Pith. sign in

REVIEW 2 major objections 4 minor 69 references

AI-boosted rare event sampling to characterize extreme weather

T0 review · 2 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A fast AI weather emulator can supply the missing score function for rare event sampling, giving unbiased heatwave return periods to 50,000 years at roughly 100x lower cost than direct simulation.

desk verdict New and well-validated idea for AI-guided rare event sampling, but the headline speed-up misses the AI emulator's cost. read the letter →

arxiv 2510.27066 v3 pith:VKSFUXP4 submitted 2025-10-31 physics.ao-ph stat.COstat.ML

Amaury Lancelin , Alex Wikner , Laurent Dubus , Clément Le Priol , Dorian S. Abbot , Freddy Bouchet , Pedram Hassanzadeh , Jonathan Weare This is my paper
classification physics.ao-phstat.COstat.ML
keywords rareeventsamplingAIweatheremulatorextremeheatwavesreturnperioddiffusionMonteCarloscorefunctionclimatemodelemulationimportance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces AI+RES, a rare-event sampling method that replaces the hand-designed persistence score function with the ensemble forecast of a fast AI weather emulator. The central claim is that this simple substitution solves the long-standing problem of choosing a score function for high-dimensional chaotic systems. With only 400 walkers of the PlaSim climate model, AI+RES produces unbiased return-period estimates for 7-day mid-latitude heatwaves out to 50,000 years, matching a 50,000-member direct simulation benchmark while cutting computational cost by one to two orders of magnitude. It also yields physically realistic trajectories of the extremes, where standard RES saturates near 100-year return periods and direct emulator sampling gives biased statistics.

What carries the argument

The load-bearing object is the score function θ(Xik) = (1/M) Σj ÂL(tf | Xik), the mean of M=100 AI-emulator forecast trajectories of the 7-day regional temperature average, evaluated from each walker state at each resampling time. This score decides which walkers are duplicated or killed in the Diffusion Monte Carlo algorithm, converting the emulator's short-term predictive skill into an importance-sampling signal. Unbiasedness does not depend on the AI emulator; it is carried by the DMC weights through Eq. (4), so the emulator only needs to rank trajectories, not to be accurate in the tail.

What would settle it

Measure the correlation between the emulator's ensemble-mean forecast of AL(tf) and the true PlaSim AL(tf) at the earliest resampling step (tf − tk = 10 days) for events with return periods above 100 years. The paper's SI Fig. S6 already shows this correlation is weak for France, which it attributes to missing soil moisture. If this correlation is near zero for a given region or event type, the variance speed-up over DNS should fall toward 1, falsifying the central practical claim of an O(100) gain while leaving the unbiasedness claim intact.

Watch

Extended reading notes

Core claim

The paper's central discovery is that an AI emulator does not need to simulate extreme events accurately on its own to be useful; it only needs to rank ongoing simulations well. At each resampling step, AI+RES computes, for every walker, the ensemble-mean forecast of the regional 7-day average temperature at the target time and uses that value as the score in a diffusion Monte Carlo splitting algorithm. Because the emulator is cheap, 100-member forecasts can be run for each of 400 walkers at negligible cost. The result is unbiased estimates of the probability that the heatwave index AL(tf) exceeds very high thresholds, verified against a 50,000-member PlaSim direct simulation for France and

Load-bearing premise

The O(100) speed-up rests on the AI emulator's roughly 10-day-ahead ensemble forecasts ranking PlaSim trajectories correctly even for events far beyond the emulator's training distribution; if that ranking skill vanishes in the extreme tail, the method stays unbiased but the practical efficiency gain collapses.

Editorial extensions

If this is right

  • Rare-event probability estimates no longer require simulation lengths many times the return period; a few hundred physics-model walkers can reach 50,000-year events in the tested model.
  • The persistence score function used by standard RES is the bottleneck: it saturates near 100-year return periods, while the AI-based score continues to sample rarer events.
  • AI emulators used directly as samplers (AI-DNS) give biased tail statistics; their proper role is as a cheap guide for the physics model, not a replacement for it.
  • Because the method generates full trajectories, it can be used to study the dynamics and precursors of extremes, not only their probabilities.
  • The framework transfers to any high-dimensional dynamical system for which an AI surrogate can be trained, as long as the surrogate has ranking skill on the observable of interest.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct consequence the authors do not spell out: the practical requirement on the emulator is rank correlation in the extreme tail, not forecast accuracy; research on AI emulators for extremes should therefore be evaluated by ranking skill, not by tail RMSE.
  • If this ranking requirement generalizes, AI+RES offers a path to estimate how return periods of rare heatwaves shift under climate change, by running the same algorithm on GCMs under different forcing scenarios rather than waiting for long control runs.
  • A testable prediction is that speed-up factors should track the emulator's early-lead tail correlation: regions where soil moisture or other unobserved land states degrade that correlation (as the paper's France case suggests) should show smaller or vanishing speed-ups.
  • A probabilistic emulator, or one coupled to land-surface state, could further reduce estimator variance; the paper identifies deterministic forecasts as a limitation, which is a concrete avenue for improvement.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper introduces AI+RES, a rare-event sampling framework that couples a diffusion Monte Carlo (DMC) splitting algorithm in the PlaSim GCM with an AI weather emulator used to define the score function at resampling times. The score is the ensemble-mean forecast of the 7-day regional temperature anomaly AL(tf) from the emulator. The authors report unbiased return-period estimates for mid-latitude heatwaves over France and Chicago out to 50,000-year return times using only N=400 PlaSim walkers, with claimed computational speed-ups of O(100) over direct numerical simulation (DNS), while standard persistence-based RES saturates at ~100-year return periods and AI-only direct sampling is biased. Validation is performed against a 50,000-member PlaSim DNS ground truth, with additional controls including a perfect-forecast-system RES upper bound and extremal-value-theory baselines.

