Uncertainty Analysis of SVM Model Parameters for Estimating Suspended and Bed Sediment Load at Sierra Station in Karaj by Monte-Carlo Simulation Method

Estimation of sediment transported by the streamflow is important for planning and storing water resources of dam reservoirs and river bed changes, watershed management, coastal protection and the environment. Sediment transport in the river is an inherently uncertain and complex phenomenon. Incomplete knowledge of processes and data create uncertainty in estimating sediment transport. Parameters uncertainty is one of the main sources of uncertainty in estimating the suspended and bed sediment load. In this paper, the Monte Carlo (MC) simulation method is used to estimate the uncertainty of suspended and bed sediment load due to uncertainty in the parameters of the support vector machine (SVM) model in the Karaj Dam Basin. The partial mutual information (PMI) algorithm was used to select the efficient input variables in the SVM model to estimate the suspended and bed sediment load. The results of using PMI algorithm show that the only efficient variable in estimating the suspended and bed sediment loads is the current stream discharge. The results show that the uncertainty in estimating the suspended sediment load with SVM model for training, test and total data is equal to 12.8%, 17% and 13.5%, respectively. Also, the uncertainty in estimating the bed sediment load with SVM model for training, test and total data is equal to 23.5%, 36.8% and 27.2%, respectively. Therefore, the uncertainty in estimating the bed sediment load with SVM model is more than the one in estimating the suspended sediment load. Therefore, the use of optimization methods can be useful for accurate estimation of parameter values and reducing uncertainty in estimating the suspended and bed sediment load.