Pith. sign in

REVIEW 3 major objections 4 minor 94 references

Synthesizing at the parameter level—before computing attribution measures—gives extreme-event attribution with far smaller bias and inference across any threshold.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 12:30 UTC pith:UVZFEWY3

load-bearing objection A serious, well-executed methodological paper: it gives the first systematic statistical critique of the WWA synthesis and a genuinely new parameter-level alternative, though the headline bias advantage is shown only under a simulation design aligned with its own assumptions. the 3 major comments →

arxiv 2607.19516 v1 pith:UVZFEWY3 submitted 2026-07-21 stat.ME stat.AP

Evidence Synthesis in Probabilistic Extreme Event Attribution: From Attribution Measures to Model Parameters

Erik Haufs , Axel B\"ucher , Jonas Schr\"oter This is my paper
classification stat.ME stat.AP MSC 62G3262F4062P12
keywords extreme event attributionevidence synthesisdistributional regressiongeneralized extreme value distributionrandom-effects meta-analysisprobability rationonstationary extremesclimate change
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Probabilistic extreme event attribution usually estimates a probability ratio or intensity change separately from each observational product and climate model, then meta-analyzes those estimates. This paper argues that this measure-level synthesis is statistically fragile: it can be substantially biased, produce intervals that do not contract as more models are added, and often generate infinite estimates. The paper's alternative is parameter-level synthesis: first combine each source's fitted extreme-value (GEV) regression parameters using random-effects generalized least squares, then derive probability ratios or intensity changes from the combined model. In controlled simulations, the parameter-level estimator reduces squared bias to 0.83e-3 versus 9.0e-3 to 14.2e-3 for the benchmark, with competitive mean squared error, and its bootstrap intervals reach near-nominal coverage. The Storm Boris case study shows the choice matters: under the assumed GMST-driven model, parameter-level and modified measure-level syntheses detect a significant anthropogenic signal, while the unmodified benchmark does not.

Core claim

The central claim is that evidence synthesis in extreme event attribution should be performed on the parameters of the nonstationary distributional regression model rather than on the attribution measures derived from them. Concretely, the paper assumes each data source j has a source-specific GEV parameter vector ϑ_j centered around a common target ϑ, and that its estimator satisfies ϑ̂_j = ϑ + ε_j + δ_j with a centered representation error ε_j and estimation error δ_j. The sources are combined by a GLS estimator of ϑ using the stacked covariance matrix V = Σ + I_m ⊗ Ξ, where Σ is the within-source covariance estimated by joint bootstrap and Ξ the between-source covariance estimated by a mu

What carries the argument

The central object is the generalized least squares (GLS) estimator for the common GEV regression parameter ϑ, applied to the stacked parameter estimates from all data sources: ϑ̂_W = (X^T V^{-1} X)^{-1} X^T V^{-1} ϑ̂, with X = 1_m ⊗ I_d and V = Σ + I_m ⊗ Ξ. Σ is the bootstrap covariance of within-source estimation error; Ξ is the between-source representation-error covariance, estimated by a multivariate method-of-moments (DerSimonian–Laird-type) estimator. A structural assumption Σ_j = |J_j|^{-1} Σ_ref is imposed for model synthesis to prevent a weighting bias that would down-weight sources with large estimated scale parameters. The role of this machinery is to produce a combined parameter

Load-bearing premise

The load-bearing premise is that all data sources share one correctly specified GEV family and that, after bias correction, each source's fitted parameter vector is an unbiased estimate of a source-specific truth drawn from a centered random-effects distribution around a common target; if the parametric family is wrong, or the small set of sources is not a centered sample, the claimed bias reduction and interval coverage are not guaranteed.

What would settle it

Run the simulation study with data generated from a distribution not in the GEV family (e.g., a Gamma or log-normal process with the same covariate link) and compare the bias of parameter-level versus measure-level synthesis. If parameter-level bias is no longer smaller, the bias reduction is an artifact of correct specification. A second check: on the Storm Boris data, examine whether the estimated between-source covariance Ξ is close to rank-one and stable across bootstrap resamples; if the dominant eigenvector is unstable, the GLS weights and the significance statement may not be reproducib

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • A single synthesized parameter vector yields probability-ratio and intensity-change estimates at any event threshold, exceedance probability, and counterfactual GMST, so attribution statements no longer depend on a handful of pre-selected values.
  • Parameter-level synthesis largely eliminates the infinite probability-ratio estimates that occur in measure-level synthesis for short records and small ensembles.
  • In the paper's simulation design, full parameter-level synthesis has roughly an order of magnitude smaller squared bias than the benchmark—0.83e-3 versus 9.0e-3 to 14.2e-3—while staying competitive in MSE.
  • The Storm Boris case study changes its conclusion depending on synthesis method: the original measure-level synthesis finds no significant anthropogenic influence, while parameter-level and modified measure-level syntheses do.
  • No single method dominates: modified measure-level synthesis sometimes has lower MSE or interval score, so the paper recommends either modifying the benchmark or adopting the parameter-level approach.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: because the parameter-level framework requires a correctly specified common parametric family, a useful robustness check would be to repeat the case study with a different link function (e.g., location linear in GMST rather than scale-linear) and see whether the significance statement survives.
  • Editorial inference: the estimated between-source covariance matrices are close to rank-one, suggesting a single dominant 'severity' direction; a lower-dimensional random-effects model could plausibly yield tighter inference at little bias cost.
  • Editorial inference: the paper's bias advantage comes from centering random effects before the nonlinear map T; an analogous comparison under misspecified T or non-GEV data would reveal how much of the advantage is structural rather than model-aligned.
  • Editorial inference: the GLS machinery already handles cross-covariance within sources, so the method naturally extends to joint attribution of multiple variables (e.g., wind and precipitation) by stacking parameter vectors.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper develops statistical methodology for combining evidence from multiple observational products and climate-model ensembles in probabilistic extreme event attribution. It first formalizes and critically examines the World Weather Attribution (WWA) measure-level synthesis, identifying over-conservative intervals, inefficient variance estimators, and the failure of intervals to contract with more independent models; it then proposes modified WWA procedures (mWWA) and a new parameter-level (PL) synthesis that combines estimated parameters of nonstationary GEV distributional regressions via GLS/random-effects meta-analysis before mapping to attribution measures such as probability ratios. The PL method is designed to permit simultaneous inference across event thresholds, exceedance probabilities, and counterfactual climates. The methods are compared in an extensive simulation study and applied to the September 2024 Storm Boris heavy-precipitation case. The reported results show that PL substantially reduces squared bias relative to WWA variants, with competitive MSE and interval scores, and that in the case study the PL and modified WWA procedures yield statistically significant anthropogenic influence while the original WWA does not.

