Pith. sign in

REVIEW 5 major objections 5 minor 47 references

Building nonstationary extreme value model using L-moments

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper proposes an L-moment estimator for nonstationary extreme value models that standardizes residuals to a Gumbel distribution, and reports it beats the existing GN16 method when variance trends upward.

desk verdict A genuinely new two-step L-moment estimator for nonstationary GEV models, with simulation evidence backing its central claim, but the evidence base is narrower than the title suggests. read the letter →

arxiv 2506.00977 v1 pith:J47R7ME2 submitted 2025-06-01 stat.ME stat.CO

Yire Shin , Yonggwan Shin , Jeong-Soo Park This is my paper
classification stat.MEstat.CO MSC 62G3262F3562M10
keywords Cross-validatedL-momentdistanceExpectednumberofeventsSolvingnonlinearequationsTransformationtostationarityStandardGumbeldistributionnonstationaryGEVmodelreturnlevelestimation
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

Maximum likelihood for nonstationary extreme value models can be thrown off by a few large values near the end of the sample, and an earlier L-moment fix (the GN16 method) performs worse than stationary estimation when variance increases over time. This paper proposes a two-step estimator that first fits the trend with robust regression and then transforms the residuals to a standard Gumbel distribution and matches their L-moments to the Gumbel's population L-moments. In simulations of the GEV11 model with decreasing location and increasing scale, the paper reports that the proposed method performed best in return-level RMSE for $\xi \le 0.15$ and overcame the GN16 weakness. If correct, hydrologists and climate analysts gain an L-moment-based estimator for nonstationary extremes that is less sensitive to influential observations than maximum likelihood.

What carries the argument

The load-bearing device is the transformation $\tilde{Z}_t = -(1/\xi)\log\left(1 - \xi (Z_t - \hat{\mu}_t)/\hat{\sigma}_t\right)$, which maps the fitted GEV residuals onto a standard Gumbel variable when the parameters are correct, together with the known population L-moments of that distribution ($\lambda_1 = 0.5772\ldots$, $\lambda_2 = \log 2$, $\tau_3 = 0.169925$). The algorithm fixes the trend coefficients $\hat{\mu}_1$ and $\hat{\sigma}_1$ from a robust regression and solves the three L-moment equations against the transformed data to update $\mu_0$, $\sigma_0$, and $\xi$. This second usage of L-moments, on top of the initial robust fit, is the mechanism that corrects the poor fits the paper saw in the GN16 method.

What would settle it

Re-run the paper's GEV11 simulation with a larger positive scale trend (for example, $\sigma_1 = 0.1$ instead of 0.02); if the proposed method no longer has the lowest RMSE for return levels, the paper's central claim that it overcomes the positive-variance-trend weakness would be refuted for stronger trends.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the GEV11 model's return-level estimation under a positive variance trend improves if, instead of applying L-moments to detrended pseudo-observations, one applies L-moment estimation to residuals that have been transformed to a standard Gumbel distribution. The algorithm starts from the GN16 estimates obtained with robust regression, fixes the time-trend coefficients $\mu_1$ and $\sigma_1$, and solves the three equations $\lambda_1 = l_1(\tilde{Z}_t)$, $\lambda_2 = l_2(\tilde{Z}_t)$, $\tau_3 = t_3(\tilde{Z}_t)$ for $\mu_0$, $\sigma_0$, and $\xi$. This second L-moment pass is what the paper credits for the gain, and it is presented as overcoming the weakness of the GN16 method that motivated the study.

Load-bearing premise

The method's accuracy rests on assuming the trend coefficients from the initial robust regression are good enough to hold fixed while only the intercepts and shape parameter are re-estimated, an assumption supported by simulation at one parameter setting rather than by a proof.

Editorial extensions

If this is right

  • For GEV11 models with a positive trend in scale, the proposed method yields lower RMSE for return levels than MLE, WLS, and GN16 in the simulated heavy-tailed and near-Gumbel range; this is the paper's headline result.
  • The method extends to nonstationary GEV models with physical covariates by applying the same two-step logic: robust regression on the covariates, then L-moment matching on the standardized residuals.
  • The cross-validated generalized L-moment distance provides a model-selection criterion that does not rely on likelihoods; in the Fremantle sea-level example it selects the time-plus-SOI model that a likelihood analysis also supports.
  • Because the final estimates are produced from a standard-Gumbel-transformed series, the method inherits the robustness of L-moments against outliers while also correcting the initial trend fit.

