Drought Prediction Using GEP-GARCH Hybrid Model (Case Study: Salmas Synoptic Station)

Drought prediction plays an important role in designing drought adaptation systems and implementation of relief operations. Hydrological data is a combination of a definite and random section. Given the fact that the production data of intelligent models are definite, application of a new approach, using the random part in predicting this data can increase the certainty of the model. In this research, it was attempted to provide a hybrid model for prediction of drought using a combination of the Gene Expression Programming model (GEP) and the Generalized Auto Regressive Conditional Heteroscedasticity (GARCH) time series model. For this purpose, drought prediction in Salmas station using SPEI drought index at different time scales was investigated during 35 years statistical period and with 5 different input models. The results showed that the GEP method does not have the appropriate accuracy in short-term time scale of SPEI index and it will be improved with increasing time scale. The results of the hybrid model showed that the error of GEP model decreases in all time scales, and this performance improvement is more tangible in the short-time scales, so that the correlation coefficient in three-month time scale in the GEP model has increased from 0.622 to 0.891 in the hybrid model.