Pith. sign in

REVIEW 2 major objections 5 minor 1 cited by

Boosting Ensembles for Statistics of Tails at Conditionally Optimal Advance Split Times

T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Boosting plus a well-timed split samples rare tails accurately

desk verdict A careful, honest empirical study of AST selection for ensemble boosting; the main gap is that the recommended entropy rule is only tested with ground-truth bin boundaries. read the letter →

arxiv 2507.22310 v3 pith:3DAIAPYX submitted 2025-07-30 physics.ao-ph math.DS

Justin Finkel , Paul A. O'Gorman This is my paper · ORCID
classification physics.ao-phmath.DS
keywords rareeventsamplingensembleboostingadvancesplittimethresholdedentropytailestimationquasigeostrophicflowpassivetracerreturnperiods
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 tries to establish that ensemble boosting, when launched at the right advance split time, can faithfully estimate the extreme tail of a climatological distribution that a short simulation alone cannot resolve. It proposes thresholded entropy as a practical, ground-truth-free objective for choosing that split time, and shows in a quasigeostrophic flow with a passive tracer that the entropy-selected advance split times track the true optimal values. If correct, climate modelers would have a principled way to set perturbation timing for rare-event sampling without running a very long reference simulation first. The paper also reports that the optimal advance split time falls near one to three eddy turnover timescales and varies with target location.

What carries the argument

The load-bearing object is the thresholded entropy functional, $S[(R^* - \mu)^+] = -\sum_k \Delta Q_k \log \Delta Q_k$, defined on the conditional severity distribution of a boosted ensemble restricted to values above a threshold $\mu$. It is designed to maximize when the ensemble's extreme severities are both abundant and diverse, thereby selecting a conditionally optimal advance split time without access to ground truth. The argument is carried by two probability estimators, MoCTail and PoPTail, which aggregate conditional tail distributions across ancestors, and by a quadratic response model that maps the low-dimensional perturbation parameter $\omega$ to the resulting event severity.

What would settle it

Run the selection protocol twice at one target latitude: once with thresholded-entropy bins fixed by the long-run quantiles as in the paper, and once with bins estimated from the short simulation alone, then compare each selected advance split time against the optimum found by exhaustive search. If the short-run bins shift the thresholded-entropy maximum outside the region where chi-squared divergence is near its minimum, the practical claim that thresholded entropy is a usable selection rule fails.

Watch

Extended reading notes

Core claim

The central claim is that a boosted ensemble, built by perturbing observed extreme events at a well-chosen advance split time and then reweighting the conditional tails, can reproduce the climatological tail of local tracer fluctuations more accurately than a direct simulation of equal computational cost. The paper introduces the MoCTail estimator, a mixture of conditional tail distributions, and compares it with the PoPTail estimator, and finds both accurately approximate the long-run ground truth when the split time is chosen near the optimum. It further claims that the thresholded entropy functional, which rewards ensembles whose above-threshold severities spread across many bins, selects an advance split time close to the one that minimizes chi-squared divergence from the true tail, and that this selection rule captures how the optimal split time changes across latitudes.

Load-bearing premise

The thresholded entropy proxy is defined using bin boundaries taken from quantiles of the true long-run distribution, and the paper only demonstrates that it selects good advance split times when those exact quantiles are available.

Editorial extensions

If this is right

  • Boosted ensembles with a thresholded-entropy-selected advance split time yield tail estimates closer to a long reference simulation than an equal-number direct simulation, with modest speedups relative to equal-cost direct simulation.
  • The optimal advance split time is strictly positive and approximately one to three eddy turnover timescales in the quasigeostrophic system, indicating that splitting too early or too late both degrade tail accuracy.
  • The thresholded entropy rule selects different optimal split times for different target locations, mirroring the variation found by exhaustive search, which suggests the rule is sensitive to local predictability rather than being a fixed global prescription.
  • Optimization-based selection rules like thresholded entropy avoid the arbitrary threshold choices required by uniform or correlation-threshold rules, making them more suitable for adaptive deployment in a rare-event sampling algorithm.

Reading between the lines

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

  • If thresholded entropy generalizes, it could replace heuristic rules such as the 3/8-dispersion rule, since it requires no arbitrary threshold and naturally adapts to the target event and initial condition.
  • The location dependence of the optimal advance split time hints that the COAST is tied to the local growth of extreme fluctuations, so one testable extension is whether the entropy-selected split time correlates with local Lyapunov-type predictability timescales across different flow regimes.
  • A practical deployment would need to estimate the bin boundaries of thresholded entropy from the short simulation rather than from the long reference run; the paper does not test this, so the most direct extension is to compare the selected split times under short-run bins against the true optima.
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 / 5 minor

Summary. Finkel and O'Gorman study the choice of advance split time (AST) in ensemble boosting for rare-event tail estimation. Using a two-layer quasigeostrophic model with a passive tracer, they identify extreme events from a short direct numerical simulation (DNS), launch perturbed ensembles at ASTs from 2 to 40 days, fit quadratic response surfaces for event severity, and aggregate conditional tail distributions into climatological estimates via two estimators, 'MoCTail' and 'PoPTail'. They evaluate several AST selection rules: a uniform AST, a pattern-correlation threshold, expected improvement, and thresholded entropy, comparing the resulting tail CCDFs against a long DNS through chi-squared divergence. They report that boosting improves over equal-N DNS and is competitive at fixed cost, that an optimal AST exists at intermediate times (roughly 1-3 eddy turnover times), and that thresholded entropy and expected improvement track this optimum. They recommend thresholded entropy as a generic selection rule and provide open-source code for the experiments.

