{"id":"d51eb8b7-d8d1-4bb8-bd23-ad2871951976","arxiv_id":"2609.34685","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A multi-spacecraft persistence model ballistically and longitudinally propagates in-situ solar wind data into 2D maps, producing speed and density reconstructions at Earth that modestly beat the 27-day persistence benchmark.","lead":"This paper introduces P2D, a solar wind forecasting model that takes in-situ measurements from multiple spacecraft, propagates them longitudinally with the solar rotation and radially at the measured stream speed, and composes them into two-dimensional maps of solar wind speed and density.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Speed and density error relations used for weighting are fitted on data overlapping the validation window, so the reported MAE improvement may be partly in-sample.","rationale":"The reader identified the empirical error relations as the weakest assumption, focusing on their validity outside the fitted range. I agree that these relations are load-bearing, but I find a more immediate and concrete problem: the relations are fitted on data that overlaps the validation period, creating a potential in-sample bias in the weighting that directly affects the reported MAE numbers. The paper acknowledges the range limitation (b305) but does not address the temporal overlap. My proposed test, a temporal out-of-sample recalibration, would settle whether the measured improvement holds when the error relations are evaluated outside the fitting period. If it holds, the central claim stands; if not, the CONDITIONAL verdict is strengthened. This does not change the overall verdict, so I recommend UNCHANGED.","tokens_in":28644,"tokens_out":4314,"duration_ms":31481,"concrete_test":"Refit Eq. 3 (and, if possible, Eq. 2 from Milosic et al. 2026) using only data strictly before the validation window (e.g., speed: before Jan 2012; density: before Jan 2014), then apply the refitted relations to compute the weights for the point-to-point analyses and recompute the MAE for the 2012–2017 speed window and the 2008–2024 density window. If the resulting MAE exceeds the reported values by more than ~5% (e.g., speed MAE > 68 km/s), the headline improvement is partly due to in-sample fitting of the error relations.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reported headline MAE (65.41 km/s, Table 2) is evaluated for May 2012–May 2017. The speed error relation (Eq. 2) is quoted from Milosic et al. (2026), who fitted it on persistence-propagated data including the same period and scheme. The density relation (Eq. 3) is newly fitted in Appendix A on data from 2014–2024, which overlaps the density validation window of Jan 2008–Jan 2024 (Table 5). The weights in Eq. 4 and the uncertainty in Eq. 5 are then used to combine spacecraft profiles. If the error relations are optimistically calibrated on the validation period, the inverse-variance weighting will favor spacecraft that happen to perform well in-sample, inflating the measured improvement over the 27-day persistence benchmark. The paper does not report any temporal out-of-sample check, and the authors' stated limitation (b305, b374) concerns only the 15-degree latitude range and clustering near 1 AU, not the in-sample fit. This is distinct from the acknowledged approximations; it directly affects the quantitative claim that P2D beats persistence by ~11% in speed MAE and ~10% in density MAE, because those numbers are computed on the same data used to derive the weights.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper presents P2D, a two-dimensional persistence model for the ambient solar wind. In-situ speed and density measurements from OMNI, STEREO-A/B, PSP, and Solar Orbiter are propagated longitudinally at the 27.27-day solar rotation rate and radially at the measured bulk speed, with an inelastic-collision scheme (Eq. 1) preventing stream overtaking. Each spacecraft's map is built independently and sampled at Earth, and the resulting profiles are combined by inverse-variance weighting (Eqs. 4-5) using empirical error relations for speed (Eq. 2, from Milosic et al. 2026) and for density (Eq. 3, fitted in Appendix A). The reconstruction is validated against OMNI with the Reiss et al. (2023) metric suite and compared to the 27-day persistence benchmark, ESWF, HUXt, WSA 2.2, and AmbSoWi-ML. Headline claims: speed MAE 65.41 km/s and density MAE 3.51 cm-3, second only to AmbSoWi-ML on point-to-point metrics; best binary-metric performance on high-speed enhancements; and a roughly 35% speed-MAE improvement over 27-day persistence during periods with a spacecraft located between Earth and L5.","tokens_in":28946,"tokens_out":7321,"duration_ms":44394,"significance":"If the headline results hold, P2D is a computationally trivial (about 45 s per spacecraft-year on a laptop) multi-spacecraft persistence tool that modestly outperforms the standard 27-day persistence benchmark and several established models, and substantially outperforms it under favorable spacecraft geometry, while remaining second to AmbSoWi-ML. The design has real strengths: validation against external OMNI data (not against the model's own output), a standard metric suite following the unified Reiss et