Presenting a New Approach to Increase the Efficiency of the Sediment Rating Curve Model in Estimating Suspended Sediment Load in Watersheds (Case Study: Mahabad-Chai River, Lake Urmia Basin, West Azarbayejan Province, Iran)

  The estimation of the correct amount of suspended sediment has an important role in the optimal design of water structures, erosion studies and water quality studies. The sediment rating curve (SRC) is a conventional and well-known regression model. However, due to logarithmic transformations in calibrating this model, its estimated values ​​are often less than actual values. In the present study, using the instantaneous flow discharge and suspended sediment load of Beytas hydrometric station in the Mahabad-Chai River, the SRC model was calibrated, and then using Non-dominated Sorting Genetic Algorithm II (NSGA-II), the coefficients of this model optimized again. This algorithm is an automatic procedure and can use different objective functions in the calibration process simultaneously. In this regard, in the calibration process of the model, four objective functions RMSE, MAE, NSE, and LOGE were used as pairwise combinations. According to the results of the model evaluation, the NSE and LOGE objective functions were selected as the best objective functions for optimization of the model. In order to increase the power of the model's generalization, the self-organizing map (SOM) neural network was used to cluster data and form two homogeneous data sets (calibration and evaluation sets) of 70% and 30% respectively. The results showed that the use of the NSGA II algorithm resulted in improved model efficiency so that the results are much more favorable than the other results of conventional SRC models (such as the rating curve of mean load within discharge classes, SRC models corrected by correction factors). In this regard, the error value (RMSE) of the test data set in the best model of the conventional SRC models was 383.65 tons/day, which was reduced by using the NSGA II algorithm to 102.94 tons/day. In sum, using the NSGA-II algorithm, we can optimize the coefficients of the SRC model, which is more efficient than the other conventional models.