Significance. If the claims hold, this is a useful contribution to rare-event sampling for climate applications: it offers an intermediate-complexity testbed between Lorenz-96 and a general circulation model, a careful comparison of two probability estimators, and a candidate objective (thresholded entropy) for AST selection that could inform larger-scale deployments. The paper's strengths include the explicit formulation of AST selection as an optimization problem, the use of longitudinal rotation and bootstrapping to quantify uncertainty, the consistent boosting advantage over equal-N DNS, and the public code. The main caveat is that the advertised 'ground-truth-free' status of thresholded entropy is not established by the present experiments, and that parts of the rule comparison are oracle-tuned. These issues are fixable but currently limit the strength of the central recommendation.

major comments (2)
  1. [Section 6, Fig. 13; Section 7, conclusion 3] The thresholded entropy criterion S = -sum_k Delta Q_k log Delta Q_k uses bin boundaries r_k that, as specified in Section 2.3 and used throughout Figs. 10-12 and 15, are quantiles of the ground-truth long-DNS severity distribution, Q^Theta_k = (1/2)^(5+k). The threshold mu is likewise the ground-truth (1/2)^5 complementary quantile (Sect. 3.3). Thus the paper's central practical claim that TE can select the optimal AST without ground truth is not supported by the evidence presented: the reported success of TE is conditional on having the very quantity that the proxy is meant to replace. The paper itself notes in Sect. 2.4 that TE 'would change if the bins were changed.' In a real deployment the bins would need to be estimated from the short DNS, which contains only 14-32 peaks per latitude, with especially noisy far-tail bins; a change in bin boundaries can change the objective and its maximizer. I therefore request a limited-data test: recompute TE with bins and threshold estimated from the short DNS (or from the pooled boosted ensembles) and report how often the TE-selected AST remains close to the chi-squared-optimal AST. This is an empirical gap rather than a logical contradiction, but it directly affects the abstract's claim that a ground-truth-free proxy objective is proposed.
  2. [Section 6, Fig. 13; Section 7, conclusion 3] The comparison of AST selection rules is partly oracle-based. In Fig. 13, the uniform AST A_U and the pattern-correlation thresholds are chosen post hoc to minimize chi-squared divergence from ground truth for each subsample; the caption states that the reported chi-squared values are therefore 'practical lower bounds,' and the conclusion in Sect. 7 acknowledges that thresholds were selected 'with knowledge of the ground truth.' As a result, the finding that 'no single selection rule is superior' and the impression that TE performs comparably to A_U and A_PC are not yet a fair practical comparison: A_U and A_PC have been given access to the validation target, while EI is not tuned in this way (though TE still uses ground-truth bins). A deployment-oriented comparison should set all thresholds using short-DNS information only, and then evaluate skill against the long DNS. This does not invalidate the boosting-versus-DNS result, but it does weaken the paper's claims about the relative merits of the selection rules.
minor comments (5)
  1. [Section 2.4, Eq. (28)] The entropy is undefined when a bin probability is zero; please state the convention 0 log 0 = 0 or otherwise exclude empty bins.
  2. [Fig. 15] The caption lists panels (a)-(d) and then (g) for topography; the missing panel labels or the reference to panel (g) should be corrected for consistency.
  3. [Section 5.2, Eqs. (42)-(43)] The quadratic response model is fitted to only 21 quasi-Monte Carlo points plus one ancestor per (n, j); a cross-validation check or residual summary at the selected ASTs would make the probability estimates more transparent, especially because the estimators rely on extrapolation of the fitted response beyond the sampled impulses.
  4. [Section 3.3] It is not explicitly stated whether the threshold mu[(1/2)^5] is computed from the short DNS or the long DNS; please clarify, as this matters for the deployment of the thresholded-entropy criterion.
  5. [Section 2.2, Eqs. (17)-(18)] The notation for MoCTail and PoPTail is introduced in sequence across two paragraphs; consider defining both estimators in one place with parallel notation to improve readability.

Circularity Check

2 steps flagged · score 6.0 of 10

Thresholded entropy is advertised as a ground-truth-free proxy but is defined using ground-truth-derived bin boundaries, so the central AST-selection recommendation is partially circular; A_U and A_PC thresholds are also tuned post hoc against the same ground truth.

  1. self definitional [Abstract; Sect. 2.4 Eq. (28); Sect. 2.3 (bin construction); Sect. 3.3 (threshold)]
    "Abstract: 'Since ground truth is not known in practice, we propose a proxy objective function of thresholded entropy.' Sect. 2.4: 'The thresholded entropy is thus defined based on probability over discrete bins (with the bin boundaries r_k set based on quantiles of the ground-truth distribution) and would change if the bins were changed.' Sect. 2.3: 'Note the same set of r_k's based on the climatological distribution is used also for evaluating estimated distributions.'"

    Eq. (28) is not a self-contained proxy: its bin boundaries r_k are the long-DNS quantiles Q^Theta_k = (1/2)^(5+k) (Sect. 2.3), and its starting bin mu is also a ground-truth complementary quantile (Sect. 3.3). The abstract promises a proxy for use when 'ground truth is not known in practice,' but the objective is constructed from that ground truth. Moreover, the chi^2 skill metric in Eq. (19) uses the same ground-truth r_k's, so the reported agreement between TE-maximizing AST and chi^2-optimal AST (Figs. 10-12, 15) is evaluated under the very knowledge the proxy is supposed to replace. The maximizer is not fully forced, because the boosted conditional distributions still enter Eq.

  2. fitted input called prediction [Sect. 2.4 (A_U, A_PC); Fig. 13 caption; Sect. 6]
    "Sect. 2.4: 'Both A_U and A_PC ... both unfortunately require a threshold choice, which there is no established method for selecting. Here we selected thresholds post hoc with knowledge of the ground truth.' Fig. 13 caption: 'Because this requires ground truth knowledge, the chi^2 divergences must be interpreted as practical lower bounds.'"