al. (2023) scheme, comparison against multiple best-practice benchmarks, a running-window analysis over two solar cycles, and an honest treatment of weak density performance (CC of 0.21) and of the model's dependence on spacecraft geometry. The paper also ships an explicit list of acknowledged approximations (rigid corotation, inelastic collisions, no ICME removal, 15-degree latitude validity bound). The weaknesses are concentrated in the quantitative load-bearing claims: the spacecraft-combination weights derive from error relations fitted on data windows that overlap the validation windows, and the 35% improvement claim is defined nowhere precisely.","major_comments":[{"comment":"The weighting relations are calibrated on data that overlaps the evaluation windows, and this is not one of the acknowledged limitations. The density relation (Eq. 3) is fitted in Appendix A on 2014-2024 data, which lies entirely inside the density validation window of January 2008-January 2024 reported in Table 5; the paper states this fit window explicitly in the Appendix A text but does not flag the overlap. The speed relation (Eq. 2) is taken from Milosic et al. (2026), which uses the same propagation scheme over a largely overlapping period. Since the weights in Eq. 4 redistribute weight toward whichever spacecraft had the smallest fitted error, an in-sample calibration can inflate the measured MAE advantage over the 27-day persistence benchmark reported in Tables 2 and 5. This concern is distinct from the acknowledged 15-degree latitude bound (b305). Please provide a temporal out-of-sample check (fit Eqs. 2-3 on data strictly preceding the evaluation window and recompute Tables 2 and 5) or a weight-sensitivity analysis showing that plausible perturbations of the coefficients leave the reported MAE advantage over persistence essentially unchanged.","section":"§2.2.4 and Appendix A, Eqs. 2-3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The 35% improvement claim is central to the abstract but is not tied to any explicit computation. The paper states that during periods with a spacecraft between Earth and L5 the speed MAE 'decreases by roughly 35% in comparison to the benchmark 27-day persistence model,' but no table or equation gives the definition of this quantity: which time intervals count as 'between Earth and L5,' how many days are included, the MAE values for each model over those intervals, or an uncertainty. The statement appears to be read off the running-window curves in Figure 6. Please state the precise computation (including the period selection rule and the two MAE numbers) so the headline number is reproducible and verifiable.","section":"Abstract and §4.6 (also §4.5)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Several of the claimed advantages over the 27-day persistence benchmark are numerically small (e.g., CC 0.51 vs 0.45 in Table 2; TPR 0.64 vs 0.61 and TS 0.48 vs 0.44 in Table 3; density hit rate 0.48 vs 0.45 in Table 6), and no uncertainty or significance assessment is given for any metric difference. Given that the paper's core message is that P2D 'consistently outperforms' persistence, please add a bootstrap or equivalent confidence statement for at least the headline MAE, CC, and hit-rate differences so the reader can judge whether the advantages are statistically distinguishable from chance.","section":"Tables 2, 3, 5, 6","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"In the RMSE-speed row the units are garbled: '84.57 d' should be km/s (not days), and the HUXt entry '93.21' is missing units entirely.","section":"Table 4","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The MSE entries are labeled 'km s−1' but have units of (km/s)^2; adjust the header or the text in §3.2.1.","section":"Table 2","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The abstract states the 'peak hit rate is 50% and 48% for speed and density,' but Tables 4 and 6 report 0.51 and 0.48. Please clarify that these numbers refer to the CIR peak analysis (not the binary high-speed-enhancement metrics, which have different values) and reconcile the rounding.","section":"Abstract vs Tables 4 and 6","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The logarithmic longitude terms 9.1·ln(Δφ) and 0.23·ln(Δφ) diverge as Δφ approaches zero and become negative for Δφ < 1 degree. Since Figure 3 shows intervals where longitudinal separation approaches zero, please state the domain on which the relations are defined and how near-zero separations are handled in the weighting scheme.","section":"Eqs. 2 and 3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The running-window figures are captioned 2008-2026 while Table 5's density window is January 2008-January 2024; a sentence in the text (b187) also says the speed window starts October 15, 2008. Please make the exact time intervals per figure and per table consistent and explicit in the captions.","section":"Fig. 6 and Fig. 8 captions and §3.2.1","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision","confidential_remarks":"The core design is sound and the external validation is credible; the paper is a useful contribution to the ambient-solar-wind forecast literature. The deciding issue is the overlap between the windows used to