Significance. If the claims hold, this is a valuable contribution to the statistical foundations of extreme event attribution. The paper is among the first to subject the widely used WWA synthesis to systematic statistical scrutiny, and it offers a conceptually appealing alternative: synthesizing at the parameter level of the distributional regression rather than at the level of individual attribution measures. The derivations—e.g., the method-of-moments estimator for Ξ in Remark 4.1 and the variance decompositions in Section 3—are clear and appear correct. The simulation study is carefully designed and extensive, the code is publicly available, and the case study illustrates a practically consequential difference between synthesis methods. The main limitation is that the empirical evidence for PL's bias advantage is obtained under a data-generating process that is explicitly aligned with PL assumptions; the manuscript is transparent about this, but the central comparative claim would be considerably strengthened by robustness checks under misspecification and heterogeneous covariance structure.

major comments (3)
  1. [§5.1, Table 2] The headline bias comparison—PL squared bias 0.83×10^-3 versus 9.0–14.2×10^-3 for the WWA variants—is obtained under a DGP that the authors describe as 'more directly aligned with the assumptions underlying PL synthesis': random effects are centered at the parameter level inside the fitted GEV family. The accompanying check that E[T(ϑ_j)]−T(ϑ_0) is small addresses AML-level centeredness, but not the stronger PL requirement that every source is exactly GEV with the specified link f_scale. Section 4.1 explicitly concedes that shared misspecification is not captured, and Section 7 lists misspecification as future work. Because the central claim is a comparative bias reduction, the current simulation evidence covers only the favorable corner. I recommend adding a misspecification robustness experiment (e.g., GPD tails, non-GEV deviations, or link/location misspecification) that reports bias,
  2. [§4.2, Eq. (4.7)–(4.8)] The GLS weights rely on the structural covariance assumption Σ_j = |J_j|^{-1} Σ_ref. This assumption is used to remove the described 'weighting bias' and is therefore load-bearing for the point estimator and for the bootstrap interval procedures. The simulation design—identical covariance structure across models with only sample size differing—satisfies the assumption by construction. Real climate models have different internal variability, ensemble member dependence, and structural biases, so this proportionality is unlikely to hold. Please include a simulation setting with heterogeneous Σ_j shapes (e.g., model-specific covariance matrices not proportional to inverse sample size) and report whether the gPar/hPar procedures remain calibrated and whether the bias advantage of PL persists.
  3. [§7, Eq. (2.4)] The paper acknowledges 'parameter-scale non-invariance' as a limitation, but this is central to the definition of PL synthesis. The two mathematically equivalent scale-link parametrizations in (2.4) induce different random-effects distributions on ϑ and, hence, potentially different synthesized estimates and attribution statements. The WWA measure-level approach is not subject to this arbitrariness. Since the synthesis target ϑ is not invariant to reparametrization, a sensitivity analysis across mathematically equivalent parametrizations—or a substantive argument for a preferred scale—is needed to establish that the case-study conclusions are not artifacts of the chosen parametrization.
minor comments (4)
  1. [§4.1] The assumption E[δ_j]=0 is used to justify the GLS estimator, but the two-stage PWM estimator used in the simulations (Section 5.1) is not exactly unbiased for GEV parameters in small samples. A short diagnostic reporting the empirical mean of \hatϑ_j − ϑ_j for the sample sizes considered would help readers gauge how much of the PL bias reduction depends on this centeredness assumption.
  2. [§5.1, 'Alignment' paragraph] The first sentence claims that the shared GEV assumption 'does not, by itself, favor either method.' Since the random effects are imposed and centered at the parameter level, the design does favor PL to some degree; the authors qualify this in the following sentences, but the initial wording could mislead.
  3. [Tables 1 and 2] The hPar rows are blank for variance, squared bias, and MSE because hPar uses the same point estimator as gPar and differs only in the bootstrap interval construction. A footnote to that effect would avoid confusion for readers comparing the tables row by row.
  4. [Figure 2] The caption states that the synthesized interval 'does not contract' as the number of models increases, which is a key point of the critique. Adding a horizontal axis label indicating the number of models would improve interpretability.