    The A_U and A_PC thresholds are not independent selection rules; they are tuned post hoc against the long-DNS ground truth, and their resulting chi^2 errors are then presented as evidence that these rules 'improve substantially' and are 'equally effective' (Sect. 6, Figs. 13-14). Because the tuning target and evaluation target are the same ground-truth CCDF, the reported skill and error bars are lower bounds rather than predictions, as the Fig. 13 caption concedes. The paper is transparent about this, but the across-the-board comparison still mixes fitted thresholds with genuine prediction and does not support a claim that these rules work without ground-truth knowledge.

full rationale

The core estimator comparison (MoCTail and PoPTail against the long-DNS ground truth) is a legitimate external evaluation: the long DNS is independent of the short-DNS boosted ensembles, and the fixed-cost comparison uses a stated cost model. The 3/8 rule from Finkel and O'Gorman (2024) is tested rather than assumed, so self-citation is not load-bearing. The circularity is concentrated in the AST-selection rules. Thresholded entropy is advertised as the practical ground-truth-free objective, yet its definition (Eq. 28) requires bin boundaries that are quantiles of the ground-truth distribution; the paper explicitly says the bin boundaries are 'set based on quantiles of the ground-truth distribution' and that the same r_k's are used in the chi^2 evaluation. This makes the demonstration that TE identifies the chi^2-optimal AST partially a construction: the proxy has been handed the target's own tail quantiles. A real deployment would have to estimate those bins from limited data, which is never tested. The A_U and A_PC rules are additionally selected post hoc with ground-truth knowledge, so their success is a fitted lower bound. These issues make the paper's headline 'proxy' claim partially circular, while the numerical method itself remains a legitimate empirical study.

Assumptions & free parameters 8 free parameters · 6 assumptions · 2 invented entities

The central claim rests on several data-derived choices and modeling assumptions. The most concerning is that the thresholded-entropy criterion, presented as ground-truth-free, is evaluated with bin boundaries from the long DNS. Additionally, several AST selection rules (A_U, A_PC) are tuned post hoc with ground truth, which makes their displayed performance optimistic. The QG model with a single passive tracer and a one-mode perturbation space is an idealized testbed; generalizability to full climate models is asserted but not demonstrated.

free parameters (8)
  • Threshold mu = 0.52 at y0=26/64 L
    Set as the (1/2)^5 complementary quantile of the target intensity distribution from the long DNS (Sect. 3.3). This is a data-derived threshold, not derived from theory.
  • TE bin boundaries r_k = Quantiles of ground-truth severity distribution at (1/2)^(5+k)
    Used in the thresholded entropy criterion (Eq. 28), these bin boundaries come from the long DNS climatology, creating circularity for the claim that TE is a ground-truth-free proxy.
  • Pattern correlation threshold rho_U = Selected post hoc (e.g., 0.92 local, 0.86 nominal 3/8 rule)
    For A_PC rules, the correlation threshold is chosen post hoc using ground truth (Sect. 2.4, Conclusion point 3).
  • Perturbation scale s = 0.24 chosen as nominal
    Swept over 0.06 to 0.9; the nominal value 0.24 is chosen by hand after inspecting stability of results (Sect. 5.5).
  • Perturbation amplitude bound W = 0.3
    Set ad hoc to limit perturbation size relative to streamfunction amplitude (Sect. 4.1).
  • Argmax drift delta_t* = 5 days
    Chosen as about half an eddy turnover timescale to restrict peak timing shifts (Sect. 2.1).
  • Buffer times A_max and B = 40 and 20 days
    Chosen to exceed the mixing timescale so cluster maxima are independent (Sect. 3.3).
  • Quadratic response model coefficients = Fitted per ancestor and AST via OLS
    The severity response functions are quadratic regressions on 21 impulses plus the ancestor (Sect. 5.2); the resulting conditional CCDFs depend on these fitted parameters.
assumptions (6)
  • domain assumption The 2-layer quasigeostrophic model with a passive tracer is a representative intermediate-complexity testbed for midlatitude storm-track extreme statistics.
    The paper motivates this by the model hierarchy argument (Sect. 1.2) and notes it is an idealized surrogate for precipitation extremes.
  • domain assumption The long DNS (44 years) provides a reliable ground truth for tail statistics.
    Used as the reference for all chi2 divergence evaluations; its own sampling error is acknowledged but not propagated into the main conclusions (Sect. 3.2).
  • domain assumption Cluster maxima separated by buffers A_max and B are independent events.
    Required for the mixture-of-conditional-tails aggregation (Sect. 2.1).
  • ad hoc to paper A single linearly unstable Fourier mode perturbation is sufficient to probe the extreme tail.
    The perturbation space is one complex dimension; the authors conjecture low-dimensional kicks can excite extremes (Sect. 2.1, Sect. 4.1).
  • ad hoc to paper The quadratic response model is adequate for the ASTs near the optimum.
    The paper shows quadratic fits degrade for AST > 26 days but optimal ASTs are shorter (Sect. 5.2, 5.3).
  • ad hoc to paper Bin boundaries for TE can be estimated from the short DNS in practice.
    The paper only tests TE with ground-truth quantiles; the practical version would require estimating these from limited data (Sect. 2.4).
invented entities (2)
  • MoCTail estimator
    purpose: Aggregates conditional tail CCDFs across ancestors as a uniform mixture to estimate the climatological tail (Eq. 17).
    A new statistical estimator, not a physical entity; validated only on the QG model.
  • COAST (conditionally optimal advance split time)
    purpose: The AST maximizing statistical fidelity; a conceptual quantity the paper aims to characterize.
    Defined in the paper as a function of the estimator and ground truth; not independently measurable.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Boosting Ensembles for Statistics of Tails at Conditionally Optimal Advance Split Times." pith.science (2026). https://pith.science/paper/3DAIAPYX

@misc{pith2026250722310,
  author       = {Pith},
  title        = {Pith review of: Boosting Ensembles for Statistics of Tails at Conditionally Optimal Advance Split Times},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3DAIAPYX}},
  note         = {Machine review of arXiv:2507.22310}
}
read the original abstract

