area of the associated spaxels, yielding values in mag arcsec−2 . Orbital Crossmatching For each streak we search for a catalogued object whose propagated orbit reproduces the observed track, using archival two-line elements (TLEs) from the SpaceTrack gp history service,16 which is the only source spanning the full 2017–2024 PDR1 baseline, since CelesTrak’s public feed serves current elements only and the IAU CPS SatChecker archive (M.

6′ to 17.2′ (median 11. p. 0
Where the paper says it

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s ∼6 minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median gAB = 16.

1′). The median gAB = 550 km gives g(550 km) = 4. p. 9
Where the paper says it

topocentric slant range, by orbit class. The dashed curve is the inverse-square locus anchored to the sample median g(550 km).

55 mag at the median (the 6th–84th percentile: 3. p. 5
Where the paper says it

The median fill factor is 0. 27 (16th–84th percentile 0.

p. 5
Where the paper says it

The median fill factor is 0. 27 (16th–84th percentile 0.

25–6.12), rising from 3. p. 4
Where the paper says it

These times are tabulated as MJDopen and MJDclos, and the MJD column carries the shutter-open time of the first dither (Table 1). A coarse pass propagates every cataloged object with an element set for that night on a 20 s grid with a median of ∼21,400 objects per shot, rising to ∼25,000 by 2024 as the public catalog grows.

p. 4
Where the paper says it

These times are tabulated as MJDopen and MJDclos, and the MJD column carries the shutter-open time of the first dither (Table 1). A coarse pass propagates every cataloged object with an element set for that night on a 20 s grid with a median of ∼21,400 objects per shot, rising to ∼25,000 by 2024 as the public catalog grows.

86 for GEO to 5. p. 0
Where the paper says it

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s ∼6 minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median gAB = 16.

p. 0
Where the paper says it

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s ∼6 minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median gAB = 16.

91 for LEO, which is comparable to published photometry by A Mallama (2021). p. 1
Where the paper says it

2024) already occupy a measurable fraction of the sky, and forecasting that the roughly hundredfold growth projected for the LEO constellation population would push contamination probabilities for short, wide-field exposures toward unity (A. Mallama et al.

p. 1
Where the paper says it

2024) already occupy a measurable fraction of the sky, and forecasting that the roughly hundredfold growth projected for the LEO constellation population would push contamination probabilities for short, wide-field exposures toward unity (A. Mallama et al.

The instantaneous magnitudes span ginst = 3. p. 0
Where the paper says it

1 for GEO). However, once transit time is corrected for, instantaneous magnitudes differ widely: ginst = 7.

9–14.6 mag (median 10. p. 5
Where the paper says it

41), set by the gaps between IFUs along a streak’s path; without it tcross is overestimated and ginst too faint by ∼1. 4 mag at the median.

0 mag) and separate the classes cleanly, with medians of 7. p. 4
Where the paper says it

Vallado 2007): A match placing the object in Earth’s umbra is physically impossible and sets a lower limit on the false-positive rate (Section 3. Orbit classes follow from the propagated perigee hp, apogee ha and eccentricity e: LEO (ha < 2000 km), geostationary Earth orbit (GEO; hp > 34,000 km and ha < 37,500 km), highly elliptical orbit (HEO; e > 0.

4 (LEO), 11. p. 0
Where the paper says it

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s ∼6 minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median gAB = 16.

p. 0
Where the paper says it

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s ∼6 minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median gAB = 16.

4 (MEO), 11. p. 4
Where the paper says it

Orbit classes follow from the propagated perigee hp, apogee ha and eccentricity e: LEO (ha < 2000 km), geostationary Earth orbit (GEO; hp > 34,000 km and ha < 37,500 km), highly elliptical orbit (HEO; e > 0. 25 with ha > 2000 km, or ha ≥ 37,500 km), and medium Earth orbit (MEO) otherwise.

p. 4
Where the paper says it

Orbit classes follow from the propagated perigee hp, apogee ha and eccentricity e: LEO (ha < 2000 km), geostationary Earth orbit (GEO; hp > 34,000 km and ha < 37,500 km), highly elliptical orbit (HEO; e > 0. 25 with ha > 2000 km, or ha ≥ 37,500 km), and medium Earth orbit (MEO) otherwise.

