Scheduling Agile Earth Observation Satellites with Onboard Processing and Real-Time Monitoring

Antonio Jurado-Navas; Antonio M. Mercado-Mart\'inez; Beatriz Soret

REVIEW 4 major objections 5 minor 12 references

Scheduling Agile Earth Observation Satellites with Onboard Processing and Real-Time Monitoring

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A constructive heuristic refined by local search can schedule agile Earth-observation satellites for continuous monitoring, improving average frame resolution by up to 10% and cutting variance in monitoring frequency by up to 83% compared…

desk verdict Reasonable heuristic, but the headline gains over FIFO are baked into the objective; without real baselines and repeated trials the claims don't hold. read the letter →

arxiv 2506.11556 v1 pith:34VK2MS5 submitted 2025-06-13 cs.NI cs.RO

Antonio M. Mercado-Martínez , Beatriz Soret , Antonio Jurado-Navas This is my paper · ORCID

classification cs.NIcs.RO

keywords agileearthobservationsatelliteschedulingconstructiveheuristiclocalsearchageofinformationgroundsampledistanceonboardprocessingcontinuousmonitoring

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that a centralized constructive heuristic, improved by a local-search cleanup, can handle the scheduling problem of agile Earth-observation satellites when data are compressed on board and delivered to ground stations for continuous monitoring. The central move is to price each possible observation time window by a single profit number that multiplies image sharpness by data freshness. The authors report that this approach schedules every target in all tested instances with 1,000 to 1,800 targets, improves average frame resolution by up to 10%, and lowers the variance of monitoring frequency by up to 83% relative to a FIFO scheduler. If it holds, the practical result is that a small constellation could keep a large target set in fresher and sharper view without waiting for raw image downloads.

What carries the argument

The engine is the profit indicator $\rho_{s,t,o,w} = (GSD_{\mathrm{nadir}}/GSD_{s,t,o,w})(\delta_t/\delta_{\max})$: a number in $[0,1]$ that is large when a frame is sharp (GSD near nadir) and the target has waited long (large $\delta_t$). Supporting it are three priority indicators—staleness $\delta_t$, assignment flexibility $FL_t$, and opportunity cost—which order targets and time windows in both the constructive pass and the local-search insertion/removal pass. The $\delta_t/\delta_{\max}$ factor injects Age-of-Information thinking into the objective, and the feasibility checks fold compression time and energy directly into the schedule.

What would settle it

A concrete experiment: rerun the reported 1000–1800-target instances with a second baseline that already maximizes the same GSD and staleness priorities but uses a different ranking rule; if that baseline also matches the 10% resolution gain and 83% variance drop, the claim that this particular constructive heuristic causes the gains is falsified. Alternatively, set the freshness factor in Eq. (16) to a constant and check whether the variance reduction disappears.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that continuous-monitoring AEOSSP becomes solvable by a fast greedy construct-and-repair method once profit is defined as $\rho_{s,t,o,w} = (GSD_{\mathrm{nadir}}/GSD_{s,t,o,w})(\delta_t/\delta_{\max})$. This ratio makes physical image quality and information staleness commensurable in one number between 0 and 1. The algorithm sorts targets by their unobserved periods $\delta_t$, breaks ties by assignment flexibility, and chooses time windows by opportunity cost; then a local search tries to insert unscheduled targets by removing lower-profit conflicting observations. Feasibility checks explicitly include the CPU time of onboard compression and the satellite's energy budget. The reported consequence is that, compared with FIFO, the method observes every target in the simulated 1000-1800-target instances, reduces the mean GSD by up to 10%, and cuts the variance of average Age of Information across targets by up to 83%.

Load-bearing premise

The entire comparison rests on taking FIFO as the reference scheduler and on a small set of synthetic instances with hand-picked CPU, energy, and discretization values; if those choices are not representative, the reported margins may not transfer to real operations.

