Daily streamflow prediction using support vector machine-artificial flora (SVM-AF) hybrid model

Abstract

Precise estimation of river flow in catchment areas has a significant role in managing water resources and, particularly, making firm decisions during flood and drought crises. In recent years, different procedures have been proposed for estimating river flow, among which hybrid artificial intelligence models have garnered notable attention. This study proposes a hybrid method, so-called support vector machine–artificial flora (SVM-AF), and compares the obtained results with outcomes of wavelet support vector machine models and Bayesian support vector machine. To estimate discharge value of the Dez river basin in the southwest of Iran, the statistical daily watering data recorded by hydrometric stations located at upstream of the dam over the years 2008–2018 were investigated. Four performance criteria of coefficient of determination (R2), root-mean-square error, mean absolute error, and Nash–Sutcliffe efficiency were employed to evaluate and compare performances of the models. Comparison of the models based on the evaluation criteria and Taylor’s diagram showed that the proposed hybrid SVM-AF with the correlation coefficient R2 = 0.933–0.985, root-mean-square error RMSE = 0.008–0.088 m3/s, mean absolute error MAE = 0.004–0.040 m3/s, and Nash-Sutcliffe coefficient NS = 0.951–0.995 had the best performance in estimating daily flow of the river. The estimation results showed that the proposed hybrid SVM-AF model outperformed other models in efficiently predicting flow and daily discharge.