Abstract: Piecewise trend is an important form of variations in nonstationary hydrological series. This paper describes a new method (namely, R detection method) for detecting piecewise trends in hydrological series and evaluating the significance of corresponding variations based on correlation coefficients. The results of statistical experiments show that this method features a very low rate of false detection of the series with no piecewise trend and a high rate of positive detection of the series with different types of piecewise trends, indicating its high accuracy in the detection. We use the method to detect the piecewise trends in the annual peak flood series from five gauge stations in Dongting Lake. The results show that piecewise trends occur at different time points of the series at the stations of Shadaoguan, Guanjiapu and Kangjiagang, and are featured with different forms, all showing a certain decrease in the peak flood flow as a response to the variations in the runoff regulated by the Three Gorges reservoir and caused by the cutting of the lower Jingjiang.

Key words: piecewise trend, correlation coefficient, hydrological variability, Dongting Lake, annual peak flood