Bivariate Flood Frequency Analysis of Nonstationary Flood Characteristics

Abstract

Modeling the simultaneous behavior of flood characteristics, namely peak, volume, and duration, is essential for water resource planning and management. A multivariate probability approach, which provides a comprehensive understanding of flood characteristics and their relationship, may estimate flood magnitude more accurately than a univariate approach. Most previous studies related to the multivariate frequency analysis of extreme events assumed temporal stationarity. However, several recent studies show that flood characteristics exhibit nonstationary behavior due to climate change, urbanization, land-use change, or water resource structures. Therefore, it is necessary to perform multivariate frequency analysis in a nonstationary condition. In this study, nonstationary bivariate models, where the parameters of the marginal distribution vary with possible physical covariates (i.e., precipitation, urbanization, and deforestation), are developed to understand/model the nonstationary behavior of the flood characteristics of the Dongnai River in Vietnam. This study indicates that the assumption of temporal stationarity in flood characteristics leads to an underestimation of flood risk. For example, the flood characteristics’ quantiles estimated for a 50-year return period in a stationary condition is nearly equal to flood characteristics’ quantiles estimated for a 10-year return period in a nonstationary condition. Specifically, the volume and peak pair calculated for a nonstationary condition for the 10-year joint return period (OR) is (625.8, 147.8). The volume and peak pair calculated for a stationary condition for the 50-year joint return period (OR) is (620.1, 148.8).