Editorial extensions

If this is right

Targets that have waited longest jump to the front of the schedule, so revisit intervals become more even across the whole set rather than concentrated on a few targets.
Because sharper (lower-GSD) windows earn more profit, the schedule tends to capture each target near nadir, which is what produces the average resolution gain.
Feasibility constraints treat onboard compression as a time and energy cost per frame, so schedules are valid for satellites that process data before downlink.
The local-search step improves the greedy schedule by swapping in unscheduled targets only when total profit rises, keeping the final plan consistent with the same quality-plus-freshness objective.
In the tested constellation the method covers all targets in instances up to 1,800 targets, suggesting the approach scales beyond small demonstration sizes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper does not run any earlier priority-based heuristic as a second baseline; a direct follow-up would be to replace FIFO with a scheduler that uses the same staleness or GSD priorities alone, which would isolate how much of the gain comes from combining both terms in $\rho$.
Because $\delta_t/\delta_{\max}$ is relative to the current worst-off target in the instance, the same profit formula could drive online replanning after cloud cover or a missed pass without changing the objective.
The 83% variance reduction is reported on average Age of Information; the paper's own 99th-percentile PAoI tables suggest a prediction that peak-age tails also tighten, which is not separately tested.
Operators could tune the quality-freshness tradeoff by changing the exponents on the two factors of Eq. (16); the paper uses equal weight one and does not explore that parameter sweep.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper is a competent but not convincing AEOSSP scheduling heuristic that adds onboard processing and AoI-based freshness to the profit function. The evaluation, however, doesn't support the headline numbers: the two reported gains—GSD improvement and variance reduction—are exactly the terms in Eq. (16)'s profit, so comparing against FIFO only shows that optimizing a metric beats ignoring it. Without running the cited priority heuristics [5], [12] as baselines, the claim that the algorithm is effective is unsubstantiated.

What's actually new: integrating data processing into the scheduling loop and defining a time-dependent profit combining GSD and AoI is a sensible and relevant extension for continuous monitoring constellations. The priority indicators—δ_t, assignment flexibility, opportunity cost—are reasonable, and the local search refinement is standard but sound. The system model includes attitude transitions, energy, and processing time, which is more realistic than many scheduling papers.

Where it falls short: the simulation uses one random instance per size, no repeated trials, no error bars, so the 10% and 83% numbers could be noise. The FIFO baseline is a straw man. More importantly, because the profit function literally multiplies the GSD ratio and the freshness ratio, the improvement over FIFO is by construction, not evidence of algorithmic skill. The MILP in Eqs. (10)-(15) is also under-specified: Eqs. (12)-(13) reference y_t,w,t',w' but there's no linking constraint forcing y to be consistent with x, so the formal model is incomplete. The results section gives no runtime or sensitivity analysis, and the LS improvement is asserted rather than shown in detail.

That said, the paper is coherent on its own terms and the combination of AoI and GSD in a multi-satellite scheduler is worth addressing in the literature. It's not a waste of time, but it needs a real baseline and more careful experiments before the claims hold. I'd send it to review, but expect major revisions.

Referee Report

4 major / 5 minor

Summary. The paper addresses the Agile Earth Observation Satellite Scheduling Problem (AEOSSP) with onboard processing and continuous monitoring. It defines a profit function combining ground sample distance (GSD) and age-of-information (AoI), proposes priority indicators (freshness, flexibility, opportunity cost), and builds a constructive heuristic with a local search refinement. The simulation compares the proposed method against a FIFO scheduler and reports up to 10% better average GSD and up to 83% lower variance in monitoring frequency. The central claim is that the proposed heuristic provides high-quality, fresh information for all targets.

Significance. If the claimed results were substantiated, the paper would make a useful contribution by integrating onboard processing and AoI into a multi-satellite scheduling framework and by showing that freshness-aware priority rules can improve monitoring regularity. The problem formulation is relevant, the priority indicators are clearly motivated, and the pseudocode makes the algorithm easy to reimplement. However, the experimental design currently provides only a weak basis for the central claims: the comparison is limited to FIFO, there are no repeated random instances or statistical tests, and the evaluated metrics are largely the same as the optimized objective.

major comments (4)

