{"id":"0168dd40-c227-4b52-9d37-5f7c635402b2","arxiv_id":"2411.17626","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":2.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"A cannonball-model solar radiation pressure calculator is paired with linear regression to predict simulated LEO satellite position changes, with only self-generated validation.","lead":"The paper combines the standard cannonball model for sunlight pressure on a low Earth orbit satellite with a linear regression that predicts how the satellite's position changes. It is a small simulation study aimed at quick-look mission planning, but its validation is mostly internal and its code is not yet available.","discovery_kind":"incremental","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reported 0.0266% MAPE does not validate the model: training and test data both come from the same deterministic Eq. 10 generator, so linear regression is only fitting its own synthetic curve.","rationale":"The paper's central quantitative claim rests on the 0.0266% MAPE. My stress-test focuses on the validity of that number as evidence. Section 4 reveals that the dataset is generated internally by Eq. 10 and then split randomly; thus the test set is not independent. Because the output variable Z is nearly linear in the sole varying input (a_SRP), linear regression fitting its own generator is expected to produce a near-zero error. This circularity is sufficient to reject the claim that the model has been validated, regardless of the physical assumptions. The reader's weakest_assumption (perpendicular-only SRP, circular orbit) is also a valid concern and would undermine the physical realism of the synthetic generator; I partially agree with it. My proposed test—evaluation on an independent high-fidelity propagator—would settle whether the ML model has any real predictive power. The paper does nothing to support the central claim beyond the circular metric and an unsubstantiated HPOP mention. Credit should be given for the standard cannonball formula and the Orbitron ground-track comparison, but these do not validate the SRP perturbation or ML components. Therefore the reader's REJECT verdict stands unchanged.","tokens_in":11150,"tokens_out":9710,"duration_ms":84435,"concrete_test":"Retrain or load the paper's linear regression model (features: a_SRP, A/M, M) and test it on position vectors from an independent high-fidelity propagator—e.g., GMAT or STK HPOP—for the same 550 km, 98.6° inclination orbit over the same year, using the real Sun ephemeris, the full satellite-to-Sun SRP vector from Eq. 9, J2, drag, and no artificial a_SRP increment. Compute the MAPE between the predicted and propagated Z (or x,y,z) coordinates. If the MAPE is orders of magnitude larger than 0.0266%, the reported error is a circularity artifact and the central claim fails; if the MAPE remains comparably small, the model would have demonstrated genuine predictive skill.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 4, the authors generate the ML dataset by taking the computed a_SRP values, artificially incrementing them, and applying Eq. 10 to obtain perturbed orbital elements and position vectors. This dataset is then randomly split 80/20 into training and test sets. Because both sets are outputs of the same deterministic semi-analytical model, the test set is not an independent check. Figure 8 shows the target Z is nearly a straight line as a function of a_SRP, the only varying feature, so a linear regressor will trivially achieve a tiny MAPE on this self-generated data. Thus the quoted MAPE of 0.02662153353165422% demonstrates only that the regressor can fit the generating function; it says nothing about whether the SRP model or the ML model can predict real satellite positions. The only claimed external validation is a comparison with MATLAB HPOP, but the paper provides no comparison figure, data, or error metric, so it cannot break the circularity. This is load-bearing because the MAPE is the central claim's only quantitative evidence. Additionally, the generator itself is physically incomplete: Eq. 9 defines SRP acceleration along the satellite-to-Sun vector, which has radial, tangential, and normal components, but the perturbation analysis disregards radial and tangential components and keeps only the perpendicular W, so even the synthetic ground truth is not a faithful SRP-perturbed trajectory.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a custom Python module for computing solar radiation pressure (SRP) acceleration on a LEO satellite using the cannonball model, then analyzes orbital perturbations (restricted to inclination change under the assumption of a circular orbit and perpendicular-only SRP), compares ground station revisit times, and trains a linear regression model to predict position-vector changes from the SRP acceleration. The central quantitative claim is a MAPE of 0.0266% for predicted Z position, and the authors state validation against MATLAB's HPOP and Orbitron ground tracks.","tokens_in":11483,"tokens_out":5620,"duration_ms":42680,"significance":"If fully supported, the work would offer a lightweight, accessible SRP module and a fast ML surrogate for trajectory perturbations, which could be useful for LEO mission planning and education. Strengths include the use of standard astrodynamics references, a complete pipeline from ephemeris data to ground-track visualization, and detailed revisit-time calculations. However, the paper's main load-bearing validation