NON-STATIONARY FLOOD FREQUENCY ANALYSIS: REFINEMENT THROUGH ENSEMBLE EMPIRICAL MODE DECOMPOSITION

Abstract

Flood frequency analysis is a technique commonly used to relate the frequency of occurrence of extreme runoff events with their probability of occurrence by using suitable probability distribution functions. Traditionally, flood frequency was estimated by considering the parameters of the probability distribution to be stationary. However, subsequent research has demonstrated that climatic circumstances fluctuate and the assumption of stationarity does not hold any more. In many areas, increased development, changes in land use, trends in rainfall patterns and other anthropogenic causes have caused alterations in the hydrological cycle. As a result, non-stationary model is required to account for changes in parameters over time using covariates such as precipitation, temperature, urbanisation, teleconnections, and more. Currently, the non-stationary flood frequency analysis is done by modelling the time varying parameters of the streamflow as the linear function of covariates. The past studies on non-stationary flood frequency analysis have used either only the climatic factors, such as precipitation or only environmental factors such as the land use changes as the potential covariates for their study. However the combined effect of climatic as well as environmental factors may also be responsible for the non-stationarity in the streamflow. The time series of climatic parameters, such as precipitation and temperature, exhibit two distinct components. The first component represents the slow-varying component, which indicates the long-term trend. The second component consists of fluctuations caused by seasonal variations, which introduce noise into the time series. The primary goal of non-stationary flood frequency analysis is to model the time-varying parameters, which exhibit slow variations. Therefore, it is more reasonable to model these slow-varying dynamic parameters using the slow-varying components of the covariates