• Journal of the Korean Statistical Society
• /
• v.33 no.3
• /
• pp.339-351
• /
• 2004

In this paper we generalize and improve the refined Pickands estimator of Drees (1995) for the extreme value index. The finite-sample performance of the refined Pickands estimator is not good particularly when the sample size n is small. For each fixed k = 1,2,..., a new estimator is defined by a convex combination of k different generalized Pickands estimators and its asymptotic normality is established. Optimal weights defining the estimator are also determined to minimize the asymptotic variance of the estimator. Finally, letting k depend upon n, we see that the resulting estimator has a better finite-sample behavior as well as a better asymptotic efficiency than the refined Pickands estimator.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2020.06a
• /
• pp.279-279
• /
• 2020

최근 copula 모형은 여러 확률변수를 갖는 수문현상에 대해 빈도해석을 수행할 경우 결합확률분포형으로 유용하게 사용되고 있다. 하나의 자료를 확률변수로 사용하는 단변량 빈도해석에 비해 여러 수문자료를 동시에 각각 확률변수로 취하여 결합확률분포형을 추정할 수 있는 다변량 빈도해석은 수문자료의 상관성을 고려하면서 확률분포형을 추정할 수 있다는 장점이 있다. Copula 모형 중 Gumbel copula는 extreme-value 확률분포형으로 극치사상에 적합한 확률분포형이다. 본 연구에서는 Gumbel copula를 이용하여 우리나라 기상청 64개 종관기상관측소의 강우자료로부터 극치 강우사상을 추출하고, 이를 이용하여 빈도해석을 수행하였다. 극치 강우사상은 전체 강우사상 중 각 년도별로 최대강우량을 갖는 연최대강우량사상(annual maximum volume event)을 사용하였다. 각 확률변수의 주변분포형으로는 gamma, Gumbel, generalized extreme value, generalized logistic, Weibull 등 5개 확률분포형을 검토하였으며 각각 적합한 주변분포형을 적용하고 copula 모형의 매개변수는 의사최우도법(maximum pseudo-likelihood method)를 사용하여 추정하였다. 또한 추정된 copula 모형은 Cramer-von Mises 함수와 경험적 copula를 이용하여 적합도 검정을 수행하였다. 이를 통해 극치강우사상에 대하여 Gumbel copula 모형의 적용성을 검토하였으며 추정된 결합확률분포형을 이용하여 빈도별 확률강우사상을 2차원 등치선(contour line)형태로 제시하였다.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.20 no.5
• /
• pp.482-490
• /
• 2008

For a coastal or harbor structure design, one of the most important environmental factors is the appropriate extreme highest tide level condition. Especially, the information of extreme highest tide level distribution is essential for reliability design. In this paper, 23 set of extreme highest tide level data obtained from National Oceanographic Research Institute(NORI) were analyzed for extreme highest tide levels. The probability distributions considered in this research were Generalized Extreme Value(GEV), Gumbel, and Weibull distribution. For each of these distributions, three parameter estimation methods, i.e. the method of moments, maximum likelihood and probability weighted moments, were applied. Chi-square and Kolmogorov-Smirnov goodness-offit tests were performed, and the assumed distribution was accepted at the confidence level 95%. Gumbel distribution which best fits to the 22 tidal station was selected as the most probable parent distribution, and optimally estimated parameters and extreme highest tide level with various return periods were presented. The extreme values of Incheon, Cheju, Yeosu, Pusan, and Mukho, which estimated by Shim et al.(1992) are lower than that of this result.

• Proceedings of the Korean Society of Agricultural Engineers Conference
• /
• 2005.10a
• /
• pp.284-287
• /
• 2005

The objective of this study is to check into variation trends of design rainfall according to change of the number of years for observed data. To make comparative study of the relation between design rainfall and recorded year, this study was used maximum rainfall for 24-hr consecutive duration at Gangneung, Seoul, Incheon, Chupungnyeong, Pohang, Daegu, Jeonju, Ulsan, Gwangju, Busan, Mokpo and Yeosu rainfall stations. The tests for Independence, Homogeneity and detection of outliers were used Wald-Wolfowitz's test, Mann-Whitney's test and Grubbs and Beck test respectively. To select appopriate distribution, the distribution of genaralized pareto(GPA), generalized extreme value(GEV), generalized logistic(GLO), lognormal and pearson type 3 distribution is judged by L-moment ratio diagram and Kolmogorov-Smirnov (K-S) test. Design rainfall was estimated by at-site frequency analysis using L-moments and Generalized extreme value(GEV) distribution according to change of the number of years for observed data. Through the comparative analysis for design rainfall induced by L-moments and GEV distribution, relationship between design rainfall and recorded year is provided.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.54 no.6
• /
• pp.133-142
• /
• 2012

The objective of this study was to correct the bias of the Representative Concentration Pathways (RCP)-based future precipitation data using a quantile mapping method. This method was adopted to correct extreme values because it was designed to adjust simulated data using probability distribution function. The Generalized Extreme Value (GEV) distribution was used to fit distribution for precipitation data obtained from the Korea Meteorological Administration (KMA). The resolutions of precipitation data was 12.5 km in space and 3-hour in time. As the results of bias correction over the past 30 years (1976~2005), the annual precipitation was increased 16.3 % overall. And the results for 90 years (divided into 2011~2040, 2041~2070, 2071~2100) were that the future annual precipitation were increased 8.8 %, 9.6 %, 11.3 % respectively. It also had stronger correction effects on high value than low value. It was concluded that a quantile mapping appeared a good method of correcting extreme value.

