Flood Forecasting Using Artificial Neural Networks

Abstract

Floods, as natural disasters, cause huge material damages and often result in loss of human lives. Their early prediction is necessary in order to take appropriate actions to reduce economic losses and risks for people. Albeit there is no universal method for flood modeling, significant advances in the technology of flood modeling techniques provide opportunities for flood prediction. Their modeling usually requires a large amount of data. In cases where only a specific part of the river basin is explored for more accurate modeling, the time and the effort to implement such complicated models is not justified. Therefore, the use of intelligent techniques such as Artificial Neural Networks (ANN) can be a practical alternative. The purpose of the investigations presented in this paper has been to make flood forecast for part of the Polog region using ANN. The forecast has been based on a model developed for this purpose. Modeling has been performed by use of four artificial neural networks in time series: Support Vector Machine (SVM), Radial Basis Function Neuron Network (RBFNN), General Regression Neural Network (GRNN) and Multilayer Perception (MP). Data on maximum annual flows of Vardar river, recorded at the Radusha measuring station throughout a period of 58 years, have been used as an input for the models and the output of the ANN is the maximum annual flow forecast for a 5-year (1951-2008) period. The results presented show that the ANN method, in this case the GRNN, can be useful and can provide sufficient accuracy in solving problems related to hydrological extremes.