REVIEW 3 major objections 3 minor 12 references
Solar Cycles: Can They Be Predicted?
T0 review · 3 major / 3 minor · reviewed 2026-08-09 · deepseek-v4-flash
Pith's one-line read A review of more than 100 published predictions finds that most underestimated the amplitude of Solar Cycle 25, while timing predictions performed better.
desk verdict Useful status report on cycle 25 predictions, but its claim that only late-stage predictions work rests on comparing raw monthly sunspot numbers to smoothed prediction targets. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central machinery is the collection of more than 100 published Cycle 25 predictions, which the paper compares against the observed sunspot-number series (smoothed and raw) and against the historical baseline average peak of 178.7 across cycles 1-24. The comparison separates amplitude predictions from timing predictions, and uses the observed values (154.9 smoothed in July 2024; 216 raw in August 2024) to classify which forecasts were consistent with the cycle's behavior. The distinction between smoothed and raw sunspot numbers is structurally important: smoothing helps modelling but hides month-to-month variability that predictions cannot capture.
What would settle it
After Cycle 25 ends, consult the final revised sunspot-number series: if the final smoothed peak is revised to at or below 127 (the average prediction), or if the true maximum occurs well after 2025, then the conclusion that most predictions underestimated the cycle and that only late predictions were consistent would need to be revisited.
Extended reading notes
Core claim
The paper's central claim is that, judged against sunspot-number observations through January 2025, most published predictions for Solar Cycle 25's amplitude were too low: the average prediction was 127 sunspot units, while the highest smoothed value so far is 154.9 (July 2024) and the highest raw monthly value is 216 (August 2024). It argues this pattern, together with the fact that the few consistent amplitude predictions all appeared between 2019 and 2023, supports the view that solar cycles are at best semi-predictable and that only late-stage forecasts of a few years are likely to be reliable.
Load-bearing premise
The assessment assumes that the sunspot numbers observed through January 2025—smoothed 154.9 for July 2024 and raw 216 for August 2024—are sufficient to judge Cycle 25's amplitude, even though the paper grants that the true timing of maximum will not be clear until months later.
Editorial extensions
If this is right
- Forecasters should expect that amplitude predictions made more than a few years before solar maximum will tend to miss on the low side during active cycles, because most Cycle 25 predictions did.
- Resources for operational forecasting may be better spent on short-horizon methods and on tracking the current cycle's rise than on long-lead amplitude predictions.
- Timing predictions should be treated as comparatively more reliable, since most included 2024/25 in their ranges and the observed peak is consistent with those ranges.
- Machine-learning and neural-network forecasts that predicted a weak Cycle 25 should be treated cautiously unless they demonstrate predictive validity across multiple cycles.
- The Cycle 26 predictions already being made should be read in light of Cycle 25's outcome: if the pattern holds, many will likely underestimate the next cycle.
Reading between the lines
- A quantitative extension would be to plot forecast error against lead time across all published predictions; the paper's evidence suggests error grows sharply beyond roughly two to four years, but it does not compute this curve.
- The apparent success of timing predictions may be partly an artifact of wide prediction ranges; a testable inference is that when normalized by range width, timing forecasts may not outperform amplitude forecasts.
- If the 2-4 year predictability horizon holds, the solar dynamo's state is likely dominated by short-correlation stochastic processes, making long-horizon amplitude prediction fundamentally limited rather than merely currently inaccurate.
- A direct extension for Cycle 26 would be to pre-register amplitude and timing predictions with their uncertainty estimates before the cycle begins, then evaluate them against the revised sunspot-number series when the cycle ends.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper reviews more than 100 published predictions for solar cycle 25 (C25), comparing their amplitude and timing forecasts against observed sunspot numbers available through January 2025. The paper reports that most amplitude predictions underestimated the cycle, that timing predictions appear to be performing better, and that the few amplitude predictions consistent with the observed peak so far were all published between 2019 and 2023, leading the author to suggest that only late-stage predictions may be possible. The paper also discusses reasons for inaccurate prediction and suggests directions for future work.