Climate science needs more efficient ways to study high-impact, low-probability extreme events. Ensemble boosting, a form of rare event sampling, offers a novel strategy to extract more information from those occasional simulated events, by perturbing them slightly to probe alternative scenarios immediately instead of waiting many simulation-years for the next event. But statistical accuracy and efficiency depend on the perturbation details. In particular for sudden and transient events like precipitation, performance of boosting depends sensitively on the \emph{advance split time} (AST), which must be long enough before the event to let the ensemble diversify, but not so much as to destroy the event. In pursuit of principled guidelines, we study the effect of AST for sampling tracer fluctuations in a quasigeostrophic flow, an idealized but informative model of midlatitude storm track dynamics. We formulate AST selection as an optimization problem for statistical fidelity with a ground truth. Since ground truth is not known in practice, we propose a proxy objective function of \emph{thresholded entropy}, which rewards ensembles with both a high mean and a large spread. We show that ensemble boosting, when given a well-chosen AST and equipped with methods to estimate probabilities, can accurately sample extremes at long return periods. We furthermore find evidence that thresholded entropy successfully identifies an optimal AST, which is roughly 1-3 eddy turnover timescales in the quasigeostrophic system. Moreover, this proxy captures the \emph{variation} of AST with the target location of the tracer within the flow field, suggesting generalizability to climate models. Large-scale deployment of our method will require further development in adaptive optimization strategies, but our work here is an essential first step for establishing what must be optimized.

Figures

Figures reproduced from arXiv: 2507.22310 by the authors.

Figure 1
Figure 1. Schematic summarizing the ensemble boosting and tail estimation procedure, using a simple Langevin dynamics with a potential that is quadratic for x ∈ (−0.25,0.25)—the blue-shaded region in (a)—and logarithmic outside this range. Appendix A specifies the system completely. The position variable X(t) exhibits intermittent, transient extremes (a.i) and power law tails P{|X| > |x|} ∼ |x| −3.1 (a.ii). We set a threshold… view at source ↗
Figure 2
Figure 2. Snapshots of the QG system configuration in the upper layer. Contours indicate the anomaly streamfunction ψ, which varies over a non-dimensional range of approximately ±18, dashed contours indicating negative anomalies. Colors indicate (a) tracer concentration c, (b) zonal wind velocity u = U−∂yψ, where U = 1 is the basic background shear, and (c) meridional velocity v = ∂xψ. The timestamps increase from left to rig… view at source ↗
Figure 3
Figure 3. Hovmöller diagrams of anomalies (departures from time-means) of zonal-mean concentration (a.i) and zonal-mean zonal wind (b.i). Contours indicate zonal-mean streamfunction anomaly (range ±10, negatives values dashed). Column (ii) shows bottom topography, which directly affects the lower layer only, but indirectly sets the preferred jet positions in the upper layer as well. For the same quantities, column (iii) shows… view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Summary statistics of latitude-dependent climatological tail distributions of local tracer concentrations, also called “intensities”, which are denoted R and defined as the average concentration c over a box (x, y) ∈ (x0, y0) + [−ℓ, ℓ] 2 . x0 = 1 2L and ℓ = 1 32L are f…
Figure 5
Figure 5. Figure 5: Probability distributions of local tracer concentrations at latitude y0 = 26 64L and averaged over a box of half-width ℓ = 2 64L. (a) The full PDF of intensity R. (b) The CCDF (tail integral) of intensity R, restricted to R > µ[ 1 2 ]. (c) Further zoomed-in CCDF of the…
Figure 6
Figure 6. Figure 6: Structure of perturbations and their probability distribution. (a) Level sets of each considered input distribution from scales s = 0.06 (red) to s = 0.9 (blue), each scale restricted to 1 15 of the circle each so that all scales may be seen. Labels on the outer edge o…
Figure 7
Figure 7. Figure 7: Boosted ensembles of two selected events: (a) time t ∗ = 4152 at latitude y0 = 38 64L, and (b) time t ∗ = 2702 at latitude y0 = 26 64L. These are times when the intensity function R(x(t)) from the short DNS (dashed black curves) achieved a peak value (horizontal dashed…
Figure 8
Figure 8. Figure 8: The response of an extreme event to perturbations: magnitude, phase, and timing. The event is the same as in Fig. 7b. Row (a) represents impulses as in [PITH_FULL_IMAGE:figures/full_fig_p033_8.png]
Figure 9
Figure 9. Figure 9: Severities and their conditional distributions for the same case study as Fig. 7b. For six ASTs (same as [PITH_FULL_IMAGE:figures/full_fig_p035_9.png]
Figure 10
Figure 10. Figure 10: Ensemble dispersion indicators as a function of AST, again for the same case study as Fig. 7b: (a) expected improvement EI, (b) thresholded entropy TE, (c) local and (d) global correlations. Colors indicate input scales s, from small (red: s = 0.06) to large (blue: s …
Figure 11
Figure 11. Figure 11: Ensemble dispersion metrics averaged across ancestors at y0 = 26/64L. (a) Coefficients of determination for linear (cyan) and quadratic (orange) regressions, averaged across ancestors. (b-e) same quantities as in in [PITH_FULL_IMAGE:figures/full_fig_p037_11.png]
Figure 12
Figure 12. Figure 12: Three optimization landscapes as joint functions of AST and input scale for y0 = (26/64)L: (a) expected improvement (EI), (b) thresholded entropy (TE), and (c) χ 2 divergence between the MoCTail and ground truth. Lighter gray indicates better performance—smaller χ 2 d…
Figure 13
Figure 13. Figure 13: CCDF approximations by various mixing criteria and associated errors, at the latitude y0 = 26 64L and input scale choice s = 0.24. (a.i-v) Tail CCDFs by various estimates using only N = 11 ancestors, with lines showing medians and bands showing interquartile ranges ac…
Figure 14
Figure 14. Figure 14: Performance of all AST selection criteria, measured by χ 2 divergence, across all latitudes for s = 0.24 and N = 10 or 11, whichever is nearest to 1/3 the number of ancestors found for the latitude in question (sometimes less than 32). Black line and gray envelope rep…
Figure 15
Figure 15. Figure 15: Optimization landscapes and optimal ASTs across latitudes, again fixing the input scale to s = 0.24. (a) Frequencies of conditionally optimal ASTs (COASTs), in the maximum-thresholded entropy sense, at each latitude, with whiter shading indicating higher frequency. E.…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. AI-boosted rare event sampling to characterize extreme weather

    physics.ao-ph 2025-10 conditional novelty 7.0 of 10

    AI+RES uses AI weather-forecast ensembles as a guide for rare-event simulation, yielding accurate return-period statistics for 1-in-50,000-year heatwaves at roughly 100× lower computational cost.