Reading between the lines

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

  • A natural next experiment is to vary the size and sign of the variance trend; the paper only simulates $\sigma_1 = 0.02$, so its claim to overcome the GN16 weakness is established for one trend strength, not for the whole positive-trend regime.
  • If the Gumbel-standardization step is the source of the gain, the same correction could be grafted onto other initial estimators such as penalized MLE or Bayesian fits, and might reduce their sensitivity to end-of-sample outliers.
  • The paper's model-selection criterion (CV GLD) can be applied to any L-moment fit, so it could serve as a likelihood-free selection rule for comparing WLS, GN16, and the proposed method on the same data.
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

5 major / 5 minor

Summary. The paper proposes an L-moment-based estimation method for nonstationary generalized extreme value (NS-GEV) models, specifically targeting the GEV11 model with a time trend in both location and log-scale. The key idea is to first obtain initial trend estimates via robust (MM) regression, then transform the data to standard Gumbel residuals using the GEV probability integral transform, and finally solve a system of three L-moment equations for the intercept parameters and the shape parameter while holding the time-slope parameters fixed. The method is compared with MLE, stationary L-moment estimation, WLS, and the GN16 method in a Monte Carlo study, applied to peak streamflow data from Trehafod, UK, and extended to physical-covariate models and a cross-validated generalized L-moment distance (CV GLD) criterion for model selection. The main claimed finding is that the proposed method overcomes the known weakness of GN16 for GEV11 models with a positive variance trend, giving lower RMSE for return levels in heavy-tailed and near-Gumbel cases.

Significance. If the central claim is robust, the proposed estimator is a useful addition to the toolbox for nonstationary extreme value analysis, especially for small samples and samples with influential observations, where MLE is known to be unstable. The paper's strengths include a clearly described algorithm, reproducible R code and data links, a comparison against several standard estimators, and a novel extension to model selection via cross-validated L-moment distance. The Fremantle sea-level example and the Trehafod flood application illustrate practical implementation. However, the main empirical claim rests on a very narrow simulation design, and several methodological points concerning the two-step estimation procedure and the root-selection rule are not fully resolved. The significance is therefore conditional on the method being more thoroughly validated.

major comments (5)
  1. [§4.1, §4.2, Tables 1–2] The central claim that the proposed method overcomes the positive-variance-trend weakness of GN16 is supported by only one GEV11 simulation configuration (mu0=0, mu1=-0.1, sigma0=1, sigma1=0.02), with Tables 1 and 2 reporting results for n=50 only. The paper states in §4.1 that sample sizes n=25 and n=50 are used, but no GEV11 results for n=25 are presented. More importantly, the strength of the variance trend (sigma1) is not varied, even though this is the very feature the method is designed to handle; the sign of the location trend is also fixed. Without a sensitivity analysis over sigma1, mu1, and sample size, the abstract and Section 4.2 overstate the generality of the simulation finding.
  2. [§3.3, Step 2] The proposed estimator is a two-step procedure: the first-step robust regression supplies mu1 and sigma1, and the second step solves for mu0, sigma0, and xi conditional on those fixed slope estimates. The paper provides no consistency or asymptotic-normality argument for this two-step estimator, and the final properties necessarily inherit the unknown properties of the first-stage robust regression estimates. This gap is load-bearing because the method is recommended over competitors on the basis of finite-sample RMSE, yet no theoretical or bootstrap-based justification is offered for the effect of fixing mu1 and sigma1.
  3. [§7, Eq. (25)] The root-finding step of the proposed algorithm can return multiple solutions, and the final estimate is selected by the data-dependent chi criterion in Eq. (25), with T_max set to 80 for n=50 and starting values chosen heuristically. The sensitivity of the reported RMSEs to this selection rule, to T_max, and to the number and placement of starting values is not examined. Because this selection is part of the estimator definition, the reported simulation results do not cleanly measure the statistical procedure as it would be applied by other users.
  4. [§4.2, Table 2] The statement that 'the proposed method performed best for xi <= .15 in terms of the RMSE' is not supported by Table 2 for the redefined return level: at xi=-0.35 the WLS method has RMSE 14.38 while the proposed method has RMSE 15.27. The claim should be restricted to the conventional return level in Table 1, or the text should acknowledge the exception, since the abstract and Section 4.2 do not make this distinction.
  5. [§6.2, §6.3, Eq. (22)] The CV GLD model selection criterion is computed using a covariance matrix V that is estimated from the pooled full data of all candidate models and then fixed across all cross-validation folds. This uses test-fold information in the scoring metric, so the reported CV GLD values are not a clean out-of-sample measure. Since the model selection example concludes that Model 3 is selected because it has the smallest CV GLD, this leakage could affect the validity of that comparison.
minor comments (5)
  1. [Throughout] There are numerous typographical errors, including 'Consquently', 'leaded', 'scond', 'assummed', 'neigbohood', 'soultions', 'limitted', 'represends', and 'bahavior'; a careful proofread is needed.
  2. [§3.3] The text refers to 'the Supplementary Information' for a proof that the transformed variable follows a standard Gumbel distribution; the supplementary material should be made available and the proof should be included in the submission package.
  3. [§4.1] The description of the simulation setting mentions sample sizes n=25 and n=50, but Tables 1 and 2 report only n=50 for GEV11; the text should clarify where the n=25 GEV11 results are or remove the reference.
  4. [§6.3] In the comparison of coefficients of variation for mu1 estimates for time and SOI, the text first notes the unit difference and then compares coefficients of variation; since these are unit-free, the comparison is valid, but the discussion could be clearer about what the comparison is meant to show.
  5. [References] Several references have spelling errors, e.g., 'genaralized', and the reference to Shin et al. (2025) is a preprint; please update the final version with the published reference if available.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the proposed estimator is a standard two-stage moment-matching procedure, and the simulation comparisons are externally benchmarked.