8 (HEO), and 13. p. 4
Where the paper says it

Vallado 2007): A match placing the object in Earth’s umbra is physically impossible and sets a lower limit on the false-positive rate (Section 3. Orbit classes follow from the propagated perigee hp, apogee ha and eccentricity e: LEO (ha < 2000 km), geostationary Earth orbit (GEO; hp > 34,000 km and ha < 37,500 km), highly elliptical orbit (HEO; e > 0.

p. 4
Where the paper says it

Vallado 2007): A match placing the object in Earth’s umbra is physically impossible and sets a lower limit on the false-positive rate (Section 3. Orbit classes follow from the propagated perigee hp, apogee ha and eccentricity e: LEO (ha < 2000 km), geostationary Earth orbit (GEO; hp > 34,000 km and ha < 37,500 km), highly elliptical orbit (HEO; e > 0.

0 (GEO).

Survey Impact

Summing the masked spaxels over all 527 streaks yields a total flagged area of 1.03 × 106 arcsec2 (0. p. 5

Where the paper says it

Summing the masked spaxels over all 527 streaks yields a total flagged area of 1. 03 × 106 arcsec2 (0.

08 deg2), corresponding to just 0. p. 5
Where the paper says it

03 × 106 arcsec2 (0. 08 deg2 ), corresponding to just 0.

09% of the PDR1 effective survey area. p. 5
Where the paper says it

09% of the 86. 67 deg2 PDR1 effective survey area.

The contamination fraction rises from a few percent early in the survey to monthly values above 20% by 2023, broadly tracking the growth of payloads in orbit from the General Catalog of Artificial Space Objects (GCAT). p. 6
Where the paper says it

Latter epochs show the highest contamination values near twilight, but the values remain significant all night. tion rising from a few percent in 2017 to monthly values exceeding 20% by 2023, broadly tracking the growth of payloads in orbit from the General Catalog of Artificial Space Objects (GCAT18; J.

The dominance of older, spent objects in our sample is not evidence that modern satellite constellations pose a lesser threat to astronomy. p. 0
Where the paper says it

Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft. This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline.

p. 0
Where the paper says it

Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft. This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline.

The brighter LEO population is demonstrably increasing over time. p. 6
Where the paper says it

5% baseline). The growth of the post-twilight peak thus tracks the increasing LEO satellite population, even as streaks remain detectable at all hours of the night.

p. 6
Where the paper says it

5% baseline). The growth of the post-twilight peak thus tracks the increasing LEO satellite population, even as streaks remain detectable at all hours of the night.

Conclusion

The streaks trace six decades of accumulated launch activity rather than any one constellation, reflecting HETDEX’s baseline observational coverage (from early 2017 to mid-2024) and sensitivity, not a reprieve from constellation growth. The brighter LEO population is demonstrably increasing over time. Nearly half (46%) are in geostationary or highly elliptical orbits—a selection effect of the telescope rather than the sky, since the Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s moderately long, ∼6 min exposures bring distant, slow-moving GEO and HEO objects within reach at nearly the same observed brightness as much closer LEO objects. p. 0

Where the paper says it

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s ∼6 minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median gAB = 16.

The GEO detections are well below the naked-eye visibility brightness limit targeted by the IAU (IAU CPS et al. p. 1
Where the paper says it

Most existing studies have focused on imaging surveys, where satellite trails primarily affect a fraction of pixels that can, in principle, be masked or interpolated. A recent census of 32 million wide-field photometric observations from the Mini-MegaTORTORA network quantifies this directly, finding that satellites, rockets, and debris brighter than IAU interference thresholds ( IAU CPS et al.

2024), indicating that even fully compliant satellites can register as contamination in deep, large-aperture-telescope surveys. p. 0
Where the paper says it

This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline. The Hobby-Eberly Telescope’s large 10 m aperture and HETDEX’s ∼6 minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median gAB = 16.

The per-streak brightness shows no significant evolution, indicating that the rising impact is driven by the growing frequency of contamination rather than by any change in the streaks themselves.