Significance. The paper addresses a timely and practically important question—whether solar cycles can be reliably predicted—and it does so by evaluating a large ensemble of published forecasts against external SILSO observations, which avoids the circularity that would arise if the forecasts were used to construct the benchmark. The observation that most amplitude predictions underestimated C25 is a useful, if preliminary, empirical finding. However, the paper's most consequential conclusion—that only late-stage predictions are possible—rests on a comparison between predicted smoothed sunspot numbers and a raw monthly value, and the paper does not supply the underlying prediction list or quantitative evaluation criteria. As a result, the central claim is not currently established to the standard the paper's framing requires.
major comments (3)
- [Cycle 25 predictions, second paragraph] The claim that "Only a handful of predictions gave maximum sunspot numbers consistent with the August 2024 peak" and the later inference that "the few amplitude predictions that were consistent with the August 2024 peak were all published between 2019 and 2023" use the raw monthly SSN of 216 (August 2024) as the benchmark. According to the paper itself, the highest smoothed value, which is the quantity that prediction targets are conventionally compared against, was 154.9 (July 2024). Raw monthly values routinely exceed the smoothed envelope by tens of sunspot units, so a prediction of a smoothed peak in the range, say, 140–180 would not be inconsistent with the actual cycle even though it lies below 216. Using 216 as the criterion therefore excludes many predictions whose predicted smoothed maxima are in fact consistent with the observed cycle and biases the 'successful' subset toward high-amplitude, late-issuing forecasts. This metric mismatch undermines the paper's central late-stage-prediction claim.
- [Cycle 25 predictions and Figure 1] The paper does not provide the full list of the more than 100 predictions, their bibliographic sources, the target quantities (smoothed vs. raw, SSN v1 vs. v2, peak amplitude vs. average amplitude), or the inclusion and exclusion criteria. Consequently, the reader cannot verify the quoted average prediction (127), the range (50–233), the number of predictions consistent with the observed peak, or the statement that late-issuing predictions were the only successful ones. This lack of transparency is load-bearing because the paper's conclusions are entirely empirical summaries of this unpublished dataset.
- [Cycle 25 predictions, timing paragraph] The statement that "timing predictions seem to be performing better than amplitude predictions" is not quantitatively established. The paper itself notes that "the timing will not be clear until months after," so the assessment is necessarily preliminary, and no metric (e.g., fraction of predictions whose range contains the eventual maximum month, or a proper-scoring rule) is defined. Without a defined target time and an associated evaluation window, the claim that timing predictions are performing better is not falsifiable as presented.
minor comments (3)
- [Discussion] The phrase "this comports with Werner [10]'s conclusion" could be improved by stating whether Werner's conclusion was based on a quantitative threshold (e.g., a specific forecast horizon) and by giving the corresponding statistic from the present data, rather than relying on a general reference.
- [Cycle 25 predictions, first paragraph] The text says "The Astrophysics Data System and others return more than a hundred C25 predictions published between 1983 and 2024," but does not describe the search protocol, deduplication, or how predictions were extracted (e.g., from abstracts, tables, or figures). A brief description of the selection procedure would improve reproducibility.
- [Figure 1] The figure caption mentions lower-left and lower-right panels for amplitude and timing predictions, but the figure as described does not include error bars or a legend for the prediction markers; adding these would make the figure more informative.
Circularity Check
No circularity: the paper evaluates external forecasts against SILSO observations and does not use any forecast to construct its target.
full rationale
The paper's claims are an external performance evaluation, not a derivation. It compiles more than 100 published C25 predictions from the literature and compares them with independently observed sunspot numbers from SILSO, specifically a smoothed value of 154.9 for July 2024 and a raw value of 216 for August 2024. The target quantity, maximum sunspot number amplitude and timing of maximum, comes from observations and not from any fitted model, and no parameter is fit to a subset of predictions and then validated on the rest. The observation that most predictions underestimated the cycle is a direct comparison of published numbers with external data, so it cannot reduce to its inputs by construction. The conclusion that only late-stage predictions were consistent with the August 2024 peak is a statement about the publication dates of the subset satisfying that criterion; even if one disputes whether raw monthly sunspot number should be compared with predicted smoothed amplitudes, that is a metric-consistency and correctness concern, not circularity, because the criterion is not defined in terms of the predictions themselves. There is no load-bearing self-citation: the references to Schüssler, Pesnell, Werner, and others are background context, and the paper's assessment does not depend on a uniqueness theorem or an ansatz imported from the author's prior work. The paper is self-contained as a survey and external benchmark evaluation, so no circular step can be exhibited.