Circularity Check

0 steps flagged

No significant circularity: the PL estimator is derived from stated moment identities and validated against simulated ground truth and an external case study; the disclosed simulation-DGP alignment is a robustness limitation, not a circular step.

full rationale

The derivation chain is self-contained. The parameter-level (PL) estimator is built from explicit moment identities: (4.1) defines V = Sigma + I (x) Xi from the assumptions that the stacked estimates equal theta + epsilon + delta with E[epsilon_j]=0 and E[delta_j]=0; the GLS weighting (4.2)/(4.7) follows from Gauss-Markov; Xi-hat in (4.4) is derived as a method-of-moments estimator (Remark 4.1); and the multivariate DerSimonian-Laird estimator (B.4) is standard meta-analysis machinery (Chen et al. 2012), not a self-citation. Bootstrap uncertainty (Algorithms 1-4) is checked against known truth in Section 5, so coverage claims are not self-fulfilling. The central performance claims (Tables 1-2) are computed from simulations with known ground truth T(theta_0)=0.9439, and the Storm Boris case study (Section 6) is external data whose significance finding (gPar/hPar/mWWA significant, classical WWA not) does not depend on the simulation. The genuinely load-bearing caveat is the disclosed simulation design: Section 5.1 states 'the random effects are, however, imposed and centered at the parameter level, so the data-generating process is more directly aligned with the assumptions underlying PL synthesis,' and Appendix D shows the simulation covariance matrices Xi_obs/Xi_mod in (D.1) are estimated from the same Storm Boris case study used for illustration. This is a calibration/design-favorability limitation: the simulation exercises PL under its own assumptions, which limits external robustness, but it does not make any reported result equal an input by construction - the squared-bias values (0.83e-3 vs 9.0e-3-14.2e-3) are empirical, and Section 5.1 shows the induced AML-level centeredness gap (E[T(theta_j)] = 0.9308/0.9727 vs 0.9439; squared deviations about 1.7e-4 and 8.3e-4) is far smaller than the observed AML bias. Misspecification and parameter-scale non-invariance are explicitly conceded (Sections 4.1, 7); these are correctness risks, not circular steps. Self-citation is limited to a code repository (Haufs 2026) and is not load-bearing.

Axiom & Free-Parameter Ledger

4 free parameters · 7 axioms · 0 invented entities

The central claims rest on standard extreme-value theory, the scientific premise that GMST is an adequate forcing proxy, and — for the PL method itself — correct specification and centered parameter-level random effects across sources. The ad hoc items are the Σj ∝ 1/|Jj| structural covariance assumption and the plug-in bootstrap shortcuts, both disclosed. No invented entities (no new particles, forces, or dimensions).

free parameters (4)
  • Simulation ground truth ϑ0 = (0.1, 10, 2, 0.1) = (γ, μ, σ, α) = (0.1, 10, 2, 0.1)
    Chosen by hand to mirror the Storm Boris case-study setting; sets the scale of all reported biases and MSEs but is not fitted to outcome data.
  • Gaussian copula correlation ρ = 0.98 between observational products = 0.98
    Chosen to reproduce the strong cross-dependence of real observational products (Section 5.1); affects the observational-synthesis results.
  • Eigenvalue flooring threshold κ = 1e-12 = 1e-12
    Computational constant used to enforce positive semidefiniteness and invertibility of estimated covariance matrices (Section 4.1).
  • Six-sigma replacement rules for infinite WWA values
    Ad hoc surrogate rules (Section 5.1, 'Handling Infinite Estimates') applied only to the WWA model synthesis; they affect the WWA benchmarking.
axioms (7)
  • standard math Extremal Types Theorem justifies GEV for block maxima (Example 2.1)
    Invoked without proof as the justification for modeling X|G=g with the GEV family; standard background.
  • domain assumption GMST anomaly is an adequate proxy for the anthropogenic forced signal (G is the only covariate)
    All attribution statements (simulation and Storm Boris) are conditional on this; cited to Eyring et al. 2021, Section 2. If GMST does not capture the regional forced response, the causal reading of the probability ratios fails.
  • domain assumption Correct specification: (X|G=g) lies exactly in the chosen parametric family for every source
    Section 4.1 makes the parametric family an explicit component of the PL synthesis; shared misspecification is acknowledged (Sections 4.1 and 7) as not captured by centered representation errors.
  • domain assumption Centered, iid parameter-level representation errors εj = ϑj − ϑ with E[εj] = 0 and common covariance Ξ
    Stated after eq. (4.1) and in Section 3.1; the authors themselves note centeredness 'may not be fully justified in practice'. Load-bearing for the unbiasedness of the GLS combination.
  • domain assumption Independence of the δj across climate models, and of the observational and model syntheses
    Working assumption in Sections 3.2, 4.2 and 4.3; the paper calls it a working assumption and notes shared model biases are not identifiable.
  • ad hoc to paper Structural covariance assumption Σj = |Jj|^{-1} Σref
    Introduced in Section 4.2 to remove the GLS weighting bias; if covariance shape is not explained by sample size, the precision weights in (4.8) are misspecified.
  • ad hoc to paper Feasible plug-in bootstrap: Σ (and the Σ̂j) kept fixed across outer bootstrap iterations
    The authors state the procedure avoids a double bootstrap and 'should be interpreted as a feasible plug-in bootstrap approximation' (Sections 4.1–4.2, Algorithms 1–2).

pith-pipeline@v1.3.0-alltime-deepseek · 36961 in / 16229 out tokens · 137249 ms · 2026-08-01T12:30:33.269699+00:00 · methodology