Significance. If the demonstrated statistical accuracy holds, the paper is significant: it addresses the known bottleneck of score-function design in rare-event sampling by leveraging modern AI weather emulators, and its unbiasedness claim is supported by the DMC theory and by a strong 50,000-member DNS validation. The AI-DNS control is particularly valuable, as it shows that a biased emulator can still serve as an effective ranking function for splitting, and the PFS+RES control usefully bounds the achievable variance reduction. The methodology is potentially transferable to other extremes and to more expensive GCMs. However, the central quantitative claim of an O(100) computational speed-up is not supportable from the reported cost accounting, because the emulator inference cost is nontrivial in the demonstrated PlaSim setting and is omitted from Eqs. (10) and (13).

major comments (2)
  1. [Methods, 'Derivation of the computational speed-up factors', Eqs. (10) and (13); also Abstract and Introduction] The headline 'two orders-of-magnitude lower computational cost' and 'numerical speed-up of up to O(100)' are computed from Eqs. (10) and (13), which count only PlaSim trajectories and rely on the key assumption that AI emulator forecasts cost nothing. The Methods immediately report, however, that the emulator is only about 10x faster than PlaSim on the hardware used. With N=400, M=100 and three or six resampling forecasts (the paper is ambiguous), the emulator overhead is roughly 120,000-240,000 member forecasts, each spanning 12-20 days, equivalent to about 4,000-13,000 PlaSim runs versus 400 for the RES trajectories. The total cost is therefore ~10-33x larger than counted. A variance speed-up of 100 at the 10,000-year return period then corresponds to an actual total-compute speed-up of roughly 3-10, and for France, where VSUF is about 10, AI+RES is not faster than DNS. The statement '
  2. [SI S6; Methods 'Return period curves'; Abstract] The paper repeatedly refers to 'return periods up to 50,000 years' and 'once per millennium heatwaves' without qualification. SI S6 states that these are conditional return times, obtained by repeatedly simulating the same summer from the same initial state at the beginning of the season. This is an important limitation: the quoted return periods are not unconditional climatological return periods. The main text should state this qualification wherever return-period values are presented, and the abstract's wording should be adjusted accordingly. The comparison with DNS remains fair because the same conditional protocol is used, but the practical interpretation of the headline numbers is materially different from what a reader would normally infer.
minor comments (4)
  1. [SI Fig. S6 caption] The caption uses 'PFS-RES' and 'AI-RES' inconsistently with the main text's 'PFS+RES' and 'AI+RES'. Please unify notation.
  2. [SI Fig. S4 caption] Typo: 'GDP distributions' should be 'GPD distributions'.
  3. [Introduction, ref. [17]] Reference [17] is cited as 'in prep'; if possible, replace with a preprint or remove the citation to avoid relying on unpublished work.
  4. [Methods, Eq. (8) and surrounding text] The index convention in Eq. (8) (using \bar{w}_K and e^{-V_K(\hat{X}^i_K)}) is slightly nonstandard relative to Eq. (4) and could be clarified; consider aligning the notation for the final resampling step.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: AI+RES uses a standard unbiased DMC estimator; the AI score function only biases sampling, and the central claims are validated against an external 50,000-member DNS ground truth.