Acknowledgments

HETDEX is led by the University of Texas at Austin McDonald Observatory and Department of Astronomy with participation from the Ludwig-Maximilians-Universit¨at M¨unchen, Max-Planck-Institut f¨ur Extraterrestrische Physik (MPE), Leibniz-Institut f¨ur Astrophysik Potsdam (AIP), Texas A&M University, The Pennsylvania State University, Institut f¨ur Astrophysik G¨ottingen, The University of Oxford, Max-PlanckInstitut f¨ur Astrophysik (MPA), The University of Tokyo, and Missouri University of Science and Technology. Observations for HETDEX were obtained with the Hobby-Eberly Telescope (HET), which is a joint project of the University of Texas at Austin, the Pennsylvania State University, Ludwig-Maximilians-Universit¨at M¨unchen, and Georg-August-Universit¨at Gottingen. p. 0

Where the paper says it

Milne Centre for Astrophysics University of Hull, Cottingham Road, Hull, HU6 7RX, UK 4Centre of Excellence for Data Science, Artificial Intelligence & Modelling (DAIM), University of Hull, Cottingham Road, Hull, HU6 7RX, UK 5Department of Physics and Astronomy, Baylor University, Waco, TX 76798-7316 6Department of Physics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, UK 7South-Western Institute for Astronomy Research, Key Laboratory of Survey Science of Yunnan Province, Yunnan University, Kunming, Yunnan 650500, People’s Republic of China 8Department of Physics, School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo 169-8555, Japan 9 Institute for Multi-messenger Astrophysics and Cosmology, Department of Physics, Missouri University of Science and Technology, 1315 N. , Rolla MO 65409, USA 10Kavli Institute for the Physics and Mathematics of the Universe (WPI), The University of Tokyo Institutes for Advanced Study (UTIAS), The University of Tokyo, Chiba 277-8583, Japan 11Department of Astronomy & Astrophysics, The Pennsylvania State University, University Park, PA 16802, USA 12Institute for Gravitation and the Cosmos, The Pennsylvania State University, University Park, PA 16802, USA 13Department of Astronomy, University of Washington, Seattle, 3910 15th Ave NE, Room C319, Seattle WA 98195-0002 ABSTRACT We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX), covering 6778 observations and ∼90 deg2 of sky from January 2017 to July 2024.

The Visible Integral-field Replicable Unit Spectrograph (VIRUS) was used for HETDEX observations. p. 1
Where the paper says it

Gebhardt et al. HETDEX obtains low-resolving-power (R ∼ 800–950) optical spectroscopy over 3500–5500 ˚A using the Visible Integral-field Replicable Unit Spectrograph (VIRUS, G.

VIRUS is a joint project of the University of Texas at Austin, Leibniz-Institut f¨ur Astrophysik Potsdam (AIP), Texas A&M University (TAMU), Max-Planck-Institut f¨ur Extraterrestrische Physik (MPE), LudwigMaximilians-Universit¨at M¨unchen, Pennsylvania State University, Institut f¨ur Astrophysik Gottingen, University of Oxford, and the Max-Planck-Institut f¨ur Astrophysik (MPA). p. 0
Where the paper says it

Schneider , 11, 12 and Sarah Tuttle 13 1Department of Astronomy, The University of Texas at Austin, 2515 Speedway Boulevard, Austin, TX 78712, USA 2McDonald Observatory, The University of Texas at Austin, 2515 Speedway Boulevard, Austin, TX 78712, USA 3E. Milne Centre for Astrophysics University of Hull, Cottingham Road, Hull, HU6 7RX, UK 4Centre of Excellence for Data Science, Artificial Intelligence & Modelling (DAIM), University of Hull, Cottingham Road, Hull, HU6 7RX, UK 5Department of Physics and Astronomy, Baylor University, Waco, TX 76798-7316 6Department of Physics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, UK 7South-Western Institute for Astronomy Research, Key Laboratory of Survey Science of Yunnan Province, Yunnan University, Kunming, Yunnan 650500, People’s Republic of China 8Department of Physics, School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo 169-8555, Japan 9 Institute for Multi-messenger Astrophysics and Cosmology, Department of Physics, Missouri University of Science and Technology, 1315 N.

The authors acknowledge the Texas Advanced Computing Center (TACC) at The University of Texas at Austin for providing high performance computing, visualization, and storage resources that have contributed to the research results reported within this paper. p. 0
Where the paper says it

Schneider , 11, 12 and Sarah Tuttle 13 1Department of Astronomy, The University of Texas at Austin, 2515 Speedway Boulevard, Austin, TX 78712, USA 2McDonald Observatory, The University of Texas at Austin, 2515 Speedway Boulevard, Austin, TX 78712, USA 3E. Milne Centre for Astrophysics University of Hull, Cottingham Road, Hull, HU6 7RX, UK 4Centre of Excellence for Data Science, Artificial Intelligence & Modelling (DAIM), University of Hull, Cottingham Road, Hull, HU6 7RX, UK 5Department of Physics and Astronomy, Baylor University, Waco, TX 76798-7316 6Department of Physics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, UK 7South-Western Institute for Astronomy Research, Key Laboratory of Survey Science of Yunnan Province, Yunnan University, Kunming, Yunnan 650500, People’s Republic of China 8Department of Physics, School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo 169-8555, Japan 9 Institute for Multi-messenger Astrophysics and Cosmology, Department of Physics, Missouri University of Science and Technology, 1315 N.

