Applying hybrid artificial algorithms to the estimation of river flow: a case study of Karkheh catchment area

This study was conducted on the Karkheh catchment area in the country of Iran on the basis of data from four stations of Cham Anjir, Kashkan, Pole-zal, and Jologir. The data were monthly collected for the years from 1997 to 2017, from which 70% were used for calibration and 30% for test validation. Artificial neural networks (ANNs) were used in the prediction of river flow. To optimize the weight coefficients of the network, the meta-heuristic genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization (ACO), harmonic balance algorithm (HB), and algorithm of the innovative gunner (AIG) were employed. To assess the modeling performance of the statistic parameters, the coefficient of determination ( R 2 ), root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE), mean absolute error (MAE), and percentage of bias (PBIAS) were adopted. The achieved best input combination was different for each algorithm. In summary, variables with lower correlations had poorer performances. Hybrid algorithms improved the predicting power of the considered independent model. The results showed that for the studied four stations, among the meta-heuristic algorithms, AIG had the highest predictive power correlation coefficient R 2 = 0.985–0.995, root mean square error (RMSE) = 0.036–0.057 m 3 /s, mean absolute error (MAE) = 0.017–0.036 m 3 /s, Nash-Sutcliffe coefficient (NSE) = 0.984–0.994, and bias = 0.008–0.024 and had the best performance in estimating the daily flow of the river.