A hybrid group method of data handling with discrete wavelet transform for river flow forecasting

Abstract

River flow forecasting is important because it can assist an organization to make better plans and decision makings. One of the major goals in river flow forecasting is to improve the planning, design, operation and management of hydrology and water resources system. This study proposes designing a hybridization model using Group Method of Data Handling (GMDH) and Discrete Wavelet Transform (DWT) for forecasting monthly river flow in three catchment areas in Malaysia. The monthly data of river flow in the form of monthly means are collected from the Department of Irrigation and Drainage, Malaysia. The hybrid model is a GMDH model that uses sub-time series components obtained using DWT on original data. The original data is represented by its features, which term the wavelet coefficients that are then iterated into GMDH model. The individual GMDH is used to forecast the river flow for each single catchment area. The experiments compare the performances of a hybrid model and a single model of Wavelet-Linear Regression (WR), ANN, and conventional GMDH. The results show that the hybrid model performs better than other models for river flow forecasting. It is shown that the proposed model can provide a promising alternative technique in river flow forecasting.