[Section V, Eqs. (12)-(13)] The mathematical formulation is incomplete: the binary variables x and y are never linked. There is no constraint enforcing that y_{t,w,t',w'} can be 1 only when both corresponding x variables are 1 and the two observations are consecutive on the same satellite. As written, the energy constraint (13) can always be satisfied by setting all y to 0, and Eq. (12) is a universal condition over all OTW pairs that does not depend on whether either observation is scheduled. This means the MILP does not correctly represent the AEOSSP, and the feasibility checks inside Algorithm 1 are not grounded in the stated model. The formulation needs explicit linking constraints (e.g., y ≤ x, y ≤ x', and y ≥ x + x' − 1) and sequencing constraints before it can be claimed that the algorithm solves the modeled problem.
[Section VIII (Results)] The evaluation is based on a single randomly generated instance per problem size (1000, 1200, 1400, 1600, 1800 targets), with no repeated trials, no error bars over independent instances, and no statistical tests. The reported gains of 'up to 10%' in GSD and 'up to 83%' in variance hence cannot be assessed for significance or stability. The paper should run multiple random instances per size, report confidence intervals or boxplots across instances, and ideally compare runtimes.
[Section VIII vs. Eq. (16)] The comparison against FIFO is not sufficient to support the claim of algorithmic superiority. The objective in Eq. (16) is exactly a product of the GSD ratio and the freshness ratio δ_t/δ_max, and Section VI declares δ_t the primary sorting criterion. The evaluation then reports GSD and variance of monitoring frequency as the headline results. Against FIFO, which ignores both factors, the observed improvements are largely a consequence of optimizing a profit function that incorporates those very metrics. To substantiate the claim, the paper should compare against state-of-the-art priority-based heuristics (e.g., the methods in [5] and [12]) that also aim to maximize profit, and/or evaluate with external metrics not contained in the objective. Otherwise, the paper only demonstrates that the heuristic maximizes its own objective, not that it outperforms informed alternatives.
[Section VIII (FIFO baseline)] The paper introduces FIFO as the sole reference but does not justify why FIFO is the appropriate baseline for a continuous-monitoring scenario. Since the cited related work includes several priority-based constructive heuristics designed for the AEOSSP, at least one of those should be implemented as a baseline. Without such a comparison, the central assertion that the proposed algorithm provides 'high-quality information' is not established relative to the existing state of the art.

minor comments (5)

[Abstract and Section VIII] The phrase 'increasing the resolution ... by up to 10% on average' is imprecise: an average is a single number, not a range. The text should report the actual mean and, if desired, the range over instances.
[Section VIII] The typo 'constructive heuristicc' should be corrected to 'constructive heuristic'.
[Algorithm 1, line 5] The loop 'for STP in STH' should be written 'for each STP in the set of STPs' to avoid treating a time horizon as an iterable set.
[Section V, Eq. (16)] The definition of δ_max as 'the δt value of the target that has remained unobserved for the longest time in the instance' should specify whether δ_max is updated per STP or fixed for the whole horizon, since the normalization affects every profit value in the schedule.
[Section VIII, Figs. 4-6] The boxplots appear to summarize within-instance variation across targets or frames, not variation across independent simulations. The captions should state exactly what distribution each boxplot represents.

Circularity Check

1 steps flagged · score 6.0 of 10

The reported GSD and monitoring-frequency gains recapitulate the optimized profit in Eq. (16), so the headline results are partially circular and do not independently validate the algorithm.

self definitional [Section V, Eq. (16); Section VI; Section VIII, Figs. 4-5]
"ρs,t,o,w = GSDnadir / GSDs,t,o,w * δt / δmax ... In our scenario, the primary priority indicator is δt, as it directly correlates with the AoI. A higher δt value indicates higher priority. ... The solution of the proposed method exhibits lower variance ... Specifically, a variance reduction of up to 83 % is achieved."