is circular, and the missing independent comparison makes the claimed accuracy unverifiable; the modeling assumptions exclude radial and tangential SRP components, which are physically important in LEO.","major_comments":[{"comment":"The reported MAPE of 0.0266% does not validate the model, because the training and test data are both generated by the same deterministic semi-analytical model (Section 4, Eq. 10) from the same a_SRP feature, so the linear regressor is only fitting the generator's own curve; the near-linear relation in Figure 8 makes this result tautological. The claimed comparison with MATLAB HPOP (Section 3, Figure 6 caption) is not shown anywhere in the manuscript, so it cannot serve as an independent check.","section":"Section 4, Figure 8"},{"comment":"The notation is internally inconsistent: the text states that R_geos denotes 'the geocentric position of the Earth,' but the denominator |r - R_geos|^3 and the SRP context require R_geos to be the geocentric position of the Sun; moreover, the roles of the barycentric Earth and Moon coordinates in the bullets are not explained, making the calculation irreproducible from the text.","section":"Section 3, Eq. (9)"},{"comment":"The perturbation analysis discards radial and tangential SRP components and assumes a perfectly circular orbit, so only inclination changes; however, Eq. (9) produces an acceleration along the Sun-satellite vector that has nonzero radial and tangential components, which in LEO drive significant changes in semi-major axis and eccentricity. The training dataset is therefore not a faithful SRP-perturbed trajectory, and the ML model trained on it cannot capture the dominant SRP effects.","section":"Section 4, Eq. (10)"},{"comment":"The statement that the state vectors and trajectory 'were validated through simulations in MATLAB using identical parameters' is not supported by any figure, table, or error metric in the paper, so the trajectory validation is unverifiable; the comparison with Starlink-4566/Orbitron is also only visual.","section":"Section 2.1"},{"comment":"The ML feature set is limited to a_SRP (with mass and area-to-mass ratio held constant), and the target is the Z position for a circular orbit with only inclination perturbed; this setup does not support the abstract's claim that the model 'can adaptively learn and refine predictions based on historical data and real-time input,' and no generalization test is provided.","section":"Section 4"}],"minor_comments":[{"comment":"There are numerous typographical errors, including 'effect' for 'effect', 'pertuberbed' in Section 4, 'alogorithm' and 'comaprison' in Section 2.1, and 'the the JPL' in Section 3.","section":"Throughout"},{"comment":"Abstract: 'it's effects' should be 'its effects.'","section":"Abstract"},{"comment":"Section 3: 'The ephemeris was obtained from the the Jet Propulsion Laboratory' contains a duplicated 'the.'","section":"Section 3"},{"comment":"Section 4: The MAPE is quoted to an excessive number of significant figures (0.02662153353165422%) without any uncertainty estimate.","section":"Section 4"},{"comment":"The GitHub link in the abstract is public, but Section 1 states that access will be granted upon request after review; this inconsistency prevents reviewers from checking the code during the review process.","section":"Section 1"},{"comment":"Section 4: The value a_SRP = 0.00994 km/day^2 is used in Figure 7, but the units of Eq. (9) are not stated; consistent SI or specified units should be used throughout.","section":"Section 4"}],"recommendation":"reject","confidential_remarks":"The central flaw is the circular validation of the ML model and the absence of the claimed HPOP comparison. The physical model's restriction to perpendicular SRP, while noted in the text, is not justified and omits major SRP effects. The authors would need substantial new validation work (independent numerical propagators, real TLE data, and inclusion of radial/tangential components) to support the claims; this is beyond a minor revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a tidy walkthrough of orbital elements, ground track generation, and a revisit-time comparison for Indian ground stations, with a visual check against a Starlink TLE. That part is fine as a tutorial. But the core claim - that the 0.0266% MAPE validates a machine-learning position predictor - does not hold up. The ML dataset is generated by incrementing the same a_SRP values and feeding them through the same Eq. 10 used to produce the ground truth, so the test set is not independent. Figure 8 shows Z varying nearly linearly with a_SRP, so linear regression trivially fits its own generator. The MAPE is a self-consistency check, not a validation. The physical model also has a load-bearing inconsistency: Eq. 9 gives the full SRP acceleration vector, but the perturbation analysis keeps only the perpendicular component W and says radial and tangential are disregarded. The paper never explains how W is extracted from Eq. 9 or why those components would vanish for realistic Sun-satellite geometry. The claim that inclination is the only affected element follows only under that assumption, which is not justified for LEO. The HPOP validation is asserted but never shown - no figure, no error metric - so the only external check is the ground track comparison, which is visual. There is also a notation slip: R_geos is described as the geocentric position of the Earth rather than the Sun, and the motivating SpaceX example