{"id":"a2137b7c-15e2-4f27-896a-f4c85d51f09f","arxiv_id":"2507.10275","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"AfriTEC tracks diurnal TEC patterns at East African stations, but the paper's claimed mean error below 1.5 TECU is contradicted by its own tables, which show errors up to 6.96 TECU.","lead":"This paper evaluates the AfriTEC regional ionospheric model against GPS-based TEC measurements at five East African stations during 2016-2017, and compares it with the global NeQuick model. The abstract claims mean errors below 1.5 TECU, but the paper's own tables show errors mostly between 1.7 and 6.9 TECU, so the headline result is not supported.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim 'MAE generally below 1.5 TECU' is contradicted by the paper's own Tables 2–6, where most of the 38 non-NaN AfriTEC MAE values exceed 1.5 TECU.","rationale":"The most load-bearing concern is the direct numerical contradiction between the paper's central claim and its own results tables. The reader's weakest_assumption concerned GNSS reference data quality and possible training overlap; that is a legitimate methodological gap, but it is not needed to reject the paper. Even taking the tables at face value, the statement 'MAE values generally below 1.5 TECU' is unsupported: roughly 84% of the reported AfriTEC MAE values exceed 1.5 TECU. This is an internal inconsistency, not a matter of disagreeing with external consensus. The correlation part of the claim (r > 0.80) is actually well supported by the tables, all of which exceed 0.90, so the problem is specifically the MAE threshold. The manuscript also contains mutually incompatible statements about whether equinoxes or solstices are the better seasons, and the Weaknesses section cites an outlier value (10.583 TECU) that is absent from the corresponding table. These contradictions mean the central argument cannot be accepted as written; a correction would need to either revise the reported MAE claim or resubmit the underlying metrics. Since the reader already reached REJECT, my independent concern leaves that verdict unchanged.","tokens_in":12014,"tokens_out":4080,"duration_ms":39734,"concrete_test":"Tally the AfriTEC MAE values in Tables 2–6 (38 valid entries) and compute the fraction below 1.5 TECU. If that fraction is not a clear majority or at least consistent with 'generally below 1.5,' the Abstract and Conclusion claims are disproven by the paper's own tables. As a secondary check, recompute MAE for one discrepant station-season (e.g., MOIU March Equinox 2016) from the archived ICTP GNSS data and AfriTEC toolbox output to confirm whether the true value is 4.583 TECU and whether any 10.583 TECU outlier exists.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The Abstract and Conclusion assert that AfriTEC captures TEC with MAE values generally below 1.5 TECU. The paper's Tables 2–6 contain 38 non-NaN AfriTEC MAE entries across five stations and eight seasons. Only six entries are at or below 1.5 TECU (ZAMB June 2016: 0.974; ZAMB September 2016: 1.372; ADIS March 2017: 1.353; ZAMB June 2017: 1.115; ZAMB September 2017: 1.318; ZAMB December 2017: 1.500). That is roughly 16% of all entries. The other 32 values range from 1.690 to 6.964 TECU, including MBAR December 2016 at 6.964 TECU and MBAR September 2016 at 6.307 TECU. The Discussion text itself states MAE ranged between about 1.2 and 4.6 TECU, which is inconsistent with the tables. The seasonal narrative is also self-contradictory: the Abstract emphasizes equinox performance, while the Seasonal Variations section states that AfriTEC 'exhibits stronger performance during solstice periods,' which the tables broadly support. Additionally, the Weaknesses section cites a MOIU outlier of 10.583 TECU that does not appear anywhere in Table 2 (March Equinox 2016 is listed as 4.583 TECU). Because the headline claim is falsified by the paper's own reported numbers, the central argument is internally inconsistent, independent of any external validation concerns.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript evaluates the AfriTEC model against GNSS-derived TEC at five East African stations (MOIU, MAL2, ZAMB, ADIS, MBAR) during the descending phase of Solar Cycle 24 (2016-2017), using MAE and the Pearson correlation coefficient, and it compares AfriTEC with NeQuick. The paper's central claim is that AfriTEC effectively captures the diurnal and seasonal behavior of TEC, with MAE values generally below 1.5 TECU and correlation coefficients exceeding 0.80, and that performance is best during equinoxes. The reported tables, however, show most MAE values between 1.69 and 6.96 TECU, the discussion states a range of 1.2 to 4.6 TECU, and the seasonal narrative is internally inconsistent.","tokens_in":12322,"tokens_out":5862,"duration_ms":61027,"significance":"A clean regional validation of AfriTEC would be valuable for African space-weather applications, and the explicit NeQuick comparison and reliance on public data are strengths. The paper, however, is not usable in its current form: the headline numerical claims are contradicted by the paper's own tables, and the manuscript does not