{"id":"edf5168d-246b-4784-8f56-e33f87b6a482","arxiv_id":"2504.17072","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"The CODET v1.1 model reproduces SDO/EVE measurements of solar EUV emission at 28.4 nm and 21.1 nm with about 15 percent error, and provides daily irradiance estimates from 1996 to 2024.","lead":"This paper updates the CODET model, which estimates extreme ultraviolet sunlight from the Sun's magnetic field, to produce daily irradiance at two iron emission lines. The updated model matches spacecraft observations during 2010 to 2014 within about 20 percent, then produces estimates for 1996 to 2024 where direct measurements are missing.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'reliable estimate' claim is undermined by the model's acknowledged failure on complex active regions; the out-of-sample validation is not stratified by solar activity, so the post-2014 high-activity gap may be exactly where errors are largest.","rationale":"The paper's central claim is that CODET v1.1 provides reliable SSI estimates in EUV wavelengths where observational data are absent, with the post-2014 MEGS-A gap as the motivating example. The reader's verdict (CONDITIONAL) already identifies the load-bearing premise: the density and temperature scalings fitted to 2010–2014 are assumed stable across solar cycles. My stress-test converges on a sharper, overlapping concern: the model's acknowledged poor performance on days with large, complex active regions (Section 4, Figure 4) directly targets the high-activity epochs that dominate the post-2014 gap and the 1996–2024 interval. The out-of-sample validation with GOES/EUVS and AIA is not stratified by activity, so the reported 26% and 42% average MAPEs may conceal a systematic degradation at solar maximum. This is not a disagreement with the reader's weakest assumption; it is a more concrete route to the same conclusion, and it points to a simple, decisive test. I also give credit where due: the model update is a plausible physics-based approach, the in-sample EVE fit is reasonable, and the GOES/EUVS comparison is a genuine out-of-sample check. However, the averaged metrics are insufficient to support the 'reliable estimate' language for every unobserved period, especially the high-activity gap. Since the concern is addressable with existing data and the reader already recommended CONDITIONAL, my verdict remains UNCHANGED rather than moving to ACCEPT or REJECT.","tokens_in":12092,"tokens_out":4176,"duration_ms":39191,"concrete_test":"Recompute the out-of-sample MAPE and relative-error distributions for GOES/EUVS 28.4 nm (and AIA 21.1 nm, with the stated scaling) separately for solar-activity bins—e.g., quartiles of daily F10.7 or sunspot number over 2017–2024, and separately for the 2018–2020 minimum versus the 2023–2024 maximum. If the high-activity bin MAPE is significantly larger than the low-activity bin (e.g., more than 1.5× the period-average of about 26%), the model is not reliable in the high-activity gap after MEGS-A, and the central claim should be restricted. This test requires only the already-used GOES/EUVS and F10.7 data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (abstract, §4) is that CODET v1.1 reliably estimates SSI at 28.4 nm and 21.1 nm where no observations exist, specifically after the SDO/EVE MEGS-A era. This rests on a single set of scaling-law parameters (γ, α, N0, T0, Bs; Appendix A, Eqs. A2–A4) fitted to 2010–2014 EVE data and applied to 1996–2024. The paper itself reports that days with large, complex active regions are poorly reproduced (§4, Fig. 4), attributing this to the PFSS extrapolation's inability to describe such regions. Since the post-2014 gap includes the approach to solar maximum (cycle 25) and the 1996–2024 interval includes cycle 23 maximum, the model's known failure mode is concentrated in the very epochs the 'reliable estimate' claim targets. The out-of-sample metrics (GOES/EUVS 28.4 nm MAPE ≈ 26%; AIA 21.1 nm MAPE ≈ 42% after an empirical 7 × 10^5 scaling, §3.2, Fig. 9) are reported as period averages; they are not stratified by activity level. A model that misses complex-AR days by more than 2σ in-sample can still yield a modest average MAPE while being systematically biased at solar maximum.