Comparison of Artificial Neural Networks, Bayesian Network and Gene Expression Programming in Drought Prediction(Case Study: Maragheh Synoptic Station)

Drought is an inseparable part of any climate that has significant effects on different parts of the community and it increases the stress on water resources. Therefore, predicting its future status can help planners and decision makers in different sectors. In this study, for predicting drought in different time scales of the SPEI drought index, from 5 different inputs, including SPEI values with a lag of 1 to 5 months, then three intelligent methods including Gene Expression Programming (GEP), Bayesian Network (BN) and Artificial Neural Networks (ANNs) were used to predict future values. The results showed that all three methods in the short-term time-scale of the SPEI index are not appropriate so that the best performance in the one-month time scale is related to the Bayesian network model with a correlation coefficient of 0.142 and in the 3-month time-scale is related to the ANN model with correlation coefficient of 0.704. The results also showed that predictive accuracy of the model has a direct correlation with the SPEI calculation scale and, with increasing SPEI time scale, predictive accuracy increases. Also, all three methods have good performance in long-term time-scales.