misattributes the Starlink losses to SRP when the public record points to atmospheric drag from a geomagnetic storm. Credit where due: the authors are transparent about their simplifications (no eclipses, circular orbit), the code is on GitHub, and the revisit-time table could be useful for a first-pass mission design exercise. But the central validation is circular, and the physical inconsistency between Eq. 9 and the perpendicular-only assumption is not minor. Who gets value from this? Maybe a student learning how to string together SPICE, RK4, and a basic SRP calculation. As a research contribution, the evidence does not support the claims. I would desk-reject rather than spend referee time on a revision. If the authors fixed the circularity by testing on an independent propagator and justifying the perpendicular-only assumption, the tutorial portion might merit a teaching-oriented outlet.","headline":"A clean tutorial on SRP basics and ground tracks, but the ML validation is circular, the HPOP comparison is missing, and a misattributed Starlink example undercuts confidence.","tokens_in":696,"tokens_out":1975,"would_cite":false,"duration_ms":36148,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that solar radiation pressure on a LEO satellite can be computed with a simple cannonball model and that a linear-regression model can predict the resulting position shift to within 0.027% mean absolute error.","keywords":["Astrodynamics","Orbital Perturbations","Solar Radiation Pressure","Trajectory changes","Machine Learning","LEO satellites","cannonball model","linear regression"],"falsifier":"Propagate the same 550 km circular sun-synchronous orbit over several days with a full force model that includes radial and tangential SRP components, J2, and atmospheric drag, and record whether semi-major axis and eccentricity change; if they change measurably while the paper's inclination-only model predicts no change, the central perturbation premise is contradicted.","tokens_in":10929,"feed_emoji":"🛰️","tokens_out":7952,"duration_ms":61578,"temperature":0.7,"pith_summary":"The paper claims that the perturbation of a low-Earth-orbit satellite by solar radiation pressure can be captured by a lightweight 'cannonball' model and that the resulting change in position can be predicted with a simple linear-regression machine-learning model. Using a 550 km circular, sun-synchronous orbit and a 15 kg satellite, the authors compute the SRP acceleration from a year of ephemeris data, apply it as a perpendicular perturbation so that only the orbital inclination changes, and train a regressor on the SRP acceleration plus mass and area-to-mass ratio to output the perturbed Z position. They report a mean absolute percentage error of 0.0266%, which they take as validation of the model. The motivation is practical: accurate, fast trajectory-change estimates for the growing number of LEO satellites using a computationally light pipeline that can run alongside mission planning.","feed_headline":"Solar-pressure model and ML pin satellite drift to 0.027%","feed_subtitle":"A cannonball-model SRP calculation plus linear regression tracks the perturbed Z position of a 550 km LEO satellite.","key_machinery":"The load-bearing object is the cannonball model of solar radiation pressure, which treats the satellite as a sphere with one effective cross-sectional area and a single reflectivity coefficient. Its acceleration formula is $\\vec{a}_{\\mathrm{SRP}} = \\frac{C_r P_0 A\\, \\mathrm{AU}^2}{M} \\frac{\\vec{r} - \\vec{R}_{\\mathrm{geos}}}{|\\vec{r} - \\vec{R}_{\\mathrm{geos}}|^3}$ (with an eclipse shadow factor $\\nu$ available but set to the no-eclipse value), and the companion perturbation equation $\\delta i(t) = \\frac{1}{n a}\\int W \\cos(u)\\,dt$ converts the perpendicular SRP acceleration into an inclination change. These feed a generated dataset whose features are $\\vec{a}_{\\mathrm{SRP}}$, satellite mass $M$, and area-to-mass ratio $A/M$, with the $X,Y,Z$ position vectors as targets, on which the paper trains linear regression with gradient descent. That combination is what makes the pipeline semi-analytical: the physics of SRP is handled by closed-form equations, while the machine-learning step learns the mapping from SRP input to perturbed position.","core_discovery":"On its own terms, the paper's central finding is that a custom software module implementing the cannonball model for solar radiation pressure, with a reflectivity coefficient $C_r = 1 + \\varepsilon$ and emissivity $\\varepsilon = 0.30$, produces SRP acceleration values that agree with an established high-precision orbit propagator, and that when this acceleration is applied purely perpendicular to a circular orbit, the only perturbed Keplerian element is the inclination, per $\\delta i(t) = \\frac{1}{n a}\\int_{t_0}^{t} W \\cos(u)\\,dt$. A linear-regression model trained on $\\vec{a}_{\\mathrm{SRP}}$, mass, and area-to-mass ratio predicts the resulting $Z$ position vector with a MAPE of $0.02662153353165422\\%$, which the authors present as evidence that the combined analytical-machine-learning pipeline works for rapid satellite trajectory assessment. The paper also validates its ground-track computation against a reference LEO satellite's TLE and compares ground-station revisit times across Indian subcontinent sites.","pith_inferences":["An extension the paper leaves implicit is to activate the eclipse shadow factor, which is present in the model but set to the no-eclipse case; testing