• Communications for Statistical Applications and Methods
• /
• v.24 no.5
• /
• pp.493-505
• /
• 2017

Maximum likelihood estimation (MLE) of the generalized extreme value distribution (GEVD) is known to sometimes over-estimate the positive value of the shape parameter for the small sample size. The maximum penalized likelihood estimation (MPLE) with Beta penalty function was proposed by some researchers to overcome this problem. But the determination of the hyperparameters (HP) in Beta penalty function is still an issue. This paper presents some data adaptive methods to select the HP of Beta penalty function in the MPLE framework. The idea is to let the data tell us what HP to use. For given data, the optimal HP is obtained from the minimum distance between the MLE and MPLE. A bootstrap-based method is also proposed. These methods are compared with existing approaches. The performance evaluation experiments for GEVD by Monte Carlo simulation show that the proposed methods work well for bias and mean squared error. The methods are applied to Blackstone river data and Korean heavy rainfall data to show better performance over MLE, the method of L-moments estimator, and existing MPLEs.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.4 no.4
• /
• pp.250-260
• /
• 1992

Two methods for computing extreme sea levels, which are the extreme probability method and the joint probability method, are examined at five different ports (Incheon, Cheju, Yeosu, Pusan, Mukho). The extreme probability mothod estimates the extreme sea levels from three different probability papers of Gumbel, Weibull and generalized extreme value(GEV) using the least square method, conventional moment method and probability weighted moment method. respectively. The results showed that the extreme sea levels estimated by the Gumbel paper or the least square method appeared higher than those calculated by other papers or methods. The extreme values estimated by the extreme probability method are approximately 5-10 cm lower than the values by the joint probability method.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.29 no.3B
• /
• pp.259-267
• /
• 2009

Generalized Extreme Value (GEV) distribution is recommended for flood frequency and extreme rainfall distribution in many country. L-moment method is the most common estimation procedure for the GEV distribution. In this study, the relationships between the cross-site correlations between extreme events and the cross-correlation of estimators of L-moment ratios (L-moment Coefficient of Variation (L-CV) and L-moment Coefficient of Skewness (L-CS)) for data generated from GEV distribution were derived by Monte Carlo simulation. Those relationships were fit to the simple power function. In this Monte Carlo simulation, GEV+ distribution were employed wherein unrealistic negative values were excluded. The simple power models provide accurate description of the relationships between cross-correlation of data and cross-correlation of L-moment ratios. Estimated parameters and accuracies of the power functions were reported for different GEV distribution parameters combinations. Moreover, this study provided a description about regional regression approach using Generalized Least Square (GLS) regression method which require the cross-site correlation among L-moment estimators. The relationships derived in this study allow regional GLS regression analyses of both L-CV and L-CS estimators that correctly incorporate the cross-correlation among GEV L-moment estimators.

• Communications for Statistical Applications and Methods
• /
• v.31 no.1
• /
• pp.37-53
• /
• 2024

In this study, flood records from 79 sites across Thailand were analyzed to estimate flood indices using the regional frequency analysis based on the L-moments method. Observation sites were grouped into homogeneous regions using k-means and Ward's clustering techniques. Among various distributions evaluated, the generalized extreme value distribution emerged as the most appropriate for certain regions. Regional growth curves were subsequently established for each delineated region. Furthermore, 20- and 100-year return values were derived to illustrate the recurrence intervals of maximum rainfall across Thailand. The predicted return values tend to increase at each site, which is associated with growth curves that could describe an increasing long-term predictive pattern. The findings of this study hold significant implications for water management strategies and the design of flood mitigation structures in the country.

• The Korean Journal of Applied Statistics
• /
• v.23 no.5
• /
• pp.835-844
• /
• 2010

Tools for statistical analysis of extreme values include the classical annual maximum method, the modern threshold method and variants improving the second one. While the annual maximum method is to t th generalized extreme value distribution to the annual maxima of a time series, the threshold method is to the generalized Pareto distribution to the excesses over a high threshold from the series. In this paper we deal with the Poisson-GPD method, a variant of the threshold method with a further assumption that the total number of exceedances follows the Poisson distribution, and apply it to the daily percentage increases and decreases computed from the spot prices of West Texas Intermediate, which were collected from January 4th, 1988 until December 31st, 2009. According to this analysis, the distribution of daily percentage increases as well as decreases turns out to have a heavy tail, unlike the normal distribution, which coincides well with the general phenomenon appearing in the analysis of lots of nowaday nancial data.