Author(s): Jin-Young Kim, Hyun-Han Kwon, Jin-Guk Kim, Suk-Hwan Jang
Linked Author(s): Hyun-han Kwon
Keywords: Trivariate Copula, Bayesian, climate change scenarios, uncertainty, return period
Abstract: Drought can be defined as an extended period of precipitation deficiency and is often driven by natural climate variability. Drought frequency analysis has been typically based on univariate approach. However, extreme events of interest in hydrologic design show multivariate dependencies that may be described by a set of dependent random variables. In particular, drought events are generally characterized by three dependent attributes: drought duration, severity and intensity. It has been reported that the univariate models often underestimate drought risk compared to the multivariate models. Therefore, the univariate model may not be appropriate for complex hydrological events. Various studies have been conducted to estimate the joint or conditional probability of extreme events using multivariate joint distributions. In this regard, this study aims to propose a novel approach to estimate the parameters of the copula function and their uncertainty within the Bayesian framework. The proposed nonstationary trivariate drought frequency model is used to estimate drought risk in South Korea. As an experimental study, we evaluated the stationary and time-varying joint return periods of the drought duration and severity for both the historical runs and future projection using GCM model output. It is clearly identified that the climate model shows a limited capability to accurately model the underlying probabilities in the drought variables, which could lead to misinterpretation for climate variability associated drought patterns in the future
Year: 2017