Evaluation of intelligent prediction models towards precision of flood peak flows

Due to the lack of hydrological and meteorological stations, the use of data-based models is essential. Herein efficiency of Gene Expression Programming Models and Support Vector Machine are evaluated involving peak flood discharge prediction of Mahneshan-Angoran basin as a case study, central Iran. For this purpose, observational data of 36 annual maximum daily flow (1975-2011), corresponding rainfall and average monthly temperature of three stations including Mehrabad, Yangikand and Qarahgoni were used. The observed and predicted peak discharge flows in both models were compared based on the RMSE, explanatory coefficient (R^2) and Nash-Sutcliffe (NSE) criteria. The mean values of RMSE in the validation stage for the GEP model in Yingikand, Qarahgooni and Mehrabad stations are equal to 0.049, 0.080 and 0.050, respectively, and in the training stage are equal to 0.042, 0.060 and 0.047, the mean values of R^2 in the validation stage in above mentioned stations are equal to 0.88, 0.86 and 0.87, respectively, and for the training stage estimated equal to 0.89, 0.89 and 0.92. NSE values in the validation stage is equal to 0.75 for all three stations. However that is equal to 0.77, 0.76 and 0.80 for the mentioned stations in the training stage. Also, the RMSE values in the SVM model for the validation stage are equal to 0.042, 0.040, 0.054, respectively, and in the training stage are equal to 0.053, 0.064 and 0.044. R^2 values in the validation stage are equal to 0.66, 0.85 and 0.73. Also for the training stage are equal to 0.86, 0.88 and 0.91. NSE values for validation are equal to 0.56, 0.75 and 0.61 and for the training stage are equal to 0.71, 0.77 and 0.80. According to the evaluation criteria, the GEP model performed relatively better and this model is more suitable for predicting floods in Mahneshan-Angoran basin.