full rationale

The paper's derivation chain is self-contained. In Step 1, the trend parameters (mu_1, sigma_1) are obtained from a robust regression of the raw data. In Step 2, the remaining parameters (mu_0, sigma_0, xi) are solved from the moment equations lambda_1 = l_1(Ztilde_t), lambda_2 = l_2(Ztilde_t), and tau_3 = t_3(Ztilde_t), where Ztilde_t is the GEV-to-Gumbel transformation (13). These equations match sample L-moments of the transformed data to the fixed population L-moments of a standard Gumbel distribution. This is a standard method-of-moments construction: the data enter through the sample L-moments, and the target distribution enters through fixed constants (Euler constant, log 2, 0.169925). No target return level is used as an input, and no fitted parameter is renamed as a prediction. The claim that the proposed method 'performed best for xi <= .15' is an empirical Monte Carlo result obtained by comparing RMSEs against MLE, WLS, GN16, and stationary LME under an explicitly stated simulation setting; it is not forced by the estimating equations. The two-stage choice to fix mu_1 and sigma_1 while updating mu_0, sigma_0, and xi is a model-selection detail and a potential robustness limitation, but it is not circular. The only self-citation is the reference to Shin et al. (2025) in Section 6.2 as a prior user of GLD, and it is not load-bearing: the GLD criterion itself is attributed to Elamir and Seheult (2004), and the CV GLD proposal is new. Therefore the paper's central derivation has independent content and does not reduce to its inputs.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The method introduces no new physical entities. It relies on standard extreme value theory and L-moment theory, plus the domain assumption of correct model specification. Two algorithmic constants (starting values, T_max) are hand-chosen and could affect results in rare cases.

free parameters (3)
  • number of root-finding starting values = 20
    Chosen by hand in Section 3.3; affects convergence and possible multiple solutions.
  • T_max in chi selection criterion = 80
    Set in Section 7 for n=50 to choose among multiple solutions when they exist; a data-dependent selection constant.
  • bootstrap replicates B = 300
    Used for standard errors in Section 6.3; arbitrary but conventional.
assumptions (5)
  • domain assumption The NS GEV model with linear location and log-linear scale is correctly specified for the data.
    The transformation (13) yields standard Gumbel only under this model; invoked in Section 3.3 and throughout.
  • standard math Sample L-moments are unbiased estimators of population L-moments.
    Hosking (1990); used to justify matching l1, l2, t3 to lambda1, lambda2, tau3.
  • domain assumption MM robust regression yields consistent estimates of trend coefficients.
    Used in Step 1; no proof in the paper that the final estimator inherits consistency.
  • standard math Newton's method in nleqslv converges to a root of the three-equation system.
    Relied on in Step 2; the paper reports failure rate below 0.05 percent but no guarantee.
  • standard math Elamir and Seheult (2004) formula for the covariance matrix of sample L-moments is correct.
    Used to build the GLD in equation (22).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Building nonstationary extreme value model using L-moments." pith.science (2026). https://pith.science/paper/J47R7ME2

