Predicting River Floods Using Discrete Wavelet Transform.

The paper demonstrates the efficiency of Wavelet Regression (WR) in estimating floods in rivers when the only data available is historical flow series. Discrete Wavelet Transform (DWT) decomposes the flow series into constituent wavelet components, i.e., approximations and details. A modified flow series is then constructed after removing the most fluctuating components and recombining other wavelet components. The modified flow series forms the input basis for WR implementation. Autoregressive (AR) models, developed for the comparison purpose, were implemented on the original flow series. A case study of developed models was made using monsoon flood data of the Kosi River at Birpur gauge site in the Bihar state of India. Based on various performance indices, it can be concluded that WR models forecast floods with greater accuracy than AR models. [ABSTRACT FROM AUTHOR]