Equation (16) defines the observation profit that Eq. (10) maximizes as the product of the GSD ratio and the freshness ratio δt/δmax. Section VI makes δt the primary sorting indicator, and the algorithm groups targets by δt and sorts OTWs by opportunity cost computed from ρ. The two headline metrics in Section VIII — GSD improvement of up to 10% and variance reduction of up to 83% — are exactly the two components of ρ. A scheduler that explicitly maximizes ρ will, by construction, improve those metrics relative to FIFO, which ignores both GSD and freshness. Thus the numerical results do not independently demonstrate algorithmic quality; they restate the objective being optimized.

full rationale

The central circular step is the alignment between the optimized objective and the reported success metrics. The profit in Eq. (16) is deliberately defined as a product of spatial-resolution quality (GSD_nadir/GSD) and information freshness (δt/δmax), and the constructive heuristic sorts primarily by δt. The results then report GSD reduction and monitoring-frequency variance reduction as evidence of high-quality information; these are the same quantities that the algorithm is built to maximize, so the improvements over FIFO are entailed by the objective design rather than being independent evidence of algorithmic superiority. This is partial circularity, but it is not complete: the algorithm still has to produce feasible schedules under attitude, energy, and processing constraints, and the comparison to FIFO is a real, if weak, baseline. There is no load-bearing self-citation chain: the AoI definition cites the authors' prior work [8], but AoI is a standard metric also supported by [9], so that citation is not what forces the conclusion. The absence of the cited priority-based heuristics [5], [12] as baselines is a correctness and benchmarking limitation rather than a circularity by itself. Overall, the headline quantitative claims reduce to the objective definition, warranting a score of 6 rather than a higher score.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

No parameters are statistically fitted to data; all listed constants are hand-set simulation inputs. The central claim relies chiefly on the ad hoc profit definition in Eq. (16) and on modeling assumptions about processing, attitude dynamics, and target distribution.

free parameters (5)

compression_factor_sigma = 10
Hand-set in Section VIII; used in Eq. (3) to compute processed data D'.
compression_complexity_C = 100 cycles/bit
Hand-set in Section VIII; used in Eq. (4) to compute processing time and affects schedule feasibility.
energy_budget_Emax = 5000 units
Hand-set in Section VIII; acts as the bound in Eq. (13) and determines which schedules are feasible.
OTW_discretization_prc = 10 seconds
Hand-set in Section VIII; defines the OTW grid and therefore the opportunity cost in Eq. (17).
cpu_cores_and_frequency = 8 cores, 1.8 GHz
Hand-set in Section VIII; used in Eq. (4) for processing time.

assumptions (5)

domain assumption Storage constraints are neglected because data is processed dynamically and transmitted when ground stations are reachable.
Section V states storage constraints are neglected; this fails if downlink opportunities are sparse.
domain assumption The piecewise-linear attitude transition function in Eq. (1) accurately models agile satellite maneuvers.
Eq. (1) is the sole source of transition times; if it is inaccurate, schedule feasibility and profits change.
ad hoc to paper The observation profit in Eq. (16) is a valid measure of value for continuous monitoring.
The product of the GSD ratio and delta_t/delta_max is introduced without justification and embeds the evaluation metrics.
domain assumption Onboard processing time follows Eq. (4) with a constant complexity per bit.
This abstracts real compression algorithms but is not validated against actual codecs in the paper.
domain assumption Targets are uniformly distributed over the coverage area.
Section VIII uses uniform distribution; realistic clustered targets could change scheduling outcomes.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Scheduling Agile Earth Observation Satellites with Onboard Processing and Real-Time Monitoring." pith.science (2026). https://pith.science/paper/34VK2MS5

@misc{pith2026250611556,
  author       = {Pith},
  title        = {Pith review of: Scheduling Agile Earth Observation Satellites with Onboard Processing and Real-Time Monitoring},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/34VK2MS5}},
  note         = {Machine review of arXiv:2506.11556}
}

read the original abstract

The emergence of Agile Earth Observation Satellites (AEOSs) has marked a significant turning point in the field of Earth Observation (EO), offering enhanced flexibility in data acquisition. Concurrently, advancements in onboard satellite computing and communication technologies have greatly enhanced data compression efficiency, reducing network latency and congestion while supporting near real-time information delivery. In this paper, we address the Agile Earth Observation Satellite Scheduling Problem (AEOSSP), which involves determining the optimal sequence of target observations to maximize overall observation profit. Our approach integrates onboard data processing for real-time remote monitoring into the multi-satellite optimization problem. To this end, we define a set of priority indicators and develop a constructive heuristic method, further enhanced with a Local Search (LS) strategy. The results show that the proposed algorithm provides high-quality information by increasing the resolution of the collected frames by up to 10% on average, while reducing the variance in the monitoring frequency of the targets within the instance by up to 83%, ensuring more up-to-date information across the entire set compared to a First-In First-Out (FIFO) method.