@misc{pith2026250600977,
  author       = {Pith},
  title        = {Pith review of: Building nonstationary extreme value model using L-moments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J47R7ME2}},
  note         = {Machine review of arXiv:2506.00977}
}
read the original abstract

The maximum likelihood estimation for a time-dependent nonstationary (NS) extreme value model is often too sensitive to influential observations, such as large values toward the end of a sample. Thus, alternative methods using L-moments have been developed in NS models to address this problem while retaining the advantages of the stationary L-moment method. However, one method using L-moments displays inferior performance compared to stationary estimation when the data exhibit a positive trend in variance. To address this problem, we propose a new algorithm for efficiently estimating the NS parameters. The proposed method combines L-moments and robust regression, using standardized residuals. A simulation study demonstrates that the proposed method overcomes the mentioned problem. The comparison is conducted using conventional and redefined return level estimates. An application to peak streamflow data in Trehafod in the UK illustrates the usefulness of the proposed method. Additionally, we extend the proposed method to a NS extreme value model in which physical covariates are employed as predictors. Furthermore, we consider a model selection criterion based on the cross-validated generalized L-moment distance as an alternative to the likelihood-based criteria.

Figures

Figures reproduced from arXiv: 2506.00977 by the authors.

Figure 1
Figure 1. Simulation results: Root mean squared error (RMSE) of the 50-year return level [PITH_FULL_IMAGE:figures/full_fig_p013_1.png] view at source ↗
Figure 2
Figure 2. Scatterplot and 50-year conventional return levels for the GEV11 model fitted using [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗
Figure 3
Figure 3. Quantile-per-quantile plots for standardized residuals based on the five methods, [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Modified return level plots based on the formula by Parey et al. (2010) for the five [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 46 canonical work pages

  1. [1]

    Extremes in a changing climate: Detection, analysis and uncertainty

    AghaKouchak A, Easterling D, Hsu K, Schubert S, Sorooshian S (Eds.) 2013. Extremes in a changing climate: Detection, analysis and uncertainty . Springer

    work page 2013

  2. [2]

    CreateSpace

    Asquith WH (2011) Distributional analysis with L-moment statistics using the R environment for statistical computing , First ed. CreateSpace. Available from http://www.amazon.com

    work page 2011

  3. [3]

    Nat Hazards Earth Syst Sci, 22:3663–3677

    Baldan D, Coraci E, Crosato F, Ferla M et al.(2022) Importance of non-stationary analysis for assessing extreme sea levels under sea level rise. Nat Hazards Earth Syst Sci, 22:3663–3677

    work page 2022

  4. [4]

    Bousquet N, Bernardara P (Eds.) (2021) Extreme value theory with applications to natural hazards , Springer

    work page 2021

  5. [5]

    Choi G (2021) A new approach for detecting gradual changes in non-stationary time series with seasonal effects. J. Korean Stat. Soc. 50, 419–430

    work page 2021

  6. [6]

    Hydro Process, 24:673–685

    Cannon AJ (2010) A flexible nonlinear modelling framework for nonstationary generalized extreme value analysis in hydroclimatology. Hydro Process, 24:673–685

    work page 2010

  7. [7]

    Springer, London

    Coles S (2001) An introduction to statistical modeling of extreme values . Springer, London

    work page 2001

  8. [8]

    (Eds.): Extremes in a changing climate: Detection, analysis and uncertainty

    Cooley D (2013) Return periods and return levels under climate change, in AghaKouchak A, Easterling D et al. (Eds.): Extremes in a changing climate: Detection, analysis and uncertainty . Springer

    work page 2013

Show all 47 references
  1. [9]

    J Hydro 276:210-223

    Cunderlik JM, Burn DH (2003) Non-stationary pooled flood frequency analysis. J Hydro 276:210-223

    2003

  2. [10]

    Jr and Schnabel, R.B

    Dennis, J.E. Jr and Schnabel, R.B. (1996), Numerical Methods for Unconstrained Optimization and Nonlinear Equations, SIAM, Philadelpia

    1996

  3. [11]