Assumptions & free parameters
assumptions (2)
- domain assumption The SILSO sunspot number series is a reliable measure of solar activity.
- domain assumption The collection of predictions from ADS and other sources is representative of the prediction literature.
Cite this review
Pith. "Pith review of Solar Cycles: Can They Be Predicted?." pith.science (2026). https://pith.science/paper/BRXRNXLN
@misc{pith2026250200978,
author = {Pith},
title = {Pith review of: Solar Cycles: Can They Be Predicted?},
year = {2026},
howpublished = {\url{https://pith.science/paper/BRXRNXLN}},
note = {Machine review of arXiv:2502.00978}
}
read the original abstract
The solar magnetic field, thought to be generated by the motion of plasma within the Sun, alternates on the order of 11-year cycles and is incompletely understood. Industries rely on accurate forecasts of solar activity, but can solar cycles be predicted? Of more than 100 predictions for cycle 25, most underestimated the amplitude (peak sunspot number). Fewer predictions were made for the timing of solar maximum, but timing predictions seem to be performing better than amplitude predictions. Reasons for inaccurate prediction are suggested, and perspectives are given on how future studies might improve upon the extant literature.
Reference graph
Works this paper leans on
-
[1]
Barbuzano, J.: Solar activity ramps up but solar cycle still weak. Sky & Telescope 9, 11 (2023)
work page 2023
-
[2]
Nandy, D.: Progress in solar cycle predictions: Sunspot cycles 24–25 in perspec- tive. Solar Phys. 296(54), 1–27 (2021)
work page 2021
-
[3]
Rodr ´ ıguez, J.: Predicting solar cycle 26 using the polar flux as a precursor, spec- tral analysis, and machine learning: Crossing a gleissberg minimum? Solar Phys. 299(117), 1–18 (2024)
work page 2024
-
[4]
Space Weather 16(9), 1424–1436 (2018)
Okoh, D., Seemala, G., Rabiu, A., Uwamahoro, J., Habarulema, J., Aggarwal, M.: A hybrid regression-neural network (hr-nn) method for forecasting the solar activity. Space Weather 16(9), 1424–1436 (2018)
work page 2018
-
[5]
Solar Physics 295, 1–15 (2020)
Benson, B., Pan, W., Prasad, A., Gary, G., Hu, Q.: Forecasting solar cycle 25 using deep neural networks. Solar Physics 295, 1–15 (2020)
work page 2020
-
[6]
Monthly Notices of the Royal Astronomical Society 515(4), 5062–5070 (2022)
Bizzarri, I., Barghini, D., Mancuso, S., Alessio, S., Rubinetti, S., Taricco, C.: Forecasting the solar cycle 25 using a multistep bayesian neural network. Monthly Notices of the Royal Astronomical Society 515(4), 5062–5070 (2022)
work page 2022
-
[7]
Solar Physics 294(24), 1–15 (2019) 4
Covas, E., Peixinho, N., Fernandes, J.: Neural network forecast of the sunspot butterfly diagram. Solar Physics 294(24), 1–15 (2019) 4
work page 2019
-
[8]
Brehm, N., Bayliss, A., Christl, M., Synal, H.-A., Adolphi, F., Beer, J., Kromer, B., Muscheler, R., Solanki, S.K., Usoskin, I., Bleicher, N., Bollhalder, S., Tyers, C., Wacker, L.: Eleven-year solar cycles over the last millennium revealed by radiocarbon in tree rings. Nat. Geosci. 14, 10–15 (2021)
work page 2021
Show all 12 references
-
[9]
Astrophys
Hiremath, K.M.: Prediction of solar cycle 24 and beyond. Astrophys. Space Sci. 314, 45–49 (2008)
2008
-
[10]
Sun and Geosphere 7(2), 75–80 (2012)
Werner, R.: Sunspot number prediction by an autoregressive model. Sun and Geosphere 7(2), 75–80 (2012)
2012
-
[11]
Sch¨ ussler, M.: Are solar cycles predictable? Astron. Nachr. 328(10), 1087–1091 (2007)
2007
-
[12]
Pesnell, W.D.: Lessons learned from predictions of solar cycle 24. J. Space Weather Space Clim. 10(60), 1–10 (2020) 5
2020
Reviewed August 9, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.