fit the weighting relations (Eqs. 2-3) and the windows on which the headline MAEs are reported, particularly the density fit (2014-2024) sitting inside the Table 5 validation window (2008-2024). The authors acknowledge several limitations, but this one is not acknowledged and it bears directly on the quantitative advantage claimed over the 27-day persistence benchmark. A clean temporal hold-out or weight-sensitivity analysis should settle it. The 35% claim also needs a defined computation. Both are fixable within the paper's scope, hence major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Colleague —\n\nThis paper deserves a real referee. It's an honest, workmanlike extension of the persistence modeling program: instead of single-spacecraft 27-day persistence, it ballistically propagates in-situ speed and density from five sources (OMNI, STEREO-A/B, PSP, SolO) into 2D ecliptic maps, extracts profiles at Earth, and combines them with inverse-variance weights set by empirical error relations (Eqs. 2–3). The 2D map construction plus multi-source compositing with source-specific weighting is genuinely new relative to the group's own 1D work and to Owens et al. 2013. Validation is done properly: external OMNI data, the unified metric suite, benchmarks including ESWF, HUXt, WSA, and the ML model AmbSoWi-ML, plus binary and CIR-peak analyses. The model is computationally trivial — 45 s per spacecraft-year on a laptop — which is the real selling point for operational use.\n\nThe headline numbers hold up as reported: speed MAE 65.41 vs 73.85 for persistence, second to AmbSoWi-ML in point-to-point metrics; best of all non-ML models on binary speed metrics; density modestly better than persistence though honestly acknowledged as poor in absolute terms (RMSE ≈ mean density). The 35% improvement is a best-case geometry figure, grey-shaded Earth-to-L5 windows, presented as such in the text — an honest hedge, not a bait-and-switch. The point-to-point window excluding PSP and SolO is also stated plainly (b169–b170).\n\nThe soft spots, in order. First, the one needing referee attention: Eqs. 2–3 are fitted on persistence-propagated data overlapping the validation windows — the speed relation from Milosic et al. 2026 (that period), the density relation in Appendix A on 2014–2024, overlapping the 2008–2024 density window. No temporal out-of-sample check of the weighting is reported. If the relations are optimistically calibrated, the inverse-variance weighting could favor in-sample performers and inflate the reported ~10% MAE gains. That's a real, answerable question. My guess is the effect is bounded — the improvements are modest and consistent across many metrics and an 18-year running MAE — but the authors should either fit on one half and test on the other, or perturb the weights and show the headline is robust. The acknowledged limits (15-degree latitude band, 1AU clustering, b305/b374) don't cover this. Second, no code or data is released yet; \"will be made public\" is not enough for a proposed operational baseline. Third, ICMEs are not removed, which the authors disclose for comparability; acceptable for benchmarking, but the sensitivity should be shown.\n\nNone of this breaks the central claim in my reading. The model modestly beats persistence in most metrics, and that is credible. What's missing is the out-of-sample weighting check and the data; both are fixable in revision.\n\nWho it's for: the space-weather forecasting community, CME-drag modelers, and anyone arguing for L5 (Vigil). Worth citing for the 2D maps and the density relation. I'd send it to review, with the weighting question front and center.","headline":"Honest, well-validated extension of persistence modeling; the real open question is whether the spacecraft weighting relations are in-sample, and that's referee-answerable rather than fatal.","tokens_in":29594,"tokens_out":2535,"would_cite":true,"duration_ms":18620,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"The paper builds a two-dimensional persistence model of the solar wind from multi-spacecraft in-situ data, claiming that from Earth's perspective it reconstructs bulk speed with an MAE of 65.41 km/s and density with an MAE of 3.51 cm^-3…","keywords":["solar wind","persistence model","multi-spacecraft","corotating interaction regions","space weather forecasting","in-situ data propagation","Lagrange point L5","heliospheric reconstruction"],"falsifier":"Run P2D on a held-out period in which a spacecraft (or a synthetic probe) sats at a latitudinal separation from Earth beyond the roughly 15 degrees where Equations 2 and 3 were fitted, and check whether the inverse-uncertainty weighting still lowers the MAE below the 27-day persistence benchmark; if the error relations do not extrapolate, the weighting scheme's contribution to the reported errors collapses, and the headline speed and density figures change.","tokens_in":30039,"feed_emoji":"🛰️","tokens_out":4518,"duration_ms":29119,"temperature":0.7,"pith_summary":"This paper claims that the solar wind can be reconstructed over a two-dimensional (longitude and radius) map of the inner heliosphere by propagating in-situ speed and density measurements from several spacecraft, using only rigid corotation and a simple