    Comput Stat Data Anal 43:299-314

    Elamir EAH, Scheult AH (2003) Trimmed L-moments. Comput Stat Data Anal 43:299-314

    2003

  4. [12]

    Jour Statist Plann Infer 124(2):337-359

    Elamir EA, Seheult AH (2004) Exact variance structure of sample L-moments. Jour Statist Plann Infer 124(2):337-359

    2004

  5. [13]

    (2023) Non-stationary large-scale statistics of precipitation extremes in central Europe

    Fauer, F.S., Rust, H.W. (2023) Non-stationary large-scale statistics of precipitation extremes in central Europe. Stoch Environ Res Risk Assess 37, 4417–4429

    2023

  6. [14]

    A simulation study

    Gado TA, Nguyen VTN (2016a) An at-site flood estimation method in the context of nonstationarity I. A simulation study. J Hydro 535:710-721

    2016

  7. [15]

    Statistical analysis of floods in Quebec

    Gado TA, Nguyen VTN (2016b) An at-site flood estimation method in the context of nonstationarity II. Statistical analysis of floods in Quebec. J Hydro 535:722-736

    2016

  8. [16]

    Robust Local Likelihood Estimation for Non-stationary Flood Frequency Analysis

    Grego JM, Yates PA (2024). Robust Local Likelihood Estimation for Non-stationary Flood Frequency Analysis. J Agri Bio Envir Stat, 1-20. https://doi.org/10.1007/s13253-024-00614-0

    2024 doi

  9. [17]

    J Royal Stat Soc B, 52(1):105-124

    Hosking JRM (1990) L-Moments: Analysis and Estimation of Distributions Using Linear Combinations of Order Statistics. J Royal Stat Soc B, 52(1):105-124

    1990

  10. [18]

    Technometrics, 27:251-261

    Hosking JRM, Wallis JR, Wood EF (1985) Estimation of the genaralized extreme-value distribution by the method of probabilty weighted moments. Technometrics, 27:251-261

    1985

  11. [19]

    Cambridge Univ Press, Cambridge

    Hosking JRM, Wallis JR (1997) Regional Frequency Analysis: An Approach Based on L-Moments . Cambridge Univ Press, Cambridge

    1997

  12. [20]

    & Khastagir, A (2021) Effects of estimation techniques on generalised extreme value distribution (GEVD) parameters and their spatio-temporal variations

    Hossain, I., Imteaz, M.A. & Khastagir, A (2021) Effects of estimation techniques on generalised extreme value distribution (GEVD) parameters and their spatio-temporal variations. Stoch Environ Res Risk Assess 35, 2303–2312

    2021

  13. [21]

    Stoch Environ Res Risk Assess 28(8):2063-2074

    Jeong BY, Murshed MS, Seo YA, Park J-S (2014) A three-parameter kappa distribution with hydrologic application: A generalized Gumbel distribution. Stoch Environ Res Risk Assess 28(8):2063-2074

    2014

  14. [22]

    Springer

    James G, Witten D, Hastie T, Tibshirani R (2021) An Introduction to Statistical Learning: With Applications in R , 2nd ed. Springer

    2021

  15. [23]

    J Water Clim Change, 11(4):966-979

    Jan NAM, Shabri A, Samsudin R (2020) Handling non-stationary flood frequency analysis using TL-moments approach for estimating parameter. J Water Clim Change, 11(4):966-979

    2020

  16. [24]

    et al.(Eds.) Extremes in a changing climate: Detection, analysis and uncertainty , Springer

    Katz R (2013) Statistical methods for nonstationary extremes, in AghaKouchak, A., Easterling, D. et al.(Eds.) Extremes in a changing climate: Detection, analysis and uncertainty , Springer

    2013

  17. [25]

    Water Resour Res

    Kjeldsen T.R., Prosdocimi I.(2015) A bivariate extension of the Hosking and Wallis goodness-of-fit measure for regional distributions. Water Resour Res. 51(2), 896-907

    2015

  18. [26]

    Koller M, Stahel WA (2017) Nonsingular subsampling for regression S estimators with categorical predictors, Comput Stat 32(2):631–646

    2017

  19. [27]

    Environmetrics 32(4):e2667

    Konzen E, Neves C, Jonathan P (2021) Modeling nonstationary extremes of storm severity: Comparing parametric and semiparametric inference. Environmetrics 32(4):e2667

    2021

  20. [28]