Figures

Figures reproduced from arXiv: 2506.11556 by the authors.

**Figure 1.** (a) Scenario with multiple AEOSs tasked with observing multiple targets, processing and transmitting the acquired () [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Evolution of the AoI of target t. where ξs,t,o,w is the swath width, i.e, the width of the observable area on the surface of the Earth, determined by the field of view (FoV) of the sensor and the relative position between target t and satellite s during OTWs,t,o,w; GSDs,t,o,w is the GSD for OTWs,t,o,w; RE is the Earth’s radius; ω is the satellite angular velocity; and qpx is the bit depth of a pixel in the frame. E… view at source ↗

**Figure 5.** Boxplot of the average AoI of in the targets in the [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

Figures from the paper (1 more)

**Figure 6.** Figure 6: Boxplot of the average PAoI of the targets in the [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 11 canonical work pages

[5]

Priority-based constructive algorithms for scheduling agile Earth observation satellites with total priority maximization,

R. Xu, H. Chen, X. Liang, and H. Wang, “Priority-based constructive algorithms for scheduling agile Earth observation satellites with total priority maximization,” Expert Systems with Applications , vol. 51, pp. 195–206, 2016

work page 2016
[12]

Priority-based and conflict-avoidance heuristics for multi-satellite scheduling,

X. Chen, G. Reinelt, G. Dai, and M. Wang, “Priority-based and conflict-avoidance heuristics for multi-satellite scheduling,”Applied Soft Computing, vol. 69, pp. 177–191, 2018

work page 2018
[1]

Leyva-Mayorga, M

I. Leyva-Mayorga, M. Martinez-Gost, M. Moretti, A. P ´erez-Neira, M. A. V ´azquez, P. Popovski, and B. Soret, “Satellite edge computing Fig. 5: Boxplot of the average AoI of in the targets in the instance that have been captured by the end of the STH. Fig. 6: Boxplot of the average PAoI of the targets in the instance that have been captured by the end of ...

work page 2023
[2]

Agile Earth Observation Satellite Scheduling Over 20 Years: Formulations, Methods, and Future Directions,

X. Wang, G. Wu, L. Xing, and W. Pedrycz, “Agile Earth Observation Satellite Scheduling Over 20 Years: Formulations, Methods, and Future Directions,” IEEE Systems Journal , vol. 15, no. 3, p. 3881–3892, Sep. 2021

work page 2021
[3]

Comprehensive Evaluation of the GF-4 Satellite Image Quality from 2015 to 2020,

W. Yi et al. , “Comprehensive Evaluation of the GF-4 Satellite Image Quality from 2015 to 2020,” ISPRS International Journal of Geo- Information, vol. 10, no. 6, 2021

work page 2015
[4]

H. Chen, S. Peng, C. Du, and J. Li, Earth observation satellites : task planning and scheduling , 1st ed. Singapore: Springer Nature Singapore, 2023

work page 2023
[6]

Scheduling multiple agile Earth observation satellites for oversubscribed targets using complex networks theory,

X. Wang, C. Han, R. Zhang, and Y . Gu, “Scheduling multiple agile Earth observation satellites for oversubscribed targets using complex networks theory,” IEEE Access , vol. 7, pp. 110 605–110 615, 2019

work page 2019
[7]

Optimal bandwidth and computing resource allocation in low Earth orbit satellite constellation for earth observation applications,

F. Valente, V . Eramo, and F. G. Lavacca, “Optimal bandwidth and computing resource allocation in low Earth orbit satellite constellation for earth observation applications,” Computer Networks , vol. 232, p. 109849, 2023

work page 2023

Show all 12 references

[8]