We acknowledge RAIC Labs (https://raiclabs. p. 3
Where the paper says it

In addition to astrophysical sources, the detection database contains a significant number of artifacts, including those arising from satellite streaks, which are identified and masked as part of the data reduction pipeline. Satellite Detection Satellite streaks are detected from spectrally collapsed image cutouts classified using the proprietary AI-classification software Rapid Automatic Image Categorization (RAIC; RAIC Labs 2025).

com) for providing access to the RAIC platform, which enabled the efficient classification and identification of satellite streaks within the HETDEX dataset. p. 3
Where the paper says it

In addition to astrophysical sources, the detection database contains a significant number of artifacts, including those arising from satellite streaks, which are identified and masked as part of the data reduction pipeline. Satellite Detection Satellite streaks are detected from spectrally collapsed image cutouts classified using the proprietary AI-classification software Rapid Automatic Image Categorization (RAIC; RAIC Labs 2025).

This research made use of data and/or services provided by the International Astronomical Union Centre for the Protection of the Dark and Quiet Sky (IAU CPS) SatHub. p. 0
Where the paper says it

Milne Centre for Astrophysics University of Hull, Cottingham Road, Hull, HU6 7RX, UK 4Centre of Excellence for Data Science, Artificial Intelligence & Modelling (DAIM), University of Hull, Cottingham Road, Hull, HU6 7RX, UK 5Department of Physics and Astronomy, Baylor University, Waco, TX 76798-7316 6Department of Physics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, UK 7South-Western Institute for Astronomy Research, Key Laboratory of Survey Science of Yunnan Province, Yunnan University, Kunming, Yunnan 650500, People’s Republic of China 8Department of Physics, School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo 169-8555, Japan 9 Institute for Multi-messenger Astrophysics and Cosmology, Department of Physics, Missouri University of Science and Technology, 1315 N. , Rolla MO 65409, USA 10Kavli Institute for the Physics and Mathematics of the Universe (WPI), The University of Tokyo Institutes for Advanced Study (UTIAS), The University of Tokyo, Chiba 277-8583, Japan 11Department of Astronomy & Astrophysics, The Pennsylvania State University, University Park, PA 16802, USA 12Institute for Gravitation and the Cosmos, The Pennsylvania State University, University Park, PA 16802, USA 13Department of Astronomy, University of Washington, Seattle, 3910 15th Ave NE, Room C319, Seattle WA 98195-0002 ABSTRACT We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX), covering 6778 observations and ∼90 deg2 of sky from January 2017 to July 2024.

The development of SatChecker has been supported by the National Science Foundation (NSF) under grant numbers: AST 2332735 and AST 2332736. p. 0
Where the paper says it

Schneider , 11, 12 and Sarah Tuttle 13 1Department of Astronomy, The University of Texas at Austin, 2515 Speedway Boulevard, Austin, TX 78712, USA 2McDonald Observatory, The University of Texas at Austin, 2515 Speedway Boulevard, Austin, TX 78712, USA 3E. Milne Centre for Astrophysics University of Hull, Cottingham Road, Hull, HU6 7RX, UK 4Centre of Excellence for Data Science, Artificial Intelligence & Modelling (DAIM), University of Hull, Cottingham Road, Hull, HU6 7RX, UK 5Department of Physics and Astronomy, Baylor University, Waco, TX 76798-7316 6Department of Physics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH, UK 7South-Western Institute for Astronomy Research, Key Laboratory of Survey Science of Yunnan Province, Yunnan University, Kunming, Yunnan 650500, People’s Republic of China 8Department of Physics, School of Advanced Science and Engineering, Faculty of Science and Engineering, Waseda University, 3-4-1 Okubo, Shinjuku, Tokyo 169-8555, Japan 9 Institute for Multi-messenger Astrophysics and Cosmology, Department of Physics, Missouri University of Science and Technology, 1315 N.