full rationale

The derivation chain is self-contained. The rare-event estimator in Eqs. (4)/(8) is the standard DMC importance-sampling identity; its unbiasedness holds for any score function theta, including the AI-ensemble-mean forecast of Eq. (6), because the weights e^{-V} correctly compensate for the biased selection. The paper directly demonstrates this independence: the AI-DNS baseline (50,000-member emulator ensemble) is explicitly biased ('The AI-DNS baseline is biased both in the mean and variability of the observable'), yet AI+RES using the same emulator as score function produces return-period curves that match the independent PlaSim DNS ground truth (N=50,000) up to 50,000 years (Fig. 2). The PFS+RES control confirms that the mechanism is forecast skill, not circular construction. Self-citations ([17], [18-20], [55]) are background or methodology citations and are not load-bearing; no uniqueness theorem or fitted value is imported from them. The one in-text limitation - Methods: 'We make the key assumption that the computational cost of running the AI emulator ensemble forecasts is negligible... our emulator runs approximately 10 times faster... than PlaSim' - qualifies the magnitude of the headline O(100) speed-up under total-cost accounting, but this is a computational-accounting caveat, not a circular reduction, so it does not raise the circularity score.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard DMC theory plus several domain assumptions: PlaSim as the testbed, conditional-return-period interpretation, and the emulator's cost and forecast-skill premises that support the efficiency claims. The free parameters are algorithm hyperparameters tuned to the target case, which is acceptable for a proof-of-concept but should be stress-tested in follow-ups.

free parameters (5)
  • splitting constants C_k = (0., 0., 0., 1.6, 1.8, 2.0)
    K=6 resampling steps; C_k ≥ 0 control selection strength. Described as 'tuned empirically' (Extended Data Table II); SI §S4.3 gives the equivalent C=2.0, alpha=0.02 from a non-exhaustive grid search.
  • resampling interval tau = 5 days
    Time between resampling steps; chosen to match the forecast skill horizon of the AI emulator (Extended Data Table II).
  • AI ensemble forecast size M = 100
    Number of AI ensemble members used to form the score function (Eq. 6). The paper notes no substantial gains are expected beyond this value.
  • surface pressure perturbation amplitude epsilon = 3e-3
    Perturbation added to walkers after resampling to maintain diversity in the deterministic GCM; authors tested values from 1e-4 to 1e-2 and chose 3e-3 as a balance (SI S2).
  • AI training input noise std = 2.5e-3
    Gaussian noise added to standardized atmospheric inputs during emulator training to stabilize autoregressive prediction (Methods, Training and Validation).
assumptions (5)
  • standard math DMC estimator unbiasedness: Eq. (4) gives unbiased estimates of the original process distribution for any score function that is a function of the state at resampling times.
    Invoked in Methods, 'Rare event Algorithm', citing Del Moral (2004) and Webber et al. (2019). The paper does not explicitly extend the theory to randomized scores built from an emulator ensemble, though empirical validation supports it.
  • domain assumption PlaSim is a faithful stand-in for the real climate system for the purpose of testing the methodology.
    The entire validation and the claims about rare heatwaves are based on PlaSim T42. The paper is explicit that this is a proof-of-concept; transfer to full GCMs is conjectured.
  • domain assumption Return periods are conditional on repeating the same summer initial condition each year.
    Explicitly acknowledged in SI S6: the return times correspond to the same t0 (July 2) initial state, not to unconditional climatological return periods.
  • ad hoc to paper The computational cost of running AI emulator ensemble forecasts is negligible compared to the physics-based model.
    Central to the derivation of both speed-up factors (Methods, 'Derivation of the computational speed-up factors'). For PlaSim the emulator is only ~10× faster, so this assumption is approximate for the presented validation.
  • domain assumption The AI emulator's ensemble spread (from initial perturbations) mimics the internal variability of PlaSim well enough to rank walkers.
    Relied upon in the score function construction (Methods, 'Ensemble Forecasting'); the paper validates forecast skill in Extended Data Fig. 2 but notes weaker rare-event skill for France at long lead times in Fig. S6.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AI-boosted rare event sampling to characterize extreme weather." pith.science (2026). https://pith.science/paper/VKSFUXP4

@misc{pith2026251027066,
  author       = {Pith},
  title        = {Pith review of: AI-boosted rare event sampling to characterize extreme weather},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VKSFUXP4}},
  note         = {Machine review of arXiv:2510.27066}
}
read the original abstract

Weather extremes pose major societal risks, especially in a changing climate, but due to their rarity, they are difficult to study using limited observations or complex climate models. We introduce AI+RES, a framework coupling fast AI weather forecasts with a high-fidelity physics model using a rare-event algorithm to efficiently characterize extremes. This approach enables the study of the statistics and physics of very rare events, such as once per millennium heatwaves at two orders-of-magnitude lower computational cost. AI+RES can be applied broadly across climate science and other fields concerned with rare events.

