{"id":"b7fb313d-0c58-4000-a8a5-9ea8057c8289","arxiv_id":"2502.00643","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"The paper forecasts a neutral ENSO event for 2025/26, with 69.6% probability, based on climate-network and complexity methods plus an ONI-based logistic regression.","lead":"Using two previously developed statistical methods, the authors forecast that 2025/26 will be a neutral ENSO year with about 70% probability and no El Niño with over 91% probability. The result matters because early ENSO forecasts can inform climate risk planning and expectations for global temperature.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 4 logistic regression contradicts the paper's own 'neutral never followed by La Niña' claim, making the 69.6% neutral probability unreliable.","rationale":"The central claim is the probabilistic forecast for 2025/26, which is built from two components: the probability of no El Niño (~91%) and the probability of neutral versus La Niña given no El Niño. The second component is determined by the Section 4 logistic regression, which yields 23.8% La Niña and thereby reduces the neutral probability from ~91% to 69.6%. This split is the most consequential and least secure part of the forecast. The paper's own Figure 6 observation that neutral events were never followed by La Niña directly contradicts a 23.8% La Niña probability at the current neutral OND value of -0.4°C. The regression achieves this value by pooling neutral and La Niña years, but the La Niña years have colder OND values and are not exchangeable with the current state. If the regression is conditioned on neutral OND values, the La Niña probability drops to near zero, and the neutral probability rises to about 90%, a substantial change in the headline numbers. This is not merely a small-sample concern; it is an internal inconsistency between the paper's reported data and its modeling choice. The no-El Niño component has at least some out-of-sample track record, so the qualitative 'likely neutral' conclusion may hold, but the numerical 69.6% is not well supported. Since the issue can be addressed by a better conditional analysis, conditional acceptance remains appropriate, matching the reader's verdict.","tokens_in":10461,"tokens_out":5995,"duration_ms":55547,"concrete_test":"Using the same ONI data as Figure 6, restrict the Section 4 logistic regression to years where the OND ONI is between -0.5°C and 0.5°C (neutral) and repeat the prediction at OND = -0.4°C. If the resulting La Niña probability is much lower than the reported 23.8% (e.g., near 0% or at most the Laplace estimate of 10%), the 69.6% neutral probability is not supported and should be revised upward toward the ~90% level implied by the paper's own observation. Additionally, compute the empirical frequency of La Niña following neutral OND values in a small window around -0.4°C to confirm the regression's extrapolation is not an artifact of distant La Niña years.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The final headline probability (69.6% neutral, 21.8% La Niña) depends on a logistic regression in Section 4 that is inconsistent with the paper's own data. The text states that 'neutral events were not followed by a La Niña phase' and reports 8 neutral-to-neutral transitions with zero neutral-to-La Niña transitions since 1950. The current OND ONI is -0.4°C, a neutral value. Yet the logistic regression, trained on all non-El Niño years followed by non-El Niño years (including La Niña years with OND below -0.5), predicts a 23.8% La Niña probability at OND = -0.4. This regression pools non-exchangeable data: the La Niña years in the training set have colder OND values that are more likely to persist as La Niña. The fitted logistic curve interpolates across the -0.5 boundary, producing a substantial La Niña estimate at a neutral OND despite zero historical occurrences in that regime. If one instead conditions on neutral OND values, the probability of La Niña is 0 (or 1/10 under Laplace's rule), leading to a neutral probability of roughly 90%, not 69.6%. The 69.6% is therefore an artifact of an unjustified parametric extrapolation rather than a robust estimate. This directly affects the central claim's numerical probabilities, even though the qualitative conclusion that neutral is more likely than La Niña or El Niño may survive.