0 comments
read the original abstract

Probabilistic extreme event attribution aims to quantify how anthropogenic climate change has altered the likelihood or intensity of a class of extreme events. Existing studies commonly combine evidence from observational products and climate-model ensembles by first estimating attribution measures, such as probability ratios or intensity changes, for each data source and then synthesizing the resulting estimates. We critically assess this approach, identify potential shortcomings of a respective benchmark procedure from the literature, and propose both targeted modifications and a new parameter-level synthesis method. The latter combines estimates of the underlying nonstationary distributional regression parameters, thereby enabling inference across multiple event thresholds and counterfactual climate conditions. In controlled simulation studies, the proposed modifications substantially improve upon the benchmark procedure, while parameter-level synthesis provides competitive overall performance. The practical usefulness is illustrated through a case study of the heavy precipitation associated with Storm Boris in September 2024.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

94 extracted references · 31 canonical work pages · 1 internal anchor

  1. [1]

    How climate change intensified storm Boris' extreme rainfall, revealed by near-real-time storylines , url =

    Athanase, Marylou and S. How climate change intensified storm Boris' extreme rainfall, revealed by near-real-time storylines , url =. Communications Earth & Environment , number =

  2. [2]

    Shepherd, Theodore G. , doi =. A Common Framework for Approaches to Extreme Event Attribution , url =. Current Climate Change Reports , number =. 2016 , bdsk-url-1 =

    2016

  3. [3]

    2017 , publisher =

    van Oldenborgh, Geert Jan and van der Wiel, Karin and Sebastian, Antonia and Singh, Roop and Arrighi, Julie and Otto, Friederike and Haustein, Karsten and Li, Sihan and Vecchi, Gabriel and Cullen, Heidi , title =. 2017 , publisher =. doi:10.1088/1748-9326/aa9ef2 , url =

    work page doi:10.1088/1748-9326/aa9ef2 2017

  4. [4]

    , TITLE =

    Zhang, Likun and Shaby, Benjamin A. , TITLE =. Biometrika , FJOURNAL =. 2022 , NUMBER =. doi:10.1093/biomet/asab043 , URL =

    work page doi:10.1093/biomet/asab043 2022

  5. [5]

    , TITLE =

    Zanger, Leandra and B\"ucher, Axel and Kreienkamp, Frank and Lorenz, Philip and Tradowsky, Jordis S. , TITLE =. Extremes , FJOURNAL =. 2024 , NUMBER =. doi:10.1007/s10687-023-00480-y , URL =

    work page doi:10.1007/s10687-023-00480-y 2024

  6. [6]

    Formally combining different lines of evidence in extreme-event attribution , volume=

    Otto, Friederike and Barnes, Clair and Philip, Sjoukje and Kew, Sarah and van Oldenborgh, Geert Jan and Vautard, Robert , year=. Formally combining different lines of evidence in extreme-event attribution , volume=. Advances in Statistical Climatology, Meteorology and Oceanography , publisher=. doi:10.5194/ascmo-10-159-2024 , number=

    doi:10.5194/ascmo-10-159-2024 2024

  7. [7]

    and Achuta Rao, K

    Eyring, Veronika and Gillett, Nathan P. and Achuta Rao, K. M. and Barimalala, Reto and Barreiro Parrillo, Mercedes and Bellouin, Nicolas and Cassou, Christophe and Durack, Paul and Kosaka, Yu and McGregor, Shane and Min, Seung-Ki and Morgenstern, Olaf and Sun, Yimin , title =. Climate Change 2021: The Physical Science Basis , editor =. 2021 , doi =

    2021

  8. [8]

    Framing, Context, and Methods , booktitle =

    Chen, Deliang and Rojas, Mauricio and Samset, Bj. Framing, Context, and Methods , booktitle =. 2021 , doi =

    2021

  9. [9]

    A protocol for probabilistic extreme event attribution analyses , volume =

    Philip, Sjoukje and Kew, Sarah and van Oldenborgh, Geert Jan and Otto, Friederike and Vautard, Robert and van der Wiel, Karin and King, Andrew and Lott, Fraser and Arrighi, Julie and Singh, Roop and van Aalst, Maarten , year =. A protocol for probabilistic extreme event attribution analyses , volume =. Advances in Statistical Climatology, Meteorology and ...

    doi:10.5194/ascmo-6-177-2020 2020

  10. [10]

    Clair Barnes , year =

  11. [11]

    Attribution of Extreme Events to Climate Change , volume =

    Otto, Friederike , year =. Attribution of Extreme Events to Climate Change , volume =. Annual Review of Environment and Resources , publisher =. doi:10.1146/annurev-environ-112621-083538 , number =

    doi:10.1146/annurev-environ-112621-083538

  12. [12]

    Pathways and pitfalls in extreme event attribution , volume =

    van Oldenborgh, Geert Jan and van der Wiel, Karin and Kew, Sarah and Philip, Sjoukje and Otto, Friederike and Vautard, Robert and King, Andrew and Lott, Fraser and Arrighi, Julie and Singh, Roop and van Aalst, Maarten , year =. Pathways and pitfalls in extreme event attribution , volume =. Climatic Change , publisher =. doi:10.1007/s10584-021-03071-7 , number =

    doi:10.1007/s10584-021-03071-7

  13. [13]

    Reconciling risk-based and storyline attribution with Bayes theorem , publisher =

    Buschow, Sebastian and Friederichs, Petra and Hense, Andreas , keywords =. Reconciling risk-based and storyline attribution with Bayes theorem , publisher =. 2024 , copyright =. doi:10.48550/ARXIV.2407.10776 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2407.10776 2024

  14. [14]

