Comparative study of threshold selection methods in the generalised Pareto distribution with application to rainfall datasets
DOI:
https://doi.org/10.15282/daam.v7i1.13686Keywords:
Extreme values, Threshold, Generalized Pareto distribution, Goodness of fit, p-valuesAbstract
Extreme rainfall events pose significant challenges for flood risk management and infrastructure planning, necessitating robust statistical tools for accurate risk assessment. This study rigorously compares four threshold selection methods for the generalised Pareto Distribution across five distinct rainfall datasets from Southwest England, New Zealand, Bangladesh, Singapore, and the US (Seattle). The methods evaluated include the classical mean residual life plot, a goodness-of-fit p-value-based approach, a parameter stability method, and an automated procedure that combines goodness-of-fit testing with the method of estimation. Return level estimates for 10-, 50-, and 100-year events were estimated, with uncertainties quantified via a bootstrap percentile method. The results reveal that, while each method has its merits, the approach based on goodness-of-fit criteria coupled with the method of estimation generally provides a slight edge. It consistently delivers logically interpretable thresholds that balance bias and variance effectively, particularly in managing datasets with a high prevalence of zero rainfall events. Although alternative methods occasionally yield narrower confidence intervals, they sometimes sacrifice the accurate representation of tail behaviour. Importantly, the study does not dismiss the reliability of other techniques; rather, it underscores that threshold selection is inherently dataset dependent. Overall, this study proposes that while each method offers specific advantages depending on the dataset's characteristics, the approach that integrates goodness-of-fit testing with estimation techniques consistently achieves a favourable balance between simplicity, interpretability, and statistical robustness for threshold selection in generalised Pareto modelling of rainfall extremes. These findings highlight the importance of methodological adaptability and contribute valuable insights toward improving flood risk assessments under diverse climatic conditions.
References
[1] Else H. Climate change implicated in Germany's deadly floods. Nature. 2021 Jul 20. Available from: https://www.nature.com/articles/d41586-021-01968-3.
[2] Taye MT, Ntegeka V, Ogiramoi NP, Willems P. Assessment of climate change impact on hydrological extremes in two source regions of the Nile River Basin. Hydrology and Earth System Sciences. 2011;15(1):209–222.
[3] Zhou XH, Zhou A, Shen SL. How to mitigate the impact of climate change on modern cities: lessons from extreme rainfall. Smart Construction and Sustainable Cities. 2023;1(1):7.
[4] Coles S. An introduction to statistical modeling of extreme values. London: Springer; 2001.
[5] Davison AC, Smith RL. Models for exceedances over high thresholds. Journal of the Royal Statistical Society: Series B (Statistical Methodology). 1990;52(3):393–425.
[6] Pickands J III. Statistical inference using extreme order statistics. The Annals of Statistics. 1975;3(1):119–131.
[7] Bader B, Yan J, Zhang X. Automated threshold selection for extreme value analysis via goodness-of-fit tests with application to batched return level mapping. arXiv [Preprint]. 2016. Available from: https://arxiv.org/abs/1604.02024.
[8] Liang B, Shao Z, Li H, Shao M, Lee D. An automated threshold selection method based on the characteristic of extrapolated significant wave heights. Coastal Engineering. 2019;144:22–32.
[9] Liu B, Ananda MMA. A new insight into reliability data modeling with an exponentiated composite exponential-Pareto model. Applied Sciences. 2023;13(1):645.
[10] Dupuis DJ. Exceedances over high thresholds: A guide to threshold selection. Extremes. 1999;1:251–261.
[11] Solari S, Eguen M, Polo MJ, Losada MA. Peaks Over Threshold (POT): A methodology for automatic threshold estimation using goodness of fit p-value. Water Resources Research. 2017;53(4):2833–2849.
[12] Thompson P, Cai Y, Reeve D, Stander J. Automated threshold selection methods for extreme wave analysis. Coastal Engineering. 2009;56(10):1013–1021.
[13] Gaigall D, Gerstenberg J. Cramer-von-Mises tests for the distribution of the excess over a confidence level. Journal of Nonparametric Statistics. 2023;35(3):529–561.
[14] Murphy C, Tawn JA, Varty Z. Automated threshold selection and associated inference uncertainty for univariate extremes. Technometrics. 2024;66(3):363-375.
[15] Minguez R. Automatic threshold selection for generalized Pareto and Pareto–Poisson distributions in rainfall analysis: A case study using the NOAA NCDC daily rainfall database. Atmosphere. 2025;16(1):78.
[16] Alaswed H. Graphical diagnostics for threshold selection in fitting the generalized Pareto distribution. Journal of Pure & Applied Sciences. 2024;23(1):90–95.
[17] Curceac S, Atkinson PM, Milne A, Wu L, Harris P. An evaluation of automated GPD threshold selection methods for hydrological extremes across different scales. Journal of Hydrology. 2020;585:124845.
[18] Hambuckers J, Kratz M, Usseglio-Carleve A. Efficient estimation in extreme value regression models of hedge fund tail risks. arXiv [Preprint]. 2023. Available from: https://arxiv.org/abs/2304.06950.