Goal-oriented vessel detection with distributed computing in a LEO satellite constella- tion,

A. Mercado-Mart ´ınez, B. Soret, and A. Jurado-Navas, “Goal-oriented vessel detection with distributed computing in a LEO satellite constella- tion,” in 2024 6th International Conference on Communications, Signal Processing, and their Applications (ICCSPA) , 2024, pp. 1–5

2024
[9]

Age of information: A new concept, metric, and tool,

A. Kosta, N. Pappas, and V . Angelakis, “Age of information: A new concept, metric, and tool,” F oundations and Trends® in Networking , vol. 12, no. 3, pp. 162–259, 2017

2017
[10]

Leyva-Mayorga et al

I. Leyva-Mayorga et al. , Non-Geostationary Satellite Communications Systems. Institution of Engineering and Technology, Dec. 2022

2022
[11]

An adaptive large neigh- borhood search metaheuristic for agile satellite scheduling with time- dependent transition time,

X. Liu, G. Laporte, Y . Chen, and R. He, “An adaptive large neigh- borhood search metaheuristic for agile satellite scheduling with time- dependent transition time,” Computers & Operations Research , vol. 86, pp. 41–53, 2017

2017

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [5]

Priority-based constructive algorithms for scheduling agile Earth observation satellites with total priority maximization,

R. Xu, H. Chen, X. Liang, and H. Wang, “Priority-based constructive algorithms for scheduling agile Earth observation satellites with total priority maximization,” Expert Systems with Applications , vol. 51, pp. 195–206, 2016

work page 2016

[2] [12]

Priority-based and conflict-avoidance heuristics for multi-satellite scheduling,

X. Chen, G. Reinelt, G. Dai, and M. Wang, “Priority-based and conflict-avoidance heuristics for multi-satellite scheduling,”Applied Soft Computing, vol. 69, pp. 177–191, 2018

work page 2018

[3] [1]

Leyva-Mayorga, M

I. Leyva-Mayorga, M. Martinez-Gost, M. Moretti, A. P ´erez-Neira, M. A. V ´azquez, P. Popovski, and B. Soret, “Satellite edge computing Fig. 5: Boxplot of the average AoI of in the targets in the instance that have been captured by the end of the STH. Fig. 6: Boxplot of the average PAoI of the targets in the instance that have been captured by the end of ...

work page 2023

[4] [2]

Agile Earth Observation Satellite Scheduling Over 20 Years: Formulations, Methods, and Future Directions,

X. Wang, G. Wu, L. Xing, and W. Pedrycz, “Agile Earth Observation Satellite Scheduling Over 20 Years: Formulations, Methods, and Future Directions,” IEEE Systems Journal , vol. 15, no. 3, p. 3881–3892, Sep. 2021

work page 2021

[5] [3]

Comprehensive Evaluation of the GF-4 Satellite Image Quality from 2015 to 2020,

W. Yi et al. , “Comprehensive Evaluation of the GF-4 Satellite Image Quality from 2015 to 2020,” ISPRS International Journal of Geo- Information, vol. 10, no. 6, 2021

work page 2015

[6] [4]

H. Chen, S. Peng, C. Du, and J. Li, Earth observation satellites : task planning and scheduling , 1st ed. Singapore: Springer Nature Singapore, 2023

work page 2023

[7] [6]

Scheduling multiple agile Earth observation satellites for oversubscribed targets using complex networks theory,

X. Wang, C. Han, R. Zhang, and Y . Gu, “Scheduling multiple agile Earth observation satellites for oversubscribed targets using complex networks theory,” IEEE Access , vol. 7, pp. 110 605–110 615, 2019

work page 2019

[8] [7]

Optimal bandwidth and computing resource allocation in low Earth orbit satellite constellation for earth observation applications,

F. Valente, V . Eramo, and F. G. Lavacca, “Optimal bandwidth and computing resource allocation in low Earth orbit satellite constellation for earth observation applications,” Computer Networks , vol. 232, p. 109849, 2023

work page 2023