Daily precipitation forecasting using artificial neural networks: A case study: Synoptic station of Mashhad

Rainfall forecasting as one of the most important climatological parameter has a special importance in water resources management. In other hand، there are some complex nonlinear phrases in the equations which are make precipitation modeling too hard. Therefore nowadays scientists are looking for new approaches to improve the knowledge and predict meteorology parameters، especially precipitation. One of these methods is Artificial Neural Networks (ANN) which is categorized in computational intelligence group. In this study was focused on developing a non-numeric precipitation prediction model using ANN. Using daily precipitation data of Mashhad synoptic station، for 23 years (1982-2004) of March as a month with high humidity and May and December as months with medium humidity. From 713 daily data points، 580 were randomly selected as training and testing data sets. The reminder، were taken to check the validity of the models after training. The new approach ANN type used in this study was a feed forward-back propagation-perceptron، using a gradient decent learning algorithm. Out of many models tried، two topologies with best fit parameters were selected for March، named as GS521 and GS651، GS521 and GS681 for May and GS571 and GS631 for December and MATLAB (7. 01) environment was employed to accomplish this study. After applying fitness statically factors، it was founded in the best models R، RMSE and MAE were founded equal 0. 89، 0. 14 (mm) and 1. 15 (mm) for March، 0. 85، 0. 14 (mm) and 1. 16 (mm) for May and 0. 86، 0. 15 (mm) and 1. 17 (mm) for December، respectively.