    , year =

    Shepherd, Theodore G. , year =. Storyline approach to the construction of regional climate change information , volume =. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , publisher =. doi:10.1098/rspa.2019.0013 , number =

    arXiv 2019

  15. [15]

    and van den Hurk, Bart and Hazeleger, Wilco and Martius, Olivia and Slingo, Julia and Zscheischler, Jakob , year =

    Sillmann, Jana and Shepherd, Theodore G. and van den Hurk, Bart and Hazeleger, Wilco and Martius, Olivia and Slingo, Julia and Zscheischler, Jakob , year =. Event‐Based Storylines to Address Climate Risk , volume =. Earth’s Future , publisher =. doi:10.1029/2020ef001783 , number =

    doi:10.1029/2020ef001783

  16. [16]

    and Philip, Sjoukje Y

    Tradowsky, Jordis S. and Philip, Sjoukje Y. and Kreienkamp, Frank and Kew, Sarah F. and Lorenz, Philip and Arrighi, Julie and Bettmann, Thomas and Caluwaerts, Steven and Chan, Steven C. and De Cruz, Lesley and de Vries, Hylke and Demuth, Norbert and Ferrone, Andrew and Fischer, Erich M. and Fowler, Hayley J. and Goergen, Klaus and Heinrich, Dorothy and He...

    2021

  17. [17]

    and van Oldenborgh, Geert Jan and Anslow, Faron S

    Philip, Sjoukje and Kew, Sarah F. and van Oldenborgh, Geert Jan and Anslow, Faron S. and Seneviratne, Sonia I. and Vautard, Robert and Coumou, Dim and Ebi, Kristie L. and Arrighi, Julie and Singh, Roop and van Aalst, Maarten and Pereira Marghidan, Carolina and Wehner, Michael and Yang, Wenchang and Li, Sihan and Schumacher, Dominik L. and Hauser, Mathias ...

    2021

  18. [18]

    and Davis, Richard A

    Brockwell, Peter J. and Davis, Richard A. , year =. Introduction to Time Series and Forecasting , ISBN =. doi:10.1007/978-3-319-29854-2 , journal =

    doi:10.1007/978-3-319-29854-2

  19. [19]

    Bootstrapping Estimators based on the Block Maxima Method

    B\". Bootstrapping Estimators based on the Block Maxima Method , publisher =. 2024 , copyright =. doi:10.48550/ARXIV.2409.05529 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2409.05529 2024

  20. [20]

    Strictly Proper Scoring Rules, Prediction, and Estimation , volume =

    Gneiting, Tilmann and Raftery, Adrian E , year =. Strictly Proper Scoring Rules, Prediction, and Estimation , volume =. Journal of the American Statistical Association , publisher =. doi:10.1198/016214506000001437 , number =

    doi:10.1198/016214506000001437

  21. [21]

    , year =

    Gnedenko, B. , year =. Sur La Distribution Limite Du Terme Maximum D’Une Serie Aleatoire , volume =. The Annals of Mathematics , publisher =. doi:10.2307/1968974 , number =

    doi:10.2307/1968974

  22. [22]

    Fisher, R. A. and Tippett, L. H. C. , year =. Limiting forms of the frequency distribution of the largest or smallest member of a sample , volume =. Mathematical Proceedings of the Cambridge Philosophical Society , publisher =. doi:10.1017/s0305004100015681 , number =

    doi:10.1017/s0305004100015681

  23. [23]

    On the disjoint and sliding block maxima method for piecewise stationary time series , volume =

    B\". On the disjoint and sliding block maxima method for piecewise stationary time series , volume =. The Annals of Statistics , publisher =. doi:10.1214/23-aos2260 , number =

    doi:10.1214/23-aos2260

  24. [24]

    On the maximum likelihood estimator for the Generalized Extreme-Value distribution , volume =

    B\". On the maximum likelihood estimator for the Generalized Extreme-Value distribution , volume =. Extremes , publisher =. 2017 , month = mar, pages =. doi:10.1007/s10687-017-0292-6 , number =

    work page doi:10.1007/s10687-017-0292-6 2017

  25. [25]

    Existence and consistency of the maximum likelihood estimators for the extreme value index within the block maxima framework , volume =

    Dombry, Clément , year =. Existence and consistency of the maximum likelihood estimators for the extreme value index within the block maxima framework , volume =. Bernoulli , publisher =. doi:10.3150/13-bej573 , number =

    work page doi:10.3150/13-bej573

  26. [26]

    and Guillou, A

    Ribereau, P. and Guillou, A. and Naveau, P. , year =. Estimating return levels from maxima of non-stationary random sequences using the Generalized PWM method , volume =. Nonlinear Processes in Geophysics , publisher =. doi:10.5194/npg-15-1033-2008 , number =

    work page doi:10.5194/npg-15-1033-2008 2008

  27. [27]

    Improving Probability-Weighted Moment Methods for the Generalized Extreme Value Distribution , journal =

    Diebolt, Jean and Guillou, Armelle and Naveau, Philippe and Ribereau, Pierre , language =. Improving Probability-Weighted Moment Methods for the Generalized Extreme Value Distribution , journal =. 2008 , copyright =. doi:10.57805/REVSTAT.V6I1.56 , url =

    work page doi:10.57805/revstat.v6i1.56 2008

  28. [29]