Reference graph

Works this paper leans on

69 extracted references · 31 canonical work pages · cited by 1 Pith paper

  1. [1]

    and Beck, J

    Au, S.-K. and Beck, J. L.: Estimation of small failure probabilities in high dimensions by subset simulation, Probabilistic Engineering Mechanics, 16, 263–277, https://doi.org/https://doi.org/10.1016/S0266-8920(01)00019-4,

    doi:10.1016/s0266-8920(01)00019-4

  2. [9]

    Castaing, B., Gunaratne, G., Heslot, F., Kadanoff, L., Libchaber, A., Thomae, S., Wu, X.-Z., Zaleski, S., and Zanetti, G.: Scaling of hard thermal turbulence in Rayleigh-Bénard convection, Journal of Fluid Mechanics, 204, 1–30, https://doi.org/10.1017/S0022112089001643,

    doi:10.1017/s0022112089001643

  3. [11]

    and Guyader, A.: Adaptive Multilevel Splitting for Rare Event Analysis, Stochastic Analysis and Applications, 25, 417–443, https://doi.org/10.1080/07362990601139628,

    Cérou, F. and Guyader, A.: Adaptive Multilevel Splitting for Rare Event Analysis, Stochastic Analysis and Applications, 25, 417–443, https://doi.org/10.1080/07362990601139628,

    doi:10.1080/07362990601139628

  4. [13]

    Diaconescu, E. P. and Laprise, R.: Singular vectors in atmospheric sciences: A review, Earth-Science Reviews, 113, 161–175, https://doi.org/https://doi.org/10.1016/j.earscirev.2012.05.005,

    work page doi:10.1016/j.earscirev.2012.05.005 2012

  5. [19]

    Gálfi, V . M., Bódai, T., and Lucarini, V .: Convergence of Extreme Value Statistics in a Two-Layer Quasi-Geostrophic Atmospheric Model, Complexity, 2017, 5340 858, https://doi.org/10.1155/2017/5340858,

    work page doi:10.1155/2017/5340858 2017

  6. [20]

    Gessner, C., Fischer, E. M., Beyerle, U., and Knutti, R.: Very Rare Heat Extremes: Quantifying and Understanding Using Ensemble Reinitialization, Journal of Climate, 34, 6619 – 6634, https://doi.org/10.1175/JCLI-D-20-0916.1,

    doi:10.1175/jcli-d-20-0916.1

  7. [21]

    D., Naveau, P., Soloviev, A., Friederichs, P., Keilis-Borok, V ., Kondrashov, D., Kossobokov, V ., Mestre, O., Nicolis, C., Rust, H

    Ghil, M., Yiou, P., Hallegatte, S., Malamud, B. D., Naveau, P., Soloviev, A., Friederichs, P., Keilis-Borok, V ., Kondrashov, D., Kossobokov, V ., Mestre, O., Nicolis, C., Rust, H. W., Shebalin, P., Vrac, M., Witt, A., and Zaliapin, I.: Extreme events: dynamics, statistics and prediction, Nonlinear Processes in Geophysics, 18, 295–350, https://doi.org/10....

    doi:10.5194/npg-18-295-2011

  8. [23]

    P., Clarke, J., Gharib, M., Lane, B., and Mesquita, O

    Gollub, J. P., Clarke, J., Gharib, M., Lane, B., and Mesquita, O. N.: Fluctuations and transport in a stirred fluid with a mean gradient, Phys. Rev. Lett., 67, 3507–3510, https://doi.org/10.1103/PhysRevLett.67.3507,

    work page doi:10.1103/physrevlett.67.3507

Show all 69 references
  1. [27]

    L.: Modeling of spatial extremes in environmental data science: time to move away from max-stable processes, Environmental Data Science, 4, e3, https://doi.org/10.1017/eds.2024.54,

    Huser, R., Opitz, T., and Wadsworth, J. L.: Modeling of spatial extremes in environmental data science: time to move away from max-stable processes, Environmental Data Science, 4, e3, https://doi.org/10.1017/eds.2024.54,

    2024 doi

  2. [28]

    Jalbert, J., Farmer, M., Gobeil, G., and Roy, P.: Extremes.jl: Extreme Value Analysis in Julia, Journal of Statistical Software, 109, 1–35, https://doi.org/10.18637/jss.v109.i06,

    doi

  3. [29]

    John, A., Douville, H., Ribes, A., and Yiou, P.: Quantifying CMIP6 model uncertainties in extreme precipitation projections, Weather and Climate Extremes, 36, 100 435, https://doi.org/https://doi.org/10.1016/j.wace.2022.100435,

    2022

  4. [30]

    justinfocus12: justinfocus12/COAST: Initial release for submission of BEST COAST paper to NPG, https://doi.org/10.5281/zenodo.17355215,

    doi

  5. [31]

    49 Kabir, H. M. D., Khosravi, A., Hosen, M. A., and Nahavandi, S.: Neural Network-Based Uncertainty Quantification: A Survey of Methodologies and Applications, IEEE Access, 6, 36 218–36 234, https://doi.org/10.1109/ACCESS.2018.2836917,

  6. [37]

    Lucente, D., Rolland, J., Herbert, C., and Bouchet, F.: Coupling rare event algorithms with data-based learned committor functions using the analogue Markov chain, Journal of Statistical Mechanics: Theory and Experiment, 2022, 083 201, https://doi.org/10.1088/1742- 5468/ac7aa7,