AI Disclosure: This work made use of Claude (Sonnet 4. p. 2
Where the paper says it

5, where gHETDEX is a pseudo-magnitude derived from the HETDEX source spectrum by convolving it with the SDSS g-band transmission curve and integrating the resulting flux. HETDEX Public Data Release 1 This analysis makes use of the HETDEX Public Data Release 1 (PDR1; E.

5, Opus 4.5/4.8/5; Anthropic 2026), a large language model (LLM), for literature search, text editing, and developing and documenting parts of the analysis code described in Sections 2.3 and Section 2. p. 1
Where the paper says it

Mentuch Cooper et al. In Section 2, we provide a brief overview of PDR1 and describe the satellite detection, spectral extraction, and orbital crossmatching methods used to build the catalog.

4. Photometric extraction (Section 2. p. 3
Where the paper says it

We note that five observations within this list (20190731019, 20200523028, 20220928016, 20230720009, and 20240311025) are manually removed from the analysis because the spectra revealed the emission was due to a meteor. Spectral Extraction and Photometric Measurements For each identified satellite streak, we extract a onedimensional spectrum by summing the flux of all spaxels whose sky positions fall within a fixed perpendicular distance of the fitted streak track.

3) used LLM context grounded in the HETDEX PDR1 paper (E. p. 1
Where the paper says it

2021), which uses untargeted emission-line searches over large sky areas to construct massive samples of galaxies at 1. In this paper, we present a spectral database and analysis of satellite streaks in integral-field unit (IFU) data from the Public Data Release 1 of HETDEX (PDR1; E.

Mentuch Cooper et al. p. 0
Where the paper says it

Draft version November 10, 2026 Typeset using LATEX twocolumn style in AASTeX7. 1 Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact Erin Mentuch Cooper , 1, 2 Karl Gebhardt , 1 Dustin Davis , 1 Daniel J.

p. 0
Where the paper says it

Draft version November 10, 2026 Typeset using LATEX twocolumn style in AASTeX7. 1 Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact Erin Mentuch Cooper , 1, 2 Karl Gebhardt , 1 Dustin Davis , 1 Daniel J.

p. 0
Where the paper says it

Draft version November 10, 2026 Typeset using LATEX twocolumn style in AASTeX7. 1 Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact Erin Mentuch Cooper , 1, 2 Karl Gebhardt , 1 Dustin Davis , 1 Daniel J.

2026) and its associated dexcube repos- The matching pipeline (Section 2. p. 2
Where the paper says it

Mentuch Cooper et al. 2026), which provides both calibrated data cubes and associated source catalogs for the full HETDEX survey.

4), built on sgp4 propagation of archival elements, was validated by crosschecking a subset of matches against satchecker, an independent identification service drawing on a separate element archive; diagnostic figures for each fit, available at https://github. p. 8
Where the paper says it

As a purity check, five matched streaks are assigned orbits whose SGP4 propagation places the object in Earth’s umbra at crossing time, giving a false-positive floor of ≲1. As an independent check, a random subset of 180 of the 468 Space-Track matches was reidentified through the IAU CPS SatChecker API (M.

com/eriniocentric/hetdex sats and archived at 10. p. 5
Where the paper says it

The left panel shows the monthly and bi-monthly contamination frac17 https://github. com/eriniocentric/hetdex sats

5281/zenodo.22962479, provide additional validation of the crossmatching method. p. 1
Where the paper says it

Mentuch Cooper et al. In Section 2, we provide a brief overview of PDR1 and describe the satellite detection, spectral extraction, and orbital crossmatching methods used to build the catalog.

All code, results, and text were reviewed by the authors, who take full responsibility for the content and accuracy of this paper. Facilities: HET Software: astropy (Astropy Collaboration et al. p. 7
Where the paper says it

While satellite visibility is often assumed to peak at small Sun–target angles, the HETDEX sample shows a broad distribution in both measured magnitude and surface brightness over several hours after astronomical twilight. This behavior is enabled in part by the large collecting area of the HET, which allows for detection of satellite streaks over a wider range of brightness and observing geometries than would be accessible to smaller-aperture facilities.

2013, 2018), RAIC Labs (RAIC Labs 2025), hetdex-api (E. p. 3
Where the paper says it

In addition to astrophysical sources, the detection database contains a significant number of artifacts, including those arising from satellite streaks, which are identified and masked as part of the data reduction pipeline. Satellite Detection Satellite streaks are detected from spectrally collapsed image cutouts classified using the proprietary AI-classification software Rapid Automatic Image Categorization (RAIC; RAIC Labs 2025).