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper applies two previously developed forecasting approaches to the 2025 ENSO state: a climate-network method based on mean link strength S(t) and a complexity method based on System Sample Entropy (SysSampEn). Both methods forecast the absence of an El Niño in 2025, with claimed probabilities of 91.2% and 91.7%. The authors then use a logistic regression on the Oceanic Niño Index (ONI) to split the no-El Niño outcome into neutral (69.6%) and La Niña (21.8%) probabilities for NDJ 2025/26, and compare these with the NOAA CPC forecast.","tokens_in":10817,"tokens_out":4094,"duration_ms":40041,"significance":"If the forecast is correct, the paper provides a useful early-warning demonstration for a key climate mode, and the network method has a genuine real-time track record since 2011 (12/13 correct forecasts in its original version, including the successful 2023/24 El Niño forecast). The SysSampEn method also has a published methodology with out-of-sample application. However, the quantitative probability claims are weakened by small sample sizes, in-sample parameter selection, and an internal inconsistency in the logistic-regression step. The central qualitative conclusion (no El Niño in 2025, with neutral more likely than La Niña) is plausible, but the specific numbers 91.2%, 91.7%, and 69.6% are not supported as stated.","major_comments":[{"comment":"The text states that 'neutral events were not followed by a La Niña phase' and reports 8 neutral-to-neutral transitions with zero neutral-to-La Niña transitions since 1950. Yet the logistic regression is fitted to all non-El Niño years followed by non-El Niño years, which includes La Niña years with OND values below -0.5°C. At the current OND = -0.4°C, the historical conditional frequency of La Niña is 0/8 (or 1/10 under Laplace's rule), but the fitted curve gives 23.8%. The final 69.6% neutral probability is therefore an artifact of pooling non-exchangeable data and extrapolating across the -0.5°C boundary. This is load-bearing for the headline probabilities and must be replaced with an analysis that conditions on the observed neutral OND state.","section":"Section 4, Figures 6 and 7"},{"comment":"The 91.7% probability is the in-sample hit rate of a threshold and parameter combination (m=30, p=30, γ=8, l_eff=360) selected for best hindcast skill in reference [4]. The same 12 cases are used to select the parameters and to evaluate the rule, so the reported probability is a fitted quantity rather than an out-of-sample predictive probability. No confidence interval or cross-validation is provided. The central no-El Niño conclusion may survive, but the 91.7% number is not a valid probability estimate as presented.","section":"Section 3.2, Eq. (1)"},{"comment":"The 91.2% no-El Niño probability for version (ii) is based on 31/34 absence forecasts over 1981-2024, but version (ii) was introduced in 2022 and applied retrospectively. Only a subset of those 34 cases are genuine out-of-sample forecasts; the explicitly reported real-time record is 12/13 correct for 2012-2024 using version (i). The paper should separate prospective out-of-sample performance from hindcast performance and report binomial confidence intervals for the proportions.","section":"Section 2.2"},{"comment":"The 'combined probability' of 91.4% is the arithmetic mean of 91.2% and 91.7%, not a combination of independent forecast probabilities. If the two methods were treated as independent, the no-El Niño probability would be roughly 1 - (1-0.912)(1-0.917) ≈ 0.993. The subsequent multiplication of the logistic-regression probabilities by 0.914 also treats the no-El Niño probability as a single number and double-counts it. The combination rule and any assumption about dependence between the two methods must be stated explicitly.","section":"Section 4, combination of forecasts"}],"minor_comments":[{"comment":"The sentence 'For a brief description of the approach we follow [5]' appears twice in the same subsection; the duplicate should be removed.","section":"Section 3.1"},{"comment":"The caption contains a typo: '0 for a neural event' should read '0 for a neutral event'.","section":"Figure 7 caption"},{"comment":"The phrase 'in Figure 6, 2 such cases are on top of each other' is unclear; please explain explicitly how overlapping points are counted in the total of 8 cases.","section":"Section 4, Figure 6"},{"comment":"The description of version (ii) of the network algorithm says alarms are considered only when 'the ONI remains below 0.5°C for the rest of the calendar year'; the timing of this condition relative to the alarm date needs clarification.","section":"Section 