    2022 doi

  7. [38]

    Mahesh, A., Collins, W., Bonev, B., Brenowitz, N., Cohen, Y ., Elms, J., Harrington, P., Kashinath, K., Kurth, T., North, J., OBrien, T., Pritchard, M., Pruitt, D., Risser, M., Subramanian, S., and Willard, J.: Huge Ensembles Part I: Design of Ensemble Weather Forecasts using ...

    arXiv

  8. [39]

    Mohamad, M. A. and Sapsis, T. P.: Sequential sampling strategy for extreme event statistics in nonlinear dynamical systems, Proceedings of the National Academy of Sciences, 115, 11 138–11 143, https://doi.org/10.1073/pnas.1813263115,

    doi

  9. [40]

    D., Lintner, B

    Neelin, J. D., Lintner, B. R., Tian, B., Li, Q., Zhang, L., Patra, P. K., Chahine, M. T., and Stechmann, S. N.: Long tails in deep columns of natural and anthropogenic tropospheric tracers, Geophysical Research Letters, 37, https://doi.org/https://doi.org/10.1029/2009GL041726,

    doi

  10. [42]

    O’Gorman, P. A. and Schneider, T.: Scaling of Precipitation Extremes over a Wide Range of Climates Simulated with an Idealized GCM, Journal of Climate, 22, 5676 – 5685, https://doi.org/10.1175/2009JCLI2701.1,

    doi

  11. [44]

    Penland, C. and Magorian, T.: Prediction of Niño 3 Sea Surface Temperatures Using Linear Inverse Modeling, Journal of Climate, 6, 1067 – 1076, https://doi.org/10.1175/1520-0442(1993)006<1067:PONSST>2.0.CO;2,

    1993 doi

  12. [47]

    Pons, F. M. E., Yiou, P., Jézéquel, A., and Messori, G.: Simulating the Western North America heatwave of 2021 with analogue importance sampling, Weather and Climate Extremes, 43, 100 651, https://doi.org/https://doi.org/10.1016/j.wace.2024.100651,

    2021

  13. [48]

    I., and Siggia, E

    Pumir, A., Shraiman, B. I., and Siggia, E. D.: Exponential tails and random advection, Phys. Rev. Lett., 66, 2984–2987, https://doi.org/10.1103/PhysRevLett.66.2984,

    doi

  14. [49]

    Rackauckas, C.: QuasiMonteCarlo.jl, https://github.com/SciML/QuasiMonteCarlo.jl, accessed: 2025-05-09,

    2025

  15. [50]

    Ragone, F. and Bouchet, F.: Rare Event Algorithm Study of Extreme Warm Summers and Heatwaves Over Europe, Geophysical Research Letters, 48, e2020GL091 197, https://doi.org/https://doi.org/10.1029/2020GL091197, e2020GL091197 2020GL091197,

    doi

  16. [51]

    Ragone, F., Wouters, J., and Bouchet, F.: Computation of extreme heat waves in climate models using a large deviation algorithm, Proceedings of the National Academy of Sciences, 115, 24–29, https://doi.org/10.1073/pnas.1712645115,

    doi

  17. [52]

    Rampal, N., Gibson, P. B., Sherwood, S., Abramowitz, G., and Hobeichi, S.: A Reliable Generative Adversarial Network Approach for Climate Downscaling and Weather Generation, Journal of Advances in Modeling Earth Systems, 17, e2024MS004 668, https://doi.org/https://doi.org/10.1...

    doi

  18. [53]

    Rolland, J.: Collapse of transitional wall turbulence captured using a rare events algorithm, Journal of Fluid Mechanics, 931, A22, https://doi.org/10.1017/jfm.2021.957,

    2021 doi

  19. [54]

    Saha, A. and Ravela, S.: Statistical-Physical Adversarial Learning From Data and Models for Downscaling Rainfall Extremes, Journal of Advances in Modeling Earth Systems, 16, e2023MS003 860, https://doi.org/https://doi.org/10.1029/2023MS003860, e2023MS003860 2023MS003860,

    doi

  20. [55]

    Sapsis, T. P.: Output-weighted optimal sampling for Bayesian regression and rare event statistics using few samples, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 476, 20190 834, https://doi.org/10.1098/rspa.2019.0834,

  21. [56]

    51 Sundar, R., Parashar, N., Blanchard, A., and Dodov, B.: TAUDiff: Improving statistical downscaling for extreme weather events using generative diffusion models, https://arxiv.org/abs/2412.13627,

    arXiv

  22. [57]

    P., and Link, R.: Emulating climate extreme indices, Environmental Research Letters, 15, 074 006, https://doi.org/10.1088/1748-9326/ab8332,

    Tebaldi, C., Armbruster, A., Engler, H. P., and Link, R.: Emulating climate extreme indices, Environmental Research Letters, 15, 074 006, https://doi.org/10.1088/1748-9326/ab8332,

    doi

  23. [58]

    F.: Jet Formation and Evolution in Baroclinic Turbulence with Simple Topography, Journal of Physical Oceanography, 40, 257 – 278, https://doi.org/10.1175/2009JPO4218.1,

    Thompson, A. F.: Jet Formation and Evolution in Baroclinic Turbulence with Simple Topography, Journal of Physical Oceanography, 40, 257 – 278, https://doi.org/10.1175/2009JPO4218.1,

    doi

  24. [59]

    J., Scaife, A

    Thompson, V ., Dunstone, N. J., Scaife, A. A., Smith, D. M., Slingo, J. M., Brown, S., and Belcher, S. E.: High risk of unprecedented UK rainfall in the current climate, Nature Communications, 8, 107, https://doi.org/10.1038/s41467-017-00275-3,

    doi

  25. [60]

    M.: A New Look at Weather Forecasting through Analogues, Monthly Weather Review, 117, 2230 – 2247, https://doi.org/10.1175/1520-0493(1989)117<2230:ANLAWF>2.0.CO;2,

    van den Dool, H. M.: A New Look at Weather Forecasting through Analogues, Monthly Weather Review, 117, 2230 – 2247, https://doi.org/10.1175/1520-0493(1989)117<2230:ANLAWF>2.0.CO;2,

    1989 doi

  26. [61]

    van Kekem, D. L. and Sterk, A. E.: Wave propagation in the Lorenz-96 model, Nonlinear Processes in Geophysics, 25, 301–314, https://doi.org/10.5194/npg-25-301-2018,

    2018 doi

  27. [62]

    Vandal, T., Kodra, E., Ganguly, S., Michaelis, A., Nemani, R., and Ganguly, A. R.: DeepSD: Generating High Resolution Climate Change Projections through Single Image Super-Resolution, in: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Da...

