Peaks Over Threshold (POT): A methodology for automatic threshold estimation using goodness of fit p‐value
Peaks Over Threshold (POT): A methodology for automatic threshold estimation using goodness of fit p‐value
复制标题
DOI：
10.1002/2016wr019426
复制
发表时间：
2017-04
影响因子：
5.4
通讯作者：
S. Solari;Marta Egüen;M. Polo;M. Losada
中科院分区：
文献类型：
--
作者：
S. Solari;Marta Egüen;M. Polo;M. Losada
峰值超阈值(POT)方法中的阈值估计以及该估计方法对高重现期分位数及其不确定性(或可信区间)的计算的影响仍然是尚未解决的问题。过去，基于拟合优度检验和EDF统计量的方法已经取得了令人满意的结果，但它们的使用还没有系统化。本文提出了一种基于Anderson-Darling EDF统计量和拟合优度检验的自动阈值估计方法。当与自举技术相结合时，该方法可用于量化阈值估计的不确定性及其对高回报周期分位数的不确定性的影响。将该方法应用于几个模拟序列和四个降水/河流流量数据序列。实验结果证实了该算法的鲁棒性。对于测量的序列，估计的阈值与非自动方法获得的阈值相对应。此外，尽管阈值估计的不确定性很高，但这对高回报周期分位数的可信区间宽度没有显著影响。
Threshold estimation in the Peaks Over Threshold (POT) method and the impact of the estimation method on the calculation of high return period quantiles and their uncertainty (or confidence intervals) are issues that are still unresolved. In the past, methods based on goodness of fit tests and EDF‐statistics have yielded satisfactory results, but their use has not yet been systematized. This paper proposes a methodology for automatic threshold estimation, based on the Anderson‐Darling EDF‐statistic and goodness of fit test. When combined with bootstrapping techniques, this methodology can be used to quantify both the uncertainty of threshold estimation and its impact on the uncertainty of high return period quantiles. This methodology was applied to several simulated series and to four precipitation/river flow data series. The results obtained confirmed its robustness. For the measured series, the estimated thresholds corresponded to those obtained by nonautomatic methods. Moreover, even though the uncertainty of the threshold estimation was high, this did not have a significant effect on the width of the confidence intervals of high return period quantiles.