Author(s) :

Otache, Y. M., Mohammed, A. S., Ahaneku, I. E., Egharevba, N. A..

Volume/Issue :

Abstract :

The importance of understanding the dynamics and forecasting of streamflow processes of a particular river finds relevance in the fact that the physical mechanisms governing flow dynamics act on a wide range. In view of this, this study presents a simple basis for and application of Artificial Neural Network (ANN) methodology as an alternative modelling tool for predicting flow. To this end, the main focus is the development of ANN model for short term streamflow forecasting of the Benue River using univariate time series; inter alia, evaluate its performance of extreme events. It is evident from the modelling framework that the application of the knowledge of evolution of a dynamical system in a multi-dimensional state space is a robust approach for determining the size of an ANN model input. The ANN model forecast performance showed that reliable short term forecasts, 5 day - ahead can be made for the daily streamflow series based on CE and R2 performance indexes. However, on the general question of the suitability of ANN model application for streamflow forecasting as applied in this study (i.e., daily streamflow), though the neural network could simulate the different attributes of the flow hydrograph, its relative forecast performance of high flows is robustly better than the case of low flows; it grossly under predicted and over predicted same depending on the particular network input data pre-processing schema. The forecast performance results also indicated that, for feed-forward MLP networks, with a tan-sigmoid transfer function, standardising the data by subtracting the mean and dividing by the standard deviation is better than rescaling the data to a small interval of particular range. Considering the findings, to appropriately capture the dynamics of the flow regime, it is necessary to include exogenous variables of the runoff generating process in the network input data base.

No. of Downloads :

1