establish that the five validation stations were outside AfriTEC's training data. These issues affect every derived conclusion, not just the framing.","major_comments":[{"comment":"The abstract and the first two bullets of the Conclusion state that MAE values are 'generally below 1.5 TECU' and that the model performs particularly well during equinoxes. Tables 2-6 contain 38 non-NaN AfriTEC MAE entries; only six meet the 1.5 TECU threshold (all but one at ZAMB), while the remaining 32 entries range from 1.690 to 6.964 TECU, including 6.307 and 6.964 TECU at MBAR in 2016. The headline accuracy claim is therefore not supported by the reported data.","section":"Abstract and Conclusion; Tables 2-6"},{"comment":"The Discussion states that MAE ranged approximately between 1.2 and 4.6 TECU, which contradicts Tables 4 (e.g., MBAR September 2016: 6.307 TECU; December 2016: 6.964 TECU). Furthermore, the Abstract and Conclusion claim especially strong performance during equinoxes, but the 'Seasonal Variations in MAE and Correlation Coefficient' subsection concludes 'The AfriTEC model exhibits stronger performance during solstice periods', and the tables broadly show lower MAE in June and December. The paper cannot support both statements simultaneously.","section":"Discussion ('Diurnal Variation of VTEC over East African Sector') and 'Seasonal Variations in MAE and Correlation…"},{"comment":"The first Weakness bullet cites a MOIU outlier with MAE 10.583 TECU 'likely corresponding to March Equinox 2016'. Table 2 lists MOIU March Equinox 2016 MAE as 4.583 TECU, and no 10.583 value appears in any table or figure. The provenance of this outlier, and which computation it belongs to, must be clarified; as written, the reference is unsupported by the presented data.","section":"Weaknesses of the AfriTEC Model"},{"comment":"The GNSS processing section describes the data only as 'processed using MATLAB scripts' and gives no details on differential code bias estimation and removal, elevation cutoff, cycle-slip handling, or the mapping function used to convert slant to vertical TEC. This is doubly important because AfriTEC is a neural network trained on African GNSS TEC data (Okoh et al., 2019): the manuscript never states whether any of the five validation stations (ZAMB, ADIS, MOIU, MAL2, MBAR) were part of the training set. If they were, the reported MAE and correlation values are not an independent validation. The authors should either confirm non-overlap using the station list from Okoh et al. or reframe the paper as a reproducibility/self-consistency check rather than a performance assessment.","section":"GNSS Data Processing and AfriTEC Model Data"}],"minor_comments":[{"comment":"The text refers to 'Tables 1 - 5' and to 'Table 2' when discussing ZAMB, but the seasonal metrics appear in Tables 2-6; ZAMB is Table 3, MBAR is Table 4, ADIS is Table 5, and MAL2 is Table 6. Please correct all table cross-references.","section":"Seasonal Variation Analysis and Tables 2-6"},{"comment":"The phrase 'African GNNS TEC model' should read 'African GNSS TEC model'.","section":"AfriTEC Model Data"},{"comment":"The competing-interests statement says 'The author declares no competing interests', but the manuscript lists four authors; the statement should be pluralized or signed by all authors.","section":"Competing Interests"},{"comment":"The caption and text describe the MOIU period as '10-21 January 2016', but the text later says 'January 13 and 14' as the active days; please check the date range for consistency with the actual data interval used.","section":"Figure 2 caption and Results text"}],"recommendation":"reject","confidential_remarks":"The internal contradictions are severe enough that the paper is not publishable in its present form. The authors need to reconcile the abstract and conclusion with their own tables, explain the 10.583 TECU MOIU outlier, and establish whether the validation stations overlap with AfriTEC's training data. If those issues are resolved, a revised manuscript focusing on the actual error distribution and the solstice/equinox comparison could be reconsidered."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The first thing to know: the abstract and conclusion claim that AfriTEC MAE values are generally below 1.5 TECU, but the paper's own Tables 2–6 show that only 6 of 38 non-NaN AfriTEC MAE values meet that threshold, with most ranging from 1.7 to 6.96 TECU. That is a load-bearing inconsistency, not a cosmetic one.\n\nWhat is actually new and useful: a five-station comparison of AfriTEC versus NeQuick over East Africa for 2016–2017, using publicly available GNSS data. The correlation coefficients are mostly above 0.9, so the qualitative claim that AfriTEC captures the diurnal shape of TEC is well supported. The station-by-station tables extend the authors' earlier validation work and could be a useful reference for regional modelers.