عنوان مقاله [English]
Accurate prediction of the rainfall plays a key role in effective water resources management, , especially in arid and semi-arid regions. Achieving reliable and accurate precipitation forecasts is one of today's challenging issues in water resources management and climate hazards. Even though the great deal of research has been conduted on application of computational intelligence models for climatic forecasting, but selecting the best combination of inputs to such models, modellers have faced with problems. The main objective of this study is to evaluate the effect of input variables preprocessing in choosing the best combination of variables affecting on precipitation process, for forecasting monthly rainfall by using two data-driven modeling techniques including Support Vector Regresion (SVR) and Gene Expression Programming (GEP). For this purpose, Gamma Test and correlation Test were used to preprocess the inputs of the models used in this research under a case study with monthly climate data related with Shiraz Synoptic station over 1982-2011.The performance of these models was evaluated by statistical criteria of R2,RMSE and NSE (Nash-Sutcliffe efficiency coefficient). The results showed that SVR model combined with Gamma Test forecasts monthly rainfall better than other models used ib this study.But GammaTest was not able to improve performance of Gene Expression Programme model the same as SVR model. Also, based on the obtained results, sun hours, relative humidity, rainfall in a previous month and temperature, respectively,are most significant variables in monthly rainfall forecasting.