Figures

Figures reproduced from arXiv: 2510.27066 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

69 extracted references · 7 linked inside Pith

  1. [1]

    Extreme weather and climate change: population health and health system implications,

    K. L. Ebi, J. Vanos, J. W. Baldwin, J. E. Bell, D. M. Hondula, N. A. Errett, K. Hayes, C. E. Reid, S. Saha, J. Spector, et al. , “Extreme weather and climate change: population health and health system implications,” Annual review of public health , vol. 42, no. 1, pp. 293–315, 2021

    2021

  2. [2]

    Extreme weather and climate events with ecological relevance: a review,

    C. C. Ummenhofer and G. A. Meehl, “Extreme weather and climate events with ecological relevance: a review,” Philosophical Transactions of the Royal Society B: Biological Sciences , vol. 372, no. 1723, p. 20160135, 2017

    2017

  3. [3]

    Extreme weather events on energy systems: a comprehensive review on impacts, mitigation, and adaptation measures,

    A. C. Gon¸ calves, X. Costoya, R. Nieto, and M. L. Liberato, “Extreme weather events on energy systems: a comprehensive review on impacts, mitigation, and adaptation measures,” Sustainable Energy Research, vol. 11, no. 1, p. 4, 2024

    2024

  4. [4]

    Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change

    IPCC, Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change . Cambridge University Press, 2021

    2021

  5. [5]

    Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change

    IPCC, Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change . Cambridge University Press, 2022

    2022

  6. [6]

    The effect of a short observational record on the statistics of temperature extremes,

    J. Zeder, S. Sippel, O. C. Pasche, S. Engelke, and E. M. Fischer, “The effect of a short observational record on the statistics of temperature extremes,” Geophysical Research Letters, vol. 50, no. 16, p. e2023GL104090, 2023

    2023

  7. [7]

    Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators,

    J. Pathak, S. Subramanian, P. Harrington, S. Raja, A. Chattopadhyay, M. Mardani, T. Kurth, D. Hall, Z. Li, K. Azizzadenesheli, et al. , “Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators,” arXiv preprint arXiv:2202.11214 , 2022

    arXiv 2022

  8. [8]

    Learning skillful medium-range global weather forecasting,

    R. Lam, A. Sanchez-Gonzalez, M. Willson, P. Wirnsberger, M. Fortunato, F. Alet, S. Ravuri, T. Ewalds, Z. Eaton-Rosen, W. Hu, et al. , “Learning skillful medium-range global weather forecasting,” Science, vol. 382, no. 6677, pp. 1416–1421, 2023

    2023

Show all 69 references
  1. [9]

    Accurate medium-range global weather forecasting with 3D neural networks,

    K. Bi, L. Xie, H. Zhang, X. Chen, X. Gu, and Q. Tian, “Accurate medium-range global weather forecasting with 3D neural networks,” Nature, vol. 619, no. 7970, pp. 533–538, 2023. 25

    2023

  2. [10]

    Probabilistic weather forecasting with machine learning,

    I. Price, A. Sanchez-Gonzalez, F. Alet, T. R. Andersson, A. El-Kadi, D. Masters, T. Ewalds, J. Stott, S. Mohamed, P. Battaglia, et al. , “Probabilistic weather forecasting with machine learning,” Nature, vol. 637, no. 8044, pp. 84–90, 2025

    2025

  3. [11]

    ACE2: Accurately learning subseasonal to decadal atmospheric variability and forced responses,

    O. Watt-Meyer, B. Henn, J. McGibbon, S. K. Clark, A. Kwa, W. A. Perkins, E. Wu, L. Har- ris, and C. S. Bretherton, “ACE2: Accurately learning subseasonal to decadal atmospheric variability and forced responses,” Nov. 2024

    2024

  4. [12]

    Neural general circulation models for weather and climate,

    D. Kochkov, J. Yuval, I. Langmore, P. Norgaard, J. Smith, G. Mooers, M. Kl¨ ower, J. Lottes, S. Rasp, P. D¨ uben,et al., “Neural general circulation models for weather and climate,”Nature, vol. 632, no. 8027, pp. 1060–1066, 2024

    2024

  5. [13]

    Camulator: Fast emulation of the community atmosphere model,

    W. E. Chapman, J. S. Schreck, Y. Sha, D. J. Gagne II, D. Kimpara, L. Zanna, K. J. Mayer, and J. Berner, “Camulator: Fast emulation of the community atmosphere model,” arXiv preprint arXiv:2504.06007, 2025

    2025 arXiv

  6. [14]

    Can AI weather models predict out-of-distribution gray swan tropical cyclones?,

    Y. Q. Sun, P. Hassanzadeh, M. Zand, A. Chattopadhyay, J. Weare, and D. S. Abbot, “Can AI weather models predict out-of-distribution gray swan tropical cyclones?,” Proceedings of the National Academy of Sciences , vol. 122, p. e2420914122, May 2025

    2025

  7. [15]

    Predicting beyond training data via extrapolation versus translocation: Ai weather models and dubai’s unprecedented 2024 rainfall,

    Y. Q. Sun, P. Hassanzadeh, T. Shaw, and H. A. Pahlavan, “Predicting beyond training data via extrapolation versus translocation: Ai weather models and dubai’s unprecedented 2024 rainfall,” arXiv e-prints , pp. arXiv–2505, 2025

    2024

  8. [16]

    Numerical models outperform ai weather forecasts of record-breaking extremes,

    Z. Zhang, E. Fischer, J. Zscheischler, and S. Engelke, “Numerical models outperform ai weather forecasts of record-breaking extremes,” arXiv preprint arXiv:2508.15724 , 2025

    2025 arXiv

  9. [17]

    Can ai climate emulators quantify the statistics of the rarest unseen weather extremes?