2.1"},{"comment":"All claimed probabilities should be accompanied by confidence intervals or Bayesian credibility intervals, given the small numbers of events on which they are based.","section":"Sections 3 and 4"}],"recommendation":"major_revision","confidential_remarks":"This is a forecast report rather than a methods paper. The qualitative no-El Niño forecast is credible given the network method's real-time track record, but the quantitative probabilities require substantial revision. The authors should re-estimate probabilities with genuinely out-of-sample evaluations, remove the inconsistent logistic regression in Section 4, and clearly state combination assumptions. With those changes, the paper could be publishable as a short forecast report; in its current form the headline numbers are not reliable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is an annual ENSO forecast from the established Ludescher–Bunde–Schellnhuber pipeline, not a new method. The genuinely new pieces are the 2025 target year and a logistic regression on ONI transitions meant to split the non-El Niño category into neutral vs. La Niña. The two published methods both forecast no El Niño in 2025, and that forecast is worth taking seriously because the network method has a real-time track record (12/13 correct since 2012, p≈4e-3) and the complexity method has a credible hindcast record.\n\nWhat the paper does well: it is transparent about sample sizes, states the Laplace-rule assumptions, and gives a concrete falsifiable prediction. The out-of-sample streak is the strongest evidence here, and it is real.\n\nThe soft spots are in the probabilities and in Section 4. The 91.2% and 91.7% figures are in-sample hit rates of thresholds and parameters selected on the same historical record. They are not calibrated predictive probabilities. That is a moderate issue, but it matters if you quote the headline probabilities.\n\nThe bigger problem is the La Niña split. The text says no neutral event since 1950 has been followed by La Niña, and reports 8 neutral-to-neutral transitions. Then it runs a logistic regression on all non-El Niño starts, including La Niña years with OND below -0.5. At the current neutral OND of -0.4, the fitted curve interpolates across the -0.5 boundary and returns a 23.8% La Niña probability. That contradicts the paper's own empirical claim. Conditioning on neutral OND values would give a Laplace probability of roughly 0.1 or 0, making the neutral probability around 90%, not 69.6%. The qualitative conclusion survives, but the headline number is an artifact of pooling non-exchangeable data.\n\nAlso, the 'combined probability' of 91.4% is just an average of two non-independent methods; the real uncertainty is larger.\n\nWho this is for: anyone tracking real-time ENSO forecast verification. It deserves a serious referee because the forecast is falsifiable and the track record is worth documenting, but Section 4 needs reworking or removal before publication.\n\nRecommendation: send to review, with the Section 4 issue flagged as a major revision.","headline":"Useful annual ENSO outlook from an established pipeline, but the 69.6% neutral probability depends on a logistic regression that contradicts the paper's own empirical claim about neutral-to-La Niña transitions.","tokens_in":11310,"tokens_out":3408,"would_cite":false,"duration_ms":32664,"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 paper forecasts a neutral ENSO year for 2025/26 with 69.6% probability, based on a climate-network signal and a complexity-entropy signal that both rule out an El Niño onset in 2025.","keywords":["ENSO forecasting","El Niño","La Niña","climate network","System Sample Entropy","Oceanic Niño Index","spring predictability barrier","global mean temperature"],"falsifier":"If the observed Oceanic Niño Index for November 2025 through January 2026 or the surrounding season reaches at least +0.5°C for five consecutive months, the central forecast of no El Niño in 2025 is false. A season recorded as La Niña, with ONI at or below -0.5°C for five months, would also contradict the paper's 69.6% neutral estimate, though less directly.","tokens_in":10275,"feed_emoji":"🌊","tokens_out":8219,"duration_ms":80555,"temperature":0.7,"pith_summary":"This paper predicts that the 2025/26 ENSO season will be neutral: neither an El Niño nor a La Niña. It applies two forecasting methods developed in earlier work — a climate-network measure of Pacific cooperativity and a System Sample Entropy measure of disorder in the Niño 