Predicting monthly precipitation of Kermanshah synoptic station using the hybrid model of neural network and wavelet

Precipitation is one of the most important meteorology elements which Recognition the amount of it, its variation and prediction, are necessary to have an actual planning in agricultural, economic and social management. So the hydrologists and climatologists are paying attention to it. Due to precipitation importance in planning and crisis management, the goal of this research leads to the implementation of neural network and wavelet conversion hybrid model to monthly precipitation prediction of Kermanshah synoptic station.
In this research to predicting monthly precipitation time series of Kermanshah’s synoptic station, five parameters namely: monthly precipitation average, relative humidity average, Maximum temperature average, minimum temperature average and wind speed average, used within the forty years period (1970-2010). In order to quality control of given statistic and information, sequencing test is used. The results showed that given information is significantly homogeneous and according to nonlinear specifications of multiple time scales, neural network and wavelet model is used to precipitation prediction.
In order to precipitation prediction, four parameters of relative humidity average, maximum temperature average, minimum temperature average and wind speed average, is used and using wavelet alternation decomposed into 8 sub-series and then these series are used to future monthly precipitation prediction as the entrance neural network model data. The correlation coefficient (R=0.874) of Next month prediction denoted the relatively low efficiency of the neural network while the wavelet-neural network model correlation coefficient is 0.94. Also, both model prediction precision decreases by increasing the number of output neurons delays. It should be noted that Meyer wavelet is used to wavelet- neurotic network prediction which has a high precision. Regarding the F statistic, variance analysis of homogeneity and heterogeneity indices of observed and predicted precipitation, is homogeneous in the confidence level of %99 (P The comparison between the driven results from wavelet conversion-neural network and driven results from a neural network application, is shown that wavelet neural network method has more predicting precision than neural network and also predicting precision in both models decreases through increasing the number of delays of exit neurons. It is noteworthy that Meyer wavelet was used to neural network predicting which its simulated results has high precision.