inelastic collision scheme, without any magnetohydrodynamic physics. The resulting model, P2D, is said to beat the 27-day persistence benchmark in nearly all speed metrics and in all binary high-speed-stream metrics, to add a density reconstruction that also beats the benchmark, and to be second only to a machine-learning model (AmbSoWi-ML) in point-to-point speed metrics. The central practical claim is that the reconstruction improves most when a spacecraft is positioned between Earth and L5, which the authors take as evidence that the planned Vigil L5 mission would provide a strong basis for persistence forecasts. If correct, P2D offers a computationally trivial forecasting tool whose only inputs are in-situ measurements and whose output can be evaluated at any point in the heliosphere where a spacecraft lies closer to the Sun than the target.","feed_headline":"Multi-probe persistence model rebuilds the solar wind at Earth","feed_subtitle":"P2D's speed error drops to 65 km/s overall and by 35% when a probe sits upstream between Earth and L5.","key_machinery":"The load-bearing object is the two-dimensional propagation-and-combine scheme: each spacecraft's in-situ measurements are spread into a radial-longitudinal map by rigid corotation plus ballistic radial advection, and the per-spacecraft reconstructions are merged by weighting each source inversely to an empirical uncertainty both designs and powers the argument: Equation 2 (speed) and Equation 3 (density), the error relations that assign a predicted MAE based on latitudinal and longitudinal separation to the target, and Equation 4, the inverse-uncertainty combination that converts those errors into weights. The work the scheme does is to convert spacecraft separation geometry directly into forecast contribution, so that the model's headline performance follows from the error relations rather than from any physical evolution model.","core_discovery":"The central claim is that solar wind persistence, which is normally exploited as a one-dimensional one-Carrington-rotation echo of near-Earth data, can be organised into a two-dimensional map from which reconstructions at arbitrary targets follow. In-situ speed and density samples from OMNI, STEREO-A, STEREO-B, Parker Solar Probe and Solar Orbiter are each propagated longitudinally at the rigid synodic rotation period of 27.27 days and radially at their own measured speed, with an ideal inelastic collision (Equation 1) preventing streams from overtaking one another inside each spacecraft's map. The per-spacecraft maps are then merged at the target by inverse-uncertainty weighting, where the uncertainty of each source comes from empirical error relations in latitudinal and longitudinal separation to Earth: the speed relation (Equation 2) is taken from the authors' earlier persistence study, and a new density analogue (Equation 3) is fitted in Appendix A. Validated against OMNI at Earth over the full 2008-2026 range and in a restricted 2012-2017 window against other models, P2D reports a speed MAE of 65.41 km/s, a density MAE of 3.51 $cm^{-3}$, a 64% true positive rate and best-in-class binary metrics, while remaining second to AmbSoWi-ML on point-to-point speed metrics. The paper reports that during periods when STEREO-A or STEREO-B lies between Earth and L5, the speed MAE falls to roughly 40-60 km/s, a decrease of about 35% relative to the 27-day persistence benchmark, and attributes the improvement mainly to short propagation paths from the upstream probe.","pith_inferences":["The headline 65.41 km/s figure is earned on a 2012-2017 window in which PSP and SolO do not contribute, while the 35% improvement figure comes from the full-range analysis during favorable 2023-2026 geometry; the two headline numbers describe different conditions and should not be conflated.","The density reconstruction has a low point-to-point correlation (CC 0.21) yet very accurate CIR density-peak timing (mean error under an hour), so the model is probably capturing stream-interface timing far better than amplitude; users of dynamic-pressure estimates should trust its timing, not its magnitude.","If the empirical error relations are confirmed to extrapolate beyond their fitted range, the same inverse-uncertainty weighting formalism would transfer directly to forecast targets other than Earth, making the model a generic heliospheric interpolation tool rather than an Earth-specific forecast.","The 2D output is a natural cheap boundary condition for more physical models: feeding its multi-probe maps into data-assimilation or HD/MHD schemes could replace magnetogram-driven inputs with purely in-situ-driven evolution."],"forward_implications":["P2D outperforms the 27-day persistence benchmark in nearly every speed metric and in all binary high-speed-stream metrics, so a persistence argument alone, organised in two dimensions, is a stronger baseline than the conventional one-dimensional corotation echo.","A spacecraft stationed between Earth and L5 cuts the speed MAE by roughly 35% relative to the benchmark, which is the paper's concrete quantitative case for planning future L5 missions such as Vigil as the backbone of persistence forecasts.","Because the reconstruction needs only one spacecraft closer to the Sun