3.4 region — and both point to the absence of an El Niño, at 91.2% and 91.7% probability. Combining these with a logistic regression on the Oceanic Niño Index gives a 69.6% probability of neutral conditions, 21.8% for La Niña, and 8.6% for El Niño. The authors argue that if this is right, 2025 mean global temperature is likely to fall somewhat below the 2024 level, because strong El Niños are a major driver of record global warmth.","feed_headline":"2025/26 most likely neutral, not El Niño or La Niña","feed_subtitle":"Each method puts the chance of El Niño at less than 9%, and the combined forecast favors neutral at 69.6%.","key_machinery":"The climate network consists of 14 grid points in the central and eastern equatorial Pacific linked to 193 points outside; the mean link strength is computed from cross-correlations of surface air temperature anomalies and compared with a threshold learned in 1950-1980. The System Sample Entropy is the negative log of the conditional probability that similar temperature-anomaly subsequences in the Niño 3.4 region stay similar for longer windows, and a linear regression maps its previous-year value to a forecast El Niño magnitude, with values below 1.31 forecasting absence. A logistic regression on the ONI values of non-El Niño years followed by non-El Niño years, plus Laplace's rule of succession, converts the no-El-Niño forecast into neutral-versus-La-Niña probabilities.","core_discovery":"The central claim is that the cooperative mode in the Pacific climate network and the low complexity of Niño 3.4 temperature anomalies in 2024 both signal that no El Niño will start in 2025. In the network approach the mean link strength stayed below the decision threshold through 2024, and in the entropy approach the 2024 SysSampEn value of 0.79 lies below the 1.31 threshold. Placing the current OND ONI value of -0.4°C in the historical OND-to-NDJ relationship then favors a neutral year over a La Niña. The authors present these as probabilistic forecasts, not deterministic statements.","pith_inferences":["An extension the paper does not draw: the 69.6% neutral estimate inherits considerable uncertainty from tiny samples; adding one more neutral-to-neutral transition to the eight in the record would move the Laplace estimate from 90% to 91%, and a single La Niña following a neutral year would materially change the rough split.","A natural stress test is to run the same two predictors on the 2026 target season in January 2026: stable probabilities would support the claim that these methods genuinely bypass the spring barrier, while large swings would suggest overfitting.","The same ONI-based logistic regression could be extended to predict La Niña onset specifically, where historical support is even thinner; such an extension would sharpen the actionable part of the forecast for agriculture and water management.","Because the forecast depends only on data available in January, it is falsifiable within the calendar year, unlike typical decadal projections."],"forward_implications":["If correct, the 2025/26 season is far more likely to be neutral (69.6%) than La Niña (21.8%) or El Niño (8.6%).","The forecast implies 2025 global mean temperature will likely decline from the 2024 record, though other forcings could offset part of the drop.","Both methods would again demonstrate skill at lead times beyond the spring predictability barrier, reinforcing their value as early-warning tools.","The network algorithm's version (ii) would log its second consecutive correct no-El-Niño forecast after the 2023/24 El Niño call."],"supporting_citations":[{"why":"Introduces the climate-network cooperativity measure and the threshold-learning procedure that defines the network forecast.","marker":"[1]"},{"why":"Extends the network approach to very early warning, establishing the alarm rule and the link-strength threshold band used here.","marker":"[2]"},{"why":"Evaluates the network method's real-time forecasts since 2011 and supplies the hit/false-alarm counts behind the 91.2% no-El-Niño probability.","marker":"[3]"},{"why":"Defines System Sample Entropy and its linear relation to El Niño magnitude, providing the threshold and regression used for the 91.7% forecast.","marker":"[4]"},{"why":"Documents the prior successful forecast of the 2023/24 El Niño, the benchmark against which the current application is framed.","marker":"[5]"},{"why":"Introduces the restrictive version (ii) of the network algorithm that requires the ONI to remain below 0.5°C, used to count