Mentuch Cooper & HETDEX Collaboration 2024), dexcube (E. p. 0
Where the paper says it

Draft version November 10, 2026 Typeset using LATEX twocolumn style in AASTeX7. 1 Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact Erin Mentuch Cooper , 1, 2 Karl Gebhardt , 1 Dustin Davis , 1 Daniel J.

Mentuch Cooper et al. 2026), speclite (D. p. 3
Where the paper says it

5 log10 3 = 1. Synthetic photometry is computed using the SDSS gband filter as implemented in the speclite package (D.

Kirkby et al. p. 3
Where the paper says it

Synthetic photometry is computed using the SDSS gband filter as implemented in the speclite package (D. Kirkby et al.

2023), astroplan (B. p. 0
Where the paper says it

We detect 527 streaks across 492 observations, a contamination fraction of 7. 3% that rises to monthly values above 20% by 2023–2024, although the streaks mask only 0.

M. Morris et al. 2018), sgp4 (B Rhodes 2020), satchecker (M Dadighat et al. p. 4
Where the paper says it

Elements are propagated with the sgp4 implementation of the SGP4/SDP4 theory (B. Rhodes 2020; D.

2024), Claude (Anthropic 2026). p. 0
Where the paper says it

Draft version November 10, 2026 Typeset using LATEX twocolumn style in AASTeX7. 1 Satellite Streaks in HETDEX: Spectra, Orbital Identification, and Survey Impact Erin Mentuch Cooper , 1, 2 Karl Gebhardt , 1 Dustin Davis , 1 Daniel J.

REFERENCES

Pruning redundant reasoning tokens

The contamination fraction evolves significantly over this period, reaching a peak monthly value of 22.7% in July 2023. p. 5

Where the paper says it

The contamination fraction evolves significantly over this period, reaching a peak monthly value of 22. 7% in July 2023.

The right panel shows the contamination fraction as a function of hours after astronomical twilight, split into survey epochs (2017–2024; one-hour bins, binomial errors). p. 6
Where the paper says it

Right: contamination fraction vs. hours after astronomical twilight (when the Sun falls 18 deg below the horizon for HET), split into survey epochs (2017–2020, 2021–2022, 2023–2024; one-hour bins, binomial errors).

The growth of the post-twilight peak thus tracks the increasing LEO satellite population, even as streaks remain detectable at all hours of the night. p. 6
Where the paper says it

5% baseline). The growth of the post-twilight peak thus tracks the increasing LEO satellite population, even as streaks remain detectable at all hours of the night.

How the training signal is built

Starting with an initial random sample set of ∼20K spectrally collapsed images, the images are uploaded into the RAIC classification framework. Labels are assigned to images with similar morphologies until each set has at least 1000 labeled samples. p. 3

Where the paper says it

Starting with an initial random sample set of ∼20K spectrally collapsed images, the images are uploaded into the RAIC classification framework. Labels are assigned to images with similar morphologies until each set has at least 1000 labeled samples.

These are then applied as supervised inputs to the RAIC system for training. p. 3
Where the paper says it

Labels are assigned to images with similar morphologies until each set has at least 1000 labeled samples. These are then applied as supervised inputs to the RAIC system for training.

This trained model then assigns a probability for each label, including “satellite” and those exceeding a threshold classification score of 0. p. 3
Where the paper says it

These are then applied as supervised inputs to the RAIC system for training. This trained model then assigns a probability for each label, including “satellite” and those exceeding a threshold classification score of 0.

3 are flagged accordingly in our catalogs (E. p. 3
Where the paper says it

This trained model then assigns a probability for each label, including “satellite” and those exceeding a threshold classification score of 0. 3 are flagged accordingly in our catalogs (E.

Mentuch Cooper et al. 2026).

Where the gains come from

The sensitivity provided by HET’s large aperture and HETDEX’s 360 s exposure time result in similar ranges of observed magnitudes for GEO and LEO populations even though the lower-altitude LEOs are visually brighter by eye. The trail-integrated magnitudes carry almost no distance information (Spearman ρ = 0. p. 5

Where the paper says it

Four further columns record the seeing (observation image quality) and the precise dither timing (Section 2. The remaining 22 columns carry the orbital identification: the NORAD catalog number, object name, COSPAR designator, object type, orbit class, match residuals, crossing epoch, slant range, orbital elements, and instantaneous trail magnitude.

