Application of Bayesian approach to hydrological frequency analysis

Department of Geosciences, University of Nevada, Las Vegas, NV 89154, USA
Science China Technological Sciences (Impact Factor: 1.11). 05/2011; 54(5):1183-1192. DOI: 10.1007/s11431-010-4229-4

ABSTRACT An existing Bayesian flood frequency analysis method is applied to quantile estimation for Pearson type three (P-III) probability
distribution. The method couples prior and sample information under the framework of Bayesian formula, and the Markov Chain
Monte Carlo (MCMC) sampling approach is used to estimate posterior distributions of parameters. Different from the original
sampling algorithm (i.e. the important sampling) used in the existing approach, we use the adaptive metropolis (AM) sampling
technique to generate a large number of parameter sets from Bayesian parameter posterior distributions in this paper. Consequently,
the sampling distributions for quantiles or the hydrological design values are constructed. The sampling distributions of
quantiles are estimated as the Bayesian method can provide not only various kinds of point estimators for quantiles, e.g.
the expectation estimator, but also quantitative evaluation on uncertainties of these point estimators. Therefore, the Bayesian
method brings more useful information to hydrological frequency analysis. As an example, the flood extreme sample series at
a gauge are used to demonstrate the procedure of application.

KeywordsBayesian theory–hydrological frequency analysis–Markov Chain Monte Carlo–prior distribution–posterior distribution

1 Bookmark
  • [Show abstract] [Hide abstract]
    ABSTRACT: Hydrological frequency analysis is the foundation for hydraulic engineering design and water resources management. However, the existence of uncertainty in sample representation usually causes the uncertainty in quantile or design value estimation. Standard deviation (SD) of the design value with a given non-exceedance probability, as an extremely valuable indicator, is suggested to quantify the uncertainties of hydrological frequency analysis. In order to assess the impact of the sampling uncertainty on design value, in China’s national standard for design flood calculation, an empirical formula for estimating SD of the design value was suggested; however, it is applicable only to the case that Pearson type three probability distribution function (PE3) and method of moment are used to analyze the hydrological samples. In principle, for other types of probability distribution functions and parameter estimation methods, the suggested empirical formula is not suitable. The aim of this article is to propose a general approach to estimate the SD of a design value by using the Bootstrap resampling technique. In order to testify the applicability of the approach, under the condition of PE3 distribution, three kinds of schemes were designed to calculate SD, i.e. the suggested empirical formula in China’s national standard for design flood calculation (SEF), Bootstrap technique with method of moment for parameter estimation and Bootstrap with linear moment for parameter estimation. The annual precipitation observations from 50 gauges around China were analyzed using these three approaches. Results show that the SD values of quantile estimates are significantly different among these three approaches, which means that when parameter estimation methods rather than method of moment are employed, the SD provided by SEF is inapplicable. It also indicates that, in terms of modified design value, i.e., original design value adding security correction value, the proposed method using the Bootstrap method, as a general method for SD estimation, is effective and feasible.
    Stochastic Environmental Research and Risk Assessment 02/2015; 29(2). DOI:10.1007/s00477-014-0979-z · 2.67 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Profile likelihood function is introduced to analyze the uncertainty of hydrometeorological extreme inference and the theory of estimating confidence intervals of the key parameters and quantiles of extreme value distribution by profile likelihood function is described. GEV (generalized extreme value) distribution and GP (generalized Pareto) distribution are used respectively to fit the annual maximum daily flood discharge sample of the Yichang station in the Yangtze River and the daily rainfall sample in 10 big cities including Guangzhou. The parameters of the models are estimated by maximum likelihood method and the fitting results are tested by probability plot, quantile plot, return level plot and density plot. The return levels and confidence intervals of flood and rainstorm in different return periods are calculated by profile likelihood function. The results show that the asymmetry of the profile likelihood function curve increases with the return period, which can reflect the effect of the length of sample series and return periods on confidence interval. As an effective tool for estimating confidence interval of the key parameters and quantiles of extreme value distribution, profile likelihood function can lead to a more accurate result and help to analyze the uncertainty of extreme values of hydrometeorology.
    Science China Technological Sciences 12/2013; 56(12):3151-3160. DOI:10.1007/s11431-013-5421-0 · 1.11 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: The hydrological frequency analysis (HFA) is the foundation for the hydraulic engineering design and water resources management. Hydrological extreme observations or samples are the basis for HFA; the representativeness of a sample series to the population distribution is extremely important for the estimation reliability of the hydrological design value or quantile. However, for most of hydrological extreme data obtained in practical application, the size of the samples is usually small, for example, in China about 40 ∼ 50 years. Generally, samples with small size cannot completely display the statistical properties of the population distribution, thus leading to uncertainties in the estimation of hydrological design values. In this paper, a new method based on bootstrap is put forward to analyze the impact of sampling uncertainty on the design value. By bootstrap resampling technique, a large number of bootstrap samples are constructed from the original flood extreme observations; the corresponding design value or quantile is estimated for each bootstrap sample, so that the sampling distribution of design value is constructed; based on the sampling distribution, the uncertainty of quantile estimation can be quantified. Compared with the conventional approach, this method provides not only the point estimation of a design value but also quantitative evaluation on uncertainties of the estimation.
    Mathematical Problems in Engineering 06/2013; 2013. DOI:10.1155/2013/724632 · 1.08 Impact Factor


Available from
Dec 22, 2014