{"id":"81a78cd5-201a-405a-a2d1-89cee26dbd78","arxiv_id":"2506.11556","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A constructive heuristic that schedules Earth observation satellites using a profit score based on image resolution and data freshness outperforms a FIFO baseline, but the comparison is weak and the gains are partly by construction.","lead":"A satellite scheduling algorithm that adds a data-freshness score and frame-resolution score to each observation choice can reduce how unevenly satellites revisit targets, compared with a simple first-come-first-served baseline. The paper's headline gains, however, are largely built into the objective function it optimizes, and it does not compare against state-of-the-art schedulers.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claimed gains over FIFO are largely built into the objective (Eq. 16); without comparison against the cited priority-based heuristics, the central claim of algorithm superiority is unsubstantiated.","rationale":"The reader's rejection is based on the weak comparison against FIFO and the circularity of the profit function. My examination confirms this is the most load-bearing concern: the paper's headline numbers are the very quantities embedded in the objective, and the only baseline is a naive FIFO. No prior priority-based heuristics are run, so the algorithmic contribution is not isolated. I considered other potential flaws—such as the somewhat ad hoc local-search acceptance rule in Step 2 (comparing OC with rho) or the global normalizer delta_max in Eq. (16) being time-varying—but these are secondary and do not by themselves invalidate the approach. The missing baseline comparison is decisive because it directly undermines the causal claim 'our algorithm improves resolution/freshness.' The concrete test I propose would settle the issue: if the proposed method fails to beat the cited heuristics when they are given the same objective, then the reported gains are not an algorithmic advance. Therefore the reader's REJECT verdict stands, and I recommend no change.","tokens_in":8695,"tokens_out":2936,"duration_ms":30165,"concrete_test":"Implement the priority-based constructive heuristics from Xu et al. [5] and Chen et al. [12] using the same profit function (Eq. 16), the same constraints, and the same instances as Section VIII, with multiple random seeds per instance size; then compare total profit, GSD, and AoI variance. If the proposed method does not significantly outperform both baselines on a paired statistical test, the central claim of superiority is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that the proposed constructive heuristic plus local search yields high-quality monitoring: up to 10% better GSD and up to 83% lower variance in monitoring frequency compared to FIFO. However, the objective being maximized, Eq. (16), is exactly rho = (GSD_nadir/GSD) * (delta_t/delta_max), i.e., a product of the two metrics later reported as results. FIFO schedules by ascending VTW start and ignores both GSD and freshness, so the observed improvements are almost inevitable consequences of the objective definition rather than evidence that the algorithm is effective relative to informed alternatives. The paper itself cites priority-based constructive heuristics in [5] and [12] but never implements them as baselines, so there is no comparison against the relevant state of the art. Furthermore, Section VIII uses only one randomly generated instance per size and hand-set parameters, with no repeated trials or statistical testing. Thus the reported improvements cannot be attributed to the algorithmic design (priority ordering and local search); they may simply reflect that the optimizer's objective aligns with the evaluation metrics. Without a baseline that also optimises the same profit function, the paper does not establish that its heuristic outperforms existing methods at solving the AEOSSP.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":8984,"tokens_out":4101,"duration_ms":43615,"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":[{"comment":"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":"Section V, Eqs. (12)-(13)"},{"comment":"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":"Section VIII (Results)"},{"comment":"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":"Section VIII vs. Eq. (16)"},{"comment":"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.","section":"Section VIII (FIFO baseline)"}],"minor_comments":[{"comment":"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":"Abstract and Section VIII"},{"comment":"The typo 'constructive heuristicc' should be corrected to 'constructive heuristic'.","section":"Section VIII"},{"comment":"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":"Algorithm 1, line 5"},{"comment":"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":"Section V, Eq. (16)"},{"comment":"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.","section":"Section VIII, Figs. 4-6"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and addresses a timely problem, but the experimental methodology needs substantial work before the claims can be accepted. The circularity between the optimized objective and the evaluation metrics is a real concern, but it can be addressed by adding informed baselines and external metrics. I recommend a major revision rather than rejection because the central idea is defensible and the identified issues are fixable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","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.\n\nWhat'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.\n\nWhere 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.