18 against log d). p. 4
Where the paper says it

25 with ha > 2000 km, or ha ≥ 37,500 km), and medium Earth orbit (MEO) otherwise. Streaks with no acceptable candidate are additionally queried against SatChecker, which draws on a separate element archive; identifications recovered this way are flagged by IDsrc (Table 1).

The matched sample has g(550 km) = 4. p. 9
Where the paper says it

topocentric slant range, by orbit class. The dashed curve is the inverse-square locus anchored to the sample median g(550 km).

55 mag at the median (the 6th–84th percentile: 3.25–6.12), rising from 3.86 for GEO to 5.91 for LEO, which is comparable to published photometry by A Mallama (2021). The high-orbit objects are intrinsically the brightest, as expected for geostationary buses and spent upper stages. p. 0
Where the paper says it

1%) are identified with a cataloged object using archival two-line elements: 258 (54%) lie in low or medium Earth orbit and 217 (46%) are geostationary or highly elliptical. Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft.

The 34 Starlink matches give 5. p. 0
Where the paper says it

6 mag, while the exposure-averaged magnitudes differ by only 0. Normalized to 550 km, the 34 Starlink streaks give g(550 km) = 5.

95 mag (5.23–6.94), in agreement with the mean V (550 km) = 5. p. 0
Where the paper says it

Normalized to 550 km, the 34 Starlink streaks give g(550 km) = 5. 95, consistent with the mean V (550 km) = 5.

92 ± 0.04 reported for VisorSat-design spacecraft by A Mallama (2021). p. 0
Where the paper says it

95, consistent with the mean V (550 km) = 5. 04 reported for VisorSat-design spacecraft.

The geostationary objects are the cleanest case: their slant ranges span only 0. p. 4
Where the paper says it

Instead per-dither shutter open and close times are read from the internal HETDEX survey file (from internal data release HDR5, from which PDR1 is drawn), which covers all 492 shots in this sample, and each of the three dithers is searched over [MJDopen − 60 s, MJDclose + 60 s], which comfortably exceeds the ∼35–40 s of readout and repositioning between dithers. The elapsed shot span ranges from 1169 to 2263 s.

10 mag of distance, so their 4. p. 3
Where the paper says it

For these aligned satellite detections, the fitting is done manually by performing a linear fit in sky positions. Following the HETDEX masking strategy, detections and fibers within a fixed perpendicular angular distance of 6 arcsec of the fitted track are associated with the satellite streak (E.

3 mag vertical extent in the panel is intrinsic almost by construction. The HETDEX PDR1 baseline spans 2017–2024, giving pre-constellation-era debris and rocket bodies far more cumulative opportunities to cross the field than recently launched satellites, which make up only 35% of identifications. p. 0
Where the paper says it

Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft. This reflects HETDEX’s sensitivity to faint objects and its 2017–2024 observing baseline.

The brighter LEO population is increasing over time. p. 6
Where the paper says it

5% baseline). The growth of the post-twilight peak thus tracks the increasing LEO satellite population, even as streaks remain detectable at all hours of the night.

The 527 satellite streaks across 492 observations, out of 6778 total observations in PDR1, corresponding to an overall contamination fraction of 7. p. 5
Where the paper says it

22962479; the analysis pipeline is also maintained on GitHub17 . RESULTS We identify 527 satellite streak records across 492 unique HETDEX observations, out of 6778 total observations in PDR1, corresponding to an overall contamination fraction of 7.

3%. The streaks have a median gAB ∼ 16. p. 5
Where the paper says it

5 log10 tcross 360 s , (2) tabulated as ginst. Because tcross ≪ 360 s, ginst is brighter than gAB: over the 475 identified streaks the correction has a median of 6.

4 and µg ∼ 24. p. 6
Where the paper says it

4 with a standard deviation of 1. 1 mag, while the surface brightness distribution has a median value of µg = 24.

5 mag arcsec−2, but because each ∼360 s exposure far exceeds the satellite transit time, these are strict lower limits: Instantaneous magnitudes are brighter by 6 mag on average, with class medians of ginst = 7. p. 5
Where the paper says it

Because tcross ≪ 360 s, ginst is brighter than gAB: over the 475 identified streaks the correction has a median of 6. 2 mag, ranging from 8.

4 (LEO), 11.4 (MEO), 11.8 (HEO), and 13.0 (GEO). The spectra are consistent with reflected sunlight, despite affecting many observations, the streaks mask only ∼0. p. 0
Where the paper says it

04 reported for VisorSat-design spacecraft. The spectra are consistent with reflected sunlight, with modest, signal-to-noise-driven differences in slope.