than the target, the same scheme can yield forecasts at any heliospheric point with such coverage, including other planets such as Mars.","The two-dimensional maps supply ambient-solar-wind context in which coronal mass ejection propagation can be studied and in which combined speed-and-density output enables dynamic-pressure and drag-force calculations.","With science-level data and zero lead time the model is computationally trivial, running roughly 45 seconds per spacecraft-year at 0.5-degree resolution, so operational deployment reduces to a data-availability question rather than a computing one."],"supporting_citations":[{"why":"Supplies Equation 2, the empirical speed-error relation in latitudinal and longitudinal separation that sets the inverse-uncertainty weights.","marker":"Milošić et al. (2026)"},{"why":"Established the two-dimensional radial-plus-longitudinal propagation scheme and the inelastic collision handling that P2D reuses.","marker":"Milošić et al. (2025)"},{"why":"Defines the 27-day persistence model and its autocorrelation basis, the benchmark that P2D is measured against throughout.","marker":"Owens et al. (2013)"},{"why":"Provides the OMNI database that serves both as a key input source and as the validation reference at Earth.","marker":"King and Papitashvili (2005)"},{"why":"Supplies the high-speed stream / CIR catalog used for the peak-analysis metrics reported for speed and density.","marker":"Koller et al. (2025)"},{"why":"Describes AmbSoWi-ML, the machine-learning model that P2D trails on point-to-point speed metrics and thereby defines the paper's comparative ceiling.","marker":"Bailey et al. (2021)"},{"why":"Provides the WSA model whose inelastic stream-interaction treatment is the conceptual analogue to P2D's collision scheme.","marker":"Arge et al. (2000)"}],"fun_headline_variants":["P2D maps solar wind persistence from multiple probes","Multi-probe 2D maps cut solar wind speed error by 35%","Persistence model builds 2D solar wind from five spacecraft","Solar wind persistence goes 2D with multi-spacecraft input","Upstream probes sharpen P2D solar wind forecasts at Earth"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The load-bearing premise is that the empirical error relations (Equations 2 and 3) correctly predict reconstruction error from latitudinal and longitudinal separation, because they set the spacecraft weights and hence the reported performance; the authors acknowledge these relations are fitted to their own persistence-propagated data, hold only up to roughly 15 degrees of latitudinal separation, and rest on data clustered near 1 AU.","fun_headline_variants_meta":{"raw":{"variants":["P2D maps solar wind persistence from multiple probes","Multi-probe 2D maps cut solar wind speed error by 35%","Persistence model builds 2D solar wind from five spacecraft","Solar wind persistence goes 2D with multi-spacecraft input","Upstream probes sharpen P2D solar wind forecasts at Earth"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000177,"raw_usage":{"cost":0.000177128,"is_byok":false,"cost_details":{"upstream_inference_cost":0.000177128,"upstream_inference_prompt_cost":0.000137648,"upstream_inference_completions_cost":3.948e-05},"total_tokens":1329,"prompt_tokens":1188,"completion_tokens":141,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":256,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":51}},"tokens_in":1188,"tokens_out":141,"cost_source":"openrouter_usage","duration_ms":2291,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":51,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-09-29T19:57:13.857149+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"falsifier":"Run P2D on a held-out period in which a spacecraft (or a synthetic probe) sats at a latitudinal separation from Earth beyond the roughly 15 degrees where Equations 2 and 3 were fitted, and check whether the inverse-uncertainty weighting still lowers the MAE below the 27-day persistence benchmark; if the error relations do not extrapolate, the weighting scheme's contribution to the reported errors collapses, and the headline speed and density figures change.","supporting_citations":[{"cited_title":"J., Challen, R., Methven, J., Henley, E., & Jackson, D","cited_arxiv_id":null,"evidence_quote":"Defines the 27-day persistence model and its autocorrelation basis, the benchmark that P2D is measured against throughout."},{"cited_title":"Papitashvili, 2005","cited_arxiv_id":null,"evidence_quote":"Provides the OMNI database that serves both as a key input source and as the validation reference at Earth."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the high-speed stream / CIR catalog used for the peak-analysis metrics reported for speed and density."},{"cited_title":"Using gradient boosting regression to improve ambient solar wind model predictions","cited_arxiv_id":"2006.12835","evidence_quote":"Describes AmbSoWi-ML, the machine-learning model that P2D trails on point-to-point speed metrics and thereby defines the paper's comparative ceiling."},{"cited_title":"N., & Pizzo, V","cited_arxiv_id":null,"evidence_quote":"Provides the WSA model whose inelastic stream-interaction treatment is the conceptual analogue to P2D's collision scheme."}],"review_version":1}