\n\nThat 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.","headline":"Reasonable heuristic, but the headline gains over FIFO are baked into the objective; without real baselines and repeated trials the claims don't hold.","tokens_in":9463,"tokens_out":2766,"would_cite":false,"duration_ms":26952,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"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…","keywords":["agile earth observation satellite","satellite scheduling","constructive heuristic","local search","age of information","ground sample distance","onboard processing","continuous monitoring"],"falsifier":"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.","tokens_in":8517,"feed_emoji":"🛰️","tokens_out":10198,"duration_ms":95016,"temperature":0.7,"pith_summary":"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.","feed_headline":"Satellite scheduler lifts image resolution 10%, cuts revisit gaps 83%","feed_subtitle":"A heuristic that weighs sharpness and freshness keeps hundreds of ground targets in regular, high-resolution view.","key_machinery":"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.","core_discovery":"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%.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"It supplies the satellite edge-computing model of onboard compression used to define processed-data size and processing time in the feasibility constraints.","marker":"[1]"},{"why":"It defines the Agile Earth Observation Satellite Scheduling Problem and the three-axis agility that creates flexible visible time windows.","marker":"[2]"},{"why":"It establishes ground sample distance as the standard spatial-resolution indicator used in the profit formula.","marker":"[3]"},{"why":"It provides the priority-based constructive heuristic approach and the opportunity-cost conflict measure the algorithm builds on.","marker":"[5]"},{"why":"It treats oversubscribed-target scheduling with a feedback heuristic, the scenario closest to the continuous monitoring setting addressed here.","marker":"[6]"},{"why":"It supplies the Age-of-Information freshness objective for onboard-processing satellite constellations and the compression factors used in simulations.","marker":"[8]"},{"why":"It defines Age of Information and its peak variant used to evaluate monitoring freshness and variance.","marker":"[9]"},{"why":"It gives the piecewise-linear attitude transition time used to enforce maneuver feasibility between consecutive observations.","marker":"[11]"},{"why":"It defines assignment flexibility, the tie-breaking priority indicator that ranks targets with the same staleness.","marker":"[12]"}],"fun_headline_variants":["Satellite scheduling heuristic sharpens images 10%, evens revisit gaps 83%","Agile satellite planner improves resolution 10%, cuts revisit variance 83%","Onboard-aware satellite scheduler: sharper images 10%, steadier revisits 83%","Satellite scheduler with onboard computing: +10% fidelity, -83% revisit spread"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Satellite scheduling heuristic sharpens images 10%, evens revisit gaps 83%","Agile satellite planner improves resolution 10%, cuts revisit variance 83%","Onboard-aware satellite scheduler: sharper images 10%, steadier revisits 83%","Satellite scheduler with onboard computing: +10% fidelity, -83% revisit spread"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001157,"raw_usage":{"total_tokens":4794,"prompt_tokens":948,"completion_tokens":3846,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":564,"completion_tokens_details":{"reasoning_tokens":3756}},"tokens_in":564,"tokens_out":3846,"duration_ms":25327,"temperature":1.0,"reasoning_tokens":3756,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:04:33.531242+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[{"cited_title":"Leyva-Mayorga, M","cited_arxiv_id":null,"evidence_quote":"It supplies the satellite edge-computing model of onboard compression used to define processed-data size and processing time in the feasibility constraints."},{"cited_title":"Agile Earth Observation Satellite Scheduling Over 20 Years: Formulations, Methods, and Future Directions,","cited_arxiv_id":null,"evidence_quote":"It defines the Agile Earth Observation Satellite Scheduling Problem and the three-axis agility that creates flexible visible time windows."},{"cited_title":"Comprehensive Evaluation of the GF-4 Satellite Image Quality from 2015 to 2020,","cited_arxiv_id":null,"evidence_quote":"It establishes ground sample distance as the standard spatial-resolution indicator used in the profit formula."},{"cited_title":"Priority-based constructive algorithms for scheduling agile Earth observation satellites with total priority maximization,","cited_arxiv_id":null,"evidence_quote":"It provides the priority-based constructive heuristic approach and the opportunity-cost conflict measure the algorithm builds on."},{"cited_title":"Scheduling multiple agile Earth observation satellites for oversubscribed targets using complex networks theory,","cited_arxiv_id":null,"evidence_quote":"It treats oversubscribed-target scheduling with a feedback heuristic, the scenario closest to the continuous monitoring setting addressed here."},{"cited_title":"Goal-oriented vessel detection with distributed computing in a LEO satellite constella- tion,","cited_arxiv_id":null,"evidence_quote":"It supplies the Age-of-Information freshness objective for onboard-processing satellite constellations and the compression factors used in simulations."},{"cited_title":"An adaptive large neigh- borhood search metaheuristic for agile satellite scheduling with time- dependent transition time,","cited_arxiv_id":null,"evidence_quote":"It gives the piecewise-linear attitude transition time used to enforce maneuver feasibility between consecutive observations."},{"cited_title":"Priority-based and conflict-avoidance heuristics for multi-satellite scheduling,","cited_arxiv_id":null,"evidence_quote":"It defines assignment flexibility, the tie-breaking priority indicator that ranks targets with the same staleness."}],"review_version":1}