  28. [63]

    V onich, P. T. and Hakim, G. J.: Predictability Limit of the 2021 Pacific Northwest Heatwave From Deep-Learning Sensitivity Analysis, Geophysical Research Letters, 51, e2024GL110 651, https://doi.org/https://doi.org/10.1029/2024GL110651, e2024GL110651 2024GL110651,

    2021 doi

  29. [64]

    Wang, Q., Mu, M., and Sun, G.: A useful approach to sensitivity and predictability studies in geophysical fluid dynamics: conditional non- linear optimal perturbation, National Science Review, 7, 214–223, https://doi.org/10.1093/nsr/nwz039,

    doi

  30. [65]

    Watt, R. A. and Mansfield, L. A.: Generative Diffusion-based Downscaling for Climate, https://arxiv.org/abs/2404.17752,

    arXiv

  31. [66]

    J., Plotkin, D

    Webber, R. J., Plotkin, D. A., O’Neill, M. E., Abbot, D. S., and Weare, J.: Practical rare event sampling for extreme mesoscale weather, Chaos: An Interdisciplinary Journal of Nonlinear Science, 29, 053 109, https://doi.org/10.1063/1.5081461,

    doi

  32. [67]

    Yang, Y ., Blanchard, A., Sapsis, T., and Perdikaris, P.: Output-weighted sampling for multi-armed bandits with extreme payoffs, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 478, 20210 781, https://doi.org/10.1098/rspa.2021.0781,

  33. [68]

    and Jézéquel, A.: Simulation of extreme heat waves with empirical importance sampling, Geoscientific Model Development, 13, 763–781, https://doi.org/10.5194/gmd-13-763-2020,

    Yiou, P. and Jézéquel, A.: Simulation of extreme heat waves with empirical importance sampling, Geoscientific Model Development, 13, 763–781, https://doi.org/10.5194/gmd-13-763-2020,

    2020 doi

  34. [69]

    Zuckerman, D. M. and Chong, L. T.: Weighted Ensemble Simulation: Review of Methodology, Applications, and Software, Annual Review of Biophysics, 46, 43–57, https://doi.org/10.1146/annurev-biophys-070816-033834, pMID: 28301772,

    doi

  35. [1951]

    and Held, I

    Lapeyre, G. and Held, I. M.: The Role of Moisture in the Dynamics and Energetics of Turbulent Baroclinic Eddies, Journal of the Atmospheric Sciences, 61, 1693 – 1710, https://doi.org/10.1175/1520-0469(2004)061<1693:TROMIT>2.0.CO;2,

    2004 doi

  36. [1956]

    E., and Sapsis, T

    Pickering, E., Guth, S., Karniadakis, G. E., and Sapsis, T. P.: Discovering and forecasting extreme events via active learning in neural operators, Nature Computational Science, 2, 823–833, https://doi.org/10.1038/s43588-022-00376-0,

    doi

  37. [1980]

    K., Stein, M

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

    doi

  38. [1989]

    Coles, S.: An introduction to statistical modeling of extreme values, Springer Series in Statistics, Springer, 1 edn., ISBN 978-1-85233-459-8, https://doi.org/10.1007/978-1-4471-3675-0,

    doi

  39. [1991]

    Haidvogel, D. B. and Held, I. M.: Homogeneous Quasi-Geostrophic Turbulence Driven by a Uniform Temperature Gradient, Journal of Atmospheric Sciences, 37, 2644 – 2660, https://doi.org/10.1175/1520-0469(1980)037<2644:HQGTDB>2.0.CO;2,

    1980 doi

  40. [1993]

    Phillips, N. A.: The general circulation of the atmosphere: A numerical experiment, Quarterly Journal of the Royal Meteorological Society, 82, 123–164, https://doi.org/https://doi.org/10.1002/qj.49708235202,

    doi

  41. [1998]

    and Gritsun, A.: A new mathematical framework for atmospheric blocking events, Climate Dynamics, 54, 575–598, https://doi.org/10.1007/s00382-019-05018-2,

    Lucarini, V . and Gritsun, A.: A new mathematical framework for atmospheric blocking events, Climate Dynamics, 54, 575–598, https://doi.org/10.1007/s00382-019-05018-2,

    doi

  42. [2001]

    Baars, S., Castellana, D., Wubs, F., and Dijkstra, H.: Application of adaptive multilevel splitting to high-dimensional dynamical systems, Journal of Computational Physics, 424, 109 876, https://doi.org/https://doi.org/10.1016/j.jcp.2020.109876,

  43. [2002]

    Breitung, K.: SORM, Design Points, Subset Simulation, and Markov Chain Monte Carlo, ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering, 7, 04021 052, https://doi.org/10.1061/AJRUA6.0001166,

    doi

  44. [2007]

    Dematteis, G., Grafke, T., and Vanden-Eijnden, E.: Extreme Event Quantification in Dynamical Systems with Random Components, SIAM/ASA Journal on Uncertainty Quantification, 7, 1029–1059, https://doi.org/10.1137/18M1211003,

    doi

  45. [2009]

