Monitoring Event-Based Suspended Sediment Concentration by Artificial Neural Network Models

This paper is concerned with monitoring the hourly event-based river suspended sediment concentration (SSC) due to storms in Jiasian diversion weir in southern Taiwan. The weir is built for supplying 0.3 million tons of water per day averagely for civil and industrial use. Information of suspended sediments fluxes of rivers is crucial for monitoring the quality of water. The issue of water quality is of particular importance to Jiasian area where there are high population densities and intensive agricultural activities. Therefore, this study explores the potential of using artificial neural networks (ANNs) for modeling the eventbased SSC for continuous monitoring of the river water quality. The data collected include the hourly water discharge, turbidity and SSC during the storm events. The feed forward backpropagation network (BP), generalized regression neural network (GRNN), and classical regression were employed to test their performances. From the statistical evaluation, it has been found that the performance of BP was slightly better than GRNN model. In addition, the classical regression performance was inferior to ANNs. Statistically, it appeared that both BP (r=0.930) and GRNN (r=0.927) models fit well for estimating the event-based SSC in the Jiasian diversion weir. The weir SSC estimation using a single input data with the neural networks showed the dominance of the turbidity variable over water discharge. Furthermore, using the the ANN models are more reliable than classical regression method for estimating the SSC in the area studied herein. Key-Words: -Artificial neural networks, river monitoring, suspended sediment concentration, turbidity, water discharge, modeling.