    A. Wikner, A. Lancelin, T. Arcomano, D. P. Karan Jakhar, F. Bouchet, and P. Hassanzadeh, “Can ai climate emulators quantify the statistics of the rarest unseen weather extremes?.” in prep

  10. [18]

    Computation of extreme heat waves in climate models using a large deviation algorithm,

    F. Ragone, J. Wouters, and F. Bouchet, “Computation of extreme heat waves in climate models using a large deviation algorithm,” Proceedings of the National Academy of Sciences , vol. 115, no. 1, pp. 24–29, 2018

    2018

  11. [19]

    Practical rare event sampling for extreme mesoscale weather,

    R. Webber, D. Plotkin, M. O’Neill, D. Abbot, and J. Weare, “Practical rare event sampling for extreme mesoscale weather,” Chaos: An Interdisciplinary Journal of Nonlinear Science , vol. 29, p. 053109, 05 2019

    2019

  12. [20]

    Rare event sampling improves mercury instability statistics,

    D. S. Abbot, R. J. Webber, S. Hadden, D. Seligman, and J. Weare, “Rare event sampling improves mercury instability statistics,” The Astrophysical Journal , vol. 923, no. 2, p. 236, 26 2021

    2021

  13. [21]

    Rare event simulation of extreme european winter rainfall in an intermediate complexity climate model,

    J. Wouters, R. K. Schiemann, and L. C. Shaffrey, “Rare event simulation of extreme european winter rainfall in an intermediate complexity climate model,”Journal of Advances in Modeling Earth Systems , vol. 15, no. 4, p. e2022MS003537, 2023

    2023

  14. [22]

    Using rare event algorithms to understand the statistics and dynamics of extreme heatwave seasons in south asia,

    C. Le Priol, J. M. Monteiro, and F. Bouchet, “Using rare event algorithms to understand the statistics and dynamics of extreme heatwave seasons in south asia,” Environmental Research: Climate, vol. 3, no. 4, p. 045016, 2024

    2024

  15. [23]

    Statistical and dynamical aspects of extremely hot summers in western europe sampled with a rare events algorithm,

    R. Noyelle, A. Caubel, Y. Meurdesoif, P. Yiou, and D. Faranda, “Statistical and dynamical aspects of extremely hot summers in western europe sampled with a rare events algorithm,” Journal of Climate , 2025

    2025

  16. [24]

    A meta-analysis of the total economic impact of climate change,

    R. S. Tol, “A meta-analysis of the total economic impact of climate change,” Energy Policy, vol. 185, p. 113922, 2024

    2024

  17. [25]

    Swiss Re SONAR: New emerging risk insights,

    J. Anchen, V. B. Gonzalez, M. Chatterjee, R. Egloff, A. Felderer, A. Mejler¨ o, A. Vischer, and B. Wilke, “Swiss Re SONAR: New emerging risk insights,” tech. rep., Swiss Re Institute, Zurich, Switzerland, June 2025

    2025

  18. [26]

    Embrechts, C

    P. Embrechts, C. Kl¨ uppelberg, and T. Mikosch,Modelling Extremal Events: for Insurance and Finance. Stochastic Modelling and Applied Probability, Springer Berlin Heidelberg, 2013

    2013

  19. [27]

    Estimating changes in temperature extremes from millennial-scale climate simulations using generalized extreme value (GEV) distributions,

    W. K. Huang, M. L. Stein, D. J. McInerney, S. Sun, and E. J. Moyer, “Estimating changes in temperature extremes from millennial-scale climate simulations using generalized extreme value (GEV) distributions,” Advances in Statistical Climatology, Meteorology and Oceanogra- phy, ...

