Comparison of the accuracies of Geomorphologic Artificial Neural Networks (GANN) and Regression Model (RM) for estimation of Taleghan river sediment yield

Evaluating and modeling of volume of sediment yield is one of the most important subjects in water resources management. Estimation of sediment yield based on traditional methods needs a lot of time and money and usually does not have enough accuracy while using new methods like GANN provides needed accuracy. One of the newest methods for solving hydrologic and water engineering problems is using GANN (Geographical Artificial neural network) method. GANN has structure similar human brain which performs training process and extracts internal relationship between data and then generalize them to other situations. In this research water-sediment discharge data of Gelinak Hydrometric Station on Taleghan River and a series of geomorphologic parameters of Taleghan Watershed was used to model suspend Sediment yield by GANN method. For this purpose, after correction of statistical errors and elimination of deviated data, 80 percent of data were used for training and 20 percent for network examination. After standardizing the data, a back-propagation neural network was provided using training data series. Then a regression equation between water and sediment discharge data was generated using logarithmic training data. Output data were evaluated using RMSE, MAE, R2 statistic indices. Results showed that GANN estimations have more accuracy (RMSE = 48.5, MAE = 33.25, R2 = 0.89) than regression models estimations (RMSE = 93, MAE = 54.25, R2 = 0.74). It can be concluded that ANN model have more efficiency than regression models in sediment yield estimation. In addition, by using geomorphologic parameters which are effective in sediment production as input of the model, accuracy of sediment yield is increased.