\n\nThe soft spots are serious. The internal numbers do not line up: the Discussion says MAE ranged between 1.2 and 4.6 TECU, but the tables include values up to 6.96. The seasonal narrative flips—the abstract emphasizes equinox performance, while the Seasonal Variations section says solstice performance is stronger, which the tables actually support. The Weaknesses section cites a MOIU outlier of 10.583 TECU that does not appear in Table 2 (March Equinox 2016 is listed as 4.583). The GNSS processing is under-specified: no elevation mask, no differential code bias calibration details, no mapping function description. There is also no disclosure of whether the five validation stations were part of AfriTEC's training set, which is critical for a neural-network model trained on African GNSS TEC. No error bars or significance tests are provided.\n\nTo give credit where it is due: the high correlations and the qualitative agreement are real, and the comparison with NeQuick is a legitimate contribution. But the quantitative headline is false on the paper's own evidence, and the seasonal story is confused.\n\nThis paper is for regional ionospheric modelers who care about AfriTEC validation. In its current form I would not cite it. However, the underlying dataset is real and the comparison is of interest, so a serious editor could send it to peer review with the expectation of major revision. The referee report should demand that the abstract, conclusion, and seasonal discussion be rewritten to match the tables, and that the authors specify the GNSS processing steps and the training/validation split. If the authors cannot correct the internal contradictions, the paper should be rejected.","headline":"A useful validation dataset undercut by a headline MAE claim that its own tables contradict.","tokens_in":12928,"tokens_out":2807,"would_cite":false,"duration_ms":30972,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims AfriTEC captures East African TEC with errors generally below 1.5 TECU.","keywords":["Ionosphere","NeQuick model","AfriTEC model","Total Electron Content","F10.7","GNSS","East Africa","Solar Cycle 24"],"falsifier":"Recompute AfriTEC's MAE for the same five stations and seasons using an independently processed TEC data set, computed with a documented bias-removal chain and at stations that were definitely not used to train AfriTEC; if the median equinox MAE exceeds 1.5 TECU at most stations, the abstract's headline accuracy claim is false.","tokens_in":11781,"feed_emoji":"📡","tokens_out":11208,"duration_ms":112921,"temperature":0.7,"pith_summary":"This paper tests AfriTEC, a regional African ionosphere model, against GNSS-derived total electron content (TEC) from five stations in East Africa during 2016-2017, the descending phase of Solar Cycle 24. The authors aim to show that AfriTEC captures the daily and seasonal TEC cycle in this equatorial region, with mean absolute error generally below 1.5 TECU and correlation coefficients above 0.80, and that it beats the global NeQuick model at most stations. They also document where the model struggles, mainly at solstice periods, after sunset, and when the equatorial ionization anomaly is active. If the claims hold, AfriTEC offers East African GNSS users a locally calibrated, quiet-time TEC model that can fill in for sparse ground infrastructure.","feed_headline":"AfriTEC claims under-1.5 TECU errors over East Africa","feed_subtitle":"Five-station 2016-17 test shows strong correlations and beats NeQuick, with gaps at equinox and sunset.","key_machinery":"The load-bearing object is AfriTEC, a neural-network-based regional model that outputs hourly vertical TEC for any African location and day, distributed as a MATLAB toolbox. The argument is carried by comparing that output with GNSS-derived vertical TEC at five IGS stations (MOIU, MAL2, ZAMB, ADIS, and MBAR) using two scalar metrics: mean absolute error (MAE, in TECU) and Pearson correlation coefficient r. Those two numbers are the entire quantitative bridge between model and observation in the paper, and every conclusion about model reliability is read off them.","core_discovery":"The paper's central claim is that AfriTEC, a neural-network regional model trained on African GNSS TEC data, reliably reproduces quiet-time ionospheric behavior over East Africa in 2016-2017. The abstract states the result as MAE values 'generally below 1.5 TECU' with correlations 'exceeding 0.80'; the paper's own Tables 2-6 list MAE values from 0.974 to 6.964 TECU across the five stations and seasons, and the discussion gives a typical MAE range of 1.2 to 4.6 TECU. The paper further claims AfriTEC outperforms NeQuick in most station-season cases, which it attributes to AfriTEC's regional calibration on African data, while conceding weaknesses during solstices, post-sunset hours, and events tied to the equatorial ionization anomaly. The intended takeaway is that AfriTEC is a useful regional modeling tool for the East African sector, but one that would benefit from real-time solar and geomagnetic index inputs.","pith_inferences":["A testable extension the paper leaves open is checking whether any of the five validation stations were part of AfriTEC's training set; if they were, the reported MAE values are in-sample and optimistic for unseen locations.","The gap between the abstract's 'generally below 1.5 TECU' and the tables' 0.97-6.96 TECU range suggests the