Publication

Abstract : Flood frequency analysis (FFA) is mainly used for planning and designing of water resources and hydraulic structures. FFA is usually carried out with annual maximum daily flows collected over the significant period in the past (depending on the size and importance of the structure to be designed) assuming that the data series follow the stationary. However, in the recent decades, frequent occurrence of extremely high flood events is very evident, which leads to the flow data to have more of non-stationary. Among several reasons for non-stationary behavior of the flows, the impacts of climate change are significantly high. Consequently, modelling for the design floods with stationary approach might not be valid, as it may produce under/over estimation of flood magnitude or return period due to the limitations in the existing modelling approaches, thus require special treatment. In this paper, an approach has been developed for conducting FFA to include the non-stationary behavior of data set through separately modelling the normal and extreme flood events and then integrating them to estimate the design flood. For the separation, we have used statistical approaches (i.e. simple linear regression and artificial neural network models), which identify extremely high floods from the normal floods through the information from flood hydrograph such as time to peak, volume of flood and duration. To demonstrate the proposed approach, annual maximum flow data from Bow River, Canada were used. From the results, we found that traditional FFA method consistently underestimates the design floods correspond to high return periods as compared to proposed method. Furthermore, for the given design flood, proposed method yielded a less return period, which will obviously help quantifying the increase in amount of risk. These findings suggested that the proposed method in this paper estimates design floods reasonably well for the data which mainly has non-stationary.