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript presents CODET v1.1, a physics-based model that derives coronal density and temperature from PFSS magnetic-field magnitude via power-law scalings (Appendix A, Eqs. A2-A4), computes EUV solar spectral irradiance (SSI) at Fe XV 28.4 nm and Fe XIV 21.1 nm using CHIANTI 10.0.2, fits five free parameters to SDO/EVE MEGS-A observations from April 30, 2010 to May 26, 2014, and then predicts daily SSI from July 1996 to October 2024. The in-sample fit achieves R2 = 0.849 at 28.4 nm and R2 = 0.796 at 21.1 nm, with MAPE of about 15% in both lines. Out-of-sample comparisons give MAPE of about 26% versus GOES/EUVS at 28.4 nm (2017-2024) and about 42% versus SDO/AIA at 21.1 nm after applying an empirical scaling factor of 7e5. The paper concludes that CODET v1.1 provides a reliable estimate of SSI in EUV wavelengths where no observational data exist, particularly after the SDO/EVE MEGS-A era.","tokens_in":12420,"tokens_out":5103,"duration_ms":45927,"significance":"If the predictive claims survive scrutiny, the model would provide a continuous, physically motivated daily SSI record at two coronal lines across nearly three solar cycles, with associated full-disc intensity and density maps, useful for ionosphere-thermosphere and planetary aeronomy studies. The paper has clear strengths: it uses openly available magnetogram and EVE data, a transparent optimization over a small parameter set, and a genuinely external validation against GOES/EUVS after the fitting interval. The in-sample fit is good, and the modeling ingredients (optically thin emission, CHIANTI contribution functions, PFSS extrapolations) are standard. However, the central claim of reliable estimates in unobserved periods is not yet established because the validation errors are not stratified by activity level, the 21.1 nm comparison rests on a post-hoc empirical scaling, and the model's own acknowledged failure on complex active regions coincides with the high-activity epochs that dominate the post-2014 gap.","major_comments":[{"comment":"The claim that CODET v1.1 provides reliable SSI estimates in unobserved periods (abstract and Section 4) is not supported by the presented validation because the errors are reported only as period averages and are not stratified by solar activity. The paper itself identifies days with large, complex active regions as poorly reproduced (Section 4, Figure 4), attributing this to the PFSS extrapolation's inability to describe such regions. Since the post-MEGS-A gap includes the rise to cycle 25 maximum and the 1996-2024 interval includes cycle 23 maximum, the model's known failure mode is concentrated exactly in the epochs targeted by the central claim. A period-averaged MAPE of about 26% at 28.4 nm can mask a systematic bias at solar maximum. The authors should provide validation metrics stratified by activity level (e.g., bins in F10.7 or sunspot number) and discuss whether the out-of-sample error is stable across the cycle.","section":"Section 3.2 and Section 4 (Figs. 4 and 9)"},{"comment":"The AIA-based validation at 21.1 nm is not an independent test of model accuracy because the comparison requires dividing AIA full-disc daily mean DN values by an empirical factor of 7e5, chosen to make the two time series comparable. With the scaling factor selected post hoc, the reported MAPE of about 42% conflates model error with the uncertainty in the AIA-to-irradiance conversion. To support a quantitative error claim, the scaling should be fixed independently, for example from a contemporaneous EVE/AIA overlap, or the 21.1 nm comparison should be explicitly labeled as qualitative. As it stands, the abstract's error figure for 21.1 nm rests on a non-independent calibration.","section":"Section 3.2 (Figure 9)"},{"comment":"The statement that the model 'described well observational data ... with less than 20% error' during 2010-2014 refers to the training period used to fit the five free parameters, so the reported R2 values and MAPE of about 15% measure in-sample agreement rather than predictive skill. This distinction should be made explicit in the abstract. The only genuinely external benchmark is the GOES/EUVS comparison at 28.4 nm (2017-2024, MAPE about 26%), and the 'reliable estimate' conclusion should be based primarily on that benchmark. Please revise the wording so that the abstract does not present training-data performance as evidence for predictive reliability.","section":"Abstract and Section 3 (Figures 2 and 3)"},{"comment":"The model assumes that coronal density and temperature are stationary power-law functions of the local PFSS magnetic-field magnitude, with one set of parameters fitted to 2010-2014 applied to 1996-2024. The paper provides no test of this stationarity, and its own admission that complex active regions are poorly reproduced suggests the assumption is fragile at high activity. A concrete and feasible test is to compare model density and temperature profiles against independent coronal diagnostics at different phases of the cycle, for example Hinode/EIS or Metis/SoLO, as the paper lists as future work. Without at least a basic check of parameter stability across activity levels, the extrapolation to unobserved periods carries an unquantified bias risk that directly affects the paper's core claim.","section":"Appendix A (Eqs. A2-A4) and Section 4"}],"minor_comments":[{"comment":"The caption states 'October 27, 202' with an incomplete year; it should read 'October 27, 2024'.","section":"Section 