    Simmons, A. J. and Hollingsworth, A. , year =. Some aspects of the improvement in skill of numerical weather prediction , volume =. Quarterly Journal of the Royal Meteorological Society , publisher =. doi:10.1256/003590002321042135 , number =

    work page doi:10.1256/003590002321042135

  29. [30]

    and Smith, Doug , year =

    Scaife, Adam A. and Smith, Doug , year =. A signal-to-noise paradox in climate science , volume =. npj Climate and Atmospheric Science , publisher =. doi:10.1038/s41612-018-0038-4 , number =

    doi:10.1038/s41612-018-0038-4

  30. [31]

    Cochran, William G. , year=. The Combination of Estimates from Different Experiments , volume=. Biometrics , publisher=. doi:10.2307/3001666 , number=

    doi:10.2307/3001666

  31. [32]

    Meta-analysis in clinical trials , volume=

    DerSimonian, Rebecca and Laird, Nan , year=. Meta-analysis in clinical trials , volume=. Controlled Clinical Trials , publisher=. doi:10.1016/0197-2456(86)90046-2 , number=

    doi:10.1016/0197-2456(86)90046-2

  32. [33]

    and Hartung, Joachim , year =

    Knapp, Guido and Biggerstaff, Brad J. and Hartung, Joachim , year =. Assessing the Amount of Heterogeneity in Random‐Effects Meta‐Analysis , volume =. Biometrical Journal , publisher =. doi:10.1002/bimj.200510175 , number =

    work page doi:10.1002/bimj.200510175

  33. [34]

    Bias and Efficiency of Meta-Analytic Variance Estimators in the Random-Effects Model , volume =

    Viechtbauer, Wolfgang , year =. Bias and Efficiency of Meta-Analytic Variance Estimators in the Random-Effects Model , volume =. Journal of Educational and Behavioral Statistics , publisher =. doi:10.3102/10769986030003261 , number =

    doi:10.3102/10769986030003261

  34. [35]

    , year =

    Rukhin, Andrew L. , year =. Estimating Heterogeneity Variance in Meta-Analysis , volume =. Journal of the Royal Statistical Society Series B: Statistical Methodology , publisher =. doi:10.1111/j.1467-9868.2012.01047.x , number =

    arXiv 2012

  35. [36]

    van Aert, Robbie C. M. and Jackson, Dan , year =. Multistep estimators of the between‐study variance: The relationship with the Paule‐Mandel estimator , volume =. Statistics in Medicine , publisher =. doi:10.1002/sim.7665 , number =

    work page doi:10.1002/sim.7665

  36. [37]

    Random-effects model for meta-analysis of clinical trials: An update , volume=

    DerSimonian, Rebecca and Kacker, Raghu , year=. Random-effects model for meta-analysis of clinical trials: An update , volume=. Contemporary Clinical Trials , publisher=. doi:10.1016/j.cct.2006.04.004 , number=

    work page doi:10.1016/j.cct.2006.04.004 2006

  37. [38]

    Combining information from interlaboratory evaluations using a random effects model , volume =

    Kacker, Raghu , year =. Combining information from interlaboratory evaluations using a random effects model , volume =. Metrologia , publisher =. doi:10.1088/0026-1394/41/3/004 , number =

    work page doi:10.1088/0026-1394/41/3/004

  38. [39]

    , title =

    Raudenbush, Stephen W. , title =. The Handbook of Research Synthesis and Meta-Analysis , editor =. 2009 , publisher =

    2009

  39. [40]

    and Thompson, Simon G

    Hardy, Rebecca J. and Thompson, Simon G. , year=. Detecting and describing heterogeneity in meta-analysis , volume=. Statistics in Medicine , publisher=. doi:10.1002/(sici)1097-0258(19980430)17:8<841::aid-sim781>3.0.co;2-d , number=

    doi:10.1002/(sici)1097-0258(19980430)17:8

  40. [41]

    Primary, Secondary, and Meta-Analysis of Research , volume =

    Glass, Gene , year =. Primary, Secondary, and Meta-Analysis of Research , volume =. Educational Researcher , publisher =. doi:10.3102/0013189x005010003 , number =

    doi:10.3102/0013189x005010003

  41. [42]

    and Mandel, J

    Paule, R.C. and Mandel, J. , year=. Consensus Values and Weighting Factors , volume=. Journal of Research of the National Bureau of Standards , publisher=. doi:10.6028/jres.087.022 , number=

    doi:10.6028/jres.087.022

  42. [43]

    and Knapp, Guido and Salanti, Georgia , year=

    Veroniki, Areti Angeliki and Jackson, Dan and Bender, Ralf and Kuss, Oliver and Langan, Dean and Higgins, Julian P.T. and Knapp, Guido and Salanti, Georgia , year=. Methods to calculate uncertainty in the estimated overall effect size from a random‐effects meta‐analysis , volume=. Research Synthesis Methods , publisher=. doi:10.1002/jrsm.1319 , number=

    work page doi:10.1002/jrsm.1319

  43. [44]

    An Alternative Method for Meta-Analysis , volume=

    Hartung, Joachim , year=. An Alternative Method for Meta-Analysis , volume=. Biometrical Journal , publisher=. doi:10.1002/(sici)1521-4036(199912)41:8<901::aid-bimj901>3.3.co;2-n , number=

    work page doi:10.1002/(sici)1521-4036(199912)41:8

  44. [45]

    On tests of the overall treatment effect in meta‐analysis with normally distributed responses , volume=

    Hartung, Joachim and Knapp, Guido , year=. On tests of the overall treatment effect in meta‐analysis with normally distributed responses , volume=. Statistics in Medicine , publisher=. doi:10.1002/sim.791 , number=

    doi:10.1002/sim.791

  45. [46]