    Panetta, R. L.: Zonal Jets in Wide Baroclinically Unstable Regions: Persistence and Scale Selection, Journal of Atmospheric Sciences, 50, 2073 – 2106, https://doi.org/10.1175/1520-0469(1993)050<2073:ZJIWBU>2.0.CO;2,

    1993 doi

  46. [2010]

    Norwood, A., Kalnay, E., Ide, K., Yang, S.-C., and Wolfe, C.: Lyapunov, singular and bred vectors in a multi-scale system: an empirical exploration of vectors related to instabilities, Journal of Physics A: Mathematical and Theoretical, 46, 254 021, https://doi.org/10.1088/175...

    doi

  47. [2011]

    T., Deck, K., Bischoff, T., and Souza, A.: Response Theory via Generative Score Modeling, Phys

    Giorgini, L. T., Deck, K., Bischoff, T., and Souza, A.: Response Theory via Generative Score Modeling, Phys. Rev. Lett., 133, 267 302, https://doi.org/10.1103/PhysRevLett.133.267302,

    doi

  48. [2012]

    and Sapsis, T

    Farazmand, M. and Sapsis, T. P.: A variational approach to probing extreme events in turbulent dynamical systems, Science Advances, 3, e1701 533, https://doi.org/10.1126/sciadv.1701533,

    doi

  49. [2014]

    Lestang, T., Ragone, F., Bréhier, C.-E., Herbert, C., and Bouchet, F.: Computing return times or return periods with rare event algorithms, Journal of Statistical Mechanics: Theory and Experiment, 2018, 043 213, https://doi.org/10.1088/1742-5468/aab856,

    2018 doi

  50. [2015]

    Bloin-Wibe, L., Noyelle, R., Humphrey, V ., Beyerle, U., Knutti, R., and Fischer, E.: Estimating return periods for extreme events in climate models through Ensemble Boosting, EGUsphere, 2025, 1–40, https://doi.org/10.5194/egusphere-2025-525,

    2025 doi

  51. [2016]

    and Wadsworth, J

    Huser, R. and Wadsworth, J. L.: Advances in statistical modeling of spatial extremes, WIREs Computational Statistics, 14, e1537, https://doi.org/https://doi.org/10.1002/wics.1537,

    doi

  52. [2017]

    Farrell, B. F. and Ioannou, P. J.: Generalized Stability Theory. Part I: Autonomous Operators, Journal of Atmospheric Sciences, 53, 2025 – 2040, https://doi.org/10.1175/1520-0469(1996)053<2025:GSTPIA>2.0.CO;2, 1996a. Farrell, B. F. and Ioannou, P. J.: Generalized Stability The...

    1996 doi

  53. [2018]

    Linz, M., Chen, G., Zhang, B., and Zhang, P.: A Framework for Understanding How Dynamics Shape Temperature Distributions, Geophysical Research Letters, 47, e2019GL085 684, https://doi.org/https://doi.org/10.1029/2019GL085684, e2019GL085684 10.1029/2019GL085684,

    doi

  54. [2019]

    Boulaguiem, Y ., Zscheischler, J., Vignotto, E., van der Wiel, K., and Engelke, S.: Modeling and simulating spatial extremes by combining extreme value theory with generative adversarial networks, Environmental Data Science, 1, e5, https://doi.org/10.1017/eds.2022.4,

    doi

  55. [2020]

    Lorenz, E. N. and Emanuel, K. A.: Optimal Sites for Supplementary Weather Observations: Simulation with a Small Model, Journal of the Atmospheric Sciences, 55, 399 – 414, https://doi.org/10.1175/1520-0469(1998)055<0399:OSFSWO>2.0.CO;2,

    1998 doi

  56. [2021]

    R., Ha, S.-Y ., Hacker, J

    Berner, J., Fossell, K. R., Ha, S.-Y ., Hacker, J. P., and Snyder, C.: Increasing the Skill of Probabilistic Forecasts: Understanding Performance Improvements from Model-Error Representations, Monthly Weather Review, 143, 1295 – 1320, https://doi.org/10.1175/MWR-D-14- 00091.1,

    doi

  57. [2022]

    and Majda, A

    Bourlioux, A. and Majda, A. J.: Elementary models with probability distribution function intermittency for passive scalars with a mean gradient, Physics of Fluids, 14, 881–897, https://doi.org/10.1063/1.1430736,

    doi

  58. [2023]

    M., Beyerle, U., Bloin-Wibe, L., Gessner, C., Humphrey, V ., Lehner, F., Pendergrass, A

    Fischer, E. M., Beyerle, U., Bloin-Wibe, L., Gessner, C., Humphrey, V ., Lehner, F., Pendergrass, A. G., Sippel, S., Zeder, J., and Knutti, R.: Storylines for unprecedented heatwaves based on ensemble boosting, Nature Communications, 14, 4643, https://doi.org/10.1038/s41467- 0...

    doi

  59. [2024]

    and O’Gorman, P

    Finkel, J. and O’Gorman, P. A.: Rare Event Sampling for Moving Targets: Extremes of Temperature and Daily Precipitation in a General Circulation Model, Journal of Advances in Modeling Earth Systems, 18, e2025MS005 456, https://doi.org/https://doi.org/10.1029/2025MS005456, e202...

    doi

  60. [2025]

    J., Farazmand, M., and Sapsis, T

    Blonigan, P. J., Farazmand, M., and Sapsis, T. P.: Are extreme dissipation events predictable in turbulent fluid flows?, Phys. Rev. Fluids, 4, 044 606, https://doi.org/10.1103/PhysRevFluids.4.044606,

    doi

  61. [2026]

    P., Abbot, D

    Finkel, J., Gerber, E. P., Abbot, D. S., and Weare, J.: Revealing the Statistics of Extreme Events Hidden in Short Weather Forecast Data, AGU Advances, 4, e2023A V000 881, https://doi.org/https://doi.org/10.1029/2023A V000881, e2023A V000881 2023A V000881,

    doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.