3.2, Figure 9 caption"},{"comment":"The author name appears as both 'Rodrí guez-Go mez' and 'Rodrí guez Go mez'; please standardize the spelling and spacing.","section":"Throughout"},{"comment":"In Table 1, the units for the power-law indices gamma and alpha are listed as '...' but these are dimensionless; also, the units in Equation A4 for temperature are given as [cm^-3] but should be [K].","section":"Table 1 and Appendix A, Eq. A4"},{"comment":"The statement that the mean intensity ratios EVE28.4/TIMED28.5 and EVE21.1/TIMED21.5 are about 9 and 3 lacks a figure or table; please provide the comparison period and data versions, or a citation, so that the claim can be verified.","section":"Section 2"},{"comment":"The notation '2sigma^2 = +/- 41.35%' is confusing because the standard expression for a 95% interval would be written as '+/- 2sigma'. Please clarify how the error interval was constructed.","section":"Section 3, Figure 4"},{"comment":"Equation A5 includes an integral over wavelength with an instrumental response R(lambda), but the text sets R(lambda) = 1; this simplification should be stated more prominently, because it means the modeled intensity neglects instrumental bandpass effects and this may matter when comparing with broadband or filtered observations.","section":"Appendix A, Eq. A5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's core idea is sound and the GOES/EUVS validation is a genuinely useful external check, but the abstract and conclusions overstate the reliability of the filled-in SSI record. The main revision should focus on stratifying validation errors by activity level and removing or clearly qualifying the AIA-based error claim. The paper is suitable for ApJ after these revisions, provided the claims are matched to the evidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a modest update of the author's own CODET model, but it delivers a genuinely new product—daily EUV SSI at 28.4 and 21.1 nm from 1996 to 2024, calibrated against SDO/EVE instead of TIMED/SEE. The out-of-sample comparison with GOES/EUVS at 28.4 nm is real, external to the fitting, and the ~26% error there is respectable for a physics-based model. That makes the paper worth a serious look.\n\nThe genuine improvements: using actual EVE measurements rather than TIMED/SEE's model-driven values (which differ by factors of 3–9), adding the 28.4 nm channel, and producing full-disc intensity and density maps. The in-sample fit to EVE 2010–2014 is good (R2 ~0.85 and ~0.80, MAPE ~15%). The long prediction series is exactly what ionosphere-thermosphere modelers want after MEGS-A died.\n\nThe soft spots are proportional to the claim. The paper's central sentence—that CODET provides a 'reliable estimate' of SSI where no observations exist—is stronger than the evidence. The model parameters are fitted to 2010–2014 and then applied to 1996–2024 without any check that the density-temperature scaling is stable across solar cycles. The paper itself admits days with big, complex active regions are poorly reproduced (Section 4, Figure 4). That matters because the post-2014 gap includes the approach to solar maximum of cycle 25, and cycle 23 maximum in the prediction interval—exactly the epochs when complex ARs are common. The out-of-sample errors are period averages, not stratified by activity, so the error in the gap may be larger than 26%. The 21.1 nm validation is weak: it depends on an empirical scaling factor of 7e5 and shows 42% error, so I would not treat that channel as reliable. There are also no uncertainties on the parameters or predictions, and the code/data are not released.\n\nThe paper is for a specific reader: someone who needs a continuous EUV irradiance input for ITM modeling or a proxy for solar activity. It does not reorganize the field, but it fills a real gap. It deserves peer review. The referee should ask for a softened claim, error metrics stratified by solar activity, and ideally a release of code and data. With those, it would be a solid contribution.","headline":"CODET v1.1 is a useful incremental update with a real external check at 28.4 nm, but the 'reliable estimate' claim overreaches; deserves peer review with revisions.","tokens_in":12992,"tokens_out":2756,"would_cite":false,"duration_ms":23324,"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":"The CODET model version 1.1 claims that daily EUV irradiance in two iron lines can be estimated from photospheric magnetograms alone, bridging the gap left when the SDO/EVE MEGS-A detector failed in 2014.","keywords":["solar spectral irradiance","extreme ultraviolet","coronal density","coronal temperature","Fe XV 28.4 nm","Fe XIV 21.1 nm","potential field source surface","solar cycle variability"],"falsifier":"Refit the five scaling parameters using only GOES/EUVS 28.4 nm data from June 2017 to October 2024 and compare them with the EVE-fit values $\\gamma=1.3077$, $\\alpha=-0.2781$, $N_o=8.2819\\times10^8\\,\\mathrm{cm^{-3}}$, $T_o=1.8558\\times10^6\\,\\mathrm{K}$, $B_s=7.9966\\,\\mathrm{G}$; if the refit parameters drift by more than their fitting uncertainties, the time-invariance premise is falsified. A simpler check is to split the 2017-2024 GOES comparison into quiet-Sun and active-region days: a systematic error pattern that grows with magnetic complexity would show that the fixed power laws miss a state-dependent factor rather than a constant offset.","tokens_in":11854,"feed_emoji":"☀️","tokens_out":15223,"duration_ms":114954,"temperature":0.7,"pith_summary":"CODET model version 1.1 argues that the Sun's daily extreme-ultraviolet irradiance in the coronal iron lines Fe XV at 28.4 nm and Fe XIV at 21.1 nm can be computed from maps of the photospheric magnetic field, without any EUV observations at the target date. The model learns five parameters from four years of SDO/EVE data (April 30, 2010 to May 26, 2014), reproduces those data with less than 20% mean absolute percentage error, and then predicts a continuous daily irradiance record from July 1996 to October 2024. If the claim is right, researchers get an EUV time series spanning parts of three solar cycles, including the years after the MEGS-A detector stopped taking data, plus full-disc intensity and density maps for any date with magnetic-field coverage. The prediction comparisons against independent GOES/EUVS and SDO/AIA data give errors of about 26% and 42%, respectively, so the paper's central assertion is that those error levels are reliable enough for gap filling.","feed_headline":"Magnetograms alone can fill the EUV irradiance gap after 2014","feed_subtitle":"CODET v1.1 matches EVE to under 20% in 2010-2014, then extends daily 28.4 and 21.1 nm estimates through 2024.","key_machinery":"The load-bearing object is the pair of power-law scalings $N(B)=N_o(B/B_s)^\\gamma$ and $T(B)=T_o(B/B_s)^\\alpha$ (with a floor temperature $T_o$ for weak-field regions), where $B$ is the magnitude of the magnetic field from a potential-field source-surface extrapolation of the observed photospheric magnetogram. These scalings convert each magnetic map into a three-dimensional coronal density and temperature structure; the emission-measure integral $I(\\lambda)=\\iint R(\\lambda)\\,G(\\lambda,T)\\,d\\lambda\\,N^2\\,ds$, with the contribution function $G$ from the CHIANTI atomic database, converts that structure into full-disc intensities. An evolutionary optimizer selects the five free parameters by minimizing the $\\chi^2$ difference between modeled and observed daily irradiance. Version 1.1's changes are the use of actual SDO/EVE irradiance as the fitting target instead of model-filled TIMED/SEE data, the addition of the 28.4 nm line alongside 21.1 nm, and line-of-sight integration of intensity maps from $1.0\\,R_\\odot$ to $2.5\\,R_\\odot$.","core_discovery":"On its own terms, the paper establishes that the chain magnetogram → potential-field extrapolation → power-law density and temperature scalings → emission-measure synthesis preserves enough information to reproduce daily full-disk irradiance in two coronal lines. Simultaneously fitting Fe XV 28.4 nm and Fe XIV 21.1 nm to SDO/EVE MEGS-A observations from 2010 to 2014 yields mean absolute percentage errors near 15% for both lines ($R^2 = 0.849$ and $0.796$), with the fitted parameters $\\gamma = 1.3077$, $\\alpha = -0.2781$, $N_o = 8.2819\\times10^8\\,\\mathrm{cm^{-3}}$, $T_o = 1.8558\\times10^6\\,\\mathrm{K}$, and $B_s = 7.9966\\,\\mathrm{G}$. Applied forward to every day with SOHO/MDI or SDO/HMI magnetograms, the same parameters produce daily SSI from July 1996 to October 2024, including the post-2014 MEGS-A gap; out-of-sample comparisons give errors of about 26% at 28.4 nm against GOES/EUVS and about 42% at 21.1 nm against AIA daily means. The paper therefore claims that a reliable EUV irradiance time series can be maintained where direct observations no longer exist.","pith_inferences":["If the fixed power laws hold across cycles, the same two-line fitting strategy could be extended to other iron lines with formation temperatures near $\\log_{10}T\\approx6.3$ K, widening the synthetic EUV spectrum without new calibration data.","A natural test of time invariance would be to refit the five parameters on post-2014 GOES/EUVS data alone; any drift in the best-fit values would quantify the cycle dependence that the single-fit approach hides.","Users of these irradiance series for ionosphere-thermosphere modeling should treat the reported 26% and 42% out-of-sample errors as a floor, because the current comparisons cannot separate instrumental bandpass mismatch from true coronal change.","The empirical factor of $7\\times10^5$ used to scale SDO/AIA data into irradiance units is a candidate for independent cross-calibration on MEGS-A-era days; replacing it with a measured