09% of the survey area and can be sufficiently excluded from survey data. p. 0
Where the paper says it

3% that rises to monthly values above 20% by 2023–2024, although the streaks mask only 0. 09% of the survey area.

The fraction of affected observations rises from a few percent early in the survey to monthly values above 20% by 2023. p. 6
Where the paper says it

McDowell 2020) (right axis). The contamination fraction rises from a few percent early in the survey to monthly values above 20% by 2023, broadly tracking the growth of the low-Earth-orbit population.

Contamination is most frequent near twilight but remains significant at all hours, and this enhancement has strengthened over the survey—being nearly flat in 2017–2020, reaching ∼20% in the first hour after twilight by 2023–2024. p. 6
Where the paper says it

hours after astronomical twilight (when the Sun falls 18 deg below the horizon for HET), split into survey epochs (2017–2020, 2021–2022, 2023–2024; one-hour bins, binomial errors). Latter epochs show the highest contamination values near twilight, but the values remain significant all night.

The launch dates of the identified objects span 1963–2024 with a median of 1999, so the typical streak is produced by hardware ∼23 yr old at the time of observation (the 16th–84th percentile 3–39 yr). p. 7
Where the paper says it

Panels (c) and (d) show the same quantities as a function of time after astronomical twilight. Points show individual satellite streaks, black curves show binned medians, and shaded regions indicate the 16th–84th percentile range.

The distribution is two-peaked: a broad maximum through the 1980s and 1990s (the 40%), dominated by spent upper stages and defunct payloads, and a second rise after 2015 (the 25%) driven by the LEO constellations. The dominance of older, spent objects in our sample is not evidence that modern satellite constellations pose a lesser threat to astronomy.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Precise contamination modeling based on orbital dynamics can be implemented to predict contamination fractions in future surveys, as the fractional contamination does not rise as steeply as the total number of satellites in orbit.

  2. Integration of instantaneous magnitude corrections can allow AI systems to estimate the physically relevant quantity for assessing survey impact, such as instantaneous magnitudes are brighter by 6 mag on average when correcting for transit time.

  3. Satellite identification pipelines can be refined by incorporating cross-validation against independent services, as shown by the agreement between SGP4 propagation and SatChecker, which agree (97.0%).

  4. The system can be trained to recognize spectral signatures of reflected sunlight versus other sources, as noted in Section 3.3 where all stacks display clear solar spectrum features.

  5. AI systems can utilize the derived photometric properties to assess the intrinsic brightness of different orbit classes, noting that the brighter LEO population is increasing over time.

Abstract

We present a spectroscopic database and analysis of satellite streaks in Public Data Release 1 (PDR1) from the Hobby-Eberly Telescope Dark Energy Experiment (HETDEX), covering 6778 observations and about 90 deg squared of sky from January 2017 to July 2024. We detect 527 streaks across 492 observations, a contamination fraction of 7.3% that rises to monthly values above 20% by 2023--2024, although the streaks mask only 0.09% of the survey area. Of the 527 streaks, 475 (90.1%) are identified with a cataloged object using archival two-line elements: 258 (54%) lie in low or Earth orbit and 217 (46%) are geostationary or highly elliptical. Only 35% were launched since 2010; the rest are spent rocket bodies, debris, and older spacecraft. This reflects HETDEX's sensitivity to faint objects and its 2017--2024 observing baseline. The Hobby-Eberly Telescope's large 10,m aperture and HETDEX's about 6,minute exposures result in similar observed magnitudes for low Earth orbit (LEO) and geostationary Earth orbit (GEO) objects (median g AB=16.2 plus or minus1.1 for LEO, 16.7 plus or minus1.1 for GEO). However, once transit time is corrected for, instantaneous magnitudes differ widely: g inst=7.4 for LEO versus g inst=13.0 for GEO objects, a difference of 5.6,mag, while the exposure-averaged magnitudes differ by only 0.5 mag. Normalized to 550 km, the 34 Starlink streaks give g(550 km)=5.95, consistent with the mean V(550, km)=5.92 plus or minus0.04 reported for VisorSat-design spacecraft. The spectra are consistent with reflected sunlight, with modest, signal-to-noise-driven differences in slope. The catalog, spectra, and code are archived at https://doi.org/10.5281/zenodo.22962479.

Sources

Related papers