    Jackson, Dan and Law, Martin and R\"ucker, Gerta and Schwarzer, Guido , TITLE =. Stat. Med. , FJOURNAL =. 2017 , NUMBER =. doi:10.1002/sim.7411 , URL =

    doi:10.1002/sim.7411 2017

  46. [47]

    Methods to estimate the between‐study variance and its uncertainty in meta‐analysis , volume=

    Veroniki, Areti Angeliki and Jackson, Dan and Viechtbauer, Wolfgang and Bender, Ralf and Bowden, Jack and Knapp, Guido and Kuss, Oliver and Higgins, Julian PT and Langan, Dean and Salanti, Georgia , year=. Methods to estimate the between‐study variance and its uncertainty in meta‐analysis , volume=. Research Synthesis Methods , publisher=. doi:10.1002/jrs...

    doi:10.1002/jrsm.1164

  47. [48]

    Langan, Dean and Higgins, Julian P. T. and Simmonds, Mark , year=. Comparative performance of heterogeneity variance estimators in meta‐analysis: a review of simulation studies , volume=. Research Synthesis Methods , publisher=. doi:10.1002/jrsm.1198 , number=

    work page doi:10.1002/jrsm.1198

  48. [49]

    A refined method for the meta‐analysis of controlled clinical trials with binary outcome , volume=

    Hartung, Joachim and Knapp, Guido , year=. A refined method for the meta‐analysis of controlled clinical trials with binary outcome , volume=. Statistics in Medicine , publisher=. doi:10.1002/sim.1009 , number=

    work page doi:10.1002/sim.1009

  49. [50]

    and Becker, Betsy Jane and Kalaian, Hripsime , year =

    Raudenbush, Stephen W. and Becker, Betsy Jane and Kalaian, Hripsime , year =. Modeling multivariate effect sizes. , volume =. Psychological Bulletin , publisher =. doi:10.1037/0033-2909.103.1.111 , number =

    work page doi:10.1037/0033-2909.103.1.111

  50. [51]

    and Arends, Lidia R

    van Houwelingen, Hans C. and Arends, Lidia R. and Stijnen, Theo , year =. Advanced methods in meta‐analysis: multivariate approach and meta‐regression , volume =. Statistics in Medicine , publisher =. doi:10.1002/sim.1040 , number =

    doi:10.1002/sim.1040

  51. [52]

    , TITLE =

    Jackson, Dan and Riley, Richard and White, Ian R. , TITLE =. Stat. Med. , FJOURNAL =. 2011 , NUMBER =. doi:10.1002/sim.4172 , URL =

    work page doi:10.1002/sim.4172 2011

  52. [53]

    and Thompson, Simon G

    Jackson, Dan and White, Ian R. and Thompson, Simon G. , year =. Extending DerSimonian and Laird’s methodology to perform multivariate random effects meta‐analyses , volume =. Statistics in Medicine , publisher =. doi:10.1002/sim.3602 , number =

    work page doi:10.1002/sim.3602

  53. [54]

    and Dupuis, Josée , year =

    Chen, Han and Manning, Alisa K. and Dupuis, Josée , year =. A Method of Moments Estimator for Random Effect Multivariate Meta‐Analysis , volume =. Biometrics , publisher =. doi:10.1111/j.1541-0420.2012.01761.x , number =

    arXiv 2012

  54. [55]

    KNMI Climate Explorer , year =

  55. [56]

    The Annals of Applied Statistics , volume=

    Improved inference on risk measures for univariate extremes , author=. The Annals of Applied Statistics , volume=. 2022 , publisher=

    2022

  56. [57]

    arXiv preprint arXiv:1005.3573 , year=

    Profile Likelihood Intervals for Quantiles in Extreme Value Distributions , author=. arXiv preprint arXiv:1005.3573 , year=

    Pith/arXiv arXiv

  57. [58]

    , year =

    Miralles, Ophélia and Davison, Anthony C. , year =. Timing and spatial selection bias in rapid extreme event attribution , volume =. doi:10.1016/j.wace.2023.100584 , journal =

    arXiv 2023

  58. [59]

    and Lam, Henry and Engelke, Sebastian , keywords =

    Pasche, Olivier C. and Lam, Henry and Engelke, Sebastian , keywords =. Extreme Conformal Prediction: Reliable Intervals for High-Impact Events , publisher =. 2025 , copyright =. doi:10.48550/ARXIV.2505.08578 , url =

    doi:10.48550/arxiv.2505.08578 2025

  59. [60]

    Consistency of M-estimators for non-identically distributed data: the case of fixed-design distributional regression , publisher =

    B\". Consistency of M-estimators for non-identically distributed data: the case of fixed-design distributional regression , publisher =. 2025 , copyright =. doi:10.48550/ARXIV.2511.11067 , url =

    work page doi:10.48550/arxiv.2511.11067 2025

  60. [61]

    Davison, A. C. and Hinkley, D. V. , year =. Bootstrap Methods and their Application , ISBN =. doi:10.1017/cbo9780511802843 , publisher =

    doi:10.1017/cbo9780511802843

  61. [62]

    Better Bootstrap Confidence Intervals , volume =

    Efron, Bradley , year =. Better Bootstrap Confidence Intervals , volume =. Journal of the American Statistical Association , publisher =. doi:10.1080/01621459.1987.10478410 , number =

    arXiv 1987

  62. [63]

    Central European high-resolution gridded daily data sets (HYRAS): Mean temperature and relative humidity , volume =

