Monsoon Flood Forecasting in Gandak River Using Discreet Wavelet Transform

Abstract

The Gandak River in North Bihar carries huge flood during the monsoon period. The daily variation in its flow during this period is so significant that conventional methods find it difficult to model the flow. In this paper, new models employing discreet wavelet transform (DWT) have been developed to forecast daily flows in a large river like Gandak. DWT decomposes the flow series into constituent wavelet components of ``approximations’’ and ``details.’’ After removing the most uncorrelated ``detail’’ from the original flow series and recombining other wavelet components, a modified flow series is reconstructed. Five wavelet regression (WR) models are developed for the forecasting of the current-day flood. Predictions from WR models are compared against that from autoregressive (AR) models developed for the purpose. Based on various performance indices, it can be concluded that WR models have strong generalization ability and predict large floods with greater accuracy than AR models.
Keywords: Flood forecasting, Monsoon, Gandak, Wavelet regression, Auto regression, River stage