conversion would sharpen the 21.1 nm validation."],"forward_implications":["A continuous daily SSI record at 28.4 nm and 21.1 nm now exists from July 1996 to October 2024, covering solar cycles 23 and 24 plus the rising phase of cycle 25, wherever MDI or HMI magnetograms are available.","The post-2014 gap left by the MEGS-A detector failure can be filled with model output, and the paper's GOES/EUVS comparison suggests the 28.4 nm channel carries about 26% error there.","Full-disc intensity maps and density maps are generated for any date with magnetogram coverage, giving EUV imagers a synthetic comparison product in wavelengths where observations may be missing.","The same parameter set reproduces the 21.1 nm line against AIA daily means to about 42% error, so users of the predicted record should carry that uncertainty into any derived product.","Because the two fitted lines form at nearly the same temperature ($\\log_{10}T \\approx 6.3$ K), the simultaneous fit is well constrained, which supports the model's use of a single density-temperature relation for both lines."],"supporting_citations":[{"why":"Defines the original CODET model whose density-temperature-magnetic-field relation version 1.1 updates.","marker":"Rodríguez-Gómez (2017)"},{"why":"Supplies the scaling laws, emission-measure formalism, and v1.0 fitting methodology that v1.1 refits on EVE data.","marker":"Rodríguez-Gómez et al. (2018)"},{"why":"Provides the surface flux transport model that produces the magnetic field maps used as boundary conditions.","marker":"Schrijver (2001)"},{"why":"Supplies the PFSS extrapolation procedure that converts photospheric maps into the coronal field B used by the scaling laws.","marker":"Schrijver & De Rosa (2003)"},{"why":"Provides SOHO/MDI magnetograms, the input for the 1996-2010 portion of the predicted time series.","marker":"Scherrer et al. (1995)"},{"why":"Provides SDO/HMI magnetograms, the input for the post-2010 portion of the predicted time series.","marker":"Scherrer et al. (2012)"},{"why":"Describes SDO/EVE and its calibration, the observed irradiance dataset used for fitting and in-sample validation.","marker":"Woods et al. (2012)"},{"why":"Documents the EVE instrument design and data products supporting the MEGS-A 28.4 nm and 21.1 nm channels.","marker":"Hock et al. (2012)"},{"why":"Supplies the ionization equilibrium and abundance data used with CHIANTI 10.0.2 to build the emission contribution functions.","marker":"Scott et al. (2015b,a)"},{"why":"Provides an earlier plasma-beta validation of the model's density and temperature estimates, supporting the physical plausibility of the new parameters.","marker":"Rodríguez-Gómez et al. (2019)"}],"fun_headline_variants":["Magnetogram-driven EUV irradiance model bridges post-2014 gap","CODET v1.1: 28 years of daily EUV from magnetic field maps","Physics-based EUV irradiance fills MEGS-A era gap with <20% error","Magnetic maps alone predict daily EUV solar irradiance through 2024"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire prediction rests on the premise that coronal density and temperature at each height are set by the local magnetic field strength through fixed power-law relationships, with parameters learned from 2010-2014 and then applied unchanged to 1996-2024; if that relationship drifts across solar cycles or depends on active-region complexity, the gap-filling irradiance inherits the drift.","fun_headline_variants_meta":{"raw":{"variants":["Magnetogram-driven EUV irradiance model bridges post-2014 gap","CODET v1.1: 28 years of daily EUV from magnetic field maps","Physics-based EUV irradiance fills MEGS-A era gap with <20% error","Magnetic maps alone predict daily EUV solar irradiance through 2024"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000187,"raw_usage":{"total_tokens":1475,"prompt_tokens":1236,"completion_tokens":239,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":852,"completion_tokens_details":{"reasoning_tokens":150}},"tokens_in":852,"tokens_out":239,"duration_ms":2932,"temperature":1.0,"reasoning_tokens":150,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:49:43.789205+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Refit the five scaling parameters using only GOES/EUVS 28.4 nm data from June 2017 to October 2024 and compare them with the EVE-fit values $\\gamma=1.3077$, $\\alpha=-0.2781$, $N_o=8.2819\\times10^8\\,\\mathrm{cm^{-3}}$, $T_o=1.8558\\times10^6\\,\\mathrm{K}$, $B_s=7.9966\\,\\mathrm{G}$; if the refit parameters drift by more than their fitting uncertainties, the time-invariance premise is falsified. A simpler check is to split the 2017-2024 GOES comparison into quiet-Sun and active-region days: a systematic error pattern that grows with magnetic complexity would show that the fixed power laws miss a state-dependent factor rather than a constant offset.","supporting_citations":[],"review_version":1}