Long term drought prediction by means of neural networks with wavelet method

Drought is defined as one of the biggest problems for living and it influences every aspect of life in terms of physical, economical, development, technology, agriculture, cleanwater, etc. Therefore, drought studies and preparedness works have an important place in order to take necessary precautions in water resources planning. In this study, long term drought prediction was carried out using various methods with monthly precipitation data of Karapınar, Manisa and Rize stations. Primarily, the standardized precipitation index (SPI) values of these three stations were calculated and the models were determined by using artificial neural Networks method in accordance with SPI values for twelve month period. In the modelling, frequencies of SPI12 values were seperated by using the most commonly wavelet method of Db4, Bior 3.1 and Haar via the Matlab program and models were created for long term drought predictions. Afterwards, created models were processed with Multilayer Artificial Neural Networks and it was aimed to determine the best model for each long term prediction. Consequently, best fit models were compared with the real drought index values and results were evaluated. In addition, different transfer functions and learning rules have been tried in network models, the best ones of these have been determined with various performance criteria and presented in tables and graphics.