    Frick, Claudia and Steiner, Heiko and Mazurkiewicz, Alex and Riediger, Ulf and Rauthe, Monika and Reich, Thomas and Gratzki, Annegret , year =. Central European high-resolution gridded daily data sets (HYRAS): Mean temperature and relative humidity , volume =. Meteorologische Zeitschrift , publisher =. doi:10.1127/0941-2948/2014/0560 , number =

    work page doi:10.1127/0941-2948/2014/0560 2014

  63. [64]

    doi:10.24381/CDS.151D3EC6 , url =

    E-OBS daily gridded meteorological data for Europe from 1950 to present derived from in-situ observations , publisher =. doi:10.24381/CDS.151D3EC6 , url =

    doi:10.24381/cds.151d3ec6 1950

  64. [65]

    , year =

    Lorenz, Edward N. , year =. The Mechanics of Vacillation , volume =. Journal of the Atmospheric Sciences , publisher =. doi:10.1175/1520-0469(1963)020<0448:tmov>2.0.co;2 , number =

    doi:10.1175/1520-0469(1963)020 1963

  65. [66]

    Biometrika , volume=

    Maximum likelihood estimation of the parameters of the generalized extreme-value distribution , author=. Biometrika , volume=. 1980 , publisher=

    1980

  66. [67]

    Wilks, S. S. , year =. The Large-Sample Distribution of the Likelihood Ratio for Testing Composite Hypotheses , volume =. The Annals of Mathematical Statistics , publisher =. doi:10.1214/aoms/1177732360 , number =

    arXiv

  67. [68]

    Quantile functions for multivariate analysis: approaches and applications , volume =

    Serfling, Robert , year =. Quantile functions for multivariate analysis: approaches and applications , volume =. Statistica Neerlandica , publisher =. doi:10.1111/1467-9574.00195 , number =

    arXiv

  68. [69]

    Adri and Ferro, Christopher A

    Hanel, Martin and Buishand, T. Adri and Ferro, Christopher A. T. , year =. A nonstationary index flood model for precipitation extremes in transient regional climate model simulations , volume =. Journal of Geophysical Research: Atmospheres , publisher =. doi:10.1029/2009jd011712 , number =

    work page doi:10.1029/2009jd011712

  69. [70]

    Hosking, J. R. M. and Wallis, J. R. and Wood, E. F. , year =. Estimation of the Generalized Extreme-Value Distribution by the Method of Probability-Weighted Moments , volume =. Technometrics , publisher =. doi:10.1080/00401706.1985.10488049 , number =

    arXiv 1985

  70. [71]

    Maciunas and Matalas, N

    Landwehr, J. Maciunas and Matalas, N. C. and Wallis, J. R. , year =. Probability weighted moments compared with some traditional techniques in estimating Gumbel Parameters and quantiles , volume =. Water Resources Research , publisher =. doi:10.1029/wr015i005p01055 , number =

    work page doi:10.1029/wr015i005p01055

  71. [72]

    and Uhlmann, J

    Julier, S.J. and Uhlmann, J. , year =. Unscented Filtering and Nonlinear Estimation , volume =. Proceedings of the IEEE , publisher =. doi:10.1109/jproc.2003.823141 , number =

    arXiv 2003

  72. [73]

    Uhlmann , title =

    J. Uhlmann , title =

  73. [74]

    and Uhlmann, J

    Julier, S.J. and Uhlmann, J. and Durrant-Whyte, H.F. , pages =. A new approach for filtering nonlinear systems , volume =. doi:10.1109/acc.1995.529783 , booktitle =

    arXiv 1995

  74. [75]

    and Lu, Peihuang and Nocedal, Jorge and Zhu, Ciyou , year =

    Byrd, Richard H. and Lu, Peihuang and Nocedal, Jorge and Zhu, Ciyou , year =. A Limited Memory Algorithm for Bound Constrained Optimization , volume =. SIAM Journal on Scientific Computing , publisher =. doi:10.1137/0916069 , number =

    doi:10.1137/0916069

  75. [76]

    The distribution of points in a cube and the accurate evaluation of integrals (in russian) zh , author=. Vychisl. Mat. i Mater. Phys , volume=

  76. [77]

    Allen, M. R. , year =. Liability for climate change , volume =. Nature , publisher =. doi:10.1038/421891a , number =

    doi:10.1038/421891a

  77. [78]

    and Wagner-Jacht, M

    Schröter, J. and Wagner-Jacht, M. and Sauerbrei, R. and Kreienkamp, F. , TITLE =. Bericht des Deutschen Wetterdienstes , YEAR =

  78. [79]

    and Sobie, Stephen R

    Cannon, Alex J. and Sobie, Stephen R. and Murdock, Trevor Q. , year =. Bias Correction of GCM Precipitation by Quantile Mapping: How Well Do Methods Preserve Changes in Quantiles and Extremes? , volume =. Journal of Climate , publisher =. doi:10.1175/jcli-d-14-00754.1 , number =

    doi:10.1175/jcli-d-14-00754.1

  79. [80]

    Multivariate—Intervariable, Spatial, and Temporal—Bias Correction , volume =

    Vrac, Mathieu and Friederichs, Petra , year =. Multivariate—Intervariable, Spatial, and Temporal—Bias Correction , volume =. Journal of Climate , publisher =. doi:10.1175/jcli-d-14-00059.1 , number =

    work page doi:10.1175/jcli-d-14-00059.1

  80. [81]

    and Brier, Glenn W

    Panofsky, Hans W. and Brier, Glenn W. , title =. 1968 , publisher =

    1968

Showing first 80 references.