on dawn-dusk orbits with frequent eclipses would likely show larger position errors, since the reported accuracy assumes constant illumination.","Because the regression features include mass and area-to-mass ratio as constants for a single satellite, the reported accuracy only reflects interpolation in the SRP acceleration; a stronger test would train on multiple satellites with different masses and areas to see whether the linear hypothesis generalizes.","In a realistic LEO environment, SRP is usually a secondary perturbation relative to atmospheric drag and the J2 oblateness effect; the inclination-only result should be read as a controlled physics demonstration rather than a complete trajectory model, and adding drag and J2 is the obvious next step."],"forward_implications":["SRP acceleration for a LEO satellite can be computed quickly from mass, effective area, reflectivity, and ephemeris data, without a full finite-element or N-plate model.","For circular orbits, a perpendicular-only SRP treatment reduces the perturbation analysis to a single scalar change in inclination, making trajectory correction inexpensive.","A linear-regression surrogate trained on SRP acceleration and mass/area features predicts the perturbed Z position with a MAPE of about 0.027%, suggesting that similar fast surrogates could support near-real-time trajectory updates.","The same pipeline can be reused for other LEO satellites by changing mass, area, reflectivity, and ephemeris inputs, with the ML model retrained on the new satellite's data."],"supporting_citations":[{"why":"Supplies the foundational orbital-element and state-vector relations used for trajectory propagation and the initial orbit setup.","marker":"[6]"},{"why":"Provides the high-precision orbit propagator used as a validation baseline for the computed SRP acceleration and for ground-station revisit comparisons.","marker":"[7]"},{"why":"Introduces the cannonball-model SRP formulation that the custom module implements.","marker":"[8]"},{"why":"Supplies the Kepler anomaly relations used to justify the circular-orbit simplification.","marker":"[9]"},{"why":"Provides the library used to convert Keplerian elements into state vectors for trajectory visualization.","marker":"[17]"},{"why":"Gives the perturbation equations, in particular the relation used for the SRP-induced inclination change in Eq. (10).","marker":"[21]"}],"fun_headline_variants":["ML and cannonball model pin LEO satellite drift to 0.027%","Solar pressure model plus ML predicts satellite position to 0.027%","Cannonball SRP with linear regression reaches 0.0266% MAPE","Hybrid SRP-ML model tracks LEO satellite position to 0.027%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The premise the result rests on is that solar radiation pressure acts only perpendicular to the orbital plane and that the orbit is exactly circular, so Eq. (10) produces only an inclination change; if radial or tangential SRP acts, or if atmospheric drag and J2 are present, the semi-major axis and eccentricity also evolve, and the trained model would miss those dominant perturbing effects.","fun_headline_variants_meta":{"raw":{"variants":["ML and cannonball model pin LEO satellite drift to 0.027%","Solar pressure model plus ML predicts satellite position to 0.027%","Cannonball SRP with linear regression reaches 0.0266% MAPE","Hybrid SRP-ML model tracks LEO satellite position to 0.027%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000461,"raw_usage":{"total_tokens":2353,"prompt_tokens":1035,"completion_tokens":1318,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":651,"completion_tokens_details":{"reasoning_tokens":1230}},"tokens_in":651,"tokens_out":1318,"duration_ms":9305,"temperature":1.0,"reasoning_tokens":1230,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:54:33.549267+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Propagate the same 550 km circular sun-synchronous orbit over several days with a full force model that includes radial and tangential SRP components, J2, and atmospheric drag, and record whether semi-major axis and eccentricity change; if they change measurably while the paper's inclination-only model predicts no change, the central perturbation premise is contradicted.","supporting_citations":[{"cited_title":"Dover, Mineola (1971)","cited_arxiv_id":null,"evidence_quote":"Supplies the foundational orbital-element and state-vector relations used for trajectory propagation and the initial orbit setup."},{"cited_title":"MATLAB Cen tral File Exchange","cited_arxiv_id":null,"evidence_quote":"Provides the high-precision orbit propagator used as a validation baseline for the computed SRP acceleration and for ground-station revisit comparisons."},{"cited_title":"Internat ional Journal of 18 Aerospace Engineering vol","cited_arxiv_id":null,"evidence_quote":"Introduces the cannonball-model SRP formulation that the custom module implements."},{"cited_title":"In: Keplerian Ellipses","cited_arxiv_id":null,"evidence_quote":"Supplies the Kepler anomaly relations used to justify the circular-orbit simplification."},{"cited_title":"Journal of Open Source Software 5, 2050 (2020)","cited_arxiv_id":null,"evidence_quote":"Provides the library used to convert Keplerian elements into state vectors for trajectory visualization."},{"cited_title":"(1996) 19","cited_arxiv_id":null,"evidence_quote":"Gives the perturbation equations, in particular the relation used for the SRP-induced inclination change in Eq. (10)."}],"review_version":1}