    2016

  20. [28]

    Convergence of Extreme Value Statistics in a Two- Layer Quasi-Geostrophic Atmospheric Model,

    V. M. G´ alfi, T. B´ odai, and V. Lucarini, “Convergence of Extreme Value Statistics in a Two- Layer Quasi-Geostrophic Atmospheric Model,” Complexity, vol. 2017, no. 1, p. 5340858, 2017

    2017

  21. [29]

    Rare event algorithm study of extreme warm summers and heatwaves over Europe,

    F. Ragone and F. Bouchet, “Rare event algorithm study of extreme warm summers and heatwaves over Europe,” Geophysical Research Letters, vol. 48, no. 12, p. e2020GL091197, 2021

    2021

  22. [30]

    Bringing Statistics to Storylines: Rare Event Sampling for Sudden, Transient Extreme Events,

    J. Finkel and P. A. O’Gorman, “Bringing Statistics to Storylines: Rare Event Sampling for Sudden, Transient Extreme Events,” Journal of Advances in Modeling Earth Systems , vol. 16, no. 6, p. e2024MS004264, 2024

    2024

  23. [31]

    The hot summer of 2010: redrawing the temperature record map of europe,

    D. Barriopedro, E. M. Fischer, J. Luterbacher, R. M. Trigo, and R. Garc ´ ıa-Herrera, “The hot summer of 2010: redrawing the temperature record map of europe,” Science, vol. 332, 27 no. 6026, pp. 220–224, 2011

    2010

  24. [32]

    Human contribution to the european heatwave of 2003,

    P. A. Stott, D. A. Stone, and M. R. Allen, “Human contribution to the european heatwave of 2003,” Nature, vol. 432, no. 7017, pp. 610–614, 2004

    2003

  25. [33]

    The unprecedented pacific northwest heatwave of june 2021,

    R. H. White, S. Anderson, J. F. Booth, G. Braich, C. Draeger, C. Fei, C. D. Harley, S. B. Henderson, M. Jakob, C.-A. Lau, et al. , “The unprecedented pacific northwest heatwave of june 2021,” Nature communications, vol. 14, no. 1, p. 727, 2023

    2021

  26. [34]

    Very rare heat extremes: Quantify- ing and understanding using ensemble reinitialization,

    C. Gessner, E. M. Fischer, U. Beyerle, and R. Knutti, “Very rare heat extremes: Quantify- ing and understanding using ensemble reinitialization,” Journal of Climate , vol. 34, no. 16, pp. 6619 – 6634, 2021

    2021

  27. [35]

    Estimating return periods for extreme events in climate models through ensemble boosting,

    L. Bloin-Wibe, R. Noyelle, V. Humphrey, U. Beyerle, R. Knutti, and E. Fischer, “Estimating return periods for extreme events in climate models through ensemble boosting,” EGUsphere, vol. 2025, pp. 1–40, 2025

    2025

  28. [36]

    Boosting ensembles for statistics of tails at conditionally optimal advance split times,

    J. Finkel and P. A. O’Gorman, “Boosting ensembles for statistics of tails at conditionally optimal advance split times,” arXiv preprint arXiv:2507.22310 , 2025

    2025 arXiv

  29. [37]

    The rise of data-driven weather forecasting: A first statistical assessment of machine learning–based weather forecasts in an operational-like context,

    Z. Ben Bouallegue, M. C. Clare, L. Magnusson, E. Gascon, M. Maier-Gerber, M. Janouˇ sek, M. Rodwell, F. Pinault, J. S. Dramsch, S. T. Lang, et al. , “The rise of data-driven weather forecasting: A first statistical assessment of machine learning–based weather forecasts in an o...

    2024

  30. [38]

    Huge ensembles–part 1: Design of ensemble weather forecasts using spherical fourier neural operators,

    A. Mahesh, W. D. Collins, B. Bonev, N. Brenowitz, Y. Cohen, J. Elms, P. Harrington, K. Kashinath, T. Kurth, J. North, et al., “Huge ensembles–part 1: Design of ensemble weather forecasts using spherical fourier neural operators,” Geoscientific Model Development , vol. 18, no. ...

    2025

  31. [39]

    Huge ensembles–part 2: Properties of a huge ensemble of hindcasts generated with spherical fourier neural operators,

    A. Mahesh, W. D Collins, B. Bonev, N. Brenowitz, Y. Cohen, P. Harrington, K. Kashinath, T. Kurth, J. North, T. A. O’Brien, et al. , “Huge ensembles–part 2: Properties of a huge ensemble of hindcasts generated with spherical fourier neural operators,” Geoscientific Model Develo...

    2025

  32. [40]

    The Planet Simulator: Towards a user friendly model,

    K. Fraedrich, H. Jansen, E. Kirk, U. Luksch, and F. Lunkeit, “The Planet Simulator: Towards a user friendly model,” Meteorologische Zeitschrift, pp. 299–304, July 2005

    2005

  33. [41]

    Global, regional, and national burden of mortality 28 associated with non-optimal ambient temperatures from 2000 to 2019: a three-stage modelling study,

    Q. Zhao, Y. Guo, T. Ye, A. Gasparrini, S. Tong, A. Overcenco, A. Urban, A. Schneider, A. Entezari, A. M. Vicedo-Cabrera, et al., “Global, regional, and national burden of mortality 28 associated with non-optimal ambient temperatures from 2000 to 2019: a three-stage modelling s...

