Assessing the Artificial Neural Network Efficiency to Estimate Suspended Sediment Load using Classified Data

Flow sediment load is a useful indicator in predicting soil erosion in watersheds. As a result, developing a model to estimate sediment load can be useful in the management and operations of watersheds and river engineering projects. In this study, the data classification was studied as a way to increase the accuracy of Artificial Neural Network (ANN) model for estimating suspended sediment. For this purpose, the amount of suspended sediments in the Khalifa-Tarkhãn and Chehelgazi Rivers in Gheshlagh watershed, Kurdistan, Iran was predicted in three modes using an ANN with multilayer configurations. The measured data were also modeled without such classification. Then the runoff data were classified as high and low flows and the sediment data based on sediment concentration were classified as high and low concentrations. The observed runoff and sediment data were used to calibrate the models. Then the calculated values were compared with observed data and the models efficiency was examined using statistical tests. The findings indicate the effective role of data classification in improving the ANN efficiency in sediment estimation. So that classification based on sediment concentration promoted the model efficiency in Chehelgazi and Khalifa-Tarkhān by 16.6 and 30.5% respectively. The comparison of classifications showed that in comparison with flow, the sediment classification has more affective role on models estimations. The results of this study can be used to improve the management of the basin by more accurately estimating suspended sediments transporting in the rivers leading to Gheshlagh Dam Reservoir.