| description abstract | Predictions of the discharge and the associated sediment concentration are very useful ingredients in any water resources reservoir design, planning, maintenance, and operation. Although there are many empirical relationships between the discharge and sediment concentration amounts, they need estimation of model parameters. Generally, parameter estimations are achieved through the regression method (RM), which has several restrictive assumptions. Such models are locally valid and their structures and parameter values are questionable from region to others. This paper proposes a new approach for sediment concentration prediction provided that there are measurements of discharge and sediment concentration. The basis of the methodology is a dynamic transitional model between successive time instances based on two variables, namely, discharge and sediment concentration measurements. The transition matrix elements are estimated from the measurements through a special form of the artificial neural networks as perceptrons. The sediment concentration predictions from discharge measurements are achieved through a perceptron Kalman filtering (PKF) technique. In the meantime, this technique also provides temporal predictions. A certain portion of the measurement sequence is employed for the model parameter estimations through training and the remaining part is used for the model verification. Detailed comparisons between RM and PKF approaches are presented and, finally, it is shown that the latter model works dynamically by simulating the observation scatter diagram in the best possible manner with smaller prediction errors. The application of the methodology is performed for the discharge and sediment concentration measurements obtained from the Mississippi River basin at St. Louis, Missouri. It is found that the PKF methodology has smaller average relative, root-mean-square, and absolute errors than RM. Furthermore, graphical representation, such as the scatter and frequency diagrams, indicated that the PKF approach has superiority over the RM. | |