    2000

  34. [42]

    The global costs of extreme weather that are attributable to climate change,

    R. Newman and I. Noy, “The global costs of extreme weather that are attributable to climate change,” Nature Communications, vol. 14, no. 1, p. 6103, 2023

    2023

  35. [43]

    The most at-risk regions in the world for high-impact heatwaves,

    V. Thompson, D. Mitchell, G. C. Hegerl, M. Collins, N. J. Leach, and J. M. Slingo, “The most at-risk regions in the world for high-impact heatwaves,”Nature Communications, vol. 14, no. 1, p. 2152, 2023

    2023

  36. [44]

    Probabilistic forecasts of extreme heatwaves using convolutional neural networks in a regime of lack of data,

    G. Miloshevich, B. Cozian, P. Abry, P. Borgnat, and F. Bouchet, “Probabilistic forecasts of extreme heatwaves using convolutional neural networks in a regime of lack of data,” Physical Review Fluids, vol. 8, p. 040501, Apr. 2023

    2023

  37. [45]

    Skillful joint probabilistic weather forecasting from marginals,

    F. Alet, I. Price, A. El-Kadi, D. Masters, S. Markou, T. R. Andersson, J. Stott, R. Lam, M. Willson, A. Sanchez-Gonzalez, et al., “Skillful joint probabilistic weather forecasting from marginals,” arXiv preprint arXiv:2506.10772 , 2025

    2025 arXiv

  38. [46]

    Aifs-crps: ensemble forecasting using a model trained with a loss function based on the continuous ranked probability score,

    S. Lang, M. Alexe, M. C. Clare, C. Roberts, R. Adewoyin, Z. B. Bouall` egue, M. Chantry, J. Dramsch, P. D. Dueben, S. Hahner, et al. , “Aifs-crps: ensemble forecasting using a model trained with a loss function based on the continuous ranked probability score,” arXiv preprint ...

    2024 arXiv

  39. [47]

    Archesweather & archesweathergen: a deterministic and generative model for efficient ml weather forecasting,

    G. Couairon, R. Singh, A. Charantonis, C. Lessig, and C. Monteleoni, “Archesweather & archesweathergen: a deterministic and generative model for efficient ml weather forecasting,” arXiv preprint arXiv:2412.12971 , 2024

    2024 arXiv

  40. [48]

    Reframing generative models for physical systems using stochastic interpolants,

    A. Zhou, A. Wikner, A. Lancelin, P. Hassanzadeh, and A. B. Farimani, “Reframing generative models for physical systems using stochastic interpolants,” arXiv preprint arXiv:2509.26282 , 2025

    2025

  41. [49]

    Hot and cool summers: Multiple equilibria of the continental water cycle,

    F. D’Andrea, A. Provenzale, R. Vautard, and N. De Noblet-Decoudr´ e, “Hot and cool summers: Multiple equilibria of the continental water cycle,” Geophysical Research Letters , vol. 33, no. 24, 2006

    2006

  42. [50]

    Soil Moisture– Atmosphere Interactions during the 2003 European Summer Heat Wave,

    E. M. Fischer, S. I. Seneviratne, P. L. Vidale, D. L¨ uthi, and C. Sch¨ ar, “Soil Moisture– Atmosphere Interactions during the 2003 European Summer Heat Wave,”Journal of Climate, 2007

    2003

  43. [51]

    Summertime European heat and drought waves induced by 29 wintertime Mediterranean rainfall deficit,

    R. Vautard, P. Yiou, F. D’Andrea, N. de Noblet, N. Viovy, C. Cassou, J. Polcher, P. Ciais, M. Kageyama, and Y. Fan, “Summertime European heat and drought waves induced by 29 wintertime Mediterranean rainfall deficit,” Geophysical Research Letters, vol. 34, no. 7, 2007

    2007

  44. [52]

    Extreme heat wave sampling and prediction with analog Markov chain and comparisons with deep learning,

    G. Miloshevich, D. Lucente, P. Yiou, and F. Bouchet, “Extreme heat wave sampling and prediction with analog Markov chain and comparisons with deep learning,” Environmental Data Science, vol. 3, p. e9, Jan. 2024

    2024

  45. [53]

    The sea surface temperature and sea ice concentra- tion boundary conditions for AMIP II simulations,

    K. Taylor, D. Williamson, and F. Zwiers, “The sea surface temperature and sea ice concentra- tion boundary conditions for AMIP II simulations,” PCDMI Report 60, Program for Climate Model Diagnosis and Intercomparison, Lawrence Livermore National Laboratory, Apr. 2001

    2001

  46. [54]

