Nonstationary Frequency Analysis Using a Hierarchical Bayesian Model

Abstract

In this study, a nonstationary frequency analysis model was developed using a hierarchical Bayesian model. The model consists of 13 parameters which are 10 scale parameters according to different time windows, 2 hyper-parameters and 1 scale hyper-parameter. The model took use of extreme rainfall data based on POT (Peaks Over Threshold) and the GP (Generialized Pareto) distribution. The model parameters were estimated using a Gibbs sampler and Metropolis-Hastings algorithm. The model was applied for Seoul site to estimate target year probable rainfall amount and the probable rainfall estimate for the target year of 2045 is about 14-17% lager than that of current estimate according to different return periods. Results demonstrated that nonstationary frequency analysis is necessary for hydraulic structure design.