    Mudersbach C. and Jensen J (2010) Nonstationary extreme value analysis of annual maximum water levels for designing coastal structures on the German North Sea coastline, J Flood Risk Manag 3:52-62

    2010

  21. [29]

    Springer

    Naghettini M (ed)(2017) Fundamentals of Statistical Hydrology . Springer

    2017

  22. [30]

    Risk Anal 18:497-510

    Olsen J, Lambert JH, Haimes T (1998) Risk of extreme events under nonstationary conditions. Risk Anal 18:497-510

    1998

  23. [31]

    Jour Appl Stat, 49(6):1559-1573

    Papukdee N, Park J-S, Busababodhin P (2022) Penalized likelihood approach for the four-parameter kappa distribution. Jour Appl Stat, 49(6):1559-1573

    2022

  24. [32]

    Environmetrics 21:698-718

    Parey S, Hoang TTH, Dacunha-Castelle D (2010) Different ways to compute temperature return levels in the climate change context. Environmetrics 21:698-718

    2010

  25. [33]

    Parey S, Hoang TTH (2021) Extreme values of non-stationary time series, in Bousquet N, Bernardara P (Eds) Extreme value theory with applications to natural hazards , Springer

    2021

  26. [34]

    Atmos Sci Lett, 24(4), e1145

    Prahadchai T, Shin Y, Busababodhin P, Park J-S (2023) Analysis of maximum precipitation in Thailand using non-stationary extreme value models. Atmos Sci Lett, 24(4), e1145

    2023

  27. [35]

    Weath Clim Extrem, 41, 100575

    Radfar S, Galiatsatou P, Wahl T (2023) Application of nonstationary extreme value analysis in the coastal environment -- A systematic literature review. Weath Clim Extrem, 41, 100575

    2023

  28. [36]

    Birkhauser

    Reiss R-D, Thomas M (2007) Statistical analysis of extreme values, 3rd ed. Birkhauser

    2007

  29. [37]

    Nonlin Process Geophys, 15:1033-1039

    Ribereau P, Guillou A, Naveau P (2008) Estimating return levels from maxima of non-stationary random sequences using the generalized PWM method. Nonlin Process Geophys, 15:1033-1039

    2008

  30. [38]

    J Hydro Eng, 19:554–568

    Salas JD, Obeysekera J (2014) Revisiting the concepts of return period and risk for nonstationary hydrologic extreme events. J Hydro Eng, 19:554–568

    2014

  31. [39]

    Hydrol Sci J 63(3):325-352

    Salas JD, Obeysekera J, Vogel RM (2018) Techniques for assessing water infrastructure for nonstationary extreme events: A review. Hydrol Sci J 63(3):325-352

    2018

  32. [40]

    Preprint at http://arxiv.org/abs/2505.21417

    Shin Y, Shin Y, Park J-S (2025) Model averaging with mixed criteria for estimating high quantiles of extreme values: Application to heavy rainfall. Preprint at http://arxiv.org/abs/2505.21417

    2025

  33. [41]

    Stat and Comput 10, 63–72

    Smyth, P (2000) Model selection for probabilistic clustering using cross-validated likelihood. Stat and Comput 10, 63–72

    2000

  34. [42]

    Stein ML (2021) A parametric model for distributions with flexible bahavior in both tails, Environmetrics 32(2), e2658

    2021

  35. [43]

    Strupczewski WG, Kaczmarek Z (2001) Non-stationary approach to at-site flood frequency modelling. Part II. Weighted least squares estimation. J Hydro, 48(14):143-151

    2001

  36. [44]

    Strupczewski WG, Kochanek K, Bogdanowicz E, et al. (2016) Comparison of two nonstationary flood frequency analysis methods within the context of the variable regime in the representative Polish rivers, Acta Geophysica, 64(1):206-236

    2016

  37. [45]

    Academic Press

    Wilcox RR (2021) Introduction to robust estimation and hypothesis testing , 5th Ed. Academic Press

    2021

  38. [46]

    J Appl Stat, 48(13-15):2259-2284

    Yilmaz A, Kara M, Özdemir O (2021) Comparison of different estimation methods for extreme value distribution. J Appl Stat, 48(13-15):2259-2284

    2021

  39. [47]

    (1987) High breakdown point and high efficiency robust estimates for regression

    Yohai V.J. (1987) High breakdown point and high efficiency robust estimates for regression. Ann Stat 15:642-656

    1987

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.