    Planet simulator - reference manual, version 16.0,

    F. Lunkeit, E. Kirk, H. Borth, A. Kleidon, M. B¨ ottinger, U. Luksch, K. Fraedrich, H. Jansen, P. Paiewonsky, S. Schubert, F. Sielmann, and H. Wan, “Planet simulator - reference manual, version 16.0,” 2011

    2011

  47. [55]

    Challenges of learning multi-scale dy- namics with AI weather models: Implications for stability and one solution,

    A. Chattopadhyay, Y. Q. Sun, and P. Hassanzadeh, “Challenges of learning multi-scale dy- namics with AI weather models: Implications for stability and one solution,” Dec. 2024

    2024

  48. [56]

    Del Moral, Feynman-Kac formulae: genealogical and interacting particle systems with ap- plications

    P. Del Moral, Feynman-Kac formulae: genealogical and interacting particle systems with ap- plications. Springer, 2004

    2004

  49. [57]

    Unequal probability sampling without replacement through a splitting method,

    J.-C. Deville and Y. Tille, “Unequal probability sampling without replacement through a splitting method,” Biometrika, vol. 85, no. 1, pp. 89–101, 1998

    1998

  50. [58]

    Analog forecasting of extreme-causing weather patterns using deep learning,

    A. Chattopadhyay, E. Nabizadeh, and P. Hassanzadeh, “Analog forecasting of extreme-causing weather patterns using deep learning,” Journal of Advances in Modeling Earth Systems , vol. 12, no. 2, p. e2019MS001958, 2020

    2020

  51. [59]

    Learning forecasts of rare stratospheric transitions from short simulations,

    J. Finkel, R. J. Webber, E. P. Gerber, D. S. Abbot, and J. Weare, “Learning forecasts of rare stratospheric transitions from short simulations,” Monthly Weather Review , vol. 149, no. 11, pp. 3647–3669, 2021

    2021

  52. [60]

    Revealing the statistics of extreme events hidden in short weather forecast data,

    J. Finkel, E. P. Gerber, D. S. Abbot, and J. Weare, “Revealing the statistics of extreme events hidden in short weather forecast data,” AGU Advances, vol. 4, no. 2, p. e2023A V000881, 2023

    2023

  53. [61]

    Gaussian framework and optimal pro- jection of weather fields for prediction of extreme events,

    V. Mascolo, A. Lovo, C. Herbert, and F. Bouchet, “Gaussian framework and optimal pro- jection of weather fields for prediction of extreme events,” Journal of Advances in Modeling Earth Systems , vol. 17, no. 6, p. e2024MS004487, 2025

    2025

  54. [62]

    Tackling the accuracy–interpretability trade-off in a hierarchy of machine learning models for the prediction of extreme heatwaves,

    A. Lovo, A. Lancelin, C. Herbert, and F. Bouchet, “Tackling the accuracy–interpretability trade-off in a hierarchy of machine learning models for the prediction of extreme heatwaves,” Artificial Intelligence for the Earth Systems , vol. 4, no. 3, p. 240094, 2025. 30

    2025

  55. [63]

    Path properties of atmospheric transitions: Illustration with a low-order sudden stratospheric warming model,

    J. Finkel, D. S. Abbot, and J. Weare, “Path properties of atmospheric transitions: Illustration with a low-order sudden stratospheric warming model,” Journal of the Atmospheric Sciences , vol. 77, no. 7, pp. 2327–2347, 2020

    2020

  56. [64]

    C´ erou,Genetic genealogical models in rare event analysis

    F. C´ erou,Genetic genealogical models in rare event analysis . PhD thesis, INRIA, 2006

    2006

  57. [65]

    Coles, An introduction to statistical modeling of extreme values , vol

    S. Coles, An introduction to statistical modeling of extreme values , vol. 208. Springer, 2001

    2001

  58. [66]

    SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python,

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. van der Walt, M. Brett, J. Wil- son, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Larson, C. J. Carey, ˙I. Polat, Y. Fe...

    2020

  59. [67]

    Ensemble forecasting at nmc: The generation of perturbations,

    Z. Toth and E. Kalnay, “Ensemble forecasting at nmc: The generation of perturbations,” Bulletin of the american meteorological society , vol. 74, no. 12, pp. 2317–2330, 1993

    1993

  60. [68]

    Variance estimation in the particle filter,

    A. Lee and N. Whiteley, “Variance estimation in the particle filter,” Biometrika, vol. 105, no. 3, pp. 609–625, 2018

    2018

  61. [69]

    J. S. Liu and J. S. Liu, Monte Carlo strategies in scientific computing , vol. 10. Springer, 2001. 31 EXTENDED DATA Boundary Variables Land-sea mask, surface roughness, and surface geopotential height (constant); Sea surface temperature, sea ice cover, and total incoming solar...

    2001

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.