[19] Alif FK, Ali N, Safari MAM. An assessment on threshold selection for the generalized Pareto distribution using goodness of fit. Malaysian Journal of Mathematical Sciences. 2025;19(3):871–899.
[20] Embrechts P, Kluppelberg C, Mikosch T. Modelling extremal events for insurance and finance. British Actuarial Journal. 1999;5(2):465–465.
[21] Hosking JRM. On the characterization of distributions by their L-moments. Journal of Statistical Planning and Inference. 2006;136(1):193–198.
[22] Asquith WH. Distributional analysis with L-moment statistics using the R environment for statistical computing. CreateSpace Scotts Valley, CA, USA; 2011.
[23] Simkova T, Picek J. A comparison of L-, LQ-, TL-moment, and maximum likelihood high quantile estimates of the GPD and GEV distribution. Communications in Statistics-Simulation and Computation. 2017;46(8):5991–6010.
[24] Hosking JRM. L-moments: analysis and estimation of distributions using linear combinations of order statistics. Journal of the Royal Statistical Society: Series B (Statistical Methodology). 1990;52(1):105–124.
[25] van Staden PJ, Loots MT. Method of L-moment estimation for the generalized lambda distribution. In: Proceedings of the Third Annual ASEARC Conference. 2009 Dec 1; Newcastle, Australia. pp. 7–8.
[26] Thompson P, Cai Y, Reeve D, Stander J. Automated threshold selection methods for extreme wave analysis. Coastal Engineering. 2009;56(10):1013–1021.
[27] Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. European Journal of Epidemiology. 2016;31(4):337–350.
[28] Chu J, Dickin O, Nadarajah S. A review of goodness of fit tests for Pareto distributions. Journal of Computational and Applied Mathematics. 2019;361:13–41.
[29] Stephens MA. Goodness-of-fit techniques. In: D'Agostino RB, Stephens MA Eds. Goodness-of-Fit Techniques. New York: Routledge, 2017.
[30] Luceno A. Fitting the generalized Pareto distribution to data using maximum goodness-of-fit estimators. Computational Statistics & Data Analysis. 2006;51(2):904–917.
[31] Stephens MA. EDF statistics for goodness of fit and some comparisons. Journal of the American Statistical Association. 1974;69(347):730–737.
[32] Choulakian V, Lockhart RA, Stephens MA. Cramer-von Mises statistics for discrete distributions. The Canadian Journal of Statistics. 1994;22(1):125–137.
[33] Lilliefors HW. On the Kolmogorov-Smirnov test for normality with mean and variance unknown. Journal of the American Statistical Association. 1967;62(318):399–402.
[34] Massey FJ Jr. The Kolmogorov-Smirnov test for goodness-of-fit. Journal of the American Statistical Association. 1951;46(253):68–78.
[35] R Development Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2010.
[36] Faraway J, Marsaglia G, Marsaglia J, Baddeley A, et al. goftest: Classical Goodness-of-Fit Tests for Univariate Distributions. R package version 1.2-3; 2021.
[37] Mooney CZ, Duval RD. Bootstrapping: A nonparametric approach to statistical inference. Thousand Oaks, CA: SAGE Publications; 1993.
[38] Asquith WH. Distributional analysis with L-moment statistics using the R environment for statistical computing. CreateSpace Scotts Valley, CA, USA; 2011.
[39] Sinclair CD, Spurr BD, Ahmad MI. Modified Anderson-Darling test. Communications in Statistics-Theory and Methods. 1990;19(10):3677–3686.
[40] Coles SG, Tawn JA. Modelling extremes of the areal rainfall process. Journal of the Royal Statistical Society: Series B (Methodological). 1996;58(2):329–347.
[41] Niu D, Sayed T, Fu C, Mannering F. A cross-comparison of different extreme value modeling techniques for traffic conflict-based crash risk estimation. Analytic Methods in Accident Research. 2024;44:100352.
[42] Simkova T, Picek J. A comparison of L-, LQ-, TL-moment, and maximum likelihood high quantile estimates of the GPD and GEV distribution. Communications in Statistics-Simulation and Computation. 2017;46(8):5991–6010.
[43] Zubair M, Ishtiaque Mahee MN, Reza KM, Salim MS, Ahmed N. Climate data dynamics: A high-volume real-world structured weather dataset. Data in Brief. 2024;57:111156.
[44] National Environment Agency. Historical daily weather records. Retrieved from https://data.gov.sg/datasets/d_03bb2eb67ad645d0188342fa74ad7066/view; Apr 5 2025.
[45] Lai Y, Dzombak D. Compiled historical daily temperature and precipitation data for selected 210 U.S. cities. Carnegie Mellon University [Dataset]. 2019. doi:10.1184/R1/7891151.v4
[46] Couturier DL, Victoria-Feser MP. Zero-inflated truncated generalized Pareto distribution for the analysis of radio audience data. The Annals of Applied Statistics. 2010;4(4):1824–1846.
[47] Efron B, Tibshirani RJ. An introduction to the bootstrap. New York: Chapman & Hall; 1993.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 The Author(s)
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
