Hydrologic risk analysis for nonstationary streamflow records under uncertainty

The frequency and magnitude of hydrologic extreme events is critical to water resources management. Traditional hydrologic frequency analysis approaches rely on the inappropriate assumption that hydrology is stationary. To tackle the nonstationarity in the streamflow records, we proposed a hydrologic risk analysis framework for the Xiangxi River, one of the largest tributaries of the Three Gorge Region, China. The year 1989 was identified as the change point of the 50-year flow records through a CUSUM approach combined with a Bootstrap test. Annual peak flow frequency analyses were then carried out for the 50-year time series and the records after the identified change point, respectively. The results revealed that, by taking into consideration nonstationarity, the return period of high peak flood at the Xingshan Station would actually increase. Bayesian inference combined with a MCMC sampling algorithm was also conducted to address uncertainties in parameter estimation and translate them to flow quantile estimates. It was found that the uncertainty in parameter estimation greatly affected the hydrologic design. To better support the associated risk assessment, two risk concepts, the exceedance risk and the occurrence risk, were proposed and analyzed. The results provided important insights into hydrologic nonstaionarity and uncertainty, and the proposed framework can provide scientific bases for engineering design and risk management in many other rivers in China and around the world.