alarms in Section 2.","marker":"[56,57]"},{"why":"Supplies Laplace's rule of succession, the basis for the rough 90% neutral-versus-La-Niña estimate before logistic regression.","marker":"[62]"},{"why":"Provides the operational ONI values that define El Niño and La Niña and fix the current OND 2024 value at -0.4°C.","marker":"[12]"},{"why":"Supplies the daily surface air temperature data from NCEP/NCAR Reanalysis I used to compute the network link strengths.","marker":"[58]"},{"why":"Supplies the ERA5 daily near-surface temperatures used to compute SysSampEn and its 2024 value.","marker":"[61]"}],"fun_headline_variants":["2025/26 forecast: neutral ENSO, 70% chance, no El Niño","Climate network and entropy say no El Niño in 2025","Neutral ENSO likely for 2025/26, odds 70%","Two forecasting methods rule out El Niño for 2025","ENSO 2025/26: neutral most likely, La Niña 22%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The forecast probabilities assume that the coming year behaves like the small set of past non-El Niño years used to calibrate them — 34 network cases, 12 entropy cases, and 8 neutral-to-neutral transitions — so if 2025 is not exchangeable with those samples, the quoted percentages lose their support.","fun_headline_variants_meta":{"raw":{"variants":["2025/26 forecast: neutral ENSO, 70% chance, no El Niño","Climate network and entropy say no El Niño in 2025","Neutral ENSO likely for 2025/26, odds 70%","Two forecasting methods rule out El Niño for 2025","ENSO 2025/26: neutral most likely, La Niña 22%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000289,"raw_usage":{"total_tokens":1697,"prompt_tokens":955,"completion_tokens":742,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":571,"completion_tokens_details":{"reasoning_tokens":651}},"tokens_in":571,"tokens_out":742,"duration_ms":8369,"temperature":1.0,"reasoning_tokens":651,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:39:35.287322+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"If the observed Oceanic Niño Index for November 2025 through January 2026 or the surrounding season reaches at least +0.5°C for five consecutive months, the central forecast of no El Niño in 2025 is false. A season recorded as La Niña, with ONI at or below -0.5°C for five months, would also contradict the paper's 69.6% neutral estimate, though less directly.","supporting_citations":[{"cited_title":"Proc Natl Acad Sci USA 110:11742- 11745; ibid","cited_arxiv_id":null,"evidence_quote":"Introduces the climate-network cooperativity measure and the threshold-learning procedure that defines the network forecast."},{"cited_title":"Proc Natl Acad Sci USA 111:2064-2066; ibid","cited_arxiv_id":null,"evidence_quote":"Extends the network approach to very early warning, establishing the alarm rule and the link-strength threshold band used here."},{"cited_title":"https://doi.org/10.1007/s00704-024-05035-0","cited_arxiv_id":null,"evidence_quote":"Evaluates the network method's real-time forecasts since 2011 and supplies the hit/false-alarm counts behind the 91.2% no-El-Niño probability."},{"cited_title":"Proc Natl Acad Sci USA , 117:177-183; idid","cited_arxiv_id":null,"evidence_quote":"Defines System Sample Entropy and its linear relation to El Niño magnitude, providing the threshold and regression used for the 91.7% forecast."},{"cited_title":"Very early warning of a moderate-to-strong El Ni\\~no in 2023","cited_arxiv_id":"2301.10763","evidence_quote":"Documents the prior successful forecast of the 2023/24 El Niño, the benchmark against which the current application is framed."},{"cited_title":"Cambridge university press","cited_arxiv_id":null,"evidence_quote":"Supplies Laplace's rule of succession, the basis for the rough 90% neutral-versus-La-Niña estimate before logistic regression."},{"cited_title":"https://origin.cpc.ncep.noaa.gov/products/analysis monitoring/ensostuff/ONI v5.php","cited_arxiv_id":null,"evidence_quote":"Provides the operational ONI values that define El Niño and La Niña and fix the current OND 2024 value at -0.4°C."},{"cited_title":"(1996) The NCEP/NCAR 40-year reanalysis project","cited_arxiv_id":null,"evidence_quote":"Supplies the daily surface air temperature data from NCEP/NCAR Reanalysis I used to compute the network link strengths."},{"cited_title":"https://climate.copernicus.eu/climate-reanalysis?q=products/ climate-reanalysis","cited_arxiv_id":null,"evidence_quote":"Supplies the ERA5 daily near-surface temperatures used to compute SysSampEn and its 2024 value."}],"review_version":1}