summary statistic was computed on a subset of cases; recomputing seasonal median MAE across all five stations would settle which number represents typical performance.","Because the paper does not document its TEC calibration chain, a natural follow-up is to re-run the same comparison with an openly documented bias-removal and slant-to-vertical mapping procedure, then see whether AfriTEC's daytime underestimation persists.","The paper's closing suggestion of a hybrid model can be turned into a concrete test: blend AfriTEC with NeQuick and check whether the ensemble's equinox MAE falls below the better single model at MOIU and MBAR."],"forward_implications":["If the claims hold, AfriTEC can serve as a quiet-time TEC source for East Africa, filling observational gaps where GNSS receiver coverage is sparse.","GNSS users in the region could apply AfriTEC-based corrections during low-solar-activity conditions, with error expectations in the 1-5 TECU range rather than the abstract's 1.5 TECU figure.","The comparison with NeQuick suggests that regional calibration buys real accuracy over a global model in the African low-latitude sector, at least under moderate ionospheric conditions.","The identified failure modes, especially solstice, post-sunset, and equatorial ionization anomaly periods, point to where adding real-time solar and geomagnetic indices would most improve AfriTEC."],"supporting_citations":[{"why":"Defines and trains the AfriTEC model; every AfriTEC prediction in the paper is generated from this model.","marker":"Okoh et al. (2019)"},{"why":"Prior performance evaluation of AfriTEC over East Africa in quiet and disturbed conditions; the study positions itself as an extension of this work.","marker":"Data et al. (2025)"},{"why":"Documents TEC variability and scintillation over Kenya during the descending phase of Solar Cycle 24, providing the regional behavior the model is tested against.","marker":"Olwendo et al. (2018)"},{"why":"Supplies the physical account of equatorial F-layer irregularities that motivates why East African TEC is difficult to model.","marker":"Aarons, J. (1993)"},{"why":"Describes the NeQuick model, the global benchmark the paper compares AfriTEC against.","marker":"Radicella, S. M., & Leitinger, R. (2001)"},{"why":"Provides a regional GPS TEC calibration approach for Africa, the basis for treating GNSS-derived TEC as reference values.","marker":"Yizengaw et al. (2014)"}],"fun_headline_variants":["AfriTEC beats NeQuick across East Africa","AfriTEC errors under 1.5 TECU, with caveats","AfriTEC stumbles at sunset, solstice","AfriTEC model: strong diurnal, weak post-sunset"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole evaluation depends on the GNSS-derived TEC serving as an accurate, independent ground truth; the paper does not describe how receiver and satellite biases were removed or whether the five test stations were part of AfriTEC's training data, so a flaw in that reference would shift every MAE and correlation value.","fun_headline_variants_meta":{"raw":{"variants":["AfriTEC beats NeQuick across East Africa","AfriTEC errors under 1.5 TECU, with caveats","AfriTEC stumbles at sunset, solstice","AfriTEC model: strong diurnal, weak post-sunset"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000661,"raw_usage":{"total_tokens":3063,"prompt_tokens":1025,"completion_tokens":2038,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":641,"completion_tokens_details":{"reasoning_tokens":1974}},"tokens_in":641,"tokens_out":2038,"duration_ms":18438,"temperature":1.0,"reasoning_tokens":1974,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T17:35:15.625368+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute AfriTEC's MAE for the same five stations and seasons using an independently processed TEC data set, computed with a documented bias-removal chain and at stations that were definitely not used to train AfriTEC; if the median equinox MAE exceeds 1.5 TECU at most stations, the abstract's headline accuracy claim is false.","supporting_citations":[{"cited_title":"B., Jin, S., Shiokawa, K., Otsuka, Y., Aggarwal, M., Uwamahoro, J., Mungufeni, P., Segun, B., Obafaye, A., Ellahony, N., Okonkwo, C., Tshisaphungo, M., & Shetti, D","cited_arxiv_id":null,"evidence_quote":"Defines and trains the AfriTEC model; every AfriTEC prediction in the paper is generated from this model."},{"cited_title":"O., Baki, P., Cilliers, P","cited_arxiv_id":null,"evidence_quote":"Documents TEC variability and scintillation over Kenya during the descending phase of Solar Cycle 24, providing the regional behavior the model is tested against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the physical account of equatorial F-layer irregularities that motivates why East African TEC is difficult to model."},{"cited_title":"M., & Leitinger, R","cited_arxiv_id":null,"evidence_quote":"Describes the NeQuick model, the global benchmark the paper compares AfriTEC against."},{"cited_title":"A., Retterer, J","cited_arxiv_id":null,"evidence_quote":"Provides a regional GPS TEC calibration